Long Wave/Technology

Dear Client, This Special Report is the full transcript and slides of a presentation I recently gave at the London School of Economics symposium: 'Will I Work For AI, Or Will AI Work For Me?' The presentation pulls together several years of research analyzing the impact of current technological advances on work, the economy and society. I hope you find the presentation insightful and provocative, especially the narrative surrounding Slide 12. Dhaval Joshi Slide 2 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Feature Good afternoon Thank you very much for the invitation to speak here at the London School of Economics. The specific question you asked me was: will we be able to work in the future? (Slide 1). To which my answer is yes, an emphatic yes. I'm very optimistic that we will be able to work in the future. And one reason I'm saying this is, imagine that we had this symposium 100 years ago. I suspect we might have had exactly the same fears that we have right now (Slide 2). Slide 1 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Slide 2 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Specifically, at the start of the 20th century, about 35% of all jobs were on farms and another 6% were domestic servants. At the time, you could probably also have said, "Well, these jobs aren't going to exist." More or less half of the jobs that existed at that time were going to disappear - and disappear they did. So we'd have thought there would be mass unemployment. Of course, there wasn't mass unemployment, because just as jobs were destroyed, we had an equivalent job creation (Slide 3). For example, at the start of the 20th century, less than 5% of people worked in professional and technical jobs. But by the end of the century, these jobs employed a quarter of the workforce. I guess what I'm saying is that we're very conscious of job destruction because we can see existing jobs being destroyed. But we're not very conscious of job creation, because in real time, it's difficult to visualize or imagine where these new jobs will be. In essence, what we saw in the 20th century was one major segment of employment basically collapsed from very significant to insignificant. While another segment surged from insignificant to very significant (Slide 4). Slide 3 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Slide 4 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? As you all know, there is an economic thesis that underlies this. It's called Say's Law, derived by French economist Jean-Baptiste Say in 1803. In simple terms, it says that new supply creates new demand. Think about it like this: why would you replace a human with a machine? You would only do that if it increases your productivity, right? Otherwise, it does not make sense to replace a human with any sort of machine, including AI. But because you have increased productivity, you then have extra income to spend on new goods and services. Now if those goods and services are being supplied by a machine, then you can redeploy humans to satiate new desires, desires that do not even exist at the time. In economic terms, the producer of X - as long as his products are demanded - is able to buy Y (Slide 5). The question is, what is Y? Y is the new product or service. Let me give you some examples (Slide 6). In the 19th century, we had the advent of railways. And then someone thought. "Hang on a minute. We have this way of moving things around much faster, and we've got all these people who live hundreds of miles from the coast who might want to eat fresh fish." So this was the birth of the frozen food industry. But you could not have the frozen food industry without railways. What I'm saying is that entrepreneurs will seize the new technology to satiate a desire. Or even create a new desire because maybe the people in the middle of the country never thought they could eat fresh sea fish. Until someone came along and said, "you can eat fresh fish now." Slide 5 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Slide 6 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Another example is, as technology improved the health and longevity of your teeth someone thought. "Well, hang on a minute. Maybe there's a desire to make teeth look beautiful." And we created this whole new industry called the dental cosmetics industry. We know this because prior to the 1960s, there was no job called dental technician or dental hygienist. A third example is, let's say that we have more advanced healthcare and pharmaceuticals, so humans are living longer and healthier lives. Well, then you can sort of ask. "Hang on a minute. Don't you want your dog to live the same long and healthy life that you're living?" And this is behind the explosion of the pet care industry that we're seeing at the moment. So while one segment of the economy will employ less, a new segment will come along to replace it. In the 20th century we saw farm work disappearing but professional work rising. Today, we are seeing manufacturing and driving jobs disappearing but healthcare work rising (Slide 7). Which does raise a pretty obvious question (Slide 8). Is there anything really different this time around? Slide 7 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Slide 8 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Well, the answer is yes, there is a subtle but crucial difference this time around. To see the difference, we have to look more closely at where jobs are being destroyed, and where they are being created. As you can see, the mega-sectors losing a lot of jobs are manufacturing, the auto industry, and finance (Slide 9). While on the other side of the ledger, we have job creation in health, social work and education. But now, let's look in a little more detail. Where, specifically, are the jobs being created? For this we have to look at the United States data which is much more granular than in Europe. Here are the top five subsectors of job creation this decade (Slide 10). At the top of the list is food services and drinking places, which is just a euphemistic way of describing bartenders, waitresses, and pizza delivery boys. We also have a lot of new administrative jobs and care workers. What is the common link in this job creation? Answer: these are predominantly low-income jobs. Slide 9 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Slide 10 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? So it is true that we have an enormous amount of job creation in the last decade or so, and the policymakers keep boasting about it, they say, "Well look, the unemployment rate in the U.S. is at a record low, the unemployment rate in the UK is at a record low, the unemployment rate in Germany is at a record low. We're creating loads and loads of jobs." The trouble is that these are predominantly low-income jobs. Meanwhile the job destruction is in middle-income jobs in manufacturing and finance. This means what we're seeing in the labour market is called a 'negative composition effect' - a hollowing out of middle incomes. So while we're getting loads and loads of job creation, it is not translating into wage inflation at an aggregate level. I think one of the reasons is a concept called Moravec's paradox. Professor Hans Moravec is an expert in robotics and Artificial Intelligence, and he noticed this paradox (Slide 11). He said, "Look. For AI, the things that we think are difficult are actually easy." By easy, he means they're doable. Let me give you some specific examples. Say someone could speak five languages fluently and translate between them at ease. We would think that person is a genius, a real rare specimen, and the economy would value this person extremely highly, probably pay that person hundreds of thousands of pounds at a minimum. But actually, AI can translate across five languages quite easily, and even something like Google Translate, which we all use, does a reasonably good first stab at translating from one language to another. Slide 11 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Or consider something like insurance underwriting. Pricing an insurance premium from lots of data on a risk. AI can do that extremely well, much better than a human can. Or medical diagnosis. Figuring out what's wrong with a patient from very detailed medical data. Again, AI beats humans hands down on that. What I'm saying is, these skills that we thought were difficult transpired not to be that difficult for AI, because they just amount to narrow-frame pattern recognition and repetition of algorithms. Whereas, the second part of Moravec's paradox is that AI finds the easy things very hard. Things that we think are really innate, we don't even give them a second thought like walking up some stairs, cleaning a table, moving objects around, and cleaning around them. Actually, AI finds these things incredibly difficult, almost impossible. We have a false sense of what is difficult and what is easy. The main reason is that the things that we find innate took millions and millions of years of human brain evolution for us to find them innate. And as AI is in essence trying to replicate the human brain, only now are we recognizing that things that we find innate are actually incredibly complex. If it took millions and millions of years to evolve the sensorimotor skills that allow us to walk up some stairs, recognize subtle emotional signals, and respond appropriately, then obviously AI is going to find it very, very difficult to replicate those innate human skills. Conversely, the brain's ability to do calculus, construct a grammatical structure for a language, or play chess only evolved relatively recently. So AI can do them very easily. Which brings me to quite a profound thought. If there's one thing that I want you to remember from this presentation it is this (Slide 12). Might we have completely misvalued the human brain? Might we have grossly overvalued things that are actually quite easy? And might we have undervalued things which are actually very, very difficult? And what AI is now doing is correcting this huge error. In which case, the next decade could be extremely disruptive as AI corrects this economic misvaluation of our skills. Slide 12 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? This might also explain the mystery as to why there is no wage inflation when the Phillips curve says there should be. The Phillips curve makes a simple relationship between the unemployment rate and wage pressures. And the folks at the Federal Reserve and Bank of England, they're sort of getting really perplexed. They're saying, "Look, unemployment is so low. Where is this wage inflation? It's going to kick in any time now." In fact, there's a bit of a paradox going on. For the people who are continuously employed in the same job, there has been pretty good wage inflation - at sort of three, four percent (Slide 13). But when you take the negative composition effect into account, then suddenly there's this big gap because what's happening is that the well-paid jobs are disappearing to be replaced by lower-paid jobs. So even if you give the bartender making thirty thousand a big pay rise to thirty-five thousand. Even if you hire two of them, but you're losing a finance job paying over a hundred thousand, then at the aggregate level, you won't see much wage inflation. And this problem, I think, continues for the next few years, minimum. It means that you will not get the wage pressures that a lot of economists think you're going to get from the low unemployment rate. Because you have to look at the quality of the jobs as well as the quantity. I think there is another disturbing impact from a societal perspective. Look again at where the jobs are being lost and where they're being created, and look at the percentage of male employees (Slide 14). Job destruction is occurring in sectors that are male-dominated, whereas job creation is occurring in sectors that are female-dominated. Slide 13 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Slide 14 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? AI is good at narrow-frame pattern recognition and repetition of algorithms and functions - jobs like driving, which are typically male-dominated. Whereas jobs that require emotional input, emotional understanding, and empathy in the 'caring sectors' are typically female-dominated. So if you're a male, you're in trouble. You're in a lot of trouble. Obviously, there'll be re-training, so all the guys who were driving trucks will have to retrain as nurses, or as essential carers. But if you're a female, things are looking okay. You can see that in the data (Slide 15). Female labour force participation is in a very clear uptrend. Male participation is flat to down. This varies by country by country, and in the U.S., it's catastrophic for males, especially young males. Young male participation in the U.S. is really falling off a cliff at the moment. I think the other thing to say from a societal perspective is that the so-called 'Superstar Economy' is booming - both superstar individuals and superstar firms. One way of seeing this is in this index called 'the cost of living extremely well' calculated every year by Forbes (Slide 16). Whereas the ordinary CPI includes the cost of bread and milk, the CPI index for the extremely rich includes the cost of Petrossian caviar and Dom Perignon champagne. And a Learjet 70, a Sikorsky S-76D helicopter. I think there's a pedigree racehorse in there too. Anyway, we're seeing the CPI for the extremely rich rising at a dramatically faster pace than the CPI for society as a whole. So it would seem that superstar individuals and superstar firms are really thriving. Slide 15 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Slide 16 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Let's explain this dynamic in terms of a superstar we all recognise - Roger Federer. Roger Federer was unknown initially, but as he went up the tennis rankings and became a superstar, his income grew exponentially. The other aspect is, how long can he stay a superstar? Because all superstars are eventually displaced by a new superstar. So there's two aspects to the dynamics of superstar incomes (Slide 17). First, how exponential is your income growth? And second, how long do you stay a superstar? What I'm saying is that the rise of AI, by hollowing out the middle jobs, actually allows a few superstars to have this exponential rise in their income. Let's think about it in terms of the legal profession. The top lawyer will be in huge demand. Technology really boosts him. Not just AI, but things like the internet, the fact that social media will reinforce his position, whereby everyone will know who he is. Even if he can't service you directly, he will have a team with his brand on it. And he can stay there for longer before he is displaced. So this is the mechanism by which technology can increase income inequality by hollowing out the middle. In the legal profession, the assistant lawyer who just checks a document for simple legal principle, well the machine can do that. But the guy who knows all the oddities, who knows all the loopholes that can win you the case, the machine won't be able to do that. Essentially what I'm saying is that the technological revolution - it's not just AI, it's technology in aggregate, including the internet and social media, and so on - it increases the rate of income growth for a few superstar individuals and firms. And it increases their longevity (Slide 18). And these are the two drivers for the Pareto distribution of incomes. You can actually go through the mathematics of this to show that it does increase the polarization of incomes. Slide 17 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Slide 18 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Let's sum up (Slide 19). First of all, yes, we will be able to work in the future. I don't think there's any doubt about that because there will be new jobs created, the nature of which we can only guess because we're going to get new industries to satiate our new desires. However, in the coming years, middle-income work will suffer high disruption because of Moravec's Paradox. Some things that we thought were difficult are actually quite easy for AI. But things like gardening, plumbing, nursing, and childcare are very difficult for machines to replicate. Which means that low-income work will suffer much less disruption and, of course, low-income work will get paid better over time - though the gap is so large at the moment that it's preventing overall wage inflation from kicking in. And that, I think, will persist for the next few years at a minimum. Slide 19 The Impact Of AI: Will We Be Able To Work In The Future? The Impact Of AI: Will We Be Able To Work In The Future? Men are going to suffer much more disruption than women because of the nature of the job destruction versus the job creation. And the final point is that superstars will thrive. All of this has a lot of implications for how we respond as a society, and maybe we will need some support mechanisms in this period of disruption. I think the most intense disruption will be in the next decade. After that we will reach a new equilibrium once we have actually corrected this misvaluation of the brain, this misvaluation of what it is that makes us truly human. Thank you very much. Dhaval Joshi, Senior Vice President Chief European Investment Strategist dhaval@bcaresearch.com

Dear Client, In addition to this Special Report written by my colleagues Mark McClellan and Brian Piccioni, we are sending you an abbreviated weekly report. Best regards, Peter Berezin, Chief Global Strategist Global Investment Strategy Highlights Media reports warn of a "Robot Apocalypse" that is already laying waste to jobs and depressing wages on a broad scale. Technological advance in the past has not prevented improving living standards or led to ever rising joblessness over the decades, but pessimists argue that recent advances are different. The issue is important for financial markets. If structural factors such as automation are holding back inflation by more than in previous decades, then the Fed will have to proceed very slowly in raising rates. We see no compelling evidence that the displacement effect of emerging technologies is any stronger than in the past. Robot usage has had a modest positive impact on overall productivity. Despite this contribution, overall productivity growth has been dismal over the past decade. If automation is increasing 'exponentially' and displacing workers on a broad scale as some claim, one would expect to see accelerating productivity growth, robust capital spending and more violent shifts in occupational shares. Exactly the opposite has occurred. Periods of strong growth in automation have historically been associated with robust, not lackluster, wage gains, contrary to the consensus view. The Fed was successful in meeting the 2% inflation target on average from 2000 to 2007, when the impact of the IT revolution on productivity (and costs) was stronger than that of robot automation today. This and other evidence suggest that it is difficult to make the case that robots will make it tougher for central banks to reach their inflation goals than did previous technological breakthroughs. For investors, this means that we cannot rely on automation to keep inflation depressed irrespective of how tight labor markets become. Feature Recent breakthroughs in technology are awe-inspiring and unsettling. These advances are viewed with great trepidation by many because of the potential to replace humans in the production process. Hype over robots is particularly shrill. Media reports warn of a "Robot Apocalypse" that is already laying waste to jobs and depressing wages on a broad scale. In the first in our series of Special Reports focusing on the structural factors that might be preventing central banks from reaching their inflation targets, we demonstrated that the impact of Amazon is overstated in the press. We estimated that E-commerce is depressing inflation in the U.S. by a mere 0.1 to 0.2 percentage points. This Special Report tackles the impact of automation. We are optimistic that robot technology and artificial intelligence will significantly boost future productivity, and thus reduce costs. But, is there any evidence at the macro level that robot usage has been more deflationary than technological breakthroughs in the past and is, thus, a major driver of the low inflation rates we observe today across the major countries? The question matters, especially for the outlook for central bank policy and the bond market. If structural factors are indeed holding back inflation by more than in previous decades, then the Fed will have to proceed very slowly in raising rates. However, if low inflation simply reflects long lags between wages and the tightening labor market, then inflation may suddenly lurch to life as it has at the end of past cycles. The bond market is not priced for that scenario. Are Robots Different? A Special Report from BCA's Technology Sector Strategy service suggested that the "robot revolution" could be as transformative as previous General Purpose Technologies (GPT), including the steam engine, electricity and the microchip.1 GPTs are technologies that radically alter the economy's production process and make a major contribution to living standards over time. The term "robot" can have different meanings. The most basic definition is "a device that automatically performs complicated and often repetitive tasks," and this encompasses a broad range of machines: From the Jacquard Loom, which was invented over 200 years ago, on to Numerically Controlled (NC) mills and lathes, pick and place machines used in the manufacture of electronics, Autonomous Vehicles (AVs), and even homicidal robots from the future such as the Terminator. Our Technology Sector report made the case that there is nothing particularly sinister about robots. They are just another chapter in a long history of automation. Nor is the displacement of workers unprecedented. The industrial revolution was about replacing human craft labor with capital (machines), which did high-volume work with better quality and productivity. This freed humans for work which had not yet been automated, along with designing, producing and maintaining the machinery. Agriculture offers a good example. This sector involved over 50% of the U.S. labor force until the late 1800s. Steam and then internal combustion-powered tractors, which can be viewed as "robotic horses," contributed to a massive rise in output-per-man hour. The number of hours worked to produce a bushel of wheat fell by almost 98% from the mid-1800s to 1955. This put a lot of farm hands out of work, but these laborers were absorbed over time in other growing areas of the economy. It is the same story for all other historical technological breakthroughs. Change is stressful for those directly affected, but rising productivity ultimately lifts average living standards. Robots will be no different. As we discuss below, however, the increasing use of robots and AI may have a deeper and longer-lasting impact on inequality. Strong Tailwinds Chart 1Robots Are Getting Cheaper Robots Are Getting Cheaper Robots Are Getting Cheaper Factory robots have improved immensely due to cheaper and more capable control and vision systems. As these systems evolve, the abilities of robots to move around their environment while avoiding obstacles will improve, as will their ability to perform increasingly complex tasks. Most importantly, robots are already able to do more than just routine tasks, thus enabling them to replace or aid humans in higher-skilled processes. Robot prices are also falling fast, especially after quality-adjusting the data (Chart 1). Units are becoming easier to install, program and operate. These trends will help to reduce the barriers-to-entry for the large, untapped, market of small and medium sized enterprises. Robots also offer the ability to do low-volume "customized" production and still keep unit costs low. In the future, self-learning robots will be able to optimize their own performance by analyzing the production of other robots around the world. Robot usage is growing quickly according to data collected by the International Federation of Robotics (IFR) that covers 23 countries. Industrial robot sales worldwide increased to almost 300,000 units in 2016, up 16% from the year before (Chart 2). The stock of industrial robots globally has grown at an annual average pace of 10% since 2010, reaching slightly more than 1.8 million units in 2016.2 Robot usage is far from evenly distributed across industries. The automotive industry is the major consumer of industrial robots, holding 45% of the total stock in 2016 (Chart 3). The computer & electronics industry is a distant second at 17%. Metals, chemicals and electrical/electronic appliances comprise the bulk of the remaining stock. Chart 2Global Robot Usage Global Robot Usage Global Robot Usage Chart 3Global Robot Usage By Industry (2016) The Impact Of Robots On Inflation The Impact Of Robots On Inflation As far as countries go, Japan has traditionally been the largest market for robots in the world. However, sales have been in a long-term downtrend and the stock of robots has recently been surpassed by China, which has ramped up robot purchases in recent years (Chart 4). Robot density, which is the stock of robots per 10 thousand employed in manufacturing, makes it easier to compare robot usage across countries (Chart 5, panel 2). By this measure, China is not a heavy user of robots compared to other countries. South Korea stands at the top, well above the second-place finishers (Germany and Japan). Large automobile sectors in these three countries explain their high relative robot densities. Chart 4Stock Of Robots By Country (I) Stock Of Robots By Country (I) Stock Of Robots By Country (I) Chart 5Stock Of Robots By Country (II) (2016) The Impact Of Robots On Inflation The Impact Of Robots On Inflation While the growth rate of robot usage is impressive, it is from a very low base (outside of the automotive industry). The average number of robots per 10,000 employees is only 74 for the 23 countries in the IFR database. Robot use is tiny compared to total man hours worked. In the U.S., spending on robots is only about 5% of total business spending on equipment and software (Chart 6). To put this into perspective, U.S. spending on information, communication and technology (ICT) equipment represented 35-40% of total capital equipment spending during the tech boom in the 1990s and early 2000s.3 Chart 6U.S. Investment In Robots U.S. Investment In Robots U.S. Investment In Robots The bottom line is that there is a lot of hype in the press, but robots are not yet widely used across countries or industries. It will be many years before business spending on robots approaches the scale of the 1990s/2000s IT boom. A Deflationary Impact? As noted above, we view robotics as another chapter in a long history of technological advancements. Pessimists suggest that the latest advances are different because they are inherently more threatening to the overall job market and wage share of total income. If the pessimists are right, what are the theoretical channels though which this would have a greater disinflationary effect relative to previous GPT technologies? Faster Productivity Gains: Enhanced productivity drives down unit labor costs, which may be passed along to other industries (as cheaper inputs) and to the end consumer. More Human Displacement: The jobs created in other areas may be insufficient to replace the jobs displaced by robots, leading to lower aggregate income and spending. The loss of income for labor will simply go to the owners of capital, but the point is that the labor share of income might decline. Deflationary pressures could build as aggregate demand falls short of supply. Even in industries that are slow to automate, just the threat of being replaced by robots may curtail wage demands. Inequality: Some have argued that rising inequality is partly because the spoils of new technologies over the past 20 years have largely gone to the owners of capital. This shift may have undermined aggregate demand because upper income households tend to have a high saving rate, thereby depressing overall aggregate demand and inflationary pressures. The human displacement effect, described above, would exacerbate the inequality effect by transferring income from labor to the owners of capital. 1. Productivity It is difficult to see the benefits of robots on productivity at the economy-wide level. Productivity growth has been abysmal across the major developed countries since the Great Recession, but the productivity slowdown was evident long before Lehman collapsed (Chart 7). The productivity slowdown continued even as automation using robots accelerated after 2010. Chart 7Productivity Collapsed Despite Automation Productivity Collapsed Despite Automation Productivity Collapsed Despite Automation Some analysts argue that lackluster productivity is simply a statistical mirage because of the difficulties in measuring output in today's economy. We will not get into the details of the mismeasurement debate here. We encourage interested clients to read a Special Report by the BCA Global Investment Strategy service entitled "Weak Productivity Growth: Don't Blame The Statisticians." 4 Our colleague Peter Berezin makes the case that the unmeasured utility accruing from free internet services is large, but so was the unmeasured utility from antibiotics, radio, indoor plumbing and air conditioning. He argues that the real reason that productivity growth has slowed is that educational attainment has decelerated and businesses have plucked many of the low-hanging fruit made possible by the IT revolution. Cyclical factors stemming from the Great Recession and financial crisis are also to blame, as capital spending has been slow to recover in most of the advanced economies. Some other factors that help to explain the decline in aggregate productivity are provided in Appendix 1. Nonetheless, the poor aggregate productivity performance does not mean that there are no benefits to using robots. The benefits are evident at the industrial level, where measurement issues are presumably less vexing for statisticians (i.e., it is easier to measure the output of the auto industry, for example, than for the economy as a whole). Chart 8 plots the level of robot density in 2016 with average annual productivity growth since 2004 for 10 U.S. manufacturing industries (robot density is presented in deciles). A loose positive relationship is apparent. Chart 8U.S.: Productivity Vs. Robot Density The Impact Of Robots On Inflation The Impact Of Robots On Inflation Academic studies estimate that robots have contributed importantly to economy-wide productivity growth. The Centre for Economic and Business Research (CEBR) estimated that labor productivity growth rises by 0.07 to 0.08 percentage points for every 1% rise in the rate of robot density.5 This implies that robots accounted for roughly 10% of the productivity growth experienced since the early 1990s in the major economies. Another study of 14 industries across 17 countries by the Centre for Economic Performance (CEP) found that robots boosted annual productivity growth by 0.36 percentage points over the 1993-2007 period.6 This is impressive because, if this estimate holds true for the U.S., robots' contribution to the 2½% average annual U.S. total productivity growth over the period was 14%. To put the importance of robotics into historical context, its contribution to productivity so far is roughly on par with that of the steam engine (Chart 9). It falls well short of the 0.6 percentage point annual productivity contribution from the IT revolution. The implication is that, while the overall productivity performance has been dismal since 2007, it would have been even worse in the absence of robots. What does this mean for inflation? According to the "cost push" model of the inflation process, an increase in productivity of 0.36% that is not accompanied by associated wage gains would reduce unit labor costs (ULC) by the same amount. This should trim inflation if the cost savings are passed on to the end consumer, although by less than 0.36% because robots can only depress variable costs, not fixed costs. There indeed appears to be a slight negative relationship between robot density and unit labor costs at the industrial level in the U.S., although the relationship is loose at best (Chart 10). Chart 9GPT Contribution To Productivity The Impact Of Robots On Inflation The Impact Of Robots On Inflation Chart 10U.S.: Unit Labor Costs Vs. Robot Density The Impact Of Robots On Inflation The Impact Of Robots On Inflation In theory, divergences in productivity across industries should only generate shifts in relative prices, and "cost push" inflation dynamics should only operate in the short term. Most economists believe that inflation is a purely monetary phenomenon in the long run, which means that central banks should be able to offset positive productivity shocks by lowering interest rates enough that aggregate demand keeps up with supply. Indeed, the Fed was successful in meeting the 2% inflation target on average from 2000 to 2007, when the impact of the IT revolution on productivity (and costs) was stronger than that of robot automation today. Also, note that inflation is currently low across the major advanced economies, irrespective of the level of robot intensity (Chart 11). From this perspective, it is hard to see that robots should take much of the credit for today's low inflation backdrop. Chart 11Inflation Vs. Robot Density The Impact Of Robots On Inflation The Impact Of Robots On Inflation 2. Human Displacement A key question is whether robots and humans are perfect substitutes. If new technologies introduced in the past were perfect substitutes, then it would have led to massive underemployment and all of the income in the economy would eventually have migrated to the owners of capital. The fact that average real household incomes have risen over time, and that there has been no secular upward trend in unemployment rates over the centuries, means that new technologies were at least partly complementary with labor (i.e., the jobs lost as a direct result of productivity gains were more than replaced in other areas of the economy over time). Rather than replacing workers, in many cases tech made humans more productive in their jobs. Rising productivity lifted income and thereby led to the creation of new jobs in other areas. The capital that workers bring to the production process - the skills, know-how and special talents - became more valuable as interaction with technology increased. Like today, there were concerns in the 1950s and 1960s that computerization would displace many types of jobs and lead to widespread idleness and falling household income. With hindsight, there was little to worry about. Some argue that this time is different. Futurists frequently assert that the pace of innovation is not just accelerating, it is accelerating 'exponentially'. Robots can now, or will soon be able to, replace humans in tasks that require cognitive skills. This means that they will be far less complementary to humans than in the past. The displacement effect could thus be much larger, especially given the impressive advances in artificial intelligence. However, Box 1 discusses why the threat to workers posed by AI is also heavily overblown in the media. The CEP multi-country study cited above did not find a large displacement effect; robot usage did not affect the overall number of hours worked in the 23 countries studied (although it found distributional effects - see below). In other words, rather than suppressing overall labor input, robot usage has led to more output, higher productivity, more jobs and stronger wage and income growth. A report by the Economic Policy Institute (EPI)7 takes a broader look at automation, using productivity growth and capital spending as proxies. Automation is what occurs as the implementation of new technologies is incorporated along with new capital equipment or software to replace human labor in the workplace. If automation is increasing 'exponentially' and displacing workers on a broad scale, one would expect to see accelerating productivity growth, robust capital spending, and more violent shifts in occupational shares. Exactly the opposite has occurred. Indeed, the report demonstrates that occupational employment shifts were far slower in the 2000-2015 period than in any decade in the 1900s (Chart 12). Box 1 The Threat From AI Is Overblown Media coverage of AI/Deep Learning has established a consensus view that we believe is well off the mark. A recent Special Report from BCA's Technology Sector Strategy service dispels the myths surrounding AI.8 We believe the consensus, in conjunction with warnings from a variety of sources, is leading to predictions, policy discussions, and even career choices based on a flawed premise. It is worth noting that the most vocal proponents of AI as a threat to jobs and even humanity are not AI experts. At the root of this consensus is the false view that emerging AI technology is anything like true intelligence. Modern AI is not remotely comparable in function to a biological brain. Scientists have a limited understanding of how brains work, and it is unlikely that a poorly understood system can be modeled on a computer. The misconception of intelligence is amplified by headlines claiming an AI "taught itself" a particular task. No AI has ever "taught itself" anything: All AI results have come about after careful programming by often PhD-level experts, who then supplied the system with vast amounts of high quality data to train it. Often these systems have been iterated a number of times and we only hear of successes, not the failures. The need for careful preparation of the AI system and the requirement for high quality data limits the applicability of AI to specific classes of problems where the application justifies the investment in development and where sufficient high-quality data exists. There may be numerous such applications but doubtless many more where AI would not be suitable. Similarly, an AI system is highly adapted to a single problem, or type of problem, and becomes less useful when its application set is expanded. In other words, unlike a human whose abilities improve as they learn more things, an AI's performance on a particular task declines as it does more things. There is a popular misconception that increased computing power will somehow lead to ever improving AI. It is the algorithm which determines the outcome, not the computer performance: Increased computing power leads to faster results, not different results. Advanced computers might lead to more advanced algorithms, but it is pointless to speculate where that may lead: A spreadsheet from 2001 may work faster today but it still gives the same answer. In any event, it is worth noting that a tool ceases to be a tool when it starts having an opinion: there is little reason to develop a machine capable of cognition even if that were possible. Chart 12U.S. Job Rotation Has Slowed The Impact Of Robots On Inflation The Impact Of Robots On Inflation The EPI report also notes that these indicators of automation increased rapidly in the late 1990s and early 2000s, a period that saw solid wage growth for American workers. These indicators weakened in the two periods of stagnant wage growth: from 1973 to 1995 and from 2002 to the present. Thus, there is no historical correlation between increases in automation and wage stagnation. Rather than automation, the report argues that it was China's entry into the global trading system that was largely responsible for the hollowing out of the U.S. manufacturing sector. We have also made this argument in previous research. The fact that the major advanced economies are all at, or close to, full employment supports the view that automation has not been an overwhelming headwind for job creation. Chart 13 demonstrates that there has been no relationship between the change in robot density and the loss of manufacturing jobs since 1993. Japan is an interesting case study because it is on the leading edge of the problems associated with an aging population. Interestingly, despite a worsening labor shortage, robot density among Japanese firms is falling. Moreover, the Japanese data show that the industries that have a high robot usage tend to be more, not less, generous with wages than the robot laggard industries. Please see Appendix 2 for more details. Chart 13Global Manufacturing Jobs Vs. Robot Density The Impact Of Robots On Inflation The Impact Of Robots On Inflation The bottom line is that it does not appear that labor displacement related to automation has been responsible in any meaningful way for the lackluster average real income growth in the advanced economies since 2007. 3. Inequality That said, there is evidence suggesting that robots are having important distributional effects. The CEP study found that robot use has reduced hours for low-skilled and (to a lesser extent) middle-skilled workers relative to the highly skilled. This finding makes sense conceptually. Technological change can exacerbate inequality by either increasing the relative demand for skilled over unskilled workers (so-called "skill-biased" technological change), or by inducing companies to substitute machinery and other forms of physical capital for workers (so-called "capital-biased" technological change). The former affects the distribution of labor income, while the latter affects the share of income in GDP that labor receives. A Special Report appearing in this publication in 2014 focused on the relationship between technology and inequality.9 The report highlighted that much of the recent technological change has been skill-biased, which heavily favors workers with the talent and education to perform cognitively-demanding tasks, even as it reduces demand for workers with only rudimentary skills. Moreover, technological innovations and globalization increasingly allow the most talented individuals to market their skills to a much larger audience, thus bidding up their wages. The evidence suggests that faster productivity growth leads to higher average real wages and improved living standards, at least over reasonably long horizons. Nonetheless, technological change can, and in the future almost certainly will, increase income inequality. The poor will gain, but not as much as the rich. The fact that higher-income households tend to maintain a higher savings rate than low-income households means that the shift in the distribution of income toward the higher-income households will continue to modestly weigh on aggregate demand. Can the distribution effect be large enough to have a meaningful depressing impact on inflation? We believe that it has played some role in the lackluster recovery since the Great Recession, with the result that an extended period of underemployment has delivered a persistent deflationary impulse in the major developed economies. However, as discussed above, stimulative monetary policy has managed to overcome the impact of inequality and other headwinds on aggregate demand, and has returned the major countries roughly to full employment. Indeed, this year will be the first since 2007 that the G20 economies as a group will be operating slightly above a full employment level. Inflation should respond to excess demand conditions, irrespective of any ongoing demand headwind stemming from inequality. Conclusions Technological change has led to rising living standards over the decades. It did not lead to widespread joblessness and did not prevent central banks from meeting their inflation targets over time. The pessimists argue that this time is different because robots/AI have a much larger displacement effect. Perhaps it will be 20 years before we will know the answer. But our main point is that we have found no evidence that recent advances in robotics and AI, while very impressive, will be any different in their macro impact. There is little evidence that the modern economy is less capable in replacing the jobs lost to automation, although the nature of new technologies may be affecting the distribution of income more than in the past. Real incomes for the middle- and lower-income classes have been stagnant for some time, but this is partly due to productivity growth that is too low, not too high. Moreover, it is not at all clear that positive productivity shocks are disinflationary beyond the near term. The link between robot usage and unit labor costs over the past couple of decades is loose at best at the industry level, and is non-existent when looking across the major countries. The Fed was able to roughly meet its 2% inflation target in the 1990s and the first half of the 2000s, despite IT's impressive contribution to productivity growth during that period. For investors, this means that we cannot rely on automation to keep inflation depressed irrespective of how tight labor markets become. The global output gap will shift into positive territory this year for the first time since the Great Recession. Any resulting rise in inflation will come as a shock since the bond market has discounted continued low inflation for as far as the eye can see. We expect bond yields and implied volatility to rise this year, which may undermine risk assets in the second half. Mark McClellan Senior Vice President The Bank Credit Analyst Brian Piccioni Vice President Technology Sector Strategy Appendix 1 Why Is Productivity So Low? A recent study by the OECD10 reveals that, while frontier firms are charging ahead, there is a widening gap between these firms and the laggards. The study analyzed firm-level data on labor productivity and total factor productivity for 24 countries. "Frontier" firms are defined to be those with productivity in the top 5%. These firms are 3-4 times as productive as the remaining 95%. The authors argue that the underlying cause of this yawning gap is that the diffusion rate of new technologies from the frontier firms to the laggards has slowed within industries. This could be due to rising barriers to entry, which has reduced contestability in markets. Curtailing the creative-destruction process means that there is less pressure to innovate. Barriers to entry may have increased because "...the importance of tacit knowledge as a source of competitive advantage for frontier firms may have risen if increasingly complex technologies were to increase the amount and sophistication of complementary investments required for technological adoption." 11 The bottom line is that aggregate productivity is low because the robust productivity gains for the tech-savvy frontier companies are offset by the long tail of firms that have been slow to adopt the latest technology. Indeed, business spending has been especially weak in this expansion. Chart 14 highlights that the slowdown in U.S. productivity growth has mirrored that of the capital stock. Chart 14U.S. Capex Shortfall Partly To Blame For Poor Productivity U.S. Capex Shortfall Partly To Blame For Poor Productivity U.S. Capex Shortfall Partly To Blame For Poor Productivity Appendix 2 Japan - The Leading Edge Japan is an interesting case study because it is on the leading edge of the problems associated with an aging population. The popular press is full of stories of how robots are taking over. If the stories are to be believed, robots are the answer to the country's shrinking workforce. Robots now serve as helpers for the elderly, priests for weddings and funerals, concierges for hotels and even sexual partners (don't ask). Prime Minister Abe's government has launched a 5-year push to deepen the use of intelligent machines in manufacturing, supply chains, construction and health care. Indeed, Japan was the leader in robotics use for decades. Nonetheless, despite all the hype, Japan's stock of industrial robots has actually been eroding since the late 1990s (Chart 4). Numerous surveys show that firms plan to use robots more in the future because of the difficulty in hiring humans. And there is huge potential: 90% of Japanese firms are small- and medium-sized (SME) and most are not currently using robots. Yet, there has been no wave of robot purchases as of 2016. One problem is the cost; most sophisticated robots are simply too expensive for SMEs to consider. This suggests that one cannot blame robots for Japan's lack of wage growth. The labor shortage has become so acute that there are examples of companies that have turned down sales due to insufficient manpower. Possible reasons why these companies do not offer higher wages to entice workers are beyond the scope of this report. But the fact that the stock of robots has been in decline since the late 1990s does not support the view that Japanese firms are using automation on a broad scale to avoid handing out pay hikes. Indeed, Chart 15 highlights that wage deflation has been the greatest in industries that use almost no robots. Highly automated industries, such as Transportation Equipment and Electronics, have been among the most generous. This supports the view that the productivity afforded by increased robot usage encourages firms to pay their workers more. Looking ahead, it seems implausible that robots can replace all the retiring Japanese workers in the years to come. The workforce will shrink at an annual average pace of 0.33% between 2020 and 2030, according to the Japan Institute for Labour Policy and Training. Productivity growth would have to rise by the same amount to fully offset the dwindling number of workers. But that would require a surge in robot density of 4.1, assuming that each rise in robot density of one adds 0.08% to the level of productivity (Chart 16). The level of robot sales would have to jump by a whopping 2½ times in the first year and continue to rise at the same pace each year thereafter to make this happen. Of course, the productivity afforded by new robots may accelerate in the coming years, but the point is that robot usage would likely have to rise astronomically to offset the impact of the shrinking population. Chart 15Japan: Earnings Vs. Robot Density The Impact Of Robots On Inflation The Impact Of Robots On Inflation Chart 16Japan: Where Is The Flood Of Robots? Japan: Where Is The Flood Of Robots? Japan: Where Is The Flood Of Robots? The implication is that, as long as the Japanese economy continues to grow above roughly 1%, the labor market will continue to tighten and wage rates will eventually begin to rise. 1 Please see Technology Sector Strategy Special Report "The Coming Robotics Revolution," dated May 16, 2017, available at tech.bcaresearch.com 2 Note that this includes only robots used in manufacturing industry, and thus excludes robots used in the service sector and households. However, robot usage in services is quite limited and those used in households do not add to GDP. 3 Note that ICT investment and capital stock data includes robots. 4 Please see BCA Global Investment Strategy Special Report "Weak Productivity Growth: Don't Blame The Statisticians," dated March 25, 2016, available at gis.bcaresearch.com 5 Centre for Economic and Business Research (January 2017) "The Impact of Automation." A Report for Redwood. In this report, robot density is defined to be the number of robots per million hours worked. 6 Graetz, G., and Michaels, G. (2015): "Robots At Work." CEP Discussion Paper No 1335. 7 Mishel, L., and Bivens, J. (2017): "The Zombie Robot Argument Lurches On," Economic Policy Institute. 8 Please see BCA Technology Sector Strategy Special Report "Bad Information - Why Misreporting Deep Learning Advances Is A Problem," dated January 9, 2018, available at tech.bcaresearch.com 9 Please see The Bank Credit Analyst, "Rage Against the Machines: Is Technology Exacerbating Inequality?" dated June 2014, available at bca.bcaresearch.com. 10 OECD Productivity Working Papers, No. 05 (2016) "The Best Versus the Rest: The Global Productivity Slowdown, Divergence Across Firms and the Role of Public Policy." 11 Please refer to page 8.

Media reports warn of a "Robot Apocalypse" that is already laying waste to jobs and depressing wages on a broad scale. Technological advance in the past has not prevented improving living standards or led to ever rising joblessness over the decades, but pessimists argue that recent advances are different. The issue is important for financial markets. If structural factors such as automation are holding back inflation by more than in previous decades, then the Fed will have to proceed very slowly in raising rates. We see no compelling evidence that the displacement effect of emerging technologies is any stronger than in the past. Robot usage has had a modest positive impact on overall productivity. Despite this contribution, overall productivity growth has been dismal over the past decade. If automation is increasing 'exponentially' and displacing workers on a broad scale as some claim, one would expect to see accelerating productivity growth, robust capital spending and more violent shifts in occupational shares. Exactly the opposite has occurred. Periods of strong growth in automation have historically been associated with robust, not lackluster, wage gains, contrary to the consensus view. The Fed was successful in meeting the 2% inflation target on average from 2000 to 2007, when the impact of the IT revolution on productivity (and costs) was stronger than that of robot automation today. This and other evidence suggest that it is difficult to make the case that robots will make it tougher for central banks to reach their inflation goals than did previous technological breakthroughs. For investors, this means that we cannot rely on automation to keep inflation depressed irrespective of how tight labor markets become. Recent breakthroughs in technology are awe-inspiring and unsettling. These advances are viewed with great trepidation by many because of the potential to replace humans in the production process. Hype over robots is particularly shrill. Media reports warn of a "Robot Apocalypse" that is already laying waste to jobs and depressing wages on a broad scale. In the first in our series of Special Reports focusing on the structural factors that might be preventing central banks from reaching their inflation targets, we demonstrated that the impact of Amazon is overstated in the press. We estimated that E-commerce is depressing inflation in the U.S. by a mere 0.1 to 0.2 percentage points. This Special Report tackles the impact of automation. We are optimistic that robot technology and artificial intelligence will significantly boost future productivity, and thus reduce costs. But, is there any evidence at the macro level that robot usage has been more deflationary than technological breakthroughs in the past and is, thus, a major driver of the low inflation rates we observe today across the major countries? The question matters, especially for the outlook for central bank policy and the bond market. If structural factors are indeed holding back inflation by more than in previous decades, then the Fed will have to proceed very slowly in raising rates. However, if low inflation simply reflects long lags between wages and the tightening labor market, then inflation may suddenly lurch to life as it has at the end of past cycles. The bond market is not priced for that scenario. Are Robots Different? A Special Report from BCA's Technology Sector Strategy service suggested that the "robot revolution" could be as transformative as previous General Purpose Technologies (GPT), including the steam engine, electricity and the microchip.1 GPTs are technologies that radically alter the economy's production process and make a major contribution to living standards over time. The term "robot" can have different meanings. The most basic definition is "a device that automatically performs complicated and often repetitive tasks," and this encompasses a broad range of machines: From the Jacquard Loom, which was invented over 200 years ago, on to Numerically Controlled (NC) mills and lathes, pick and place machines used in the manufacture of electronics, Autonomous Vehicles (AVs), and even homicidal robots from the future such as the Terminator. Our Technology Sector report made the case that there is nothing particularly sinister about robots. They are just another chapter in a long history of automation. Nor is the displacement of workers unprecedented. The industrial revolution was about replacing human craft labor with capital (machines), which did high-volume work with better quality and productivity. This freed humans for work which had not yet been automated, along with designing, producing and maintaining the machinery. Agriculture offers a good example. This sector involved over 50% of the U.S. labor force until the late 1800s. Steam and then internal combustion-powered tractors, which can be viewed as "robotic horses," contributed to a massive rise in output-per-man hour. The number of hours worked to produce a bushel of wheat fell by almost 98% from the mid-1800s to 1955. This put a lot of farm hands out of work, but these laborers were absorbed over time in other growing areas of the economy. It is the same story for all other historical technological breakthroughs. Change is stressful for those directly affected, but rising productivity ultimately lifts average living standards. Robots will be no different. As we discuss below, however, the increasing use of robots and AI may have a deeper and longer-lasting impact on inequality. Strong Tailwinds Chart II-1Robots Are Getting Cheaper Robots Are Getting Cheaper Robots Are Getting Cheaper Factory robots have improved immensely due to cheaper and more capable control and vision systems. As these systems evolve, the abilities of robots to move around their environment while avoiding obstacles will improve, as will their ability to perform increasingly complex tasks. Most importantly, robots are already able to do more than just routine tasks, thus enabling them to replace or aid humans in higher-skilled processes. Robot prices are also falling fast, especially after quality-adjusting the data (Chart II-1). Units are becoming easier to install, program and operate. These trends will help to reduce the barriers-to-entry for the large, untapped, market of small and medium sized enterprises. Robots also offer the ability to do low-volume "customized" production and still keep unit costs low. In the future, self-learning robots will be able to optimize their own performance by analyzing the production of other robots around the world. Robot usage is growing quickly according to data collected by the International Federation of Robotics (IFR) that covers 23 countries. Industrial robot sales worldwide increased to almost 300,000 units in 2016, up 16% from the year before (Chart II-2). The stock of industrial robots globally has grown at an annual average pace of 10% since 2010, reaching slightly more than 1.8 million units in 2016.2 Robot usage is far from evenly distributed across industries. The automotive industry is the major consumer of industrial robots, holding 45% of the total stock in 2016 (Chart II-3). The computer & electronics industry is a distant second at 17%. Metals, chemicals and electrical/electronic appliances comprise the bulk of the remaining stock. Chart II-2Global Robot Usage Global Robot Usage Global Robot Usage Chart II-3Global Robot Usage By Industry (2016) February 2018 February 2018 As far as countries go, Japan has traditionally been the largest market for robots in the world. However, sales have been in a long-term downtrend and the stock of robots has recently been surpassed by China, which has ramped up robot purchases in recent years (Chart II-4). Robot density, which is the stock of robots per 10 thousand employed in manufacturing, makes it easier to compare robot usage across countries (Chart II-5, panel 2). By this measure, China is not a heavy user of robots compared to other countries. South Korea stands at the top, well above the second-place finishers (Germany and Japan). Large automobile sectors in these three countries explain their high relative robot densities. Chart II-4Stock Of Robots By Country (I) Stock Of Robots By Country (I) Stock Of Robots By Country (I) Chart II-5Stock Of Robots By Country (II) (2016) February 2018 February 2018 While the growth rate of robot usage is impressive, it is from a very low base (outside of the automotive industry). The average number of robots per 10,000 employees is only 74 for the 23 countries in the IFR database. Robot use is tiny compared to total man hours worked. Chart II-6U.S. Investment In Robots U.S. Investment in Robots U.S. Investment in Robots In the U.S., spending on robots is only about 5% of total business spending on equipment and software (Chart II-6). To put this into perspective, U.S. spending on information, communication and technology (ICT) equipment represented 35-40% of total capital equipment spending during the tech boom in the 1990s and early 2000s.3 The bottom line is that there is a lot of hype in the press, but robots are not yet widely used across countries or industries. It will be many years before business spending on robots approaches the scale of the 1990s/2000s IT boom. A Deflationary Impact? As noted above, we view robotics as another chapter in a long history of technological advancements. Pessimists suggest that the latest advances are different because they are inherently more threatening to the overall job market and wage share of total income. If the pessimists are right, what are the theoretical channels though which this would have a greater disinflationary effect relative to previous GPT technologies? Faster Productivity Gains: Enhanced productivity drives down unit labor costs, which may be passed along to other industries (as cheaper inputs) and to the end consumer. More Human Displacement: The jobs created in other areas may be insufficient to replace the jobs displaced by robots, leading to lower aggregate income and spending. The loss of income for labor will simply go to the owners of capital, but the point is that the labor share of income might decline. Deflationary pressures could build as aggregate demand falls short of supply. Even in industries that are slow to automate, just the threat of being replaced by robots may curtail wage demands. Inequality: Some have argued that rising inequality is partly because the spoils of new technologies over the past 20 years have largely gone to the owners of capital. This shift may have undermined aggregate demand because upper income households tend to have a high saving rate, thereby depressing overall aggregate demand and inflationary pressures. The human displacement effect, described above, would exacerbate the inequality effect by transferring income from labor to the owners of capital. 1. Productivity It is difficult to see the benefits of robots on productivity at the economy-wide level. Productivity growth has been abysmal across the major developed countries since the Great Recession, but the productivity slowdown was evident long before Lehman collapsed (Chart II-7). The productivity slowdown continued even as automation using robots accelerated after 2010. Chart II-7Productivity Collapsed Despite Automation Productivity Collapsed Despite Automation Productivity Collapsed Despite Automation Some analysts argue that lackluster productivity is simply a statistical mirage because of the difficulties in measuring output in today's economy. We will not get into the details of the mismeasurement debate here. We encourage interested clients to read a Special Report by the BCA Global Investment Strategy service entitled "Weak Productivity Growth: Don't Blame The Statisticians." 4 Our colleague Peter Berezin makes the case that the unmeasured utility accruing from free internet services is large, but so was the unmeasured utility from antibiotics, radio, indoor plumbing and air conditioning. He argues that the real reason that productivity growth has slowed is that educational attainment has decelerated and businesses have plucked many of the low-hanging fruit made possible by the IT revolution. Cyclical factors stemming from the Great Recession and financial crisis are also to blame, as capital spending has been slow to recover in most of the advanced economies. Some other factors that help to explain the decline in aggregate productivity are provided in Appendix II-1. Nonetheless, the poor aggregate productivity performance does not mean that there are no benefits to using robots. The benefits are evident at the industrial level, where measurement issues are presumably less vexing for statisticians (i.e., it is easier to measure the output of the auto industry, for example, than for the economy as a whole). Chart II-8 plots the level of robot density in 2016 with average annual productivity growth since 2004 for 10 U.S. manufacturing industries (robot density is presented in deciles). A loose positive relationship is apparent. Chart II-8U.S.: Productivity Vs. Robot Density February 2018 February 2018 Academic studies estimate that robots have contributed importantly to economy-wide productivity growth. The Centre for Economic and Business Research (CEBR) estimated that labor productivity growth rises by 0.07 to 0.08 percentage points for every 1% rise in the rate of robot density.5 This implies that robots accounted for roughly 10% of the productivity growth experienced since the early 1990s in the major economies. Another study of 14 industries across 17 countries by the Centre for Economic Performance (CEP) found that robots boosted annual productivity growth by 0.36 percentage points over the 1993-2007 period.6 This is impressive because, if this estimate holds true for the U.S., robots' contribution to the 2½% average annual U.S. total productivity growth over the period was 14%. To put the importance of robotics into historical context, its contribution to productivity so far is roughly on par with that of the steam engine (Chart II-9). It falls well short of the 0.6 percentage point annual productivity contribution from the IT revolution. The implication is that, while the overall productivity performance has been dismal since 2007, it would have been even worse in the absence of robots. What does this mean for inflation? According to the "cost push" model of the inflation process, an increase in productivity of 0.36% that is not accompanied by associated wage gains would reduce unit labor costs (ULC) by the same amount. This should trim inflation if the cost savings are passed on to the end consumer, although by less than 0.36% because robots can only depress variable costs, not fixed costs. There indeed appears to be a slight negative relationship between robot density and unit labor costs at the industrial level in the U.S., although the relationship is loose at best (Chart II-10). Chart II-9GPT Contribution To Productivity February 2018 February 2018 Chart II-10U.S.: Unit Labor Costs Vs. Robot Density February 2018 February 2018 In theory, divergences in productivity across industries should only generate shifts in relative prices, and "cost push" inflation dynamics should only operate in the short term. Most economists believe that inflation is a purely monetary phenomenon in the long run, which means that central banks should be able to offset positive productivity shocks by lowering interest rates enough that aggregate demand keeps up with supply. Indeed, the Fed was successful in meeting the 2% inflation target on average from 2000 to 2007, when the impact of the IT revolution on productivity (and costs) was stronger than that of robot automation today. Also, note that inflation is currently low across the major advanced economies, irrespective of the level of robot intensity (Chart II-11). From this perspective, it is hard to see that robots should take much of the credit for today's low inflation backdrop. Chart II-11Inflation Vs. Robot Density February 2018 February 2018 2. Human Displacement A key question is whether robots and humans are perfect substitutes. If new technologies introduced in the past were perfect substitutes, then it would have led to massive underemployment and all of the income in the economy would eventually have migrated to the owners of capital. The fact that average real household incomes have risen over time, and that there has been no secular upward trend in unemployment rates over the centuries, means that new technologies were at least partly complementary with labor (i.e., the jobs lost as a direct result of productivity gains were more than replaced in other areas of the economy over time). Rather than replacing workers, in many cases tech made humans more productive in their jobs. Rising productivity lifted income and thereby led to the creation of new jobs in other areas. The capital that workers bring to the production process - the skills, know-how and special talents - became more valuable as interaction with technology increased. Like today, there were concerns in the 1950s and 1960s that computerization would displace many types of jobs and lead to widespread idleness and falling household income. With hindsight, there was little to worry about. Some argue that this time is different. Futurists frequently assert that the pace of innovation is not just accelerating, it is accelerating 'exponentially'. Robots can now, or will soon be able to, replace humans in tasks that require cognitive skills. This means that they will be far less complementary to humans than in the past. The displacement effect could thus be much larger, especially given the impressive advances in artificial intelligence. However, Box II-1 discusses why the threat to workers posed by AI is also heavily overblown in the media. The CEP multi-country study cited above did not find a large displacement effect; robot usage did not affect the overall number of hours worked in the 23 countries studied (although it found distributional effects - see below). In other words, rather than suppressing overall labor input, robot usage has led to more output, higher productivity, more jobs and stronger wage and income growth. A report by the Economic Policy Institute (EPI)7 takes a broader look at automation, using productivity growth and capital spending as proxies. Automation is what occurs as the implementation of new technologies is incorporated along with new capital equipment or software to replace human labor in the workplace. If automation is increasing 'exponentially' and displacing workers on a broad scale, one would expect to see accelerating productivity growth, robust capital spending, and more violent shifts in occupational shares. Exactly the opposite has occurred. Indeed, the report demonstrates that occupational employment shifts were far slower in the 2000-2015 period than in any decade in the 1900s (Chart II-12). Box II-1 The Threat From AI Is Overblown Media coverage of AI/Deep Learning has established a consensus view that we believe is well off the mark. A recent Special Report from BCA's Technology Sector Strategy service dispels the myths surrounding AI.8 We believe the consensus, in conjunction with warnings from a variety of sources, is leading to predictions, policy discussions, and even career choices based on a flawed premise. It is worth noting that the most vocal proponents of AI as a threat to jobs and even humanity are not AI experts. At the root of this consensus is the false view that emerging AI technology is anything like true intelligence. Modern AI is not remotely comparable in function to a biological brain. Scientists have a limited understanding of how brains work, and it is unlikely that a poorly understood system can be modeled on a computer. The misconception of intelligence is amplified by headlines claiming an AI "taught itself" a particular task. No AI has ever "taught itself" anything: All AI results have come about after careful programming by often PhD-level experts, who then supplied the system with vast amounts of high quality data to train it. Often these systems have been iterated a number of times and we only hear of successes, not the failures. The need for careful preparation of the AI system and the requirement for high quality data limits the applicability of AI to specific classes of problems where the application justifies the investment in development and where sufficient high-quality data exists. There may be numerous such applications but doubtless many more where AI would not be suitable. Similarly, an AI system is highly adapted to a single problem, or type of problem, and becomes less useful when its application set is expanded. In other words, unlike a human whose abilities improve as they learn more things, an AI's performance on a particular task declines as it does more things. There is a popular misconception that increased computing power will somehow lead to ever improving AI. It is the algorithm which determines the outcome, not the computer performance: Increased computing power leads to faster results, not different results. Advanced computers might lead to more advanced algorithms, but it is pointless to speculate where that may lead: A spreadsheet from 2001 may work faster today but it still gives the same answer. In any event, it is worth noting that a tool ceases to be a tool when it starts having an opinion: there is little reason to develop a machine capable of cognition even if that were possible. Chart II-12U.S. Job Rotation Has Slowed February 2018 February 2018 The EPI report also notes that these indicators of automation increased rapidly in the late 1990s and early 2000s, a period that saw solid wage growth for American workers. These indicators weakened in the two periods of stagnant wage growth: from 1973 to 1995 and from 2002 to the present. Thus, there is no historical correlation between increases in automation and wage stagnation. Rather than automation, the report argues that it was China's entry into the global trading system that was largely responsible for the hollowing out of the U.S. manufacturing sector. We have also made this argument in previous research. The fact that the major advanced economies are all at, or close to, full employment supports the view that automation has not been an overwhelming headwind for job creation. Chart II-13 demonstrates that there has been no relationship between the change in robot density and the loss of manufacturing jobs since 1993. Japan is an interesting case study because it is on the leading edge of the problems associated with an aging population. Interestingly, despite a worsening labor shortage, robot density among Japanese firms is falling. Moreover, the Japanese data show that the industries that have a high robot usage tend to be more, not less, generous with wages than the robot laggard industries. Please see Appendix II-2 for more details. Chart II-13Global Manufacturing Jobs Vs. Robot Density February 2018 February 2018 The bottom line is that it does not appear that labor displacement related to automation has been responsible in any meaningful way for the lackluster average real income growth in the advanced economies since 2007. 3. Inequality That said, there is evidence suggesting that robots are having important distributional effects. The CEP study found that robot use has reduced hours for low-skilled and (to a lesser extent) middle-skilled workers relative to the highly skilled. This finding makes sense conceptually. Technological change can exacerbate inequality by either increasing the relative demand for skilled over unskilled workers (so-called "skill-biased" technological change), or by inducing companies to substitute machinery and other forms of physical capital for workers (so-called "capital-biased" technological change). The former affects the distribution of labor income, while the latter affects the share of income in GDP that labor receives. A Special Report appearing in this publication in 2014 focused on the relationship between technology and inequality.9 The report highlighted that much of the recent technological change has been skill-biased, which heavily favors workers with the talent and education to perform cognitively-demanding tasks, even as it reduces demand for workers with only rudimentary skills. Moreover, technological innovations and globalization increasingly allow the most talented individuals to market their skills to a much larger audience, thus bidding up their wages. The evidence suggests that faster productivity growth leads to higher average real wages and improved living standards, at least over reasonably long horizons. Nonetheless, technological change can, and in the future almost certainly will, increase income inequality. The poor will gain, but not as much as the rich. The fact that higher-income households tend to maintain a higher savings rate than low-income households means that the shift in the distribution of income toward the higher-income households will continue to modestly weigh on aggregate demand. Can the distribution effect be large enough to have a meaningful depressing impact on inflation? We believe that it has played some role in the lackluster recovery since the Great Recession, with the result that an extended period of underemployment has delivered a persistent deflationary impulse in the major developed economies. However, as discussed above, stimulative monetary policy has managed to overcome the impact of inequality and other headwinds on aggregate demand, and has returned the major countries roughly to full employment. Indeed, this year will be the first since 2007 that the G20 economies as a group will be operating slightly above a full employment level. Inflation should respond to excess demand conditions, irrespective of any ongoing demand headwind stemming from inequality. Conclusions Technological change has led to rising living standards over the decades. It did not lead to widespread joblessness and did not prevent central banks from meeting their inflation targets over time. The pessimists argue that this time is different because robots/AI have a much larger displacement effect. Perhaps it will be 20 years before we will know the answer. But our main point is that we have found no evidence that recent advances in robotics and AI, while very impressive, will be any different in their macro impact. There is little evidence that the modern economy is less capable in replacing the jobs lost to automation, although the nature of new technologies may be affecting the distribution of income more than in the past. Real incomes for the middle- and lower-income classes have been stagnant for some time, but this is partly due to productivity growth that is too low, not too high. Moreover, it is not at all clear that positive productivity shocks are disinflationary beyond the near term. The link between robot usage and unit labor costs over the past couple of decades is loose at best at the industry level, and is non-existent when looking across the major countries. The Fed was able to roughly meet its 2% inflation target in the 1990s and the first half of the 2000s, despite IT's impressive contribution to productivity growth during that period. For investors, this means that we cannot rely on automation to keep inflation depressed irrespective of how tight labor markets become. The global output gap will shift into positive territory this year for the first time since the Great Recession. Any resulting rise in inflation will come as a shock since the bond market has discounted continued low inflation for as far as the eye can see. We expect bond yields and implied volatility to rise this year, which may undermine risk assets in the second half. Mark McClellan Senior Vice President The Bank Credit Analyst Brian Piccioni Vice President Technology Sector Strategy Appendix II-1 Why Is Productivity So Low? A recent study by the OECD10 reveals that, while frontier firms are charging ahead, there is a widening gap between these firms and the laggards. The study analyzed firm-level data on labor productivity and total factor productivity for 24 countries. "Frontier" firms are defined to be those with productivity in the top 5%. These firms are 3-4 times as productive as the remaining 95%. The authors argue that the underlying cause of this yawning gap is that the diffusion rate of new technologies from the frontier firms to the laggards has slowed within industries. This could be due to rising barriers to entry, which has reduced contestability in markets. Curtailing the creative-destruction process means that there is less pressure to innovate. Barriers to entry may have increased because "...the importance of tacit knowledge as a source of competitive advantage for frontier firms may have risen if increasingly complex technologies were to increase the amount and sophistication of complementary investments required for technological adoption." 11 The bottom line is that aggregate productivity is low because the robust productivity gains for the tech-savvy frontier companies are offset by the long tail of firms that have been slow to adopt the latest technology. Indeed, business spending has been especially weak in this expansion. Chart II-14 highlights that the slowdown in U.S. productivity growth has mirrored that of the capital stock. Chart II-14U.S. Capex Shortfall Partly To Blame For Poor Productivity U.S. Capex Shortfall Partly To Blame For Poor Productivity U.S. Capex Shortfall Partly To Blame For Poor Productivity Appendix II-2 Japan - The Leading Edge Japan is an interesting case study because it is on the leading edge of the problems associated with an aging population. The popular press is full of stories of how robots are taking over. If the stories are to be believed, robots are the answer to the country's shrinking workforce. Robots now serve as helpers for the elderly, priests for weddings and funerals, concierges for hotels and even sexual partners (don't ask). Prime Minister Abe's government has launched a 5-year push to deepen the use of intelligent machines in manufacturing, supply chains, construction and health care. Indeed, Japan was the leader in robotics use for decades. Nonetheless, despite all the hype, Japan's stock of industrial robots has actually been eroding since the late 1990s (Chart II-4). Numerous surveys show that firms plan to use robots more in the future because of the difficulty in hiring humans. And there is huge potential: 90% of Japanese firms are small- and medium-sized (SME) and most are not currently using robots. Yet, there has been no wave of robot purchases as of 2016. One problem is the cost; most sophisticated robots are simply too expensive for SMEs to consider. This suggests that one cannot blame robots for Japan's lack of wage growth. The labor shortage has become so acute that there are examples of companies that have turned down sales due to insufficient manpower. Possible reasons why these companies do not offer higher wages to entice workers are beyond the scope of this report. But the fact that the stock of robots has been in decline since the late 1990s does not support the view that Japanese firms are using automation on a broad scale to avoid handing out pay hikes. Indeed, Chart II-15 highlights that wage deflation has been the greatest in industries that use almost no robots. Highly automated industries, such as Transportation Equipment and Electronics, have been among the most generous. This supports the view that the productivity afforded by increased robot usage encourages firms to pay their workers more. Looking ahead, it seems implausible that robots can replace all the retiring Japanese workers in the years to come. The workforce will shrink at an annual average pace of 0.33% between 2020 and 2030, according to the Japan Institute for Labour Policy and Training. Productivity growth would have to rise by the same amount to fully offset the dwindling number of workers. But that would require a surge in robot density of 4.1, assuming that each rise in robot density of one adds 0.08% to the level of productivity (Chart II-16). The level of robot sales would have to jump by a whopping 2½ times in the first year and continue to rise at the same pace each year thereafter to make this happen. Of course, the productivity afforded by new robots may accelerate in the coming years, but the point is that robot usage would likely have to rise astronomically to offset the impact of the shrinking population. Chart II-15Japan: Earnings Vs. Robot Density February 2018 February 2018 Chart II-16Japan: Where Is The Flood Of Robots? Japan: Where Is The Flood OF Robots? Japan: Where Is The Flood OF Robots? The implication is that, as long as the Japanese economy continues to grow above roughly 1%, the labor market will continue to tighten and wage rates will eventually begin to rise. 1 Please see Technology Sector Strategy Special Report "The Coming Robotics Revolution," dated May 16, 2017, available at tech.bcaresearch.com 2 Note that this includes only robots used in manufacturing industry, and thus excludes robots used in the service sector and households. However, robot usage in services is quite limited and those used in households do not add to GDP. 3 Note that ICT investment and capital stock data includes robots. 4 Please see BCA Global Investment Strategy Special Report "Weak Productivity Growth: Don't Blame The Statisticians," dated March 25, 2016, available at gis.bcaresearch.com 5 Centre for Economic and Business Research (January 2017): "The Impact of Automation." A Report for Redwood. In this report, robot density is defined to be the number of robots per million hours worked. 6 Graetz, G., and Michaels, G. (2015): "Robots At Work." CEP Discussion Paper No 1335. 7 Mishel, L., and Bivens, J. (2017): "The Zombie Robot Argument Lurches On," Economic Policy Institute. 8 Please see BCA Technology Sector Strategy Special Report "Bad Information - Why Misreporting Deep Learning Advances Is A Problem," dated January 9, 2018, available at tech.bcaresearch.com 9 Please see The Bank Credit Analyst, "Rage Against The Machines: Is Technology Exacerbating Inequality?" dated June 2014, available at bca.bcaresearch.com 10 OECD Productivity Working Papers, No. 05 (2016): "The Best Versus the Rest: The Global Productivity Slowdown, Divergence Across Firms and the Role of Public Policy." 11 Please refer to page 27.

Dear Client, In addition to this Special Report written by my colleague Mark McClellan, we are sending you an abbreviated weekly report, which includes the Tactical Global Asset Allocation Monthly Update. Best regards, Peter Berezin, Chief Global Strategist Global Investment Strategy Highlights A "culture of profound cost reduction" has gripped the business sector since the GFC according to one school of thought, permanently changing the relationship between labor market slack and wages or inflation. If true, it could mean that central banks are almost powerless to reach their inflation targets. Amazon, Airbnb, Uber, robotics, contract workers, artificial intelligence, horizontal drilling and driverless cars are just a few examples of companies and technologies that are cutting costs and depressing prices and wages. In the first of our series on inflation, we will focus on the rise of e-commerce and the related "Amazonification" of the economy. In theory, positive supply shocks should not have more than a temporary impact on inflation if the price level is indeed a monetary phenomenon in the long term. But a series of positive supply shocks could make it appear for quite a while that low inflation is structural in nature. We are keeping an open mind and reserving judgement on the disinflationary impact of robotics, artificial intelligence and the gig economy until we do more research. But in terms of the impact of e-commerce, it is difficult to find supportive evidence at the macro level. The admittedly inadequate measures of online prices available today do not suggest that e-commerce sales are depressing the overall inflation rate by more than 0.1 or 0.2 percentage points. Moreover, it does not appear that the disinflationary impact of competition in the retail sector has intensified over the years. Today's creative destruction in retail may be no more deflationary than the shift to 'big box' stores in the 1990s. Perhaps lower online prices are forcing traditional retailers match the e-commerce vendors, allowing for a larger disinflationary effect than we estimate. However, the fact that retail margins are near secular highs outside of department stores argues against this thesis. The sectors potentially affected by e-commerce make up a small part of the CPI index. The deceleration of inflation since the GFC has been in areas unaffected by online sales. High profit margins for the overall corporate sector and depressed productivity growth also argue against the idea that e-commerce represents a large positive macro supply shock. Perhaps the main way that e-commerce is affecting the macro economy and financial markets is not through inflation, but via the reduction in the economy's capital spending requirement. This would reduce the equilibrium level of interest rates, since the Fed has to stimulate other parts of the economy to offset the loss of demand in capital spending in the retail sector. Feature Anecdotal evidence is all around us. The global economy is evolving and it seems that all of the major changes are deflationary. Amazon, Airbnb, Uber, robotics, contract workers, artificial intelligence, horizontal drilling and driverless cars are just a few examples of companies and technologies that are cutting costs and depressing prices and wages. Central banks in the major advanced economies are having difficulty meeting their inflation targets, even in the U.S. where the labor market is tight by historical standards. Based on the depressed level of bond yields, it appears that the majority of investors believe that inflation headwinds will remain formidable for a long time. One school of thought is that low inflation reflects a lack of demand growth in the post-Great Financial Crisis (GFC) period. Another school points to the supply side of the economy. A recent report by Prudential Financial highlights "...obvious examples of ... new business models and new organizational structures, whereby higher-cost traditional methods of production, transportation, and distribution are displaced by more nontraditional cost-effective ways of conducting business." 1 A "culture of profound cost reduction" has gripped the business sector since the GFC according to this school, permanently changing the relationship between labor market slack and wages or inflation (i.e., the Phillips Curve). Employees are less aggressive in their wage demands in a world where robots are threatening humans in a broadening array of industrial categories. Many feel lucky just to have a job. In a highly sensationalized article called "How The Internet Economy Killed Inflation," Forbes argued that "the internet has reduced many of the traditional barriers to entry that protect companies from competition and created a race to the bottom for prices in a number of categories." Forbes believes that new technologies are placing downward pressure on inflation by depressing wages, increasing productivity and encouraging competition. There are many factors that have the potential to weigh on prices, but analysts are mainly focusing on e-commerce, robotics, artificial intelligence, and the gig economy. In the first of our series on inflation, we will focus on the rise of e-commerce and the related "Amazonification" of the economy. The latter refers to the advent of new business models that cut out layers of middlemen between producers and consumers. Amazonification E-commerce has grown at a compound annual rate of more than 9% over the past 15 years, and now accounts for about 8½% of total U.S. retail sales (Chart 1). Amazon has been leading the charge, accounting for 43% of all online sales in 2016 (Chart 2). Amazon's business model not only cuts costs by eliminating middlemen and (until recently) avoiding expensive showrooms, but it also provides a platform for improved price discovery on an extremely broad array of goods. In 2013, Amazon carried 230 million items for sale in the United States, nearly 30 times the number sold by Walmart, one of the largest retailers in the world. Chart 1E-Commerce: Steady Increase In Market Share E-Commerce: Steady Increase In Market Share E-Commerce: Steady Increase In Market Share Chart 2Amazon Dominates Did Amazon Kill The Phillips Curve? Did Amazon Kill The Phillips Curve? With the use of a smartphone, consumers can check the price of an item on Amazon while shopping in a physical store. Studies show that it does not require a large price gap for shoppers to buy online rather than in-store. Amazon appears to be impacting other retailers' ability to pass though cost increases, leading to a rash of retail outlet closings. Sears alone announced the closure of 300 retail outlets this year. The devastation that Amazon inflicted on the book industry is well known. It is no wonder then, that Amazon's purchase of Whole Foods Market, a grocery chain, sent shivers down the spines of CEOs not only in the food industry, but in the broader retail industry as well. What would prevent Amazon from applying its model to furniture and appliances, electronics or drugstores? It seems that no retail space is safe. A Little Theory Before we turn to the evidence, let's review the macro theory related to positive supply shocks. The internet could be lowering prices by moving product markets toward the "perfect competition" model. The internet trims search costs, improves price transparency and reduces barriers to entry. The internet also allows for shorter supply chains, as layers of wholesalers and other intermediaries are removed and e-commerce companies allow more direct contact between consumers and producers. Fewer inventories and a smaller "brick and mortar" infrastructure take additional costs out of the system. Economic theory suggests that the result of this positive supply shock will be greater product market competition, increased productivity and reduced profitability. In the long run, workers should benefit from the productivity boost via real wage gains (even if nominal wage growth is lackluster). Workers may lower their reservation wage if they feel that increased competitive pressures or technology threaten their jobs. The internet is also likely to improve job matching between the unemployed and available vacancies, which should lead to a fall in the full-employment level of unemployment (NAIRU). Nonetheless, the internet should not have a permanent impact on inflation. The lower level of NAIRU and the direct effects of the internet on consumer prices discussed above allow inflation to fall below the central bank's target. The bank responds by lowering interest rates, stimulating demand and thereby driving unemployment down to the new lower level of NAIRU. Over time, inflation will drift back up toward target. In other words, a greater degree of the competition should boost the supply side of the economy and lower NAIRU, but it should not result in a permanently lower rate of inflation if inflation is indeed a monetary phenomenon and central banks strive to meet their targets. Still, one could imagine a series of supply shocks that are spread out over time, with each having a temporary negative impact on prices such that it appears for a while that inflation has been permanently depressed. This could be an accurate description of the current situation in the U.S. and some of the other major countries. We have sympathy for the view that the internet and new business models are increasing competition, cutting costs and thereby limiting price increases in some areas. But is there any hard evidence? Is the competitive effect that large, and is it any more intense than in the past? There are a number of reasons to be skeptical because most of the evidence does not support Forbes' claim that the internet has killed inflation. 1. E-commerce affects only a small part of the Consumer Price Index As mentioned above, online shopping for goods represents 8.5% of total retail sales in the U.S. E-commerce is concentrated in four kinds of businesses (Table 1): Furniture & Home Furnishings (7% of total retail sales), Electronics & Appliances (20%), Health & Personal Care (15%), and Clothing (10%). Since goods make up 40% of the CPI, then 3.2% (8% times 40%) is a ballpark estimate for the size of goods e-commerce in the CPI. Table 1E-Commerce Market Share Of Goods Sector Did Amazon Kill The Phillips Curve? Did Amazon Kill The Phillips Curve? Table 2 shows the relative size of e-commerce in the service sector. The analysis is complicated by the fact that the data on services includes B-to-B sales in addition to B-to-C.2 However, e-commerce represents almost 4% of total sales for the service categories tracked by the BLS. Services make up 60% of the CPI, but the size drops to 26% if we exclude shelter (which is probably not affected by online shopping). Thus, e-commerce in the service sector likely affects 1% (3.9% times 26%) of the CPI. Table 2E-Commerce Market Share Of Service Sector Did Amazon Kill The Phillips Curve? Did Amazon Kill The Phillips Curve? Adding goods and services, online shopping affects about 4.2% of the CPI index at most. The bottom line is that the relatively small size of e-commerce at the consumer level limits any estimate of the impact of online sales on the broad inflation rate. 2. Most of the deceleration in inflation since 2007 has been in areas unaffected by e-commerce Table 3 compares the average contribution to annual average CPI inflation during 2000-2007 with that of 2007-2016. Average annual inflation fell from 2.9% in the seven years before the Great Recession to 1.8% after, for a total decline of just over 1 percentage point. The deceleration is almost fully explained by Energy, Food and Owners' Equivalent Rent. The bottom part of Table 3 highlights that the sectors with the greatest exposure to e-commerce had a negligible impact on the inflation slowdown. Table 3Comparison Of Pre- And Post-Lehman Inflation Rates Did Amazon Kill The Phillips Curve? Did Amazon Kill The Phillips Curve? 3. The cost advantages for online sellers are overstated Bain & Company, a U.S. consultancy, argues that e-commerce will not grow in importance indefinitely and come to dominate consumer spending.3 E-commerce sales are already slowing. Market share is following a classic S-shaped curve that, Bain estimates, will top out at under 30% by 2030. First, not everyone wants to buy everything online. Products that are well known to consumers and purchased on a regular basis are well suited to online shopping. But for many other products, consumers need to see and feel the product in person before making a purchase. Second, the cost savings of online selling versus traditional brick and mortar stores is not as great as many believe. Bain claims that many e-commerce businesses struggle to make a profit. The information technology, distribution centers, shipping, and returns processing required by e-commerce companies can cost as much as running physical stores in some cases. E-tailers often cannot ship directly from manufacturers to consumers; they need large and expensive fulfillment centers and a very generous returns policy. Moreover, online and offline sales models are becoming blurred. Retailers with physical stores are growing their e-commerce operations, while previously pure e-commerce plays are adding stores or negotiating space in other retailers' stores. Even Amazon now has storefronts. The shift toward an "multichannel" selling model underscores that there are benefits to traditional brick-and-mortar stores that will ensure that they will not completely disappear. 4. E-commerce is not the first revolution in the retail sector The retail sector has changed significantly over the decades and it is not clear that the disinflationary effect of the latest revolution, e-commerce, is any more intense than in the past. Economists at Goldman Sachs point out that the growth of Amazon's market share in recent years still lags that of Walmart and other "big box" stores in the 1990s (Chart 3).4 This fact suggests that "Amazonification" may not be as disinflationary as the previous big-box revolution. 5. Weak productivity growth and high profit margins are inconsistent with a large supply-side benefit from e-commerce As discussed above, economic theory suggests that a positive supply shock that cuts costs and boosts competition should trim profit margins and lift productivity. The problem is that the margins and productivity have moved in the opposite direction that economic theory would suggest (Chart 4). Chart 3Comparison Of Pre- And Post-Lehman Inflation Rates Did Amazon Kill The Phillips Curve? Did Amazon Kill The Phillips Curve? Chart 4Incompatible With A Supply Shock Incompatible With A Supply Shock Incompatible With A Supply Shock By definition, productivity rises when firms can produce the same output with fewer or cheaper inputs. However, it is well documented that productivity growth has been in a downtrend since the 1990s, and has been dismally low since the Great Recession. A Special Report from BCA's Global Investment Strategy 5 service makes a convincing case that mismeasurement is not behind the low productivity figures. In fact, in many industries it appears that productivity is over-estimated. If e-commerce is big enough to "move the dial" on overall inflation, it should be big enough to see in the aggregate productivity figures. Chart 5Retail Margin Squeeze Only In Department Stores Retail Margin Squeeze Only In Department Stores Retail Margin Squeeze Only In Department Stores One would also expect to see a margin squeeze across industries if e-commerce is indeed generating a lot of deflationary competitive pressure. Despite dismally depressed productivity, however, corporate profit margins are at the high end of the historical range across most of the sectors of the S&P 500. This is the case even in the retailing sector outside of department stores (Chart 5). These facts argue against the idea that the internet has moved the economy further toward a disinflationary "perfect competition" model. 6. Online price setting is characterized by frictions comparable to traditional retail We would expect to observe a low price dispersion across online vendors since the internet has apparently lowered the cost of monitoring competitors' prices and the cost of searching for the lowest price. We would also expect to see fairly synchronized price adjustments; if one vendor adjusts its price due to changing market conditions, then the rest should quickly follow to avoid suffering a massive loss of market share. However, a recent study of price-setting practices in the U.S. and U.K. found that this is not the case.6 The dataset covered a broad spectrum of consumer goods and sellers over a two-year period, comparing online with offline prices. The researchers found that market pricing "frictions" are surprisingly elevated in the online world. Price dispersion is high in absolute terms and on par with offline pricing. Academics for years have puzzled over high price rigidities and dispersion in retail stores in the context of an apparently stiff competitive environment, and it appears that online pricing is not much better. The study did not cover a long enough period to see if frictions were even worse in the past. Nonetheless, the evidence available suggests that the lower cost of monitoring prices afforded by the internet has not led to significant price convergence across sellers online or offline. Another study compared online and offline prices for multichannel retailers, using the massive database provided by the Billion Prices Project at MIT.7 The database covers prices across 10 countries. The study found that retailers charged the same price online as in-store in 72% of cases. The average discount was 4% for those cases in which there was a markdown online. If the observations with identical prices are included, the average online/offline price difference was just 1%. 7. Some measures of online prices have grown at about the same pace as the CPI index The U.S. Bureau of Labor Statistics does include online sales when constructing the Consumer Price Index. It even includes peer-to-peer sales by companies such as Airbnb and Uber. However, the BLS admits that its sample lags the popularity of such services by a few years. Moreover, while the BLS is trying to capture the rising proportion of sales done via e-commerce, "outlet bias" means that the CPI does not capture the price effect in cases where consumers are finding cheaper prices online. This is because the BLS weights the growth rate of online and offline prices, not the price levels. While there may be level differences, there is no reason to believe that the inflation rates for similar goods sold online and offline differ significantly. If the inflation rates are close, then the growing share of online sales will not affect overall inflation based on the BLS methodology. The BLS argues that any bias in the CPI due to outlet bias is mitigated to the extent that physical stores offer a higher level of service. Thus, price differences may not be that great after quality-adjustment. All this suggests that the actual consumer price inflation rate could be somewhat lower than the official rate. Nonetheless, it does not necessarily mean that inflation, properly measured, is being depressed by e-commerce to a meaningful extent. Indeed, Chart 6 highlights that the U.S. component of the Billion Prices Index rose at a faster pace than the overall CPI between 2009 and 2014. The Online Price Index fell in absolute and relative terms from 2014 to mid-2016, but rose sharply toward the end of 2016. Applying our guesstimate of the weight of e-commerce in the CPI (3.2% for goods), online price inflation added to overall annual CPI inflation by about 0.3 percentage points in 2016 (bottom panel of Chart 6). There is more deflation evident in the BLS' index of prices for Electronic Shopping and Mail Order Houses (Chart 7). Online prices fell relative to the overall CPI for most of the time since the early 1990s, with the relative price decline accelerating since the GFC. However, our estimate of the contribution to overall annual CPI inflation is only about -0.15 percentage points in June 2017, and has never been more than -0.3 percentage points. This could be an underestimate because it does not include the impact of services, although the service e-commerce share of the CPI is very small. Chart 6Online Price Index Online Price Index Online Price Index Chart 7Electronic Shopping Price Index Electronic Shopping Price Index Electronic Shopping Price Index Another way to approach this question is to focus on the parts of the CPI that are most exposed to e-commerce. It is impossible to separate the effect of e-commerce on inflation from other drivers of productivity. Nonetheless, if online shopping is having a significant deflationary impact on overall inflation, we should see large and persistent negative contributions from these parts of the CPI. We combined the components of the CPI that most closely matched the sectors that have high e-commerce exposure according to the BLS' annual Retail Survey (Chart 8). The sectors in our aggregate e-commerce price proxy include hotels/motels, taxicabs, books & magazines, clothing, computer hardware, drugs, health & beauty aids, electronics & appliances, alcoholic beverages, furniture & home furnishings, sporting goods, air transportation, travel arrangement and reservation services, educational services and other merchandise. The sectors are weighted based on their respective weights in the CPI. Our e-commerce price proxy has generally fallen relative to the overall CPI index since 2000. However, while the average contribution of these sectors to the overall annual CPI inflation rate has fallen in the post GFC period relative to the 2000-2007 period, the average difference is only 0.2 percentage points. The contribution has hovered around the zero mark for the past 2½ years. Surprisingly, price indexes have increased by more than the overall CPI since 2000 in some sectors where one would have expected to see significant relative price deflation, such as taxis, hotels, travel arrangement and even books. One could argue that significant measurement error must be a factor. How could the price of books have gone up faster than the CPI? Sectors displaying the most relative price declines are clothing, computers, electronics, furniture, sporting goods, air travel and other goods. We recalculated our e-commerce proxy using only these deflating sectors, but we boosted their weights such that the overall weight of the proxy in the CPI is kept the same as our full e-commerce proxy discussed above. In other words, this approach implicitly assumes that the excluded sectors (taxis, books, hotels and travel arrangement) actually deflated at the average pace of the sectors that remain in the index. Our adjusted e-commerce proxy suggests that online pricing reduced overall CPI inflation by about 0.1-to-0.2 percentage points in recent years (Chart 9). This contribution is below the long-term average of the series, but the drag was even greater several times in the past. Chart 8BCA E-Commerce Proxy Price Index BCA E-Commerce Proxy Price Index BCA E-Commerce Proxy Price Index Chart 9BCA E-Commerce Adjusted Proxy Price Index BCA E-Commerce Adjusted Proxy Price Index BCA E-Commerce Adjusted Proxy Price Index Admittedly, data limitations mean that all of the above estimates of the impact of e-commerce are ballpark figures. Conclusions We are keeping an open mind and reserving judgement on the disinflationary impact of robotics, artificial intelligence and the gig economy until we do more research. But in terms of the impact of e-commerce, it is difficult to find supportive evidence. The available data are admittedly far from ideal for confirming or disproving the "Amazonification" thesis. Perhaps better measures of e-commerce pricing will emerge in the future. Nonetheless, the measures available today do not suggest that online sales are depressing the overall inflation rate by more than 0.1 or 0.2 percentage points, and it does not appear that the disinflationary impact has intensified by much. One could argue that lower online prices are forcing traditional retailers to match the e-commerce vendors, allowing for a larger disinflationary effect than we estimate. Nonetheless, if this were the case, then we would expect to see significant margin compression in the retail sector. The sectors potentially affected by e-commerce make up a small part of the CPI index. The deceleration of inflation since the GFC has been in areas unaffected by online sales. High corporate profit margins and depressed productivity growth also argue against the idea that e-commerce represents a large positive macro supply shock. Finally, today's creative destruction in retail may be no more deflationary than the shift to 'big box' stores in the 1990s. Perhaps the main way that e-commerce is affecting the macro economy and financial markets is not through inflation, but via the reduction in the economy's capital spending requirement. Rising online activity means that we need fewer shopping malls and big box outlets to support a given level of consumer spending. This would reduce the equilibrium level of interest rates, since the Fed has to stimulate other parts of the economy to offset the loss of demand in capital spending in the retail sector. To the extent that central banks were slow to recognize that equilibrium rates had fallen to extremely low levels, then policy was behind the curve and this might have contributed to the current low inflation environment. Mark McClellan, Senior Vice President The Bank Credit Analyst markm@bcaresearch.com 1 Robert F. DeLucia, "Economic Perspective: A Nontraditional Analysis of Inflation," Prudential Capital Group (August 21, 2017). 2 Business to business, and business to consumer. 3 Aaron Cheris, Darrell Rigby and Suzanne Tager, "The Power Of Omnichannel Stores," Bain & Company Insights: Retail Holiday Newsletter 2016-2017 (December 19, 2016) 4 "US Daily: The Internet and Inflation: How Big is the Amazon Effect?" Goldman Sachs Economic Research (August 2, 2017). 5 Please see Global Investment Strategy Weekly Report, "Weak Productivity Growth: Don't Blame the Statisticians," dated March 25, 2016, available at gis.bcaresearch.com 6 Yuriy Gorodnichenko, Viacheslav Sheremirov, and Oleksandr Talavera, "Price Setting In Online Markets: Does IT Click?" Journal of the European Economic Association (July 2016). 7 Alberto Cavallo, "Are Online and Offline Prices Similar? Evidence from Large Multi-Channel Retailers," NBER Working Paper No. 22142 (March 2016).

Highlights Financial markets have slipped into a 'risk off' phase. The upbeat second quarter earnings season in the U.S., Japan and the Eurozone was overwhelmed by a number of negative events. Equity bear markets are usually associated with recessions. On that score, we do not see any warning signs of an economic downturn. However, geopolitical risks are rising at a time when valuation measures suggest that risk assets are vulnerable. We do not see the debt ceiling or the failure of movement on U.S. tax reform as posing large risks for financial markets. However, trade protectionism and, especially, North Korea are major wildcards. We don't believe the tensions in the Korean peninsula will end the cyclical bull market in global equities. Nonetheless, investors should expect to be tested numerous times over the next year to 18 months. BCA Strategists debated trimming equity exposure to neutral. However, the majority felt that, while there will be near-term volatility, the main equity indexes are likely to be higher on a 6-12 month horizon. Riding out the volatility is a better approach than trying to time the short-term ups and downs. That said, it appears prudent to be well shy of max overweight positions and to hold some safe haven assets within diversified portfolios. On a positive note, we have upgraded our EPS growth forecasts, except in the Eurozone where currency strength will be a significant drag in the near term. The Fed faced a similar low inflation/tight labor market environment in 1999. Policymakers acted pre-emptively and began to tighten before inflation turned up. This time, the FOMC will want to see at least a small increase in inflation just to be sure. Wages may be a lagging indicator for inflation in this cycle. Watch a handful of other indicators we identify that led inflection points in inflation in previous long economic expansions. This year's euro strength is unlikely to delay the next installment of ECB tapering, which we expect in early in 2018. Investors seem to be taking an "I'll believe it when I see it" attitude toward the U.S. inflation outlook, which has led to very lopsided rate expectations. Keep duration short. Feature Chart I-1Trump Popularity Headwind For Tax Reform September 2017 September 2017 A 'risk off' flavor swept over financial markets in August. The upbeat second quarter earnings season in the U.S., Japan and the Eurozone was overwhelmed by a number of negative events, from President Trump's Charlottesville controversy to the never-ending staff changes in the White House to North Korean tensions to the Texas flood and the terror attack in Spain. Trump's popularity rating is steadily declining, even now among Republican voters (Chart I-1). This has raised concerns that none of his business-friendly policies, tax cuts or initiatives to boost growth will be successfully enacted. It is even possible that the debt ceiling will be used as a bargaining chip among the various Republican factions. The political risks are multiplying at a time when the equity and corporate bond markets are pricey. Valuation measures do not help with timing, but they do inform on the potential downside risk if things head south. At the moment, we do not see any single risk as justifying a full retreat into safe havens and a cut in risk asset allocation to neutral or below. Nonetheless, there is certainly a case to be cautious and hold some traditional safe haven assets. Timing The Next Equity Bear Market It is rare to have an equity bear market without a recession in the U.S. There have been plenty of market setbacks that did not quite meet the 20% bear-market threshold, but were nonetheless painful even in the absence of recession (Black Monday, LTCM crisis, U.S. debt ceiling showdown and euro crises). Unfortunately, these corrections are very difficult to predict. At least with recessions, investors have a fighting chance in timing the exit from risk exposure. The slope of the yield curve and the Leading Economic Indicator (LEI) are classic recession indicators, and for good reason (Chart I-2). Over the past 50 years they have both successfully called all seven recessions with just one false positive. We can eliminate the false positive signals by combining the two indicators and follow a rule that both must be in the red to herald a recession.1 Chart I-2The Traditional Recession Indicators Have Worked Well The Traditional Recession Indicators Have Worked Well The Traditional Recession Indicators Have Worked Well It will be almost impossible for the yield curve to invert until the fed funds rate is significantly higher than it is today. Thus, it may be the case that a negative reading on the LEI, together with a flattening (but not yet inverted) yield curve, will be a powerful signal that a recession is on the way. Neither of these two indicators are warning of a recession. Global PMIs are hovering at a level that is consistent with robust growth. The erosion in the Global ZEW and the drop in the diffusion index of the Global LEI are worrying signs, but at the moment are consistent with a growth slowdown at worst (Chart I-3). Financial conditions remain growth-friendly and subdued inflation is allowing central banks to proceed cautiously when tightening (in the case of the Fed and Bank of Canada) or tapering (ECB). As highlighted in last month's Overview, the global economy has entered a synchronized upturn that should persist for the next year. The U.S. will be the first major economy to enter the next recession, but that should not occur until 2019 or 2020, barring any shocks in the near term. That said, risk asset prices have been bid up sharply and are therefore vulnerable to a correction. Below, we discuss five key risks to the equity bull market. (1) Is All Lost For U.S. Tax Cuts? Our recent client meetings highlight that investors are skeptical that any fiscal stimulus or tax cuts will see the light of day in the U.S. Tax cuts and infrastructure spending appear to have been priced out of the equity market, according to the index ratios shown in Chart I-4. We still expect a modest package to eventually be passed, although time is running out for this year. Tax reform is a major component of Trump's and congressional Republicans' agenda. If it fails, Republicans will have to go to their home districts empty-handed to campaign for the November 2018 midterm elections. Chart I-3Some Worrying Signs On Growth Some Worrying Signs On Growth Some Worrying Signs On Growth Chart I-4Fiscal Stimulus Largely Priced Out Fiscal Stimulus Largely Priced Out Fiscal Stimulus Largely Priced Out One implication of Tropical Storm Harvey is that it might force Democrats and Republicans to cooperate on an infrastructure bill for rebuilding. Even a modest spending boost or tax reduction would be equity-market positive given that so little is currently discounted. The dollar should also receive a lift, especially given that the Fed might respond to any fiscally-driven growth impulse with higher interest rates. (2) Who Will Lead The Fed? There is a significant chance that either Yellen will refuse to stay on when her term expires next February or that Trump will appoint someone else anyway. In this case, we would expect the President to do everything he can to ensure that the Fed retains its dovish bias. This means that he is likely to favor a non-economist and a loyal adviser, like Gary Cohn, over any of the more traditional, and hawkish, Republican candidates. Cohn could not arrive at the Fed and change the course of monetary policy on day one. The FOMC votes on rate changes, but in reality decisions are formed by consensus (with one or two dissents). The only way Cohn could implement an abrupt change in policy is if the Administration stacks the Fed Governors with appointees that are prepared to "toe the line" (the Administration does not appoint Regional Fed Presidents). Stacking the Governorships would take time. Nonetheless, it is not clear why President Trump would take a heavy hand in monetary policy when the current FOMC has been very cautious in tightening policy. The bottom line is that we would not see Cohn's appointment to the Fed Chair as signaling a major shift in monetary policy one way or the other. (3) The Debt Ceiling A more immediate threat is the debt ceiling. Recent fights over Obamacare and tax reform have pit fiscally conservative Republicans against the moderates, and it is possible that the debt ceiling is used as a bargaining chip in this battle. While government shutdowns have occurred in the past, the debt ceiling has never been breached. At the end of the day, the debt ceiling will always be raised because no government could stand the popular pressure that would result from social security checks not being mailed out to seniors or a halt to other entitlement programs. Even the Freedom Caucus, the most fiscally conservative grouping in the House, is considerably divided on the issue. This augurs well for a clean bill to raise the debt ceiling as the Republican majority in the House is 22 and the Freedom Caucus has 31 members. Democrats will not stand in the way of passage in the Senate. The worst-case scenario for the market would be a two-week shutdown in the first half of October, just before the debt ceiling is hit. We would not expect a shutdown to have any lasting impact on the economy, although it could provide an excuse for the equity market to correct. That said, the risk of even a shutdown has been diminished by events in Houston. It would be very difficult and damaging politically to shut down the government during a humanitarian emergency. (4) Trade And Protectionism The removal of White House Chief Strategist Stephen Bannon signals a shift in power toward the Goldman clique within the Trump Administration. National Economic Council President Gary Cohn, Treasury Secretary Steven Mnuchin, and Commerce Secretary Wilbur Ross are now firmly in charge of economic policy. The mainstream media has interpreted this shift within the Administration as reducing the risk of trade friction. We do not see it that way. President Trump still sounds hawkish on trade, particularly with respect to China. Our geopolitical experts point out that there are few constraints on the President to imposing trade sanctions on China or other countries. He could use such action to boost his popularity among his base heading into next year's midterm elections. On NAFTA, the Administration took a hard line as negotiations kicked off in August. This could be no more than a negotiating tactic. Our base case is that it will be some time before investors find out if negotiations are going off the rails. That said, the situation is volatile for both NAFTA and China, and we can't rule out a trade-related risk-off phase in financial markets over the next year. (5) North Korea North Korea's missile launch over Japan highlights that the tense situation is a long way from a resolution. The U.S. is unlikely to use military force to resolve the standoff. There are long-standing constraints to war, including the likelihood of a high death toll in Seoul. Moreover, China is unlikely to remain neutral in any conflict. However, the U.S. will attempt to establish a credible threat in order to contain Kim Jong-un. From an investor's perspective, it will be difficult to gauge whether the brinkmanship and military displays are simply posturing or evidence of real preparations for war.2 We don't believe the tensions in the Korean peninsula will end the cyclical bull market in global equities. Nonetheless, investors should expect to be tested numerous times over the next year to 18 months. Adding it all up, there is no shortage of things to keep investors awake at night. We would be de-risking our recommended portfolio were it not for the favorable earnings backdrop in the major advanced economies. Profit Outlook Update Chart I-5EPS Growth Outlook EPS Growth Outlook EPS Growth Outlook Second quarter earnings season came in even stronger than our upbeat models suggested in the U.S., Eurozone and Japan. This led to upward revisions to our EPS growth forecast, except in the Eurozone where currency strength will be a significant drag in the near term. The U.S. equity market enjoyed another quarter of margin expansion in Q2 2017 and the good news was broadly based. Earnings per share were higher versus Q2 2016 in all 11 sectors. Results were particularly strong in energy, technology and financials. Looking ahead, an update of our top-down model suggests the EPS growth will peak just under 20% late this year on a 4-quarter moving average basis, before falling to mid-single digits by the end of 2018 (Chart I-5). The peak is predicted to be a little higher than we previously forecast largely due to the feed-through of this year's pullback in the dollar. In Japan, a solid 70% of reporting firms beat estimates. Chart I-6 shows that Japan led all other major stock markets in positive earnings surprises in the second quarter. Manufacturing sectors, such as iron & steel, chemicals and machinery & electronics, were particularly impressive in the quarter, reflecting yen weakness and robust overseas demand. Japanese earnings are highly geared to the rebound in global industrial production. Moreover, Japan's nominal GDP growth accelerated in the second quarter and the latest PPI report suggested that corporate pricing power has improved. Twelve-month forward EPS estimates have risen to fresh all times highs, and have outperformed the U.S. in local currencies so far this year. Corporate governance reform - a key element of Abenomics - can take some credit for the good news on earnings. The share of companies with at least two independent directors rose from 18% in 2013 to 78% in 2016. The number of companies with performance-linked pay increased from 640 to 941, while the number that publish disclosure policies jumped from 679 to 1055. Analysts have been slow to factor in these positive developments. We expect trailing EPS growth to peak at about 25% in the first half of 2018 on a 4-quarter moving total basis, before edging lower by the end of the year. This is one reason why we like the Japanese market over the U.S. in local currency terms. Second quarter results in the Eurozone were solid, although not as impressive as in the U.S. and Japan. The 6% rise in the trade-weighted euro this year has resulted in a drop in the earnings revisions ratio into negative territory. Our previous forecast pointed to a continued rise in the 4-quarter moving average growth rate into the first half of 2018. However, we now expect the growth rate to dip by year end, before picking up somewhat next year. If the euro is flat from today's level, our model suggests that the drag on EPS growth will hover at 3-4 percentage points through the first half of next year as the negative impact feeds through (Chart I-7, bottom panel). Chart I-6Japan Led In Q2 Earning Surprises September 2017 September 2017 Chart I-7Currency Effects On Eurozone EPS Currency Effects On Eurozone EPS Currency Effects On Eurozone EPS Our top-down EPS model highlights that Eurozone earnings are quite sensitive to swings in the currency. In Chart I-7, we present alternative scenarios based on the euro weakening to EUR/USD 1.10 and strengthening to EUR/USD 1.30. For demonstration purposes we make the extreme assumption that the trade-weighted value of the euro rises and falls by the same amount in percentage terms. Profit growth decelerates by the end of 2017 in all three scenarios because of the lagged effect of currency swings. The projections begin to diverge only in 2018. EPS growth surges to around 20% by the end of next year in the euro-bear case, as the tailwind from the weakening currency combines with continuing robust economic growth. Conversely, trailing earnings growth hovers in the 5-8% range in the euro bull scenario, which is substantially less than we expect in the U.S. and Japan over the next year. EPS growth remains in positive territory because the assumed strength in European and global growth dominates the drag from the euro. The strong euro scenario would be negative for Eurozone equity relative performance versus global stocks in local currencies, although Europe might outperform on a common currency basis. The bottom line is that 12-month forward earnings estimates should remain in an uptrend in the three major economies. This means that, absent a negative political shock, the equity bull phase should resume in the coming months. Monetary policy is unlikely to spoil the party for risk assets, although the bond market is a source of risk because investors seem unprepared for even a modest rise in inflation. FOMC Has Seen This Before The Minutes from the July FOMC meeting highlighted that the key debate still centers on the relationship between labor market tightness and inflation, the timing of the next Fed rate hike and how policy should adjust to changing financial conditions. Chart I-8The FOMC Has Been Here Before The FOMC Has Been Here Before The FOMC Has Been Here Before The majority of policymakers are willing for now to believe that this year's soft inflation readings are driven largely by temporary 'one-off' factors. The hawks worry that a further undershoot of unemployment below estimates of full employment could suddenly generate a surge of inflation. They also point to the risk that low bond yields are promoting excess risk taking in financial markets. Moreover, the recent easing in financial conditions is stimulative and should be counterbalanced by additional Fed tightening. The hawks are thus anxious to resume tightening, despite current inflation readings. Others are worried that inflation softness could reflect structural factors, such as restraints on pricing power from global developments and from innovations to business models spurred by advances in technology. In this month's Special Report beginning on page 18, we have a close look at the impact of "Amazonification" in holding down overall inflation. We do not find the evidence regarding e-commerce compelling, but the jury is still out on the impact of other technologies. If robots and new business strategies are indeed weighing on inflation, it would mean that the Phillips curve is very flat or that the full employment level of unemployment is lower than the Fed estimates (or both). Either way, the doves would like to see the whites-of-the-eyes of inflation before resuming rate hikes. The last time the Fed was perplexed by a low level of inflation despite a tight labor market was in the late 1990s (Chart I-8). The FOMC cut rates following the LTCM financial crisis in late 1998, and then held the fed funds rate unchanged at 4¾% until June 1999. Core inflation was roughly flat during the on-hold period at 1% to 1½%, even as the unemployment rate steadily declined and various measures pointed to growing labor shortages. The FOMC 's internal debate in the first half of 1999 sounded very familiar. The minutes from meetings at that time noted that some policymakers pointed to the widespread inability of firms to raise prices because of strong competitive pressures in domestic and global markets. Some argued that significant cost saving efforts and new technologies also contributed to the low inflation environment for both consumer prices and wages. One difference from today is that productivity growth was solid at that time. The FOMC decided to hike rates in June 1999 by a quarter point, despite the absence of any clear indication that inflation had turned up. Policymakers described the tightening as "a small preemptive move... (that) would provide a degree of insurance against worsening inflation later". The Fed went on to lift the fed funds rate to 6½% by May 2000. Interestingly, the unemployment rate in June 1999 was 4.3%, exactly the same as the current rate. There are undoubtedly important differences in today's macro backdrop. The Fed is also more fearful of making a policy mistake in the aftermath of the Great Recession and financial crisis. Nonetheless, the point is that the Fed has faced a similar low inflation/tight labor market environment before, but in the end patience ran out and policymakers acted pre-emptively. Inflation Warning Signs During Long-Expansions We have noted in previous research that inflation pressures are slower to emerge in 'slow burn' recoveries, such as the 1980s and 1990s. In Chart I-9, we compare the core PCE inflation rate in the current cycle with the average of the previous two long expansion episodes (the inflection point for inflation in the previous cycles are aligned with June 2017 for comparison purposes). The other panels in the chart highlight that, in the 1980s and 1990s, wage growth was a lagging indicator. Economic commentators often assume that inflation is driven exclusively by "cost push" effects, such that the direction of causation runs from wage pressure to price pressure. However, causation runs in the other direction as well. Households see rising prices and then demand better wages to compensate for the added cost of living. This is not to say that we should totally disregard wage information. But it does mean that we must keep an eye on a wider set of data. Indicators that provided some leading information in the previous two long cycles are shown in Chart I-10. To this list we would also add the St. Louis Fed's Price Pressure index, which is not shown in Chart I-10 because it does not have enough history. At the moment, the headline PPI, ISM Prices Paid and BCA's pipeline inflation pressure index are all warning that inflation pressures are gradually building. However, this message is not confirmed by the St. Louis Fed's index and corporate selling prices. We are also watching the velocity of money, which has been a reasonably good leading indicator for U.S. inflation since 2000 (Chart I-11). Chart I-9In The 80s & 90s Wage Growth ##br##Gave No Early Warning On Inflation In The 80s & 90s Wage Growth Gave No Early Warning On Inflation In The 80s & 90s Wage Growth Gave No Early Warning On Inflation Chart I-10Leading Indicators Of Inflation ##br##In "Slow Burn" Recoveries Leading Indicators Of Inflation In "Slow Burn" Recoveries Leading Indicators Of Inflation In "Slow Burn" Recoveries Chart I-11Money Velocity And Inflation Money Velocity And Inflation Money Velocity And Inflation Our Fed view remains unchanged from last month; the FOMC will announce its balance sheet diet plan in September and the next rate hike will take place in December. Nonetheless, this forecast hangs on the assumption that core inflation edges higher in the coming months. Some indicators are pointing in that direction and recent dollar weakness will help. Wake Me When Inflation Picks Up Investors seem to be taking an "I'll believe it when I see it" attitude toward the U.S. inflation outlook. They also believe that persistent economic headwinds mean that monetary policy will need to stay highly accommodative for a very long time. Only one Fed rate hike is discounted between now and the end of 2018, and implied forward real short-term rates are negative until 2022. While we do not foresee surging inflation, the risks for market expectations appear quite lopsided. We expect one rate hike by year end, followed by at least another 50 basis points of tightening in 2018. The U.S. 10-year yield is also about almost 50 basis points below our short-term fair value estimate (Chart I-12). Moreover, over the medium- and long-term, reduced central bank bond purchases will impart gentle upward pressure on equilibrium bond yields. Twenty-eighteen will be the first time in four years in which the net supply of government bonds available to private investors will rise, taking the U.S., U.K., Eurozone and Japanese markets as a group. This year's euro strength is unlikely to delay the next installment of ECB tapering, which we expect in early in 2018. The currency appreciation will keep a lid on inflation in the near term. However, we see the euro's ascent as reflective of the booming economy, rather than a major headwind that will derail the growth story. Overall financial conditions have tightened this year, but only back to levels that persisted through 2016 (Chart I-13). Chart I-12U.S. 10-year Yield Is Below Fair Value U.S. 10-year Yield Is Below Fair Value U.S. 10-year Yield Is Below Fair Value Chart I-13Financial Conditions Financial Conditions Financial Conditions It will take clear signs that the economy is being negatively affected by currency strength for the ECB to back away from tapering. Indeed, the central bank has little choice because the bond buying program is approaching important technical limits. European corporate and peripheral bond spreads are likely to widen versus bunds as a result. The implication is that global yields have significant upside potential relative to forward rates, especially in the U.S. market. Duration should be kept short. JGBs are the only safe place to hide if global yields shift up because the Bank of Japan is a long way from abandoning its 10-year yield peg. Treasury yields should lead the way higher, which will finally place a bottom under the beleaguered dollar. Nonetheless, we are tactically at neutral on the greenback. Conclusions Chart I-14Gold Loves Geopolitical Crises September 2017 September 2017 In light of rising geopolitical risk, the BCA Strategists recently debated trimming equity exposure to neutral. Some argued that the risk/reward balance has deteriorated; the upside is limited by poor valuation, while there is significant downside potential if the North Korean situation deteriorates alarmingly. However, the majority felt that, while there will be near-term volatility, the main equity indexes are likely to be higher on a 6-12 month horizon. Riding out the volatility is a better approach than trying to time the short-term ups and downs. That said, it appears prudent to be well shy of max overweight positions and to hold some safe haven assets within diversified portfolios. BCA research has demonstrated that U.S. Treasurys, Swiss bonds and JGBs have been the best performers in times of crisis (Chart I-14).3 The same is true for the Swiss franc and the Japanese yen, such that the currency exposure should not be hedged in these cases. The dollar is more nuanced. It tends to perform well during financial crises, but not in geopolitical crises or recessions. Gold has tended to perform well in geopolitical events and recessions, although not in financial crises. We continue to prefer Japanese to U.S. stocks in local currency terms, given that EPS growth will likely peak in the U.S. first. Japanese stocks are also better valued. Europe is a tough call because this year's currency strength will weigh on earnings in the next quarter or two. However, the negative impact on earnings will reverse if the euro retraces as we expect. EM stocks have seen the strongest positive earnings revisions this year. We continue to worry about some of the structural headwinds facing emerging markets (high debt levels, poor governance, etc.). However, the cyclical picture remains more upbeat. Chinese H-shares remain our favorite EM market, trading at just 7.5 times 2017 earnings estimates. Our dollar and duration positions have been disappointing so far this year. Much hinges on U.S. inflation. Investors appear to have adopted the idea that structural headwinds to inflation will forever dominate the cyclical pressures. This means that the bond market is totally unprepared for any upside surprises on the inflation landscape. Admittedly, a rise in bond yields may not be imminent, but the risks appear to us to be predominantly to the upside. Lastly, crude oil inventories are shrinking as our commodity strategists predicted. They remain bullish, with a price target of USD60/bbl. Mark McClellan Senior Vice President The Bank Credit Analyst August 31, 2017 Next Report: September 28, 2017 1 Please see BCA Global ETF Strategy, "A Guide To Spotting And Weathering Bear Markets," dated August 16, 2017, available at etf.bcaresearch.com 2 Please see Geopolitical Strategy Weekly Report, "Can Pyongyang Derail The Bull Market?" dated August 16, 2017, available at gps.bcaresearch.com 3 Please see BCA Special Report, "Stairway To (Safe) Haven: Investing In Times Of Crisis," dated August 25, 2016, available at bca.bcaresearch.com II. Did Amazon Kill The Phillips Curve? A "culture of profound cost reduction" has gripped the business sector since the GFC according to one school of thought, permanently changing the relationship between labor market slack and wages or inflation. If true, it could mean that central banks are almost powerless to reach their inflation targets. Amazon, Airbnb, Uber, robotics, contract workers, artificial intelligence, horizontal drilling and driverless cars are just a few examples of companies and technologies that are cutting costs and depressing prices and wages. In the first of our series on inflation, we will focus on the rise of e-commerce and the related "Amazonification" of the economy. In theory, positive supply shocks should not have more than a temporary impact on inflation if the price level is indeed a monetary phenomenon in the long term. But a series of positive supply shocks could make it appear for quite a while that low inflation is structural in nature. We are keeping an open mind and reserving judgement on the disinflationary impact of robotics, artificial intelligence and the gig economy until we do more research. But in terms of the impact of e-commerce, it is difficult to find supportive evidence at the macro level. The admittedly inadequate measures of online prices available today do not suggest that e-commerce sales are depressing the overall inflation rate by more than 0.1 or 0.2 percentage points. Moreover, it does not appear that the disinflationary impact of competition in the retail sector has intensified over the years. Today's creative destruction in retail may be no more deflationary than the shift to 'big box' stores in the 1990s. Perhaps lower online prices are forcing traditional retailers to match the e-commerce vendors, allowing for a larger disinflationary effect than we estimate. However, the fact that retail margins are near secular highs outside of department stores argues against this thesis. The sectors potentially affected by e-commerce make up a small part of the CPI index. The deceleration of inflation since the GFC has been in areas unaffected by online sales. High profit margins for the overall corporate sector and depressed productivity growth also argue against the idea that e-commerce represents a large positive macro supply shock. Perhaps the main way that e-commerce is affecting the macro economy and financial markets is not through inflation, but via the reduction in the economy's capital spending requirement. This would reduce the equilibrium level of interest rates, since the Fed has to stimulate other parts of the economy to offset the loss of demand in capital spending in the retail sector. Anecdotal evidence is all around us. The global economy is evolving and it seems that all of the major changes are deflationary. Amazon, Airbnb, Uber, robotics, contract workers, artificial intelligence, horizontal drilling and driverless cars are just a few examples of companies and technologies that are cutting costs and depressing prices and wages. Central banks in the major advanced economies are having difficulty meeting their inflation targets, even in the U.S. where the labor market is tight by historical standards. Based on the depressed level of bond yields, it appears that the majority of investors believe that inflation headwinds will remain formidable for a long time. One school of thought is that low inflation reflects a lack of demand growth in the post-Great Financial Crisis (GFC) period. Another school points to the supply side of the economy. A recent report by Prudential Financial highlights "...obvious examples of ... new business models and new organizational structures, whereby higher-cost traditional methods of production, transportation, and distribution are displaced by more nontraditional cost-effective ways of conducting business."1 A "culture of profound cost reduction" has gripped the business sector since the GFC according to this school, permanently changing the relationship between labor market slack and wages or inflation (i.e., the Phillips Curve). Employees are less aggressive in their wage demands in a world where robots are threatening humans in a broadening array of industrial categories. Many feel lucky just to have a job. In a highly sensationalized article called "How The Internet Economy Killed Inflation," Forbes argued that "the internet has reduced many of the traditional barriers to entry that protect companies from competition and created a race to the bottom for prices in a number of categories." Forbes believes that new technologies are placing downward pressure on inflation by depressing wages, increasing productivity and encouraging competition. There are many factors that have the potential to weigh on prices, but analysts are mainly focusing on e-commerce, robotics, artificial intelligence, and the gig economy. In the first of our series on inflation, we will focus on the rise of e-commerce and the related "Amazonification" of the economy. The latter refers to the advent of new business models that cut out layers of middlemen between producers and consumers. Amazonification E-commerce has grown at a compound annual rate of more than 9% over the past 15 years, and now accounts for about 8½% of total U.S. retail sales (Chart II-1). Amazon has been leading the charge, accounting for 43% of all online sales in 2016 (Chart II-2). Amazon's business model not only cuts costs by eliminating middlemen and (until recently) avoiding expensive showrooms, but it also provides a platform for improved price discovery on an extremely broad array of goods. In 2013, Amazon carried 230 million items for sale in the United States, nearly 30 times the number sold by Walmart, one of the largest retailers in the world. Chart II-1E-Commerce: Steady Increase In Market Share E-Commerce: Steady Increase In Market Share E-Commerce: Steady Increase In Market Share Chart II-2Amazon Dominates September 2017 September 2017 With the use of a smartphone, consumers can check the price of an item on Amazon while shopping in a physical store. Studies show that it does not require a large price gap for shoppers to buy online rather than in-store. Amazon appears to be impacting other retailers' ability to pass though cost increases, leading to a rash of retail outlet closings. Sears alone announced the closure of 300 retail outlets this year. The devastation that Amazon inflicted on the book industry is well known. It is no wonder then, that Amazon's purchase of Whole Foods Market, a grocery chain, sent shivers down the spines of CEOs not only in the food industry, but in the broader retail industry as well. What would prevent Amazon from applying its model to furniture and appliances, electronics or drugstores? It seems that no retail space is safe. A Little Theory Before we turn to the evidence, let's review the macro theory related to positive supply shocks. The internet could be lowering prices by moving product markets toward the "perfect competition" model. The internet trims search costs, improves price transparency and reduces barriers to entry. The internet also allows for shorter supply chains, as layers of wholesalers and other intermediaries are removed and e-commerce companies allow more direct contact between consumers and producers. Fewer inventories and a smaller "brick and mortar" infrastructure take additional costs out of the system. Economic theory suggests that the result of this positive supply shock will be greater product market competition, increased productivity and reduced profitability. In the long run, workers should benefit from the productivity boost via real wage gains (even if nominal wage growth is lackluster). Workers may lower their reservation wage if they feel that increased competitive pressures or technology threaten their jobs. The internet is also likely to improve job matching between the unemployed and available vacancies, which should lead to a fall in the full-employment level of unemployment (NAIRU). Nonetheless, the internet should not have a permanent impact on inflation. The lower level of NAIRU and the direct effects of the internet on consumer prices discussed above allow inflation to fall below the central bank's target. The bank responds by lowering interest rates, stimulating demand and thereby driving unemployment down to the new lower level of NAIRU. Over time, inflation will drift back up toward target. In other words, a greater degree of the competition should boost the supply side of the economy and lower NAIRU, but it should not result in a permanently lower rate of inflation if inflation is indeed a monetary phenomenon and central banks strive to meet their targets. Still, one could imagine a series of supply shocks that are spread out over time, with each having a temporary negative impact on prices such that it appears for a while that inflation has been permanently depressed. This could be an accurate description of the current situation in the U.S. and some of the other major countries. We have sympathy for the view that the internet and new business models are increasing competition, cutting costs and thereby limiting price increases in some areas. But is there any hard evidence? Is the competitive effect that large, and is it any more intense than in the past? There are a number of reasons to be skeptical because most of the evidence does not support Forbes' claim that the internet has killed inflation. (1) E-commerce affects only a small part of the Consumer Price Index As mentioned above, online shopping for goods represents 8.5% of total retail sales in the U.S. E-commerce is concentrated in four kinds of businesses (Table II-1): Furniture & Home Furnishings (7% of total retail sales), Electronics & Appliances (20%), Health & Personal Care (15%), and Clothing (10%). Since goods make up 40% of the CPI, then 3.2% (8% times 40%) is a ballpark estimate for the size of goods e-commerce in the CPI. Table II-1E-Commerce Market Share Of Goods Sector (2015) September 2017 September 2017 Table II-2 shows the relative size of e-commerce in the service sector. The analysis is complicated by the fact that the data on services includes B-to-B sales in addition to B-to-C.2 However, e-commerce represents almost 4% of total sales for the service categories tracked by the BLS. Services make up 60% of the CPI, but the size drops to 26% if we exclude shelter (which is probably not affected by online shopping). Thus, e-commerce in the service sector likely affects 1% (3.9% times 26%) of the CPI. Table II-2E-Commerce Market Share Of Service Sector (2015) September 2017 September 2017 Adding goods and services, online shopping affects about 4.2% of the CPI index at most. The bottom line is that the relatively small size of e-commerce at the consumer level limits any estimate of the impact of online sales on the broad inflation rate. (2) Most of the deceleration in inflation since 2007 has been in areas unaffected by e-commerce Table II-3 compares the average contribution to annual average CPI inflation during 2000-2007 with that of 2007-2016. Average annual inflation fell from 2.9% in the seven years before the Great Recession to 1.8% after, for a total decline of just over 1 percentage point. The deceleration is almost fully explained by Energy, Food and Owners' Equivalent Rent. The bottom part of Table II-3 highlights that the sectors with the greatest exposure to e-commerce had a negligible impact on the inflation slowdown. Table II-3Comparison Of Pre- and Post-Lehman Inflation Rates September 2017 September 2017 (3) The cost advantages for online sellers are overstated Bain & Company, a U.S. consultancy, argues that e-commerce will not grow in importance indefinitely and come to dominate consumer spending.3 E-commerce sales are already slowing. Market share is following a classic S-shaped curve that, Bain estimates, will top out at under 30% by 2030. First, not everyone wants to buy everything online. Products that are well known to consumers and purchased on a regular basis are well suited to online shopping. But for many other products, consumers need to see and feel the product in person before making a purchase. Second, the cost savings of online selling versus traditional brick and mortar stores is not as great as many believe. Bain claims that many e-commerce businesses struggle to make a profit. The information technology, distribution centers, shipping, and returns processing required by e-commerce companies can cost as much as running physical stores in some cases. E-tailers often cannot ship directly from manufacturers to consumers; they need large and expensive fulfillment centers and a very generous returns policy. Moreover, online and offline sales models are becoming blurred. Retailers with physical stores are growing their e-commerce operations, while previously pure e-commerce plays are adding stores or negotiating space in other retailers' stores. Even Amazon now has storefronts. The shift toward an "multichannel" selling model underscores that there are benefits to traditional brick-and-mortar stores that will ensure that they will not completely disappear. (4) E-commerce is not the first revolution in the retail sector The retail sector has changed significantly over the decades and it is not clear that the disinflationary effect of the latest revolution, e-commerce, is any more intense than in the past. Economists at Goldman Sachs point out that the growth of Amazon's market share in recent years still lags that of Walmart and other "big box" stores in the 1990s (Chart II-3).4 This fact suggests that "Amazonification" may not be as disinflationary as the previous big-box revolution. (5) Weak productivity growth and high profit margins are inconsistent with a large supply-side benefit from e-commerce As discussed above, economic theory suggests that a positive supply shock that cuts costs and boosts competition should trim profit margins and lift productivity. The problem is that the margins and productivity have moved in the opposite direction that economic theory would suggest (Chart II-4). Chart II-3Amazon Vs. Walmart: ##br##Who's More Deflationary? September 2017 September 2017 Chart II-4Incompatible With A Supply Shock Incompatible With A Supply Shock Incompatible With A Supply Shock By definition, productivity rises when firms can produce the same output with fewer or cheaper inputs. However, it is well documented that productivity growth has been in a downtrend since the 1990s, and has been dismally low since the Great Recession. A Special Report from BCA's Global Investment Strategy5 service makes a convincing case that mismeasurement is not behind the low productivity figures. In fact, in many industries it appears that productivity is over-estimated. If e-commerce is big enough to "move the dial" on overall inflation, it should be big enough to see in the aggregate productivity figures. Chart II-5Retail Margin Squeeze ##br##Only In Department Stores Retail Margin Squeeze Only In Department Stores Retail Margin Squeeze Only In Department Stores One would also expect to see a margin squeeze across industries if e-commerce is indeed generating a lot of deflationary competitive pressure. Despite dismally depressed productivity, however, corporate profit margins are at the high end of the historical range across most of the sectors of the S&P 500. This is the case even in the retailing sector outside of department stores (Chart II-5). These facts argue against the idea that the internet has moved the economy further toward a disinflationary "perfect competition" model. (6) Online price setting is characterized by frictions comparable to traditional retail We would expect to observe a low price dispersion across online vendors since the internet has apparently lowered the cost of monitoring competitors' prices and the cost of searching for the lowest price. We would also expect to see fairly synchronized price adjustments; if one vendor adjusts its price due to changing market conditions, then the rest should quickly follow to avoid suffering a massive loss of market share. However, a recent study of price-setting practices in the U.S. and U.K. found that this is not the case.6 The dataset covered a broad spectrum of consumer goods and sellers over a two-year period, comparing online with offline prices. The researchers found that market pricing "frictions" are surprisingly elevated in the online world. Price dispersion is high in absolute terms and on par with offline pricing. Academics for years have puzzled over high price rigidities and dispersion in retail stores in the context of an apparently stiff competitive environment, and it appears that online pricing is not much better. The study did not cover a long enough period to see if frictions were even worse in the past. Nonetheless, the evidence available suggests that the lower cost of monitoring prices afforded by the internet has not led to significant price convergence across sellers online or offline. Another study compared online and offline prices for multichannel retailers, using the massive database provided by the Billion Prices Project at MIT.7 The database covers prices across 10 countries. The study found that retailers charged the same price online as in-store in 72% of cases. The average discount was 4% for those cases in which there was a markdown online. If the observations with identical prices are included, the average online/offline price difference was just 1%. (7) Some measures of online prices have grown at about the same pace as the CPI index The U.S. Bureau of Labor Statistics does include online sales when constructing the Consumer Price Index. It even includes peer-to-peer sales by companies such as Airbnb and Uber. However, the BLS admits that its sample lags the popularity of such services by a few years. Moreover, while the BLS is trying to capture the rising proportion of sales done via e-commerce, "outlet bias" means that the CPI does not capture the price effect in cases where consumers are finding cheaper prices online. This is because the BLS weights the growth rate of online and offline prices, not the price levels. While there may be level differences, there is no reason to believe that the inflation rates for similar goods sold online and offline differ significantly. If the inflation rates are close, then the growing share of online sales will not affect overall inflation based on the BLS methodology. The BLS argues that any bias in the CPI due to outlet bias is mitigated to the extent that physical stores offer a higher level of service. Thus, price differences may not be that great after quality-adjustment. All this suggests that the actual consumer price inflation rate could be somewhat lower than the official rate. Nonetheless, it does not necessarily mean that inflation, properly measured, is being depressed by e-commerce to a meaningful extent. Indeed, Chart II-6 highlights that the U.S. component of the Billion Prices Index rose at a faster pace than the overall CPI between 2009 and 2014. The Online Price Index fell in absolute and relative terms from 2014 to mid-2016, but rose sharply toward the end of 2016. Applying our guesstimate of the weight of e-commerce in the CPI (3.2% for goods), online price inflation added to overall annual CPI inflation by about 0.3 percentage points in 2016 (bottom panel of Chart II-6). There is more deflation evident in the BLS' index of prices for Electronic Shopping and Mail Order Houses (Chart II-7). Online prices fell relative to the overall CPI for most of the time since the early 1990s, with the relative price decline accelerating since the GFC. However, our estimate of the contribution to overall annual CPI inflation is only about -0.15 percentage points in June 2017, and has never been more than -0.3 percentage points. This could be an underestimate because it does not include the impact of services, although the service e-commerce share of the CPI is very small. Chart II-6Online Price Index Online Price Index Online Price Index Chart II-7Electronic Shopping Price Index Electronic Shopping Price Index Electronic Shopping Price Index Another way to approach this question is to focus on the parts of the CPI that are most exposed to e-commerce. It is impossible to separate the effect of e-commerce on inflation from other drivers of productivity. Nonetheless, if online shopping is having a significant deflationary impact on overall inflation, we should see large and persistent negative contributions from these parts of the CPI. We combined the components of the CPI that most closely matched the sectors that have high e-commerce exposure according to the BLS' annual Retail Survey (Chart II-8). The sectors in our aggregate e-commerce price proxy include hotels/motels, taxicabs, books & magazines, clothing, computer hardware, drugs, health & beauty aids, electronics & appliances, alcoholic beverages, furniture & home furnishings, sporting goods, air transportation, travel arrangement and reservation services, educational services and other merchandise. The sectors are weighted based on their respective weights in the CPI. Our e-commerce price proxy has generally fallen relative to the overall CPI index since 2000. However, while the average contribution of these sectors to the overall annual CPI inflation rate has fallen in the post GFC period relative to the 2000-2007 period, the average difference is only 0.2 percentage points. The contribution has hovered around the zero mark for the past 2½ years. Surprisingly, price indexes have increased by more than the overall CPI since 2000 in some sectors where one would have expected to see significant relative price deflation, such as taxis, hotels, travel arrangement and even books. One could argue that significant measurement error must be a factor. How could the price of books have gone up faster than the CPI? Sectors displaying the most relative price declines are clothing, computers, electronics, furniture, sporting goods, air travel and other goods. We recalculated our e-commerce proxy using only these deflating sectors, but we boosted their weights such that the overall weight of the proxy in the CPI is kept the same as our full e-commerce proxy discussed above. In other words, this approach implicitly assumes that the excluded sectors (taxis, books, hotels and travel arrangement) actually deflated at the average pace of the sectors that remain in the index. Our adjusted e-commerce proxy suggests that online pricing reduced overall CPI inflation by about 0.1-to-0.2 percentage points in recent years (Chart II-9). This contribution is below the long-term average of the series, but the drag was even greater several times in the past. Chart II-8BCA E-Commerce Proxy Price Index BCA E-Commerce Proxy Price Index BCA E-Commerce Proxy Price Index Chart II-9BCA E-Commerce Adjusted Proxy Price Index BCA E-Commerce Adjusted Proxy Price Index BCA E-Commerce Adjusted Proxy Price Index Admittedly, data limitations mean that all of the above estimates of the impact of e-commerce are ballpark figures. Conclusions We are keeping an open mind and reserving judgement on the disinflationary impact of robotics, artificial intelligence and the gig economy until we do more research. But in terms of the impact of e-commerce, it is difficult to find supportive evidence. The available data are admittedly far from ideal for confirming or disproving the "Amazonification" thesis. Perhaps better measures of e-commerce pricing will emerge in the future. Nonetheless, the measures available today do not suggest that online sales are depressing the overall inflation rate by more than 0.1 or 0.2 percentage points, and it does not appear that the disinflationary impact has intensified by much. One could argue that lower online prices are forcing traditional retailers to match the e-commerce vendors, allowing for a larger disinflationary effect than we estimate. Nonetheless, if this were the case, then we would expect to see significant margin compression in the retail sector. The sectors potentially affected by e-commerce make up a small part of the CPI index. The deceleration of inflation since the GFC has been in areas unaffected by online sales. High corporate profit margins and depressed productivity growth also argue against the idea that e-commerce represents a large positive macro supply shock. Finally, today's creative destruction in retail may be no more deflationary than the shift to 'big box' stores in the 1990s. Perhaps the main way that e-commerce is affecting the macro economy and financial markets is not through inflation, but via the reduction in the economy's capital spending requirement. Rising online activity means that we need fewer shopping malls and big box outlets to support a given level of consumer spending. This would reduce the equilibrium level of interest rates, since the Fed has to stimulate other parts of the economy to offset the loss of demand in capital spending in the retail sector. To the extent that central banks were slow to recognize that equilibrium rates had fallen to extremely low levels, then policy was behind the curve and this might have contributed to the current low inflation environment. Mark McClellan Senior Vice President The Bank Credit Analyst 1 Robert F. DeLucia, "Economic Perspective: A Nontraditional Analysis Of Inflation," Prudential Capital Group (August 21, 2017). 2 Business to business, and business to consumer. 3 Aaron Cheris, Darrell Rigby and Suzanne Tager, "The Power Of Omnichannel Stores," Bain & Company Insights: Retail Holiday Newsletter 2016-2017 (December 19, 2016). 4 "US Daily: The Internet And Inflation: How Big Is The Amazon Effect?" Goldman Sachs Economic Research (August 2, 2017). 5 Please see Global Investment Strategy Weekly Report, "Weak Productivity Growth: Don't Blame The Statisticians," dated March 25, 2016, available at gis.bcaresearch.com 6 Yuriy Gorodnichenko, Viacheslav Sheremirov, and Oleksandr Talavera, "Price Setting In Online Markets: Does IT Click?" Journal of the European Economic Association (July 2016). 7 Alberto Cavallo, "Are Online And Offline Prices Similar? Evidence From Large Multi-Channel Retailers," NBER Working Paper No. 22142 (March 2016). III. Indicators And Reference Charts Stocks struggled in August on the back of intensifying geopolitical risks, such that equity returns slipped versus bonds in the month. The earnings backdrop remains constructive for global stocks. In the U.S., 12-month forward EPS estimates continue to climb, in line with upbeat net revisions and earnings surprises. Nonetheless, the risk/reward balance has deteriorated due to escalating risks inside and outside of the U.S. Allocation to risk assets should still exceed benchmark, but should be shy of maximum settings. It is prudent to hold some of the traditional safe haven assets, including gold. Our new Revealed Preference Indicator (RPI) remained at 100% in August, sending a bullish message for equities. We introduced the RPI in the July report. Quite simply, it combines the idea of market momentum with valuation and policy measures. It provides a powerful bullish signal if positive market momentum lines up with constructive signals from the policy and valuation measures. Conversely, if constructive market momentum is not supported by valuation and policy, investors should lean against the market trend. Our Willingness-to-Pay (WTP) indicators are also bullish on stocks for the U.S., Europe and Japan. These indicators track flows, and thus provides information on what investors are actually doing, as opposed to sentiment indexes that track how investors are feeling. The U.S. WTP topped out in June and the same occurred in August for the Japan and the Eurozone indexes. While the indicators are still bullish, they highlight that flows into the equity markets in the major countries are beginning to moderate. These indicators would have to clearly turn lower to provide a bearish signal for stocks. The VIX increased last month, but remains depressed by historical standards. This implies that the equity market is vulnerable to bad news. However, investor sentiment is close to neutral and our speculation index has pulled back from previously elevated levels. These suggest that investors are not overly long at the moment. Our monetary indicator is only slightly negative, but the equity technical indicator is close to breaking below the 9-month moving average (a negative technical sign). Bond valuation continues to hover near fair value, according to our long-standing model that is based on a simple regression of the nominal 10-year yield on short-term real interest rates and a moving average of inflation. Another model, presented in the Overview section, estimates fair value based on dollar sentiment, a measure of policy uncertainty and the global PMI. This model suggests that the 10-year yield is almost 50 basis points on the expensive side. We think that Fed rate expectations are far too benign, suggesting that bond yields will rise. EQUITIES: Chart III-1U.S. Equity Indicators U.S. Equity Indicators U.S. Equity Indicators Chart III-2Willingness To Pay For Risk Willingness To Pay For Risk Willingness To Pay For Risk Chart III-3U.S. Equity Sentiment Indicators U.S. Equity Sentiment Indicators U.S. Equity Sentiment Indicators Chart III-4Revealed Preference Indicator Revealed Preference Indicator Revealed Preference Indicator Chart III-5U.S. Stock Market Valuation U.S. Stock Market Valuation U.S. Stock Market Valuation Chart III-6U.S. Earnings U.S. Earnings U.S. Earnings Chart III-7Global Stock Market And ##br##Earnings: Relative Performance Global Stock Market And Earnings: Relative Performance Global Stock Market And Earnings: Relative Performance Chart III-8Global Stock Market And ##br##Earnings: Relative Performance Global Stock Market And Earnings: Relative Performance Global Stock Market And Earnings: Relative Performance FIXED INCOME: Chart III-9U.S. Treasurys And Valuations U.S. Treasurys and Valuations U.S. Treasurys and Valuations Chart III-10U.S. Treasury Indicators U.S. Treasury Indicators U.S. Treasury Indicators Chart III-11Selected U.S. Bond Yields Selected U.S. Bond Yields Selected U.S. Bond Yields Chart III-1210-Year Treasury Yield Components 10-Year Treasury Yield Components 10-Year Treasury Yield Components Chart III-13U.S. Corporate Bonds And Health Monitor U.S. Corporate Bonds And Health Monitor U.S. Corporate Bonds And Health Monitor Chart III-14Global Bonds: Developed Markets Global Bonds: Developed Markets Global Bonds: Developed Markets Chart III-15Global Bonds: Emerging Markets Global Bonds: Emerging Markets Global Bonds: Emerging Markets CURRENCIES: Chart III-16U.S. Dollar And PPP U.S. Dollar And PPP U.S. Dollar And PPP Chart III-17U.S. Dollar And Indicator U.S. Dollar And Indicator U.S. Dollar And Indicator Chart III-18U.S. Dollar Fundamentals U.S. Dollar Fundamentals U.S. Dollar Fundamentals Chart III-19Japanese Yen Technicals Japanese Yen Technicals Japanese Yen Technicals Chart III-20Euro Technicals Euro Technicals Euro Technicals Chart III-21Euro/Yen Technicals Euro/Yen Technicals Euro/Yen Technicals Chart III-22Euro/Pound Technicals Euro/Pound Technicals Euro/Pound Technicals COMMODITIES: Chart III-23Broad Commodity Indicators Broad Commodity Indicators Broad Commodity Indicators Chart III-24Commodity Prices Commodity Prices Commodity Prices Chart III-25Commodity Prices Commodity Prices Commodity Prices Chart III-26Commodity Sentiment Commodity Sentiment Commodity Sentiment Chart III-27Speculative Positioning Speculative Positioning Speculative Positioning ECONOMY: Chart III-28U.S. And Global Macro Backdrop U.S. And Global Macro Backdrop U.S. And Global Macro Backdrop Chart III-29U.S. Macro Snapshot U.S. Macro Snapshot U.S. Macro Snapshot Chart III-30U.S. Growth Outlook U.S. Growth Outlook U.S. Growth Outlook Chart III-31U.S. Cyclical Spending U.S. Cyclical Spending U.S. Cyclical Spending Chart III-32U.S. Labor Market U.S. Labor Market U.S. Labor Market Chart III-33U.S. Consumption U.S. Consumption U.S. Consumption Chart III-34U.S. Housing U.S. Housing U.S. Housing Chart III-35U.S. Debt And Deleveraging U.S. Debt And Deleveraging U.S. Debt And Deleveraging Chart III-36U.S. Financial Conditions U.S. Financial Conditions U.S. Financial Conditions Chart III-37Global Economic Snapshot: Europe Global Economic Snapshot: Europe Global Economic Snapshot: Europe Chart III-38Global Economic Snapshot: China Global Economic Snapshot: China Global Economic Snapshot: China

A "culture of profound cost reduction" has gripped the business sector since the GFC according to one school of thought, permanently changing the relationship between labor market slack and wages or inflation. If true, it could mean that central banks are almost powerless to reach their inflation targets. Amazon, Airbnb, Uber, robotics, contract workers, artificial intelligence, horizontal drilling and driverless cars are just a few examples of companies and technologies that are cutting costs and depressing prices and wages. In the first of our series on inflation, we will focus on the rise of e-commerce and the related "Amazonification" of the economy. In theory, positive supply shocks should not have more than a temporary impact on inflation if the price level is indeed a monetary phenomenon in the long term. But a series of positive supply shocks could make it appear for quite a while that low inflation is structural in nature. We are keeping an open mind and reserving judgement on the disinflationary impact of robotics, artificial intelligence and the gig economy until we do more research. But in terms of the impact of e-commerce, it is difficult to find supportive evidence at the macro level. The admittedly inadequate measures of online prices available today do not suggest that e-commerce sales are depressing the overall inflation rate by more than 0.1 or 0.2 percentage points. Moreover, it does not appear that the disinflationary impact of competition in the retail sector has intensified over the years. Today's creative destruction in retail may be no more deflationary than the shift to 'big box' stores in the 1990s. Perhaps lower online prices are forcing traditional retailers to match the e-commerce vendors, allowing for a larger disinflationary effect than we estimate. However, the fact that retail margins are near secular highs outside of department stores argues against this thesis. The sectors potentially affected by e-commerce make up a small part of the CPI index. The deceleration of inflation since the GFC has been in areas unaffected by online sales. High profit margins for the overall corporate sector and depressed productivity growth also argue against the idea that e-commerce represents a large positive macro supply shock. Perhaps the main way that e-commerce is affecting the macro economy and financial markets is not through inflation, but via the reduction in the economy's capital spending requirement. This would reduce the equilibrium level of interest rates, since the Fed has to stimulate other parts of the economy to offset the loss of demand in capital spending in the retail sector. Anecdotal evidence is all around us. The global economy is evolving and it seems that all of the major changes are deflationary. Amazon, Airbnb, Uber, robotics, contract workers, artificial intelligence, horizontal drilling and driverless cars are just a few examples of companies and technologies that are cutting costs and depressing prices and wages. Central banks in the major advanced economies are having difficulty meeting their inflation targets, even in the U.S. where the labor market is tight by historical standards. Based on the depressed level of bond yields, it appears that the majority of investors believe that inflation headwinds will remain formidable for a long time. One school of thought is that low inflation reflects a lack of demand growth in the post-Great Financial Crisis (GFC) period. Another school points to the supply side of the economy. A recent report by Prudential Financial highlights "...obvious examples of ... new business models and new organizational structures, whereby higher-cost traditional methods of production, transportation, and distribution are displaced by more nontraditional cost-effective ways of conducting business."1 A "culture of profound cost reduction" has gripped the business sector since the GFC according to this school, permanently changing the relationship between labor market slack and wages or inflation (i.e., the Phillips Curve). Employees are less aggressive in their wage demands in a world where robots are threatening humans in a broadening array of industrial categories. Many feel lucky just to have a job. In a highly sensationalized article called "How The Internet Economy Killed Inflation," Forbes argued that "the internet has reduced many of the traditional barriers to entry that protect companies from competition and created a race to the bottom for prices in a number of categories." Forbes believes that new technologies are placing downward pressure on inflation by depressing wages, increasing productivity and encouraging competition. There are many factors that have the potential to weigh on prices, but analysts are mainly focusing on e-commerce, robotics, artificial intelligence, and the gig economy. In the first of our series on inflation, we will focus on the rise of e-commerce and the related "Amazonification" of the economy. The latter refers to the advent of new business models that cut out layers of middlemen between producers and consumers. Amazonification E-commerce has grown at a compound annual rate of more than 9% over the past 15 years, and now accounts for about 8½% of total U.S. retail sales (Chart II-1). Amazon has been leading the charge, accounting for 43% of all online sales in 2016 (Chart II-2). Amazon's business model not only cuts costs by eliminating middlemen and (until recently) avoiding expensive showrooms, but it also provides a platform for improved price discovery on an extremely broad array of goods. In 2013, Amazon carried 230 million items for sale in the United States, nearly 30 times the number sold by Walmart, one of the largest retailers in the world. Chart II-1E-Commerce: Steady Increase In Market Share E-Commerce: Steady Increase In Market Share E-Commerce: Steady Increase In Market Share Chart II-2Amazon Dominates September 2017 September 2017 With the use of a smartphone, consumers can check the price of an item on Amazon while shopping in a physical store. Studies show that it does not require a large price gap for shoppers to buy online rather than in-store. Amazon appears to be impacting other retailers' ability to pass though cost increases, leading to a rash of retail outlet closings. Sears alone announced the closure of 300 retail outlets this year. The devastation that Amazon inflicted on the book industry is well known. It is no wonder then, that Amazon's purchase of Whole Foods Market, a grocery chain, sent shivers down the spines of CEOs not only in the food industry, but in the broader retail industry as well. What would prevent Amazon from applying its model to furniture and appliances, electronics or drugstores? It seems that no retail space is safe. A Little Theory Before we turn to the evidence, let's review the macro theory related to positive supply shocks. The internet could be lowering prices by moving product markets toward the "perfect competition" model. The internet trims search costs, improves price transparency and reduces barriers to entry. The internet also allows for shorter supply chains, as layers of wholesalers and other intermediaries are removed and e-commerce companies allow more direct contact between consumers and producers. Fewer inventories and a smaller "brick and mortar" infrastructure take additional costs out of the system. Economic theory suggests that the result of this positive supply shock will be greater product market competition, increased productivity and reduced profitability. In the long run, workers should benefit from the productivity boost via real wage gains (even if nominal wage growth is lackluster). Workers may lower their reservation wage if they feel that increased competitive pressures or technology threaten their jobs. The internet is also likely to improve job matching between the unemployed and available vacancies, which should lead to a fall in the full-employment level of unemployment (NAIRU). Nonetheless, the internet should not have a permanent impact on inflation. The lower level of NAIRU and the direct effects of the internet on consumer prices discussed above allow inflation to fall below the central bank's target. The bank responds by lowering interest rates, stimulating demand and thereby driving unemployment down to the new lower level of NAIRU. Over time, inflation will drift back up toward target. In other words, a greater degree of the competition should boost the supply side of the economy and lower NAIRU, but it should not result in a permanently lower rate of inflation if inflation is indeed a monetary phenomenon and central banks strive to meet their targets. Still, one could imagine a series of supply shocks that are spread out over time, with each having a temporary negative impact on prices such that it appears for a while that inflation has been permanently depressed. This could be an accurate description of the current situation in the U.S. and some of the other major countries. We have sympathy for the view that the internet and new business models are increasing competition, cutting costs and thereby limiting price increases in some areas. But is there any hard evidence? Is the competitive effect that large, and is it any more intense than in the past? There are a number of reasons to be skeptical because most of the evidence does not support Forbes' claim that the internet has killed inflation. (1) E-commerce affects only a small part of the Consumer Price Index As mentioned above, online shopping for goods represents 8.5% of total retail sales in the U.S. E-commerce is concentrated in four kinds of businesses (Table II-1): Furniture & Home Furnishings (7% of total retail sales), Electronics & Appliances (20%), Health & Personal Care (15%), and Clothing (10%). Since goods make up 40% of the CPI, then 3.2% (8% times 40%) is a ballpark estimate for the size of goods e-commerce in the CPI. Table II-1E-Commerce Market Share Of Goods Sector (2015) September 2017 September 2017 Table II-2 shows the relative size of e-commerce in the service sector. The analysis is complicated by the fact that the data on services includes B-to-B sales in addition to B-to-C.2 However, e-commerce represents almost 4% of total sales for the service categories tracked by the BLS. Services make up 60% of the CPI, but the size drops to 26% if we exclude shelter (which is probably not affected by online shopping). Thus, e-commerce in the service sector likely affects 1% (3.9% times 26%) of the CPI. Table II-2E-Commerce Market Share Of Service Sector (2015) September 2017 September 2017 Adding goods and services, online shopping affects about 4.2% of the CPI index at most. The bottom line is that the relatively small size of e-commerce at the consumer level limits any estimate of the impact of online sales on the broad inflation rate. (2) Most of the deceleration in inflation since 2007 has been in areas unaffected by e-commerce Table II-3 compares the average contribution to annual average CPI inflation during 2000-2007 with that of 2007-2016. Average annual inflation fell from 2.9% in the seven years before the Great Recession to 1.8% after, for a total decline of just over 1 percentage point. The deceleration is almost fully explained by Energy, Food and Owners' Equivalent Rent. The bottom part of Table II-3 highlights that the sectors with the greatest exposure to e-commerce had a negligible impact on the inflation slowdown. Table II-3Comparison Of Pre- and Post-Lehman Inflation Rates September 2017 September 2017 (3) The cost advantages for online sellers are overstated Bain & Company, a U.S. consultancy, argues that e-commerce will not grow in importance indefinitely and come to dominate consumer spending.3 E-commerce sales are already slowing. Market share is following a classic S-shaped curve that, Bain estimates, will top out at under 30% by 2030. First, not everyone wants to buy everything online. Products that are well known to consumers and purchased on a regular basis are well suited to online shopping. But for many other products, consumers need to see and feel the product in person before making a purchase. Second, the cost savings of online selling versus traditional brick and mortar stores is not as great as many believe. Bain claims that many e-commerce businesses struggle to make a profit. The information technology, distribution centers, shipping, and returns processing required by e-commerce companies can cost as much as running physical stores in some cases. E-tailers often cannot ship directly from manufacturers to consumers; they need large and expensive fulfillment centers and a very generous returns policy. Moreover, online and offline sales models are becoming blurred. Retailers with physical stores are growing their e-commerce operations, while previously pure e-commerce plays are adding stores or negotiating space in other retailers' stores. Even Amazon now has storefronts. The shift toward an "multichannel" selling model underscores that there are benefits to traditional brick-and-mortar stores that will ensure that they will not completely disappear. (4) E-commerce is not the first revolution in the retail sector The retail sector has changed significantly over the decades and it is not clear that the disinflationary effect of the latest revolution, e-commerce, is any more intense than in the past. Economists at Goldman Sachs point out that the growth of Amazon's market share in recent years still lags that of Walmart and other "big box" stores in the 1990s (Chart II-3).4 This fact suggests that "Amazonification" may not be as disinflationary as the previous big-box revolution. (5) Weak productivity growth and high profit margins are inconsistent with a large supply-side benefit from e-commerce As discussed above, economic theory suggests that a positive supply shock that cuts costs and boosts competition should trim profit margins and lift productivity. The problem is that the margins and productivity have moved in the opposite direction that economic theory would suggest (Chart II-4). Chart II-3Amazon Vs. Walmart: ##br##Who's More Deflationary? September 2017 September 2017 Chart II-4Incompatible With A Supply Shock Incompatible With A Supply Shock Incompatible With A Supply Shock By definition, productivity rises when firms can produce the same output with fewer or cheaper inputs. However, it is well documented that productivity growth has been in a downtrend since the 1990s, and has been dismally low since the Great Recession. A Special Report from BCA's Global Investment Strategy5 service makes a convincing case that mismeasurement is not behind the low productivity figures. In fact, in many industries it appears that productivity is over-estimated. If e-commerce is big enough to "move the dial" on overall inflation, it should be big enough to see in the aggregate productivity figures. Chart II-5Retail Margin Squeeze ##br##Only In Department Stores Retail Margin Squeeze Only In Department Stores Retail Margin Squeeze Only In Department Stores One would also expect to see a margin squeeze across industries if e-commerce is indeed generating a lot of deflationary competitive pressure. Despite dismally depressed productivity, however, corporate profit margins are at the high end of the historical range across most of the sectors of the S&P 500. This is the case even in the retailing sector outside of department stores (Chart II-5). These facts argue against the idea that the internet has moved the economy further toward a disinflationary "perfect competition" model. (6) Online price setting is characterized by frictions comparable to traditional retail We would expect to observe a low price dispersion across online vendors since the internet has apparently lowered the cost of monitoring competitors' prices and the cost of searching for the lowest price. We would also expect to see fairly synchronized price adjustments; if one vendor adjusts its price due to changing market conditions, then the rest should quickly follow to avoid suffering a massive loss of market share. However, a recent study of price-setting practices in the U.S. and U.K. found that this is not the case.6 The dataset covered a broad spectrum of consumer goods and sellers over a two-year period, comparing online with offline prices. The researchers found that market pricing "frictions" are surprisingly elevated in the online world. Price dispersion is high in absolute terms and on par with offline pricing. Academics for years have puzzled over high price rigidities and dispersion in retail stores in the context of an apparently stiff competitive environment, and it appears that online pricing is not much better. The study did not cover a long enough period to see if frictions were even worse in the past. Nonetheless, the evidence available suggests that the lower cost of monitoring prices afforded by the internet has not led to significant price convergence across sellers online or offline. Another study compared online and offline prices for multichannel retailers, using the massive database provided by the Billion Prices Project at MIT.7 The database covers prices across 10 countries. The study found that retailers charged the same price online as in-store in 72% of cases. The average discount was 4% for those cases in which there was a markdown online. If the observations with identical prices are included, the average online/offline price difference was just 1%. (7) Some measures of online prices have grown at about the same pace as the CPI index The U.S. Bureau of Labor Statistics does include online sales when constructing the Consumer Price Index. It even includes peer-to-peer sales by companies such as Airbnb and Uber. However, the BLS admits that its sample lags the popularity of such services by a few years. Moreover, while the BLS is trying to capture the rising proportion of sales done via e-commerce, "outlet bias" means that the CPI does not capture the price effect in cases where consumers are finding cheaper prices online. This is because the BLS weights the growth rate of online and offline prices, not the price levels. While there may be level differences, there is no reason to believe that the inflation rates for similar goods sold online and offline differ significantly. If the inflation rates are close, then the growing share of online sales will not affect overall inflation based on the BLS methodology. The BLS argues that any bias in the CPI due to outlet bias is mitigated to the extent that physical stores offer a higher level of service. Thus, price differences may not be that great after quality-adjustment. All this suggests that the actual consumer price inflation rate could be somewhat lower than the official rate. Nonetheless, it does not necessarily mean that inflation, properly measured, is being depressed by e-commerce to a meaningful extent. Indeed, Chart II-6 highlights that the U.S. component of the Billion Prices Index rose at a faster pace than the overall CPI between 2009 and 2014. The Online Price Index fell in absolute and relative terms from 2014 to mid-2016, but rose sharply toward the end of 2016. Applying our guesstimate of the weight of e-commerce in the CPI (3.2% for goods), online price inflation added to overall annual CPI inflation by about 0.3 percentage points in 2016 (bottom panel of Chart II-6). There is more deflation evident in the BLS' index of prices for Electronic Shopping and Mail Order Houses (Chart II-7). Online prices fell relative to the overall CPI for most of the time since the early 1990s, with the relative price decline accelerating since the GFC. However, our estimate of the contribution to overall annual CPI inflation is only about -0.15 percentage points in June 2017, and has never been more than -0.3 percentage points. This could be an underestimate because it does not include the impact of services, although the service e-commerce share of the CPI is very small. Chart II-6Online Price Index Online Price Index Online Price Index Chart II-7Electronic Shopping Price Index Electronic Shopping Price Index Electronic Shopping Price Index Another way to approach this question is to focus on the parts of the CPI that are most exposed to e-commerce. It is impossible to separate the effect of e-commerce on inflation from other drivers of productivity. Nonetheless, if online shopping is having a significant deflationary impact on overall inflation, we should see large and persistent negative contributions from these parts of the CPI. We combined the components of the CPI that most closely matched the sectors that have high e-commerce exposure according to the BLS' annual Retail Survey (Chart II-8). The sectors in our aggregate e-commerce price proxy include hotels/motels, taxicabs, books & magazines, clothing, computer hardware, drugs, health & beauty aids, electronics & appliances, alcoholic beverages, furniture & home furnishings, sporting goods, air transportation, travel arrangement and reservation services, educational services and other merchandise. The sectors are weighted based on their respective weights in the CPI. Our e-commerce price proxy has generally fallen relative to the overall CPI index since 2000. However, while the average contribution of these sectors to the overall annual CPI inflation rate has fallen in the post GFC period relative to the 2000-2007 period, the average difference is only 0.2 percentage points. The contribution has hovered around the zero mark for the past 2½ years. Surprisingly, price indexes have increased by more than the overall CPI since 2000 in some sectors where one would have expected to see significant relative price deflation, such as taxis, hotels, travel arrangement and even books. One could argue that significant measurement error must be a factor. How could the price of books have gone up faster than the CPI? Sectors displaying the most relative price declines are clothing, computers, electronics, furniture, sporting goods, air travel and other goods. We recalculated our e-commerce proxy using only these deflating sectors, but we boosted their weights such that the overall weight of the proxy in the CPI is kept the same as our full e-commerce proxy discussed above. In other words, this approach implicitly assumes that the excluded sectors (taxis, books, hotels and travel arrangement) actually deflated at the average pace of the sectors that remain in the index. Our adjusted e-commerce proxy suggests that online pricing reduced overall CPI inflation by about 0.1-to-0.2 percentage points in recent years (Chart II-9). This contribution is below the long-term average of the series, but the drag was even greater several times in the past. Chart II-8BCA E-Commerce Proxy Price Index BCA E-Commerce Proxy Price Index BCA E-Commerce Proxy Price Index Chart II-9BCA E-Commerce Adjusted Proxy Price Index BCA E-Commerce Adjusted Proxy Price Index BCA E-Commerce Adjusted Proxy Price Index Admittedly, data limitations mean that all of the above estimates of the impact of e-commerce are ballpark figures. Conclusions We are keeping an open mind and reserving judgement on the disinflationary impact of robotics, artificial intelligence and the gig economy until we do more research. But in terms of the impact of e-commerce, it is difficult to find supportive evidence. The available data are admittedly far from ideal for confirming or disproving the "Amazonification" thesis. Perhaps better measures of e-commerce pricing will emerge in the future. Nonetheless, the measures available today do not suggest that online sales are depressing the overall inflation rate by more than 0.1 or 0.2 percentage points, and it does not appear that the disinflationary impact has intensified by much. One could argue that lower online prices are forcing traditional retailers to match the e-commerce vendors, allowing for a larger disinflationary effect than we estimate. Nonetheless, if this were the case, then we would expect to see significant margin compression in the retail sector. The sectors potentially affected by e-commerce make up a small part of the CPI index. The deceleration of inflation since the GFC has been in areas unaffected by online sales. High corporate profit margins and depressed productivity growth also argue against the idea that e-commerce represents a large positive macro supply shock. Finally, today's creative destruction in retail may be no more deflationary than the shift to 'big box' stores in the 1990s. Perhaps the main way that e-commerce is affecting the macro economy and financial markets is not through inflation, but via the reduction in the economy's capital spending requirement. Rising online activity means that we need fewer shopping malls and big box outlets to support a given level of consumer spending. This would reduce the equilibrium level of interest rates, since the Fed has to stimulate other parts of the economy to offset the loss of demand in capital spending in the retail sector. To the extent that central banks were slow to recognize that equilibrium rates had fallen to extremely low levels, then policy was behind the curve and this might have contributed to the current low inflation environment. Mark McClellan Senior Vice President The Bank Credit Analyst 1 Robert F. DeLucia, "Economic Perspective: A Nontraditional Analysis Of Inflation," Prudential Capital Group (August 21, 2017). 2 Business to business, and business to consumer. 3 Aaron Cheris, Darrell Rigby and Suzanne Tager, "The Power Of Omnichannel Stores," Bain & Company Insights: Retail Holiday Newsletter 2016-2017 (December 19, 2016). 4 "US Daily: The Internet And Inflation: How Big Is The Amazon Effect?" Goldman Sachs Economic Research (August 2, 2017). 5 Please see Global Investment Strategy Weekly Report, "Weak Productivity Growth: Don't Blame The Statisticians," dated March 25, 2016, available at gis.bcaresearch.com 6 Yuriy Gorodnichenko, Viacheslav Sheremirov, and Oleksandr Talavera, "Price Setting In Online Markets: Does IT Click?" Journal of the European Economic Association (July 2016). 7 Alberto Cavallo, "Are Online And Offline Prices Similar? Evidence From Large Multi-Channel Retailers," NBER Working Paper No. 22142 (March 2016).

Feature This is the second of three Special Reports on Electric Vehicles. In the first report published two weeks ago,1 we looked at the current costs of ownership of a typical mass-market EV, including and excluding subsidies, versus a similar Internal Combustion Engine Vehicle (ICEV). Based on current manufacturing costs and battery capabilities, EVs carry a significantly higher total cost per mile, even including current subsidies. In this second report, we determine that EV-specific manufacturers (specifically, TSLA) do not hold any material manufacturing advantage over conventional auto manufacturers, and lack their financial resources and intellectual experiences managing mass production operations. In addition to the risks from increased mass-market competition, the EV market faces risks of today's EV subsidies morphing into tomorrow's EV taxes, retarding the exponential growth of adoption many EV enthusiasts are betting on today. In our forthcoming third report, we will look at the potential regional and global impacts EV adoption will have on energy, power, and commodity markets. Despite the current cost and utility disadvantages of EVs, we expect governments (especially Europe and China) will continue to provide subsidies (carrots) and mandates (sticks) to further the adoption of EVs for the purposes of reducing CO2 emissions and tailpipe particulate pollution. The longer-term hope is that by forcing the EV market to expand, meaningful technological breakthroughs on batteries will eventually enable EVs to exceed ICEVs on a cost and utility basis. In this report, we conclude that: EV-specific manufacturers (TSLA) will face increasingly stiff competition from conventional auto manufacturers, who may enjoy lower manufacturing, distribution, and service costs and have ICEV profits to subsidize near-term EV losses. Access to chargers will be a growing problem for widespread EV adoption, especially for EVs to penetrate apartment-dwellers. Government EV subsidies will become fiscally difficult to continue as adoption increases and gasoline taxes are lost (especially in Europe). The small amount of carbon saved by EVs does not justify the subsidies, further increasing the risk subsidies are reduced or allowed to phase out (especially in the U.S.). EVs: Winners And Losers Investor interest in EVs tends to focus on the only publicly traded play in the space, Tesla Motors (TSLA, Q). Tesla has an enthusiastic fan base, which seems to extend well beyond the rather modest number of people who actually own the vehicles (Chart 1). That enthusiasm is probably somewhat responsible for favorable media coverage and the company's speculatively-high market cap (Chart 2), which is currently on a par with General Motors (GM, N), despite the fact that Tesla has never made a profit. (Chart 3 and Chart 4).When we read media and analyst coverage of Tesla, we often wonder if those writing the articles know anything about automobiles besides how to drive them. An example is this Forbes article regarding Tesla as uniquely visionary, building up a big lead on its sleepy competition. Chart 1Tesla's EV Sales Are Modest Tesla's EV Sales Are Modest Tesla's EV Sales Are Modest Chart 2Tesla's Market Cap Surpasses GM's Tesla's Market Cap Surpasses GM's Tesla's Market Cap Surpasses GM's Chart 3Tesla: Financial Performance TSLA: Financial Performance TSLA: Financial Performance Chart 4GM: Financial Performance GM: Financial Performance GM: Financial Performance "[Manufacturer] complacency about electric vehicle (EV) technology is worse than perceived. Despite more talk of developing EVs for mass-market adoption, a lack of real action and strategic commitments betray their underlying conviction, with no clear pathway to high-volume EV production before the mid-2020s"2 Setting aside for a moment the question as to whether Tesla, as a serial destroyer of capital (to date), will have access to the financial resources needed to become itself a "high-volume" producer of EVs, most commentators ignore the fact that building an EV is far less complicated than building an ICEV, and the conventional car companies are likely to have cost advantages (not to mention the benefits of decades of experience with mass production) once they do commit to the EV. What's The Difference Between An EV And An ICEV? In a general sense, an automobile consists of two main components: the drivetrain and the rest of the vehicle. What differentiates an EV from an ICEV is almost entirely the drivetrain and battery pack. Although the shape and weight of the battery pack requires some alteration to the body frame of the vehicle, and many EVs include regenerative brakes, substantially everything else in the rest of the EV is very similar. Drivetrain The drivetrain of an ICEV is where the vast majority of precision parts are located. A typical ICEV has hundreds of precision parts and must be manufactured and assembled to exact tolerances in order to last beyond the typically expected 100,000+ mile trouble-free life. Engines are also subject to extremes in temperatures ranging from -40°C (-40°F) at start up in a cold winter to close to 90°C (190°F) under operation. Transmissions are similarly complicated. In contrast, the drivetrain of an EV is extremely simple, consisting essentially of an electric motor and a transmission, which is also greatly simplified due to the nature of the torque curve of electric motors (Illustration 1). Illustration 1Key Components Of A Bolt EV Drive Unit Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact Unlike an ICEV which has numerous reciprocating parts (which are hard to engineer), all parts of an EV drivetrain rotate (which are much easier to engineer). Similarly, while there are numerous parts on an ICEV which require precision machining, friction bearings, and pressurized lubrication and cooling, analogous parts on an EV drivetrain are much fewer in number, can use ball bearings, and are lubricated for life. The fact that an EV drivetrain does not require pressurized lubrication and has a much simpler cooling system further simplifies the design and reduces the number of parts. It would not be an exaggeration to suggest that the drivetrain of an EV has an order of magnitude fewer parts than an ICEV of similar size. Any automotive company capable of designing and manufacturing an ICEV drivetrain should be capable of producing an EV drivetrain or outsourcing one if necessary. Battery Pack And Electronics Similarly, the battery pack of an EV is a mechanically simple thing to make. Battery cells are assembled into modules and the modules are assembled into the final battery pack (Illustration 2). The major challenge and potential differentiator is in the battery cells, which are effectively commodities (see below), and not in the manufacture or design of the battery pack. EV battery packs can produce a lot of heat when running or charging, and the battery packs tend to have simple cooling systems which vary from manufacturer to manufacturer.3 Illustration 2Battery Packs Are Battery Cells Assembled In Groups Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact An EV requires a significant amount of power electronics for the control of the motor, charging, and so on. Such power systems have been designed and made for decades, and, besides some unusual requirements due to the need to operate at extreme temperatures, there is no great technical challenge inherent in such systems. Indeed, while the operating life of an ICEV is typically on the order of 5,000 to 10,000 hours (100,000-200,000 miles), power electronics are often designed to operate for 100,000 hours or more. The drivetrain will not be the limiting factor on the longevity of an EV. Most likely, the cost of an EV's drivetrain (excluding the battery pack) and typical features such as regenerative brakes, a more robust suspension (due to the greater weight of the EV on account of the heavy battery), and accommodation for the battery pack, is somewhat less than that of an equivalent ICEV. Although the EV drivetrain is simpler to build, high-output electric motors and related control electronics are not cheap to manufacture due to the requirement for materials such as copper and exotic alloys. The reason for the substantially higher cost of EVs is the battery pack. And The Winners Are ... Despite investor enthusiasm for the "technological revolution" EVs represent, it is actually far more complicated and technologically difficult to design and manufacture an ICEV than an EV. The EV has far fewer precision-made parts, and few such components are truly proprietary. Electric motors have been made for over a century, and their design and manufacture are not complicated - at least when compared to the vastly more complicated and precision-made ICEV. Similarly, an EV transmission is significantly simpler than the transmissions found in all ICEVs. We conclude that the design and manufacture of an EV drivetrain should be simple for a company accustomed to making ICEVs. Even the power and charging electronics are similar to the sorts of things electrical engineers have been making for a long time. Similarly, the assembly of a battery pack from commodity cells should be a relatively straightforward process for any company used to volume manufacturing. As we predicted, battery production appears to be scaling up, and sourcing commodity batteries should not be difficult if demand for EVs emerges as some predict. Although we have largely skipped over a discussion of the non-drivetrain components of an automobile, traditional manufacturers have been manufacturing these for a very long time and are capable of producing them at a reasonable cost and in vast numbers. The major difference between the non-drivetrain components of an EV and ICEV is accommodation for the shape and weight of the battery pack, which, again, should not be a substantial engineering challenge for any large auto manufacturer. For many years, auto manufacturers have developed "platforms" that allow them to mass produce standardized components that are used on what are apparently very different vehicles. Most likely, traditional vendors will produce a platform which can be used for both ICEVs and EVs, meaning that they can reuse parts produced for their ICEVs in EVs, saving money in terms of design, tooling, and volume manufacturing. Obviously, an EV-only vendor does not have that option. Finally, large automobile manufacturers have a global distribution channel as well as nearly omnipresent parts and service networks, including parts and service available from an assortment of third party providers. Developing this support system is particularly important for EVs to enter the mainstream: it is false to assume the simpler drivetrain of an EV will mean the vehicles never need repairs, as there are many failure modes. Beyond wealthy early-adopting EV enthusiasts who purchase EVs as a second or third auto, the typical consumer owns only a single vehicle, making prompt and affordable repairs critical to the utility of a mass-market vehicle, regardless of whether that vehicle is an EV or an ICEV. In summary, we conclude that there is no particular engineering challenge for existing large automakers to enter and dominate the EV business (Tables 1 and 2). Most likely, profit margins on EVs will be low or negative for some time (see Part 1), and large vendors will be in a position to use their profitable ICEV sales to subsidize their market share in the EV business. The main competitive uncertainty for EV manufacturing is how much battery performance and price can be improved from current levels. The battery cells themselves are rather commoditized, making it difficult for any single auto company to develop a substantial lead on the field in battery pack performance. Table 1Conventional Auto Manufacturers Are Ramping Up EV Penetration Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact Table 2TSLA Will Lose Market Share As Mass-Market Competition Expands Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact Rate Of Adoption As we showed in Part 1, costs of ownership of EVs are quite high compared to ICEVs over the EV's assumed 100,000 mile life. Although we believe accelerated depreciation of the EV will significantly increase the differential, most consumers are unaware of that likelihood. Governments and EV manufacturers heavily subsidize EVs; without such subsidies, consumers' costs of ownership would be materially higher. If EVs become a significant share of the vehicle market, such subsidies will have to be reduced, and high taxes would have to be applied to either the vehicle or the fuel (electricity) to make up for the loss of massive government revenues from today's gasoline taxes. The most expensive item in an EV is the battery pack (Chart 5). It appears to be an article of faith among EV advocates that existing batteries will somehow see cost reductions to below their current materials costs, and/or that revolutionary battery technology will emerge in (rapid) due course. It is interesting to speculate as to what might occur in the future. However, we prefer to be data driven. After all, why confine speculation on technological advancements only to things battery-related? Rapid technological advancements in oil production have cut gasoline prices dramatically in the past few years, while continued improvements of conventional engines can raise fuel efficiency and dramatically lower pollution/CO2 emissions of ICEVs, stiffening the competition against the rise of EVs. Chart 5As The Battery Pack Increases In Size,##BR##It Commands A Larger Share Of The Total Cost Of The EV Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact Besides cost, there are numerous compromises associated with an EV which may temper adoption. These include the limited range and slow refueling times, which are important if the owner regularly--or even occasionally--makes long trips; degraded performance in temperature extremes, and so on. An important consideration for many buyers is the size of the car: a soccer mom is not likely to find a Bolt a suitable replacement for a minivan. Larger EVs require disproportionally larger batteries: the Tesla Model S 85 has a 40% larger battery but only a 10% greater range compared to the Bolt. EVs More Likely To Be Popular In The EU Than In North America Europeans tend to drive fewer kilometers and take fewer long trips than North Americans. The average distance traveled by car is 14,000 km4 (8,700 miles) in Europe compared to 20,000 km (12,000 miles) in the U.S., so a European would likely get a few more years out an EV - though not many more kilometers. Similarly, most of the population of Europe lives in areas where temperature extremes are less severe than they are in certain areas of the U.S. and Canada, meaning some of the compromises associated with operating an EV would be less significant. Europe has a much higher population density than the U.S., making particulate pollution a larger issue, and Europeans have more concerns regarding climate change. Much higher gasoline taxes and narrow roads in Europe also incentivize drivers to own smaller vehicles, similar to the Bolt. Due to these factors and the "carrot and stick" approach of subsidies and mandates favored by some EU countries, we conclude EVs are likely to be much more popular in the EU than in the U.S. (Chart 6) Chart 6European EV Sales Are Outpacing U.S. Sales European EV Sales Are Outpacing U.S. Sales European EV Sales Are Outpacing U.S. Sales Regardless, even EV adoption in the EU is bound to be constrained by: Higher costs of EVs compared to ICEVs; Driving habits which may preclude ownership by some people; Access to both private and public chargers; Long lives of ICEVs; and Availability of EVs for purchase. In Part 1 of our EV analysis, we break down the substantially higher cost of ownership for an EV compared to an ICEV. Driving habits boil down to the question of standard deviation: although the average EU driver may travel about 70 km (43 miles) per work day, a sizeable minority may travel much more than that or regularly make round trips beyond the range of their EVs. Alternatively, some may want to pull a trailer (caravan), etc... These drivers would be less likely to purchase an EV except perhaps as a second vehicle. Access to private chargers depends on the nature of the buyer's housing: somebody living in a house with a driveway can pay to have a slow charger installed, whereby somebody who relies on street parking or a nearby parking lot does not have that option. Due to the far greater population density of Europe, access to public chargers may be more of a constraint in the EU than in the U.S. In Part 1, we explained why we believe that ICEVs will outlast EVs for the foreseeable future due to degradation inherent with all battery technologies. There may be a dramatic breakthrough in battery technology, but batteries have numerous parameters which must be acceptable before they can be used in an EV. Most likely, an EV will be scrapped rather than have its battery replaced after about 160,000 km, whereas many ICEVs are routinely kept on the road for double that range. Consumers will eventually realize this and incorporate accelerated depreciation into their costs of ownership calculation. Not only that, but many will choose to keep their ICEVs on the road as long as possible simply to save the expense of purchasing a new vehicle, especially if the inherent limitations of EVs mean they are not suitable for that particular driver. Despite still-generous government subsidies, GM is believed to lose $9,000 for every Bolt it sells. Similarly, the CEO of Fiat lamented some time ago the company was losing $14,000 for every Fiat 500 EV it sold,5 and Tesla loses money despite selling into a premium segment. There is no reason to believe any EV vendor will actually make money on EVs for many years. After all, they all have the same problems with respect to the cost of batteries. We believe auto vendors are likely to limit sales of EVs through rationing or high prices in order to limit their own losses. EVs Are Unlikely To Replace All ICEVs The compromises/deficiencies associated with EVs mean that they will not be suitable for many consumers unless a massive battery breakthrough is achieved. The limited range is an obvious issue: a consumer might, for example, travel an average of 12,000 miles (20,000 km) per year but may regularly take a drive of a few hundred miles, which would require one or more recharging stops. It is all well and good to speak of rapid charging, but even this would quickly lose its allure after long trips, especially given the issues noted in "EVs Will Require a Sizeable Charging Infrastructure" below. Almost 3 million pickup trucks are sold in the U.S. every year, out of 17.5 million vehicle sales. Light trucks, including SUVs and Crossovers, make up another 10.5 million sales. Whether or not the trucks are actually used for hauling, the battery size, and therefore cost of ownership, would have to be particularly large for a pickup truck. A 120 kWh battery would add about 1,600 pounds (720 kg) to the vehicle, which is about half the cargo capacity of a Ford F-150 full size pickup truck. Many pickup trucks have significantly oversized engines in order to tow heavy loads. It is questionable an EV pickup truck would have the range or towing capacity required by many buyers. EVs Will Require A Sizeable Charging Infrastructure First-time EV owners will either have to invest in a charging station for their homes or somehow get access to one. Charging stations come in different types. In the case of the Bolt, a typical home charger delivers 4 miles (6.5 km) of range/hour of charge or about 32 miles (52 km) of range for 8 hours. What GM calls "Fast Charging" delivers almost a full charge over 8 hours. What GM refers to as "Super Fast Charging", or true fast charging, delivers 90 miles (145 km) of range in 30 minutes or 160 miles (258 km) in 1 hour, but is only available in public locations6 and requires a special option on the vehicle. "Super Fast Charging" means that a customer planning a trip of over 238 miles will have to plan for at least one 30 minute stop for every 90 miles of additional travel. Of course, this is when the vehicle is new and under ideal conditions without any temperature extremes, etc. An older EV may require a 30 minute stop after the first 150 miles and a subsequent 30 minute stop for every hour of travel (60-70 miles) after that. Private Chargers Unless they are satisfied with multi-day charging, new EV buyers have to pay an electrician to install a high current charger outlet which is accessible to the vehicle. Not all homes have ample parking, nor is it easy to install a high current port accessible to a vehicle in all homes. A typical high current charging port required for a "slow charger" requires a 40, 50, or 60 amp outlet. Many homes have only a 100 amp service, which may pose issues if the vehicle is charging and, for example, an air conditioner starts up. Similarly, apartment/condo dwellers with access to parking may have access to EV chargers provided by the building, though the electric service to the building/parking lot may require upgrading in the event a significant number of owners buy EVs. Publicly Available Chargers The largest challenge might be for would-be EV buyers who park on the street, as is fairly common in many urban areas. The cost of installing EV chargers is not trivial, and it is hard to believe cities will accept the costs of installing a large number of chargers to ensure EV owners can charge their vehicles. This doesn't even account for the fact that somebody has to pay for the electricity, and street-side chargers are both expensive and dangerous, require maintenance and snow removal, and may be subject to vandalism. Additionally, some parking lots feature a couple of EV chargers, and most EV vendors provide access to a rather sparse assortment of chargers. On the surface, a 6:1 ratio of global EVs to publicly available chargers may not appear to be as much of a concern, however, the ratio is about 16:1 for slow chargers and 105:1 for fast chargers in the U.S., and 6:1 and 68:1 in the EU, respectively (Charts 7 and 8). Recall that the Bolt's "Fast charger" only supplies about 25 miles of range for every hour of charging, so public units would only be useful as a "top-up". Public chargers will have to become far more common as the number of EVs increases or owners risk planning a trip which assumes access to a charger only to discover the unit is in use and the EV owner who is using it is off shopping. Chart 7Globally, There Is One Public Charger ##br##Per Six EVs Globally, There Is One Public Charger Per Six EVs Globally, There Is One Public Charger Per Six EVs Chart 8Fast Chargers Are Much More Scarce ##br##Than Slow Chargers Fast Chargers Are Much More Scarce Than Slow Chargers Fast Chargers Are Much More Scarce Than Slow Chargers Fast chargers are of particular significance in the event an EV owner wishes to make a trip in excess of the vehicle's fully-charged range. "Fast charge" times - whether with a Bolt or any other EV - assume a charging station is available when the EV arrives. This may be the case on typical days, but less likely during holiday or vacation season: "A video shot yesterday at the Supercharger in Barstow, CA shows a line at the station of Teslas waiting to juice up. The driver who shot the video was number 21 in the queue, and with wait times upwards of two hours just to get to the charger, Tesla's going to have some unhappy customers on its hands."7 One can only imagine how frustrated the owner of an aged Bolt would be if they had to wait 2 hours every 60 miles. Impact Of EV Adoption On Pollution And Greenhouse Gas Emissions The production and operation of any product leaves an environmental impact in terms of pollution and Greenhouse Gas (GHG) emissions. The environmental impact associated with vehicles arises from the production of the commodities used to make the components, the manufacture of the vehicle components, the assembly of the vehicle itself, and the operation of the vehicle. EVs are not "zero emission vehicles" in any meaningful sense. It is true that they do not discharge particulate or CO2 emissions from the tailpipe, but emissions arise from the production of the vehicle platform, the battery pack, and the production of electricity used to charge the battery. The fuel mix of power generation in a particular region has a significant impact on the GHG emissions associated with electric power: countries with significant hydroelectric or nuclear power sources will have lower GHG emissions per kW than those which burn coal, oil, or natural gas. Similarly, the GHG emissions associated with the manufacture of a vehicle and its components depend on the power mix in the country in which those components are manufactured. As previously noted, an EV is very similar to an ICEV except for the drivetrain and battery. The EV's drivetrain is simpler than an ICEV's, but total GHG emissions associated with manufacturing an EV and equivalent ICEV are estimated to be quite similar, excluding the battery pack. GHG emissions associated with the manufacture and recycling of a battery pack are quite hard to pin down. The best and most recent example we found comes from IVL Swedish Environmental Research Institute, and notes: "Based on our review, greenhouse gas emissions of 150-200 kg CO2-eq/kWh battery looks to correspond to the greenhouse gas burden of current battery production."8 To put things in perspective, the GHG burden associated with the lifecycle of a 60 kWh Bolt battery pack is between 9,000 and 12,000 kg, or 9 to 12 metric tons. Because the battery pack is likely larger than advertised to limit degradation, the actual figure is probably at least 20% more, or 10.8 to 14.4 metric tons. At just 9 metric tons, assuming a 160,000 km life, the GHG burden associated manufacture and recycling of a Bolt battery pack is about 56 g CO2/km, and at 14.4 metric tons the burden is about 88 g CO2/km. To be as favorable as possible to the Bolt's potential to reduce GHG emissions, we have used the lower bound of the estimated CO2 burden of the Bolt's 60 kWh battery, 9 metric tons, in our GHG analysis in Table 3. The actual CO2 burden could be as much as 5.4 metric tons more. Note that the above calculations do not include the GHG emissions associated with recharging the battery. Recall that in Part 1, we estimated the power consumption associated with a Bolt operating for 160,000 km would be about 31,250 kWh, or ~0.20 kWh/km (0.3125 kWh/mile). The GHG burden of recharging the battery varies considerably depending on the regional mix of power generation. As shown in Table 3: Table 3EVs Will Reduce Carbon Emissions Only If Power Grid Is Green Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact In France, where power is primarily generated via carbon-free nuclear energy, recharging the Bolt will release just 2 metric tons of CO2 during its 160,000KM life (11g/km). In coal-heavy Germany (40+% coal), recharging the Bolt will generate ~18 metric tons of CO2 (109g/km), slightly more carbon than the fuel-efficient gasoline-powered ICEV Opel Astra (104g/km). In the U.S., with the current diversified mix of power generated by natural gas (34%), coal (30%), nuclear (20%), hydro (7%), wind (6%) and solar (1%), CO2 emissions from recharging the Bolt would be only 13 metric tons (83g/km), 60% lower than the 32 tons of CO2 emitted by the ICEV Chevy Sonic. As shown, despite the higher CO2 footprint associated with manufacturing the EV's battery pack, an EV may indeed lead to an overall reduction in GHG emissions in a region where electricity generation is already low-carbon; however, the EV actually emits more CO2 in Germany, a coal-heavy country (40% coal) with fuel-efficient ICEVs. This implies EVs would create even greater CO2 increases in countries like China or India, which both generate over 70% of power from coal. The carbon intensity of U.S. power generation has been reduced by roughly 23% over the past decade due to the increased displacement of coal with natural gas (~70% of the carbon reduction) and renewables. As the U.S. and other countries continue to de-carbonize their power grids, the emissions to recharge EVs will further decline. However, even where reductions are achieved, the lifecycle emissions of the EV is nothing close to what is implied by the term "Zero Emission Vehicle." Using our generous assumptions for the carbon footprint of the EV's battery, we calculate the approximate lifecycle CO2 reductions for an EV are ~9 metric tons in the U.S., and ~6 metric tons in France. In Germany, the EV actually emits ~10 metric tons more CO2 than a comparable ICEV. EVs in coal-heavy China and India would also be expected to emit more lifecycle CO2 than a fuel-efficient ICEV. Even if power generation were 100% carbon-free in the EU and in the U.S., the CO2 savings would be only 23 tons per vehicle in the U.S and 8 tons per vehicle in the EU (lower savings in the EU due to the higher fuel efficiency of the European ICEV). One area where the EV is bound to come out ahead is in reducing particulates, NOx, and other non-GHG related pollutants, at least in the areas where the vehicles are operated, which provides cleaner air in highly populated areas. EV Subsidies Are Not Justified By Carbon Emissions In order to simplify the cost/benefit debate over legislation and regulation aimed at reducing carbon emissions, the U.S. EPA and other various U.S. agencies have calculated/estimated a "Social Cost of Carbon," i.e., the estimated economic damage created by emitting a ton of CO2 in a given year.9 In the base case, the social cost of carbon was pegged at $36/metric ton in 2015, with expectations that it would rise to $50/metric ton in 2030 and $69/metric ton in 2050 as climate issues became more severe. By comparison, the "market value" for a ton of CO2 on traded exchanges in California and in the E.U. is between $5-$15/ton. Assuming an average value of $50/metric ton, the current CO2 savings of the EV will yield about an economic benefit per vehicle of ~$450 in the U.S, and ~$300 benefit in France. In Germany, where CO2 emissions for the EV are higher than the ICEV, it adds another ~$500 to the economic cost of the EV. At a value of $50/ton, the value of CO2 savings in each region are only ~4-5% of the value of the public subsidies of $7,200-$9,500/vehicle in the U.S. and France, and only 1-2% of the total ~$22,000-$27,000 total extra societal costs of the vehicles (Table 4). In other words, the subsidies alone cost 20x more than the economic benefit of the CO2 reductions, while the total extra costs of the EV are 55-75x higher than the economic value of the CO2 reductions. Germany is offering subsidies for vehicles that increase CO2 emissions. Table 4EV Carbon Reductions Are Way Too Expensive Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact Of course, industry may be able to lower emissions associated with battery manufacturing and recycling, and power generation may continue to be de-carbonized as well, leading to lower GHG emissions associated with EVs in the future. However, the same might be said regarding continuing improvements in ICEVs as well. For example: If U.S. drivers changed preferences to drive European-style cars with smaller engines and greater fuel efficiency (that is, wider adoption of technology that already exists today), that alone could save ~17 tons of carbon per vehicle in the U.S., dwarfing the ~10 tons of carbon savings achieved by owning an EV, at a much lower economic cost. Again, one area where the EV is bound to come out ahead is in reducing particulates, NOx, and other non-GHG related pollutants, at least in the areas where the vehicles are operated, which provides cleaner air in highly populated areas. This reduction/transfer of pollution from the city center to the power generation stations has a real health/quality of life value that we have not included in the above analysis, as the overwhelming amount of EV interest we read and receive is specifically based on EVs' (overestimated) ability to reduce global carbon emissions.10 Bottom Line: TSLA does not have an insurmountable technological lead on conventional car producers in the mass-production EV market, and is likely to lose market share to larger competitors that have better costs, infrastructure, and experience supporting a global fleet of mass-produced vehicles. Near-term adoption of EVs will be forced higher by governmental carrot and stick incentives, but these will become too expensive to continue as EVs' market share increases. Today's EV subsidies will turn into tomorrow's EV taxes as gasoline taxes are diminished, weighing on the longer-term arc of commonly-forecasted EV adoption. Finally, EVs do not necessarily reduce CO2 emissions, and when they do, the value of those CO2 reductions is exceedingly small compared to the added cost of the vehicles to producers, consumers, and government coffers. A modest ICEV only emits ~$2,000 worth of CO2 over 100,000 miles in the first place, elucidating how difficult it will be for an EV to reduce GHG emissions on a cost-competitive basis. For mass-market EVs to successfully displace ICEVs in the eyes of cost-conscious consumers and taxpayers, EV battery technology needs to improve massively, not incrementally. The batteries need to provide multiples of today's energy storage capacity with lower weight, lower cost, faster recharge abilities, and a lower carbon footprint. Furthermore, since an EV's battery recharging is only as green as the power source behind it, continued (expensive) greening and expansion of global power generation would also be necessary for EVs to demonstrate a positive impact on GHG emissions, as will be discussed more in Part 3 of this report series. Brian Piccioni, Vice President Technology Sector Strategy brianp@bcaresearch.com Matt Conlan, Senior Vice President Energy Sector Strategy mattconlan@bcaresearchny.com Robert P. Ryan, Senior Vice President Commodity & Energy Strategy rryan@bcaresearch.com Michael Commisso, Research Analyst michaelc@bcaresearch.com Johanna El-Hayek, Research Assistant johannah@bcaresearch.com Hugo Bélanger, Research Assistant HugoB@bcaresearch.com 1 Please see Technology Sector Strategy Special Report, "Electric Vehicles Part 1: Costs of Ownership", dated August 1, 2017, available at tech.bcaresearch.com. 2 https://www.forbes.com/sites/neilwinton/2017/06/29/tesla-focus-means-victory-versus-complacent-mainstream-in-electric-car-market-report/#4d0d4684577e 3 http://www.hybridcars.com/2017-chevy-bolt-battery-cooling-and-gearbox-details/ 4 http://www.acea.be/publications/article/cars-trucks-and-the-environment 5 http://jalopnik.com/sergio-marchionne-doesnt-want-you-to-buy-a-fiat-500e-1579578914 6 https://www.chevyevlife.com/bolt-ev-charging-guide 7 http://bgr.com/2016/12/27/tesla-supercharger-wait-times-lines-california/ 8 http://www.ivl.se/download/18.5922281715bdaebede9559/1496046218976/C243+The+life+cycle+energy+consumption+and+CO2+emissions+from+lithium+ion+batteries+.pdf (page 42) 9 https://www.epa.gov/sites/production/files/2016-12/documents/social_cost_of_carbon_fact_sheet.pdf 10 It is worth pointing out that if the incentive structure is such that entrepreneurs are rewarded for finding ways to economically reduce carbon emissions in ICEVs in a way that is cost-competitive with EVs, the principal advantage of EVs would be challenged. There is no ironclad rule of physics we are aware of that precludes such a development. Investment Views and Themes Recommendations Strategic Recommendations Tactical Trades Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact Commodity Prices and Plays Reference Table Electric Vehicles Part 2: EV Investment Impact Electric Vehicles Part 2: EV Investment Impact Trades Closed in 2017 Summary of Trades Closed in 2016

Dear Client, Over the next three weeks, much of BCA’s Geopolitical Strategy team will be traveling in Australia, New Zealand, and Asia. As such, we are taking this week off from publication and will return to our regular schedule next week. In lieu of our regular missive, we are sending you the following Special Report, penned by our colleagues in the BCA Technology Sector Strategy. The report, originally published on May 16, tackles “The Coming Robotics Revolution” in an innovative way that aligns with our own views. Clients often ask us what will be the political consequences of the revolution in artificial intelligence and robotics. Our answers are controversial because we strongly disagree with the conventional, Terminator-inspired, doom and gloom. Brian Piccioni and Paul Kantorovich agree with us, which is reassuring given that they understand the technology behind robotics far better than we do. I hope you enjoy the enclosed report and encourage you to seek out the insights of our Technology Sector Strategy. Kindest Regards, Marko Papic, Senior Vice President Chief Geopolitical Strategist Feature "The amount of technology coming at us in the next five years is probably more than we've seen in the last 50" Mark Franks, Director Of Global Automation at General Motors, Bloomberg News, April 2017 There is good reason to believe we are at the cusp of a Robot Revolution which will have a dramatic impact on our economy. Robots have been around for decades or centuries, depending on the definition. Past robots were either fixed in place, as in the case of factory robots, or supervised by operators that are near the robot, or connected through telemetry. In contrast, the robots that are coming will not be fixed in place, and will be able to perform their functions without a human operator. This opens up massive markets for robots in industry (cutting lawns, cleaning windows, delivering parcels, etc.) and, most significantly, consumer applications. Part 1: Robots - Industrial Revolution To Early 21st Century The term "robot" can have different meanings. The most basic definition is "a device that automatically performs complicated and often repetitive tasks,"1 a definition which encompasses a broad range of machines: from the Jacquard Loom,2 which was invented over 200 years ago, on to Numerically Controlled (NC) mills and lathes, pick and place machines used in the manufacture of electronics, Autonomous Vehicles (AVs), and even homicidal robots from the future such as the Terminator. For much of history, most of the labor force was involved with the production of food: over 50% of the U.S. labor force was involved in agriculture until the late 1800s (Chart 1). Agriculture has benefitted immensely from automation as inventions such as the McCormick Reaper (a wheat cutting machine pulled by horses), the cotton gin, and other mechanical systems displaced human effort. Steam and then internal combustion-powered tractors, which can be viewed as "robotic horses," accelerated the process, as engines delivered much more power more cost effectively than mechanical devices (Chart 2). This massively improved productivity: within 20 years from 1830 to 1850, the labor to produce 100 bushels of wheat dropped from 250-300 to 75-90 hours, and by 1955 it only took 6 ½ hours of labor for a net reduction of 97.5% in 125 years.3 Chart 1Farm Workers Were Disrupted In The Late 19th Century The Coming Robotics Revolution The Coming Robotics Revolution Chart 2...And So Were Horses The Coming Robotics Revolution The Coming Robotics Revolution In other words there is nothing new about automation displacing workers while improving productivity, nor is a rapid displacement unprecedented. The industrial revolution was about replacing human craft labor with capital (i.e. machines), which did high-volume work with better quality and productivity. This freed humans for work which had not yet been automated, along with designing, producing, and maintaining the machinery. Automation Frightens People Although automation is nothing new, it has always engendered anxiety among workers. The anxiety boils down to concern for continued employment as well as fear of the technology itself. We discuss below why Artificial Intelligence (AI) does not present the sort of threat to humanity or even employment that seems to be the consensus view at the moment. Will Robots Become Self-Aware? We have covered the topic of Artificial Intelligence/Deep Learning as it relates to sentient/self-aware machines in some detail in our October 18, 2016 Special Report on Artificial Intelligence. In summary, most of the discussion surrounding AI is misinformation. Although AI uses algorithms called "artificial neural networks," which are extremely useful for solving certain classes of problems, these are nothing like biological neural networks. There is no reason whatsoever to believe AI technology in its current form can become sentient, or even meaningfully intelligent, and that will not change with increased computing power. Furthermore, whether or not AI can arise to the level of a threat, there is no current or imagined power source which could keep a rampaging robot active for more than a few hours. The Terminator would have been much less threatening if he required frequent recharging. Will Robots Make Human Workers Irrelevant? Automation in agriculture occurred rapidly enough to be felt by workers at the time - and yet there were no marauding hordes of unemployed hay cutters or cowboys. Improved productivity meant markets were opened which did not previously exist, and unemployed agricultural workers moved to factory work. Media coverage of automation tends to focus on the potential job losses without mentioning the fact that the economy and its workers adapt, and overall living standards generally improve (Chart 3). Technology has displaced entire classes of jobs very rapidly in the recent past, and many products such as smartphones would be extremely difficult to assemble if the work was done by hand. Box 1 provides several other examples. Yet as is usual for many things that have happened multiple times in the past, we are told "this time is different." Chart 3The Industrial Revolution Led To A Vast Improvement In Living Standards The Industrial Revolution Led To A Vast Improvement In Living Standards The Industrial Revolution Led To A Vast Improvement In Living Standards Box 1 Automation Displaced Entire Classes Of Jobs In The Recent Past, But Brought Enormous Benefits Before calculators and word processors were available, writing and mathematical calculations were done manually. Machines such as calculators and type writers enhanced productivity, eliminating many such jobs. Software applications such as Microsoft Word and Excel further accelerated this process. Not that long ago, welding was entirely a manual job but now most welding in factories is done by robots: you can usually tell a human weld on a mass produced product by its poor quality. Robots in the modern factory have freed up workers for other roles in the economy just as the massive loss of agricultural jobs in the 20th century did. Many modern electronic products such as smartphones would be extremely difficult to assemble if the work was done by hand, as the components are so small they require microscopes to manipulate. Even if it were possible to hand assemble a smartphone, it would take hours of manual labor to produce, and the quality would be very poor. The use of automation means that smartphones cost a few hundred dollars instead of a few thousand dollars and are affordable enough to be a mass market item. Some of the anxiety around automation-related job losses centers on the possibility that this time, robots will displace workers from the service and white-collar sectors. BCA's European Investment Strategy service has written about the potential for AI to replace jobs involving tasks that require specialized education and training, such as calculating credit scores or insurance premiums, or managing stock portfolios.4 Recent developments in AI (specifically deep learning algorithms) have allowed computers to solve pattern recognition problems that they could not previously solve. However, we do not believe AI in its current form poses a widespread risk to white collar employment for the following reasons: Both service-sector and white collar employees have been subject to replacement through automation already, and the economy has adapted: ATMs are robot bank tellers, self-checkout lanes are robot checkout kiosks, and "smart" gas and electric meters that can be read remotely replace human meter readers. The legal profession has been transformed by Google searches and the accounting business by accounting software. These tools allow certain clients to avoid the use of a lawyer or accountant altogether (for example in setting up a corporation or doing bookkeeping), or allow a firm to employ less skilled workers for the task. We can offer numerous other examples of white collar jobs which have been fully or partially automated over the past couple decades. In addition, recall that AI produces high probability answers which turn out to be wrong, and it requires a lot of subject specific training. Both of these are intrinsic to the implementation of the algorithm. In contrast, humans generally are much better at assigning confidence to decisions and train very rapidly because they have cross-expertise AI lacks. An implementation of AI has to meet BOTH of the following conditions to be successful: There has to be a lot of subject-specific data available A high probability assigned to a wrong answer is either inconsequential or can be easily overruled by a human It is also important to note that although AI may reduce the demand for accountants, insurance agents, credit analysts and other skilled professionals, these are exactly the sort of people that can handle retraining. Part 2: What Makes Upcoming Robots Revolutionary Upcoming robots will be different because they will not be confined to the factory floor. We believe this is a key transition point, and that the next 20 years or so will see as dramatic a change from robotics as was caused by the Internet. Factory robots have improved immensely due to cheaper and more capable control and vision systems. Early robots performed very specific operations under carefully controlled conditions -an assembly robot which encountered a misaligned component would simply install it that way, resulting in a defective product. Eventually vision systems were developed which allowed robots to adjust to varying conditions. As camera and computing costs continue to decline, vision systems are becoming more elaborate and useful, as they gather and process more information to make increasingly complex decisions. As these systems evolve, the abilities of robots to move around their environment while avoiding obstacles will improve, as will their ability to perform increasingly complex tasks. Mobile robots will likely rely on AI to make many decisions. In order to be cost effective, for many years AI will likely be hosted in cloud data centers. This is especially the case for consumer robots, which will have to be highly capable and yet cost effective. We discuss the implications for cloud services providers in more detail in Part 3: Investment Implications. We May Be Entering A 'Virtuous Cycle' In Robotics Improvements to one domain of robotic applications can be generally applied to others. Robotics technology is concurrently moving forward on many fronts ranging from the aforementioned vacuum cleaners, lawnmowers, and logistics robots, to medical orderlies,5 farm tractors,6 mining equipment,7 transport trucks,8 and cargo ships.9 Despite enormous differences in cost and value added, all of these applications are solving essentially the same problem. As with any other technological revolution, advances between different fields in robotics will be adapted, borrowed, extended and enhanced. This, in turn, creates opportunities for ever more applications, creating a virtuous cycle (Diagram 1). Diagram 1Robotics Will Enter Into A Virtuous Cycle The Coming Robotics Revolution The Coming Robotics Revolution There are few tasks which cannot be automated, but there is a definite cost-benefit tradeoff for each one. For example, a golf course may consider spending $25,000 for a robotic lawnmower, however costs were closer to $70 - $90,000 in 2015,10 and installed cost is even higher.11 Because the incremental cost of the machines is comprised of electronics, which will drop in price rapidly, it is probably a matter of another 2 or 3 years before the price moves to the point where mass adoption by groundskeepers begins. The same improvements to industrial lawnmowers will lead to more useable, albeit still pricy, consumer models which will probably enter mass market adoption 5 to 10 years from now. The same argument can be made for almost any manual chore ranging from cleaning the carpet to delivering parcels. We predict the virtuous cycle for robots will span several decades. As the cost of automation drops, better solutions will be developed, resulting in 'early retirement' of dated but otherwise fully functional robotic systems. This is the opposite of the Feature Saturation phenomenon currently present in the smartphone and PC industries - though feature saturation will eventually hit robots as well. A Self-Driving Car Is A Robot The most important robotics technology, from a macroeconomic perspective, is the rapidly advancing field of Autonomous Vehicles (AVs). The automobile industry is a significant part of the global economy, so changes in this industry will have profound implications. We covered AVs in detail in our April 8, 2016 Special Report. Due to technical and legal obstacles that must be overcome, a vehicle which can safely travel from point to point on major roads and city streets without driver intervention is probably 20 years away, +/- 5 years. The macro impact, however, will occur much sooner than that, due to the technologies developed on the way to full AVs. Vehicles are already offering features such as forward collision warning, autobrake, lane departure warning, lane departure prevention, adaptive headlights, and blind spot detection.12 Although we have only touched the surface, robotics are being applied across many industries, making even seemingly modest advances significant when measured in aggregate, as small changes in one industry are quickly adapted by other industries. It is noteworthy that this transition will likely occur during a period where demographic shifts, in particular in the most developed economies, signal the potential for labor shortages, or at least increasing cost of labor (Chart 4).13, 14 Robots may be showing up in the nick of time to improve both the economy and quality of life in the developed world. Chart 4Advances In Robotics Will Counter Adverse##br## Demographic Trends Advances In Robotics Will Counter Adverse Demographic Trends Advances In Robotics Will Counter Adverse Demographic Trends Part 3: Investment Implications The semiconductor industry has stagnated as the PC and smartphone markets entered a largely replacement-driven era (Chart 5). Although it may not be evident until the virtuous cycle is fully engaged, robotics represents another up-leg in demand for semiconductors and therefore should result in a significant improvement to industry growth rates. There is little opportunity for startup semiconductor companies nowadays due to the high costs of developing a new chip. Well positioned, established, semiconductor companies will be the primary beneficiaries of the robotics revolution. Large firms that attempt to fit their existing product offering into the industry (e.g. by remaining PC or mobile-phone centric) will fall behind. Winners System on a Chip (SoC) Vendors: Robotics hardware will more likely be implemented as "System on a Chip" (SoC) as this provides the greatest functionality with lowest cost and power consumption. SoCs generally consist of a variety of Intellectual Property (IP) "cores" which may be licensed from third parties. Typically, IP cores consist of a microprocessor and various specialized subsystems, depending on the application. Robotics SoCs are likely to include Digital Signal Processing (DSP) or Image Processing cores to process sensor data. SoC vendors who target or encourage robot development, such as Overweight-rated Texas Instruments, are likely to be favored by early movers in the space.15 We believe it is a matter of time before Graphics Processors (GPUs) currently used in AI/Deep Learning are replaced by processors specifically designed for AI, which will be cheaper and more power efficient.16 This is one of the reasons for our Underweight rating on Nvidia. Semiconductor Foundries, Mixed Signal and Automotive Semiconductor Vendors: This environment will favor the merchant semiconductor foundries which manufacture most SoCs. In addition, firms with "mixed signal" expertise will experience increased demand for motor controls, sensor interfaces, etc. As robotics features are added to automobiles, demand for automotive semiconductors should outpace that in other sectors. A significant degree of commonality in the parts and systems used in advanced automobiles will be used in other mobile robots, so "automotive" semiconductor demand should significantly outpace automobile sales. Sensor Vendors: Robots need a variety of sensors, depending on the application. Unlike factory floor robots which can make do with cameras, mobile robots will require advanced radar, ultrasound, laser scanning and other sensor types in order to provide redundancy and cope with weather and other related issues. Important sensors on prototype AVs are currently made in low volumes and are extremely expensive. Due to the number of sensors involved, we believe there is significant opportunity for companies offering aggressive cost reduction in sensor technology. Wireless Equipment and Service Providers: Most robotic systems will include some degree of wireless connectivity and participate in the "Internet of Things" (IoT). This will present challenges and opportunities for wireless equipment and service providers,17, 18 as networks will have to adapt to increased upload bandwidth (from robot to carrier) as well as novel billing schemes. Coverage will also have to be expanded to accommodate AVs as it is non-existent or spotty in large stretches of North American roadways. Not being able to check Facebook between two cities is one thing, losing your robot driver is much more serious. Our recent downgrade of Cisco to Underweight19 may appear inconsistent with the analysis above. However, the company's valuation is extremely elevated and revenues are declining (Chart 6). Any benefit Cisco will derive from investment into wireless infrastructure is several years out, and open-source hardware initiatives are gaining momentum.20 For that reason, we see the risks as outweighing the opportunities at the moment for the company. Chart 5Long Replacement Cycles Mean Slower ##br##Semiconductor Sales Long Replacement Cycles Mean Slower Semiconductor Sales Long Replacement Cycles Mean Slower Semiconductor Sales Chart 6Cisco's Stock Price Is Close To Tech Bubble##br## Levels Despite Declining Revenue Cisco's Stock Price Is Close To Tech Bubble Levels Despite Declining Revenue Cisco's Stock Price Is Close To Tech Bubble Levels Despite Declining Revenue Cloud Service Providers: Most robots will be on line and some will likely use cloud services to offload computational effort and minimize cost. A relatively "dumb" robotic lawnmower which offloads control to a shared computational resource in the cloud would probably be cheaper than a much more capable fully autonomous system. This will increase demand for cloud services, however the challenge of declining margins (due to increased competition in the space) will offset cloud services revenue growth somewhat in the long term. On balance, Overweight-rated Microsoft and Alphabet/Google, as well as Amazon, stand to benefit. Chart 7Eastman Kodak Tried To Ignore The Shift ##br##To Digital Cameras Eastman Kodak Tried To Ignore The Shift To Digital Cameras Eastman Kodak Tried To Ignore The Shift To Digital Cameras Losers We believe companies who ignore the robotics revolution will find themselves at a significant competitive disadvantage. This is not unprecedented in the technology sector: Digital Equipment Corporation (DEC) and Kodak vanished because their business models could not accommodate an obvious shift in their core markets (Chart 7). Similarly Intel and Microsoft completely missed the smartphone revolution. As we noted in our April 8, 2016 Special Report on AVs, the frequency and severity of crashes will decrease dramatically which will lead to reduced insurance rates, fewer repairs, and less money spent on accident related healthcare and rehabilitation. The economic losses of automobile crashes were estimated $871 billion in the US in 201021 and even a modest reduction in the frequency and severity of collisions due to partial automation would have a significant economic impact. "Dumb" Auto Parts Manufacturers: Fewer collisions will result in fewer repairs to people or vehicles. Auto parts manufacturers will fall into two camps: those with significant expertise in robotics will prosper, while those without such expertise will fall behind as the demand for replacement components (fenders, bumpers, doors, windshields, etc.) will decline. AVs are also likely to include advanced diagnostic and service reminder systems which will result in more timely service, reducing wear and tear on internal components as well. The Auto Insurance Industry: While it is doubtful robotics will ever eliminate auto accidents, the rate might be reduced to such a level that the auto-insurance industry, worth $157 billion in the US alone,22 will be much smaller in 20 years than it is today. This will be offset to a degree by greater demands for product liability insurance for AVs and robots in general. Brian Piccioni, Vice President Technology Sector Strategy brianp@bcaresearch.com Paul Kantorovich, Research Analyst paulk@bcaresearch.com 1 http://www.merriam-webster.com/dictionary/robot 2 http://www.computersciencelab.com/ComputerHistory/HistoryPt2.htm 3 https://www.agclassroom.org/gan/timeline/farm_tech.htm 4 Please see European Investment Strategy Special Report, "Female Participation: Another Mega-Trend," dated April 6, 2017, available at eis.bcaresearch.com. 5 http://www.tomsguide.com/us/Forth-Valley-Royal-Robots-Serco-Medicine,news-7124.html 6 http://modernfarmer.com/2013/04/this-tractor-drives-itself/ 7 http://www.asirobots.com/mining/ 8 http://www.theaustralian.com.au/business/powering-australia/rio-rolls-out-the-robot-trucks/story-fnnnpqpy-1227090421535 9 http://www.bloomberg.com/news/articles/2014-02-25/rolls-royce-drone-ships-challenge-375-billion-industry-freight 10 http://techon.nikkeibp.co.jp/english/NEWS_EN/20141210/393619/ 11 http://www.golfcourseindustry.com/article/do-robotic-mowers-dream-of-electric-turf/ 12 http://www.iihs.org/iihs/topics/t/crash-avoidance-technologies/topicoverview 13 http://gbr.pepperdine.edu/2010/08/preparing-for-a-future-labor-shortage/ 14 http://www.imf.org/external/pubs/ft/fandd/2013/06/das.htm 15 http://www.ti.com/corp/docs/engineeringChange/robotics.html 16 Please see Technology Sector Strategy Weekly Report, "Google - AI And Cloud Strategy," dated April 25, 2017, available at tech.bcaresearch.com. 17 http://www.fiercemobileit.com/press-releases/gartner-says-internet-things-will-transform-data-center 18 http://www.computerworld.com/article/2886316/mobile-networks-prep-for-the-internet-of-things.html 19 Please see Technology Sector Strategy Weekly Report, "Networking Equipment Update ," dated March 28, 2017, available at tech.bcaresearch.com. 20 http://www.businessinsider.com/att-white-box-test-should-scare-cisco-juniper-2017-4 21 http://www.nhtsa.gov/About+NHTSA/Press+Releases/2014/NHTSA-study-shows-vehicle-crashes-have-$871-billion-impact-on-U.S.-economy,-society 22 http://www.bloomberg.c/bw/articles/2014-09-10/why-self-driving-cars-could-doom-the-auto-insurance-industry

Dear Client, I will be visiting clients in Asia over the next ten days, so we are sending you this week's report a bit ahead of schedule. In addition, at the end of this report, we are including the recommendations from our tactical asset allocation model. Going forward, we will be updating these recommendations on our website at the end of every month. Please feel free to contact us if you have any questions. Best regards, Peter Berezin, Chief Global Strategist Global Investment Strategy Highlights Productivity growth has declined in most countries. This appears to be a structural problem that will remain with us for years to come. In theory, slower productivity growth should reduce the neutral rate of interest, benefiting bonds in the process. In reality, countries with chronically low productivity growth typically have higher interest rates than faster growing economies. The passage of time helps account for this seeming paradox: Slower productivity growth tends to depress interest rates at the outset, but leads to higher rates later on. The U.S. has reached an inflection point where weak productivity growth is starting to push up both the neutral real rate and inflation. Other countries will follow. The implication for investors is that government bond yields have begun a long-term secular uptrend. The market is not at all prepared for this. Feature Slow Productivity Growth: A Structural Problem Productivity growth has fallen sharply in most developed and emerging economies (Chart 1). As we argued in "Weak Productivity Growth: Don't Blame The Statisticians," there is little compelling evidence that measurement error explains the productivity slowdown.1 Yes, the unmeasured utility accruing from free internet services is large, but so was the unmeasured utility from antibiotics, indoor plumbing, and air conditioning. No one has offered a convincing explanation for why the well-known problems with productivity calculations suddenly worsened about 12 years ago. Chart 1 If mismeasurement is not responsible for the productivity slowdown, what is? Cyclical factors have undoubtedly played a role. In particular, lackluster investment spending has curtailed the growth in the capital stock (Chart 2). This means that today's workers have not benefited from the improvement in the quality and quantity of capital to the same extent as previous generations. However, the timing of the productivity slowdown - it began in 2004-05 in most countries, well before the financial crisis struck - suggests that structural factors have been key. These include: Waning gains from the IT revolution. Recent innovations have focused more on consumers than businesses. As nice as Facebook and Instagram are, they do little to boost business productivity - in fact, they probably detract from it, given how much time people waste on social media these days. The rising share of value added coming from software relative to hardware has also contributed to the decline in productivity growth. Chart 3 shows that productivity gains in the latter category have been much smaller than in the former. Chart 2The Great Recession Hit##BR##Capital Stock Accumulation The Great Recession Hit Capital Stock Accumulation The Great Recession Hit Capital Stock Accumulation Chart 3The Shift Towards Software Has##BR##Dampened IT Productivity Gains The Shift Towards Software Has Dampened IT Productivity Gains The Shift Towards Software Has Dampened IT Productivity Gains Slower human capital accumulation. Globally, the fraction of adults with a secondary degree or higher is increasing at half the pace it did in the 1990s (Chart 4). Educational achievement, as measured by standardized test scores in mathematics and science, is edging lower in the OECD, and is showing very limited gains in most emerging markets (Chart 5). Test scores tend to be much lower in countries with rapidly growing populations (Chart 6). Consequently, the average level of global mathematical proficiency is now declining for the first time in modern history. Chart 4 Chart 5 Decreased creative destruction. The birth rate of new firms in the U.S. has fallen by half since the late 1970s and is now barely above the death rate (Chart 7). In addition, many firms in advanced economies are failing to replicate the best practices of industry leaders. The OECD reckons that this has been a key reason for the productivity slowdown.2 Chart 6 Chart 7Secular Decline In U.S. Firm Births Secular Decline In U.S. Firm Births Secular Decline In U.S. Firm Births Productivity Growth And Interest Rates Investors typically assume that long-term interest rates will converge to nominal GDP growth. All things equal, this implies that faster productivity growth should lead to higher interest rates. Most economic models share this assumption - they predict that an acceleration in productivity growth will raise the rate of return on capital and incentivize households to save less in anticipation of faster income gains.3 Both factors should cause interest rates to rise. The problem is that these theories do not accord with the data. Chart 8 shows that interest rates are far higher in regions such as Africa and Latin America, which have historically suffered from chronically weak productivity growth. In contrast, rates are lower in regions such as East Asia, which have experienced rapid productivity growth. One sees the same negative correlation between interest rates and productivity growth over time in developed economies. In the U.S., for example, interest rates rose rapidly during the 1970s, a decade when productivity growth fell sharply (Chart 9). Chart 8 Chart 9U.S. Interest Rates Soared In The 1970s##BR##While Productivity Swooned U.S. Interest Rates Soared In The 1970s While Productivity Swooned U.S. Interest Rates Soared In The 1970s While Productivity Swooned Two Reasons Why Slower Productivity Growth May Lead To Higher Interest Rates There are two main reasons why slower productivity growth may lead to higher nominal interest rates over time: Slower productivity growth may eventually lead to higher inflation; Slower productivity growth may deplete national savings, thereby raising the neutral real rate of interest. We discuss each reason in turn. Reason #1: Slower Productivity Growth May Fuel Inflation Chart 10The Fed Continuously Overstated The Magnitude##BR##Of Economic Slack In The 1970s The Fed Continuously Overstated The Magnitude Of Economic Slack In The 1970s The Fed Continuously Overstated The Magnitude Of Economic Slack In The 1970s Most economists agree that chronically weak productivity growth tends to be associated with higher inflation. Even Janet Yellen acknowledged as much, noting in a 2005 speech that "the evidence suggests that the predominant medium-term effect of a slowdown in trend productivity growth would likely be higher inflation."4 In theory, the causation between productivity and inflation can run in either direction: Weak productivity gains can fuel inflation while high inflation can, in turn, undermine growth. With respect to the latter, economists have focused on three channels: First, higher inflation may make it difficult for firms to distinguish between relative and absolute price shocks, leading to suboptimal resource allocation. Second, higher inflation may stymie capital accumulation because investors typically pay capital gains taxes even when the increase in asset values is entirely due to inflation. Third, high inflation may cause households and firms to waste time and effort on economizing their cash holdings. There are also several ways in which slower productivity growth can lead to higher inflation. For example, sluggish productivity growth may increase the likelihood that a country will be forced to inflate its way out of any debt problems. In addition, central banks may fail to recognize structural declines in productivity growth in real time, leading them to keep interest rates too low in the errant belief that weak GDP growth is due to inadequate demand when, in fact, it is due to insufficient supply. There is strong evidence that this happened in the U.S. in the 1970s. Chart 10 shows that the Fed consistently overestimated the size of the output gap during that period. Reason #2: Slower Productivity Growth May Deplete National Savings, Leading To A Higher Neutral Real Rate Imagine that you have a career where your real income is projected to grow by 2% per year, but then something auspicious happens that leads you to revise your expected annual income growth to 20%. How do you react? If you are like most people, your initial inclination might be to celebrate by purchasing a new car or treating yourself to a lavish vacation. As such, your saving rate is likely to fall at the outset. However, as the income gains pile up, you might find yourself running out of stuff to buy, resulting in a higher saving rate. This is particularly likely to be true if you grew up poor and have not yet acquired a taste for conspicuous consumption. Now consider the opposite case: One where you realize that your income will slowly contract over time as your skills become increasingly obsolete. The logic above suggests that your immediate reaction will be to hunker down and spend less - in other words, your saving rate will rise. However, as time goes by and the roof needs to be changed and the kids sent off to college, you may find it hard to pay the bills - your saving rate will then fall. The same reasoning applies to economy-wide productivity growth. When productivity growth increases, household savings are likely to decline as consumers spend more in anticipation of higher incomes. Meanwhile, investment is likely to rise as firms move swiftly to expand capacity to meet rising demand for their products. The combination of falling savings and rising investment will cause real rates to increase. As time goes by, however, it may become increasingly difficult for the economy to generate enough incremental demand to keep up with rising productive capacity. At that point, real rates will begin falling. The historic evidence is consistent with the notion that higher productivity growth causes savings to fall at the outset, but rise later on. Chart 11 shows that East Asian economies all had rapid growth rates before they had high saving rates. China is a particularly telling example. Chinese productivity growth took off in the early 1990s. Inflation accelerated over the subsequent years, while the country flirted with current account deficits - both telltale signs of excess demand. It was not until a decade later that the saving rate took off, pushing the current account into a large surplus, even though investment was also rising at the time (Chart 12). Chart 11Asian Tigers: Growth Took Off First,##BR##Followed By Higher Savings Asian Tigers: Growth Took Off First, Followed By Higher Savings Asian Tigers: Growth Took Off First, Followed By Higher Savings Chart 12China: Productivity Growth Accelerated,##BR##Then Savings Rate Took Off China: Productivity Growth Accelerated, Then Savings Rate Took Off China: Productivity Growth Accelerated, Then Savings Rate Took Off Today, Chinese deposit rates are near rock-bottom levels, and yet the household sector continues to save like crazy. This will change over time. The working-age population has peaked (Chart 13). As millions of Chinese workers retire and begin to dissave, aggregate household savings will fall. Meanwhile, Chinese youth today have no direct memory of the hardships that their parents endured. As happened in Korea and Japan, the flowering of a consumer culture will help bring down the saving rate. Meanwhile, sluggish income growth in the developed world will make it difficult for households to save much. Population aging will only exacerbate this effect. As my colleague Mark McClellan pointed out in last month's edition of the Bank Credit Analyst, elderly people in advanced economies consume more than any other age cohort once government spending for medical care on their behalf is taken into account (Chart 14).5 Our estimates suggest that population aging will reduce the household saving rate by five percentage points in the U.S. over the next 15 years (Chart 15). The saving rate could fall as much as ten points in Germany, leading to the evaporation of the country's mighty current account surplus. As saving rates around the world begin to fall, real interest rates will rise. Chart 13China's Very High Rate Of National Savings Will Face Pressure From Demographics China's Very High Rate Of National Savings Will Face Pressure From Demographics China's Very High Rate Of National Savings Will Face Pressure From Demographics Chart 14 Chart 15Aging Will Reduce##BR##Aggregate Savings Aging Will Reduce Aggregate Savings Aging Will Reduce Aggregate Savings The Two Reasons Reinforce Each Other The discussion above has focused on two reasons why chronically low productivity growth could lead to higher interest rates: 1) weak productivity growth could fuel inflation; and 2) weak productivity growth could deplete national savings, leading to higher real rates. There is an important synergy between these two reasons. Suppose, for example, that weak productivity growth does eventually raise the neutral real rate. Since central banks cannot measure the neutral rate directly and monetary policy affects the economy with a lag, it is possible that actual rates will end up below the neutral rate. This would cause the economy to overheat, resulting in higher inflation. Thus, if the first reason proves to be true, it is more likely that the second reason will prove to be true as well. The Technological Wildcard So far, we have discussed productivity growth in very generic terms - as basically anything that raises output-per-hour. In reality, the source of productivity gains can have a strong bearing on interest rates. Economists describe innovations that raise the demand for labor relative to capital goods as being "capital saving." Paul David and Gavin Wright have argued that the widespread adoption of electrically-powered processes in the early 20th century serves as "a textbook illustration of capital-saving technological growth."6 They note that "Electrification saved fixed capital by eliminating heavy shafts and belting, a change that also allowed factory buildings themselves to be more lightly constructed." In contrast, recent technological innovations have tended to be more of the "labor saving" than "capital saving" variety. Robotics and AI come to mind, but so do more mundane advances such as containerization. Marc Levinson has contended that the widespread adoption of "The Box" in the 1970s completely revolutionized international trade. Nowadays, huge cranes move containers off ships and place them onto waiting trucks or trains. Thus, the days when thousands of longshoremen toiled in the great ports of Baltimore and Long Beach are gone.7 If technological progress is driven by labor-saving innovations, real wages will tend to grow more slowly than overall productivity (Chart 16). In fact, if technological change is sufficiently biased in favour of capital (i.e., if it is extremely "labor saving"), real wages may actually decline in absolute terms (Chart 17). Owners of capital tend to be wealthier than workers. Since richer people save more of their income than poorer people, the shift in income towards the former will depress aggregate demand (Chart 18). This will result in a lower neutral rate. Chart 16U.S.: Real Wages Have Been##BR##Lagging Productivity Gains U.S.: Real Wages Have Been Lagging Productivity Gains U.S.: Real Wages Have Been Lagging Productivity Gains Chart 17 Chart 18Savings Heavily Skewed##BR##Towards Top Earners Savings Heavily Skewed Towards Top Earners Savings Heavily Skewed Towards Top Earners It is difficult to know if the forces described above will dissipate over time. Productivity growth is largely a function of technological change. We like to think that we are living in an era of unprecedented technological upheavals, but if productivity growth has slowed, it is likely that the pace of technological innovation has also diminished. If so, the impact that technological change is having on such things as the distribution of income and global savings - and by extension on interest rates - could become more muted. To use an analogy, the music might remain the same, but the volume from the speakers could still drop. Capital In A Knowledge-Based Economy Labor-saving technological change has not been the only force pushing down interest rates. Modern economies are transitioning away from producing goods towards producing knowledge. Companies such as Google, Apple, and Amazon have thrived without having to undertake massive amounts of capital spending. This has left them with billions of dollars in cash on their balance sheets. The price of capital goods has also tumbled over the past three decades, allowing companies to cut their capex budgets (Chart 19). In addition, technological advances have facilitated the emergence of "winner-take-all" industries where scale and network effects allow just a few companies to rule the roost (Chart 20). Such market structures exacerbate inequality by shifting income into the hands of a few successful entrepreneurs and business executives. As noted above, this leads to higher aggregate savings. Market structures of this sort could also lead to less aggregate investment because low profitability tends to constrain capital spending by second- or third-tier firms, while the worry that expanding capacity will erode profit margins tends to constrain spending by winning companies. The combination of higher savings and decreased investment results in a lower neutral rate. As with labor-saving technological change, it is difficult to know how these forces will evolve over time. The growth of winner-take-all industries has benefited greatly from globalization. Globalization, however, may be running out of steam. Tariffs are already extremely low in most countries, while the gains from further breaking down the global supply chain are reaching diminishing returns (Chart 21). Perhaps more importantly, political pressures for greater income distribution, trade protectionism, and stronger anti-trust measures are likely to intensify. If that happens, it may be enough to reverse some of the downward pressure on the neutral rate. Chart 19Falling Capital Goods Prices Have Allowed Companies To Slash Capex Budgets Falling Capital Goods Prices Have Allowed Companies To Slash Capex Budgets Falling Capital Goods Prices Have Allowed Companies To Slash Capex Budgets Chart 20 Chart 21The Low-Hanging Fruits Of##BR##Globalization Have Been Picked The Low-Hanging Fruits Of Globalization Have Been Picked The Low-Hanging Fruits Of Globalization Have Been Picked Investment Conclusions Is slow productivity growth good or bad for bonds? The answer is both: Slow productivity growth is likely to depress interest rates at the outset, but is liable to lead to higher rates later on. Chart 22Output Gap Has Narrowed##BR##Thanks To Lower Potential Growth Output Gap Has Narrowed Thanks To Lower Potential Growth Output Gap Has Narrowed Thanks To Lower Potential Growth The U.S. has likely reached the inflection point where slow productivity is going from being a boon to a bane for bonds. Chart 22 shows that the U.S. output gap would be over 8% of GDP had potential GDP grown at the pace the IMF projected back in 2008. Instead, it is close to zero and will likely turn negative if growth remains over 2% over the next few quarters. Other countries are likely to follow in the footsteps of the U.S. To be clear, productivity is just one of several factors affecting interest rates - demographics, globalization, and political decisions being others. However, as we argued in our latest Strategy Outlook, these forces are also shifting in a more inflationary direction.8 As such, fixed-income investors with long-term horizons should pare back duration risk and increase allocations to inflation-linked securities. Peter Berezin, Chief Global Strategist Global Investment Strategy peterb@bcaresearch.com 1 Please see Global Investment Strategy Special Report, "Weak Productivity Growth: Don't Blame The Statisticians," dated March 25, 2016, available at gis.bcaresearch.com. 2 Dan Andrews, Chiara Criscuolo, and Peter N. Gal,"The Best versus the Rest: The Global Productivity Slowdown, Divergence across Firms and the Role of Public Policy," OECD Productivity Working Papers, No. 5 (November 2016). 3 Consider the widely-used Solow growth model. The model says that the neutral real rate, r, is equal to (a/s) (n + g + d), where a is the capital share of income, s is the saving rate, n is labor force growth, g is total factor productivity growth, and d is the depreciation rate of capital. All things equal, an increase in g will result in a higher equilibrium real interest rate. The same is true in the Ramsey model, which goes a step further and endogenizes the saving rate within a fully specified utility-maximization framework. In this model, consumption growth is pinned down by the so-called Euler equation. Assuming that utility can be described by a constant relative risk aversion utility function, the Euler equation states that consumption will grow at (r-d)/h where d is the rate at which households discount future consumption and h is a measure of the degree to which households want to smooth consumption over time. In a steady state, consumption increases at the same rate as GDP, n+g. Rearranging the terms yields: r=(n+g)h+d. Notice that both models provide a mechanism by which a higher g can decrease r. In the Solow model, this comes from thinking about the saving rate not as an exogenous variable, but as something that can be influenced by the growth rate of the economy. In particular, if s rises in response to a higher g, r could fall. Likewise, in the Ramsey model, a higher g could make households more willing to forgo consumption today in return for higher consumption tomorrow (equivalent to a decrease in the rate of time preference, d). This, too, would translate into a lower neutral rate. 4 Janet L. Yellen, "The U.S. Economic Outlook," Presentation to the Stanford Institute of Economic Policy Research, February 11, 2005. 5 Please see The Bank Credit Analyst, "Beware Inflection Points In The Secular Drivers Of Global Bonds," April 28, 2017, available at bca.bcaresearch.com. 6 Paul A. David, and Gavin Wright,"General Purpose Technologies And Surges In Productivity: Historical Reflections On the Future Of The ICT Revolution," January 2012. 7 Marc Levinson, "The Box: How the Shipping Container Made the World Smaller and the World Economy Bigger," Princeton University Press, 2006. 8 Please see Global Investment Strategy, "Strategy Outlook Second Quarter 2017: A Three-Act Play," dated March 31, 2017, available at gis.bcaresearch.com. APPENDIX: Tactical Global Asset Allocation Monthly Update To complement our analysis and intuition, we use a variety of time-tested models to assess the global investment outlook. Compared to last month, our tactical (3-month) model is recommending an upgrade to global equities at the expense of government bonds. Global equities have consolidated their gains, removing some of the overbought conditions that prevailed earlier in May. Bullish equity sentiment has also waned somewhat, while net speculative positioning in U.S. stocks has moved from being net long to net short. In contrast, speculative positioning in Treasurys has jumped into net long territory (Chart A1). Our models say that government bonds in most economies remain overbought. Image Regionally, we continue to favor higher-beta developed equity markets such as Europe and Japan. Canada, Australia, and most emerging markets have also received an upgrade, owing to a more favorable near-term outlook for commodity prices. Within the bond universe, U.S. Treasurys are most vulnerable to a selloff, given that the market is pricing in only two rate hikes over the next 12 months. Image Strategy & Market Trends Tactical Trades Strategic Recommendations Closed Trades

This month's Special Report was written by Peter Berezin, Chief Global Strategist for BCA's Global Investment Strategy Service. The report is a companion piece to last month's Special Report, which argued that some of the structural factors that have depressed global interest rates are at an inflection point. These factors include demographic trends and the integration of China's massive labor supply into the global economy. Peter's report focuses on technology's impact on bond yields. He presents the non-consensus view that slow productivity growth likely depresses interest rates at the outset, but will lead to higher rates later on. Not only could sluggish productivity growth lead to higher inflation, it could also deplete national savings. Both factors would be bond bearish, reinforcing the other factors discussed in last month's Special Report. I trust that you will find the report as insightful and educational as I did. Mark McClellan Productivity growth has declined in most countries. This appears to be a structural problem that will remain with us for years to come. In theory, slower productivity growth should reduce the neutral rate of interest, benefiting bonds in the process. In reality, countries with chronically low productivity growth typically have higher interest rates than faster growing economies. The passage of time helps account for this seeming paradox: Slower productivity growth tends to depress interest rates at the outset, but leads to higher rates later on. The U.S. has reached an inflection point where weak productivity growth is starting to push up both the neutral real rate and inflation. Other countries will follow. The implication for investors is that government bond yields have begun a long-term secular uptrend. The market is not at all prepared for this. Slow Productivity Growth: A Structural Problem Productivity growth has fallen sharply in most developed and emerging economies (Chart II-1). As we argued in "Weak Productivity Growth: Don't Blame The Statisticians," there is little compelling evidence that measurement error explains the productivity slowdown.1 Yes, the unmeasured utility accruing from free internet services is large, but so was the unmeasured utility from antibiotics, indoor plumbing, and air conditioning. No one has offered a convincing explanation for why the well-known problems with productivity calculations suddenly worsened about 12 years ago. Chart II-1 If mismeasurement is not responsible for the productivity slowdown, what is? Cyclical factors have undoubtedly played a role. In particular, lackluster investment spending has curtailed the growth in the capital stock (Chart II-2). This means that today's workers have not benefited from the improvement in the quality and quantity of capital to the same extent as previous generations. However, the timing of the productivity slowdown - it began in 2004-05 in most countries, well before the financial crisis struck - suggests that structural factors have been key. These include: Waning gains from the IT revolution. Recent innovations have focused more on consumers than businesses. As nice as Facebook and Instagram are, they do little to boost business productivity - in fact, they probably detract from it, given how much time people waste on social media these days. The rising share of value added coming from software relative to hardware has also contributed to the decline in productivity growth. Chart II-3 shows that productivity gains in the latter category have been much smaller than in the former. Chart II-2The Great Recession Hit ##br##Capital Stock Accumulation The Great Recession Hit Capital Stock Accumulation The Great Recession Hit Capital Stock Accumulation Chart II-3The Shift Towards Software Has ##br##Dampened IT Productivity Gains The Shift Towards Software Has Dampened IT Productivity Gains The Shift Towards Software Has Dampened IT Productivity Gains Slower human capital accumulation. Globally, the fraction of adults with a secondary degree or higher is increasing at half the pace it did in the 1990s (Chart II-4). Educational achievement, as measured by standardized test scores in mathematics and science, is edging lower in the OECD, and is showing very limited gains in most emerging markets (Chart II-5). Test scores tend to be much lower in countries with rapidly growing populations (Chart II-6). Consequently, the average level of global mathematical proficiency is now declining for the first time in modern history. Chart II-4 Chart II-5 Chart II-6 Decreased creative destruction. The birth rate of new firms in the U.S. has fallen by half since the late 1970s and is now barely above the death rate (Chart II-7). In addition, many firms in advanced economies are failing to replicate the best practices of industry leaders. The OECD reckons that this has been a key reason for the productivity slowdown.2 Chart II-7Secular Decline In U.S. Firm Births Secular Decline In U.S. Firm Births Secular Decline In U.S. Firm Births Productivity Growth And Interest Rates Investors typically assume that long-term interest rates will converge to nominal GDP growth. All things equal, this implies that faster productivity growth should lead to higher interest rates. Most economic models share this assumption - they predict that an acceleration in productivity growth will raise the rate of return on capital and incentivize households to save less in anticipation of faster income gains.3 Both factors should cause interest rates to rise. The problem is that these theories do not accord with the data. Chart II-8 shows that interest rates are far higher in regions such as Africa and Latin America, which have historically suffered from chronically weak productivity growth. In contrast, rates are lower in regions such as East Asia, which have experienced rapid productivity growth. One sees the same negative correlation between interest rates and productivity growth over time in developed economies. In the U.S., for example, interest rates rose rapidly during the 1970s, a decade when productivity growth fell sharply (Chart II-9). Chart II-8 Chart II-9U.S. Interest Rates Soared In The ##br##1970s While Productivity Swooned U.S. Interest Rates Soared In The 1970s While Productivity Swooned U.S. Interest Rates Soared In The 1970s While Productivity Swooned Two Reasons Why Slower Productivity Growth May Lead To Higher Interest Rates There are two main reasons why slower productivity growth may lead to higher nominal interest rates over time: Slower productivity growth may eventually lead to higher inflation; Slower productivity growth may deplete national savings, thereby raising the neutral real rate of interest. We discuss each reason in turn. Reason #1: Slower Productivity Growth May Fuel Inflation Most economists agree that chronically weak productivity growth tends to be associated with higher inflation. Even Janet Yellen acknowledged as much, noting in a 2005 speech that "the evidence suggests that the predominant medium-term effect of a slowdown in trend productivity growth would likely be higher inflation."4 Chart II-10The Fed Continuously Overstated The ##br##Magnitude Of Economic Slack In The 1970s The Fed Continuously Overstated The Magnitude Of Economic Slack In The 1970s The Fed Continuously Overstated The Magnitude Of Economic Slack In The 1970s In theory, the causation between productivity and inflation can run in either direction: Weak productivity gains can fuel inflation while high inflation can, in turn, undermine growth. With respect to the latter, economists have focused on three channels: First, higher inflation may make it difficult for firms to distinguish between relative and absolute price shocks, leading to suboptimal resource allocation. Second, higher inflation may stymie capital accumulation because investors typically pay capital gains taxes even when the increase in asset values is entirely due to inflation. Third, high inflation may cause households and firms to waste time and effort on economizing their cash holdings. There are also several ways in which slower productivity growth can lead to higher inflation. For example, sluggish productivity growth may increase the likelihood that a country will be forced to inflate its way out of any debt problems. In addition, central banks may fail to recognize structural declines in productivity growth in real time, leading them to keep interest rates too low in the errant belief that weak GDP growth is due to inadequate demand when, in fact, it is due to insufficient supply. There is strong evidence that this happened in the U.S. in the 1970s. Chart II-10 shows that the Fed consistently overestimated the size of the output gap during that period. Reason #2: Slower Productivity Growth May Deplete National Savings, Leading To A Higher Neutral Real Rate Imagine that you have a career where your real income is projected to grow by 2% per year, but then something auspicious happens that leads you to revise your expected annual income growth to 20%. How do you react? If you are like most people, your initial inclination might be to celebrate by purchasing a new car or treating yourself to a lavish vacation. As such, your saving rate is likely to fall at the outset. However, as the income gains pile up, you might find yourself running out of stuff to buy, resulting in a higher saving rate. This is particularly likely to be true if you grew up poor and have not yet acquired a taste for conspicuous consumption. Now consider the opposite case: One where you realize that your income will slowly contract over time as your skills become increasingly obsolete. The logic above suggests that your immediate reaction will be to hunker down and spend less - in other words, your saving rate will rise. However, as time goes by and the roof needs to be changed and the kids sent off to college, you may find it hard to pay the bills - your saving rate will then fall. The same reasoning applies to economy-wide productivity growth. When productivity growth increases, household savings are likely to decline as consumers spend more in anticipation of higher incomes. Meanwhile, investment is likely to rise as firms move swiftly to expand capacity to meet rising demand for their products. The combination of falling savings and rising investment will cause real rates to increase. As time goes by, however, it may become increasingly difficult for the economy to generate enough incremental demand to keep up with rising productive capacity. At that point, real rates will begin falling. The historic evidence is consistent with the notion that higher productivity growth causes savings to fall at the outset, but rise later on. Chart II-11 shows that East Asian economies all had rapid growth rates before they had high saving rates. China is a particularly telling example. Chinese productivity growth took off in the early 1990s. Inflation accelerated over the subsequent years, while the country flirted with current account deficits - both telltale signs of excess demand. It was not until a decade later that the saving rate took off, pushing the current account into a large surplus, even though investment was also rising at the time (Chart II-12). Chart II-11Asian Tigers: Growth Took Off First, ##br##Followed By Higher Savings Asian Tigers: Growth Took Off First, Followed By Higher Savings Asian Tigers: Growth Took Off First, Followed By Higher Savings Chart II-12China: Productivity Growth Accelerated, ##br##Then Savings Rate Took Off China: Productivity Growth Accelerated, Then Savings Rate Took Off China: Productivity Growth Accelerated, Then Savings Rate Took Off Today, Chinese deposit rates are near rock-bottom levels, and yet the household sector continues to save like crazy. This will change over time. The working-age population has peaked (Chart II-13). As millions of Chinese workers retire and begin to dissave, aggregate household savings will fall. Meanwhile, Chinese youth today have no direct memory of the hardships that their parents endured. As happened in Korea and Japan, the flowering of a consumer culture will help bring down the saving rate. Meanwhile, sluggish income growth in the developed world will make it difficult for households to save much. Population aging will only exacerbate this effect. As my colleague Mark McClellan pointed out in last month's edition of the Bank Credit Analyst, elderly people in advanced economies consume more than any other age cohort once government spending for medical care on their behalf is taken into account (Chart II-14).5 Our estimates suggest that population aging will reduce the household saving rate by five percentage points in the U.S. over the next 15 years (Chart II-15). The saving rate could fall as much as ten points in Germany, leading to the evaporation of the country's mighty current account surplus. As saving rates around the world begin to fall, real interest rates will rise. Chart II-13China's Very High Rate Of National Savings ##br##Will Face Pressure From Demographics China's Very High Rate Of National Savings Will Face Pressure From Demographics China's Very High Rate Of National Savings Will Face Pressure From Demographics Chart II-14 Chart II-15Aging Will Reduce ##br##Aggregate Savings Aging Will Reduce Aggregate Savings Aging Will Reduce Aggregate Savings The Two Reasons Reinforce Each Other The discussion above has focused on two reasons why chronically low productivity growth could lead to higher interest rates: 1) weak productivity growth could fuel inflation; and 2) weak productivity growth could deplete national savings, leading to higher real rates. There is an important synergy between these two reasons. Suppose, for example, that weak productivity growth does eventually raise the neutral real rate. Since central banks cannot measure the neutral rate directly and monetary policy affects the economy with a lag, it is possible that actual rates will end up below the neutral rate. This would cause the economy to overheat, resulting in higher inflation. Thus, if the first reason proves to be true, it is more likely that the second reason will prove to be true as well. The Technological Wildcard So far, we have discussed productivity growth in very generic terms - as basically anything that raises output-per-hour. In reality, the source of productivity gains can have a strong bearing on interest rates. Economists describe innovations that raise the demand for labor relative to capital goods as being "capital saving." Paul David and Gavin Wright have argued that the widespread adoption of electrically-powered processes in the early 20th century serves as "a textbook illustration of capital-saving technological growth."6 They note that "Electrification saved fixed capital by eliminating heavy shafts and belting, a change that also allowed factory buildings themselves to be more lightly constructed." In contrast, recent technological innovations have tended to be more of the "labor saving" than "capital saving" variety. Robotics and AI come to mind, but so do more mundane advances such as containerization. Marc Levinson has contended that the widespread adoption of "The Box" in the 1970s completely revolutionized international trade. Nowadays, huge cranes move containers off ships and place them onto waiting trucks or trains. Thus, the days when thousands of longshoremen toiled in the great ports of Baltimore and Long Beach are gone.7 If technological progress is driven by labor-saving innovations, real wages will tend to grow more slowly than overall productivity (Chart II-16). In fact, if technological change is sufficiently biased in favour of capital (i.e., if it is extremely "labor saving"), real wages may actually decline in absolute terms (Chart II-17). Owners of capital tend to be wealthier than workers. Since richer people save more of their income than poorer people, the shift in income towards the former will depress aggregate demand (Chart II-18). This will result in a lower neutral rate. Chart II-16U.S.: Real Wages Have Been ##br##Lagging Productivity Gains U.S.: Real Wages Have Been Lagging Productivity Gains U.S.: Real Wages Have Been Lagging Productivity Gains Chart II-17 Chart II-18Savings Heavily Skewed ##br##Towards Top Earners Savings Heavily Skewed Towards Top Earners Savings Heavily Skewed Towards Top Earners It is difficult to know if the forces described above will dissipate over time. Productivity growth is largely a function of technological change. We like to think that we are living in an era of unprecedented technological upheavals, but if productivity growth has slowed, it is likely that the pace of technological innovation has also diminished. If so, the impact that technological change is having on such things as the distribution of income and global savings - and by extension on interest rates - could become more muted. To use an analogy, the music might remain the same, but the volume from the speakers could still drop. Capital In A Knowledge-Based Economy Labor-saving technological change has not been the only force pushing down interest rates. Modern economies are transitioning away from producing goods towards producing knowledge. Companies such as Google, Apple, and Amazon have thrived without having to undertake massive amounts of capital spending. This has left them with billions of dollars in cash on their balance sheets. The price of capital goods has also tumbled over the past three decades, allowing companies to cut their capex budgets (Chart II-19). In addition, technological advances have facilitated the emergence of "winner-take-all" industries where scale and network effects allow just a few companies to rule the roost (Chart II-20). Such market structures exacerbate inequality by shifting income into the hands of a few successful entrepreneurs and business executives. As noted above, this leads to higher aggregate savings. Market structures of this sort could also lead to less aggregate investment because low profitability tends to constrain capital spending by second- or third-tier firms, while the worry that expanding capacity will erode profit margins tends to constrain spending by winning companies. The combination of higher savings and decreased investment results in a lower neutral rate. Chart II-19Falling Capital Goods Prices Have ##br##Allowed Companies To Slash Capex Budgets Falling Capital Goods Prices Have Allowed Companies To Slash Capex Budgets Falling Capital Goods Prices Have Allowed Companies To Slash Capex Budgets Chart II-20 As with labor-saving technological change, it is difficult to know how these forces will evolve over time. The growth of winner-take-all industries has benefited greatly from globalization. Globalization, however, may be running out of steam. Tariffs are already extremely low in most countries, while the gains from further breaking down the global supply chain are reaching diminishing returns (Chart II-21). Perhaps more importantly, political pressures for greater income distribution, trade protectionism, and stronger anti-trust measures are likely to intensify. If that happens, it may be enough to reverse some of the downward pressure on the neutral rate. Chart II-21The Low-Hanging Fruits Of ##br##Globalization Have Been Picked The Low-Hanging Fruits Of Globalization Have Been Picked The Low-Hanging Fruits Of Globalization Have Been Picked Investment Conclusions Is slow productivity growth good or bad for bonds? The answer is both: Slow productivity growth is likely to depress interest rates at the outset, but is liable to lead to higher rates later on. The U.S. has likely reached the inflection point where slow productivity is going from being a boon to a bane for bonds. Chart II-22 shows that the U.S. output gap would be over 8% of GDP had potential GDP grown at the pace the IMF projected back in 2008. Instead, it is close to zero and will likely turn negative if growth remains over 2% over the next few quarters. Other countries are likely to follow in the footsteps of the U.S. Chart II-22Output Gap Has Narrowed Thanks ##br##To Lower Potential Growth Output Gap Has Narrowed Thanks To Lower Potential Growth Output Gap Has Narrowed Thanks To Lower Potential Growth To be clear, productivity is just one of several factors affecting interest rates - demographics, globalization, and political decisions being others. However, as we argued in our latest Strategy Outlook, these forces are also shifting in a more inflationary direction.8 As such, fixed-income investors with long-term horizons should pare back duration risk and increase allocations to inflation-linked securities. Peter Berezin, Chief Global Strategist Global Investment Strategy 1 Please see Global Investment Strategy Special Report, "Weak Productivity Growth: Don't Blame The Statisticians," dated March 25, 2016, available at gis.bcaresearch.com. 2 Dan Andrews, Chiara Criscuolo, and Peter N. Gal,"The Best versus the Rest: The Global Productivity Slowdown, Divergence across Firms and the Role of Public Policy," OECD Productivity Working Papers, No. 5 (November 2016). 3 Consider the widely-used Solow growth model. The model says that the neutral real rate, r, is equal to (a/s) (n + g + d), where a is the capital share of income, s is the saving rate, n is labor force growth, g is total factor productivity growth, and d is the depreciation rate of capital. All things equal, an increase in g will result in a higher equilibrium real interest rate. The same is true in the Ramsey model, which goes a step further and endogenizes the saving rate within a fully specified utility-maximization framework. In this model, consumption growth is pinned down by the so-called Euler equation. Assuming that utility can be described by a constant relative risk aversion utility function, the Euler equation states that consumption will grow at (r-d)/h where d is the rate at which households discount future consumption and h is a measure of the degree to which households want to smooth consumption over time. In a steady state, consumption increases at the same rate as GDP, n+g. Rearranging the terms yields: r=(n+g)h+d. Notice that both models provide a mechanism by which a higher g can decrease r. In the Solow model, this comes from thinking about the saving rate not as an exogenous variable, but as something that can be influenced by the growth rate of the economy. In particular, if s rises in response to a higher g, r could fall. Likewise, in the Ramsey model, a higher g could make households more willing to forgo consumption today in return for higher consumption tomorrow (equivalent to a decrease in the rate of time preference, d). This, too, would translate into a lower neutral rate. 4 Janet L. Yellen, "The U.S. Economic Outlook," Presentation to the Stanford Institute of Economic Policy Research, February 11, 2005. 5 Please see The Bank Credit Analyst, "Beware Inflection Points In The Secular Drivers Of Global Bonds," April 28, 2017, available at bca.bcaresearch.com. 6 Paul A. David, and Gavin Wright,"General Purpose Technologies And Surges In Productivity: Historical Reflections On the Future Of The ICT Revolution," January 2012. 7 Marc Levinson, "The Box: How the Shipping Container Made the World Smaller and the World Economy Bigger," Princeton University Press, 2006. 8 Please see Global Investment Strategy, "Strategy Outlook Second Quarter 2017: A Three-Act Play," dated March 31, 2017, available at gis.bcaresearch.com.