Sorry, you need to enable JavaScript to visit this website.
Skip to main content
Skip to main content

Bubbles

Are you sure you are in a trade war-induced selloff and/or recession? Or, is America – writ large – the bubble? Fed a steady dose of fiscal profligacy over the past five years, the US economy and its various associated assets have become fat and complacent. As such, the end of the fiscal gravy train – our signature macro call for 2025 – is what is causing the underlying weakness, with the Big Tech – the biggest beneficiary of US Exceptionalism – now buckling.

Our message? The tech bros are lying to you. The AI revolution is not positive for Big Tech, not at their current valuation. Play the rotation, even if a (mild) recession may be afoot, as we have been saying throughout 2024.

In this report, we reassess our bullish stance on crypto from early 2023, following Bitcoin’s recent all-time highs. While institutional adoption is broadening, there are also signs of excessive exuberance, speculation, and optimism. Given these conditions, a near-term correction appears likely. We are booking profits and will look to re-enter the market at $75,000.

Over the past few weeks, global equities have been hit by rising scepticism over the bullish AI narrative and increasing concerns over global growth. Stocks should stabilize in the near term, but the medium-term direction is to the downside. We expect the S&P 500 to drop to 3750 in 2025 and the 10-year Treasury yield to fall to 3%.

Special Report Highlights Investors who are optimistic about the potential for artificial intelligence (AI) to impact economic growth have several bullish private sector estimates to point to. At the same time, other credible estimates point to a minimal impact of AI on economic growth. Bullish estimates of AI’s growth potential rely significantly on expectations that AI models will be able to conduct new tasks, and/or a much higher share of existing tasks that can be profitably performed by AI than existing studies show. The information technology revolution of the 1990s appears to be the most obvious comparable episode to the potential for generative AI. The IT revolution boosted real potential growth by roughly a percentage point for a few years, which we regard as a very high-end estimate for AI’s possible economic impact. We doubt, however, that AI will end up truly boosting economic activity by this magnitude. We model the fair value of a sustained 1% boost to real output for the US equity market using a discounted cash flow (DCF) approach. Using a single-stage DCF model with either a 10- or 20-year time horizon, we find that a 1% improvement in real GDP growth is worth between $3 and $10 trillion for the corporate sector, depending on the margin and multiple assumed. US growth stocks have seen a $4.3 trillion increase in market capitalization since late 2022 from multiple expansion, and the broad market has seen a $7 trillion increase. This suggests that the US equity market is significantly overvalued, unless the deployment of AI technology causes a 10-to-20 year productivity surge in line with what occurred during the IT revolution of the 1990s, with persistently high margins on the revenue generated from the improvement in growth. A “sudden-stop” shift in sentiment about AI is certainly a possible trigger that could pop the bubble in growth stocks, but funding will not be a factor like it was in the late-1990s. A recession remains the most likely trigger for the AI bubble to burst. The bottom line for investors is that the US equity market may continue to rise over the coming few months, potentially strongly, until signs of a recession are unambiguous. But the market’s extremely optimistic expectations about AI’s impact on growth underscore that the US equity market selloff during the next recession is likely to be outsized relative to the impact on employment and GDP growth. Since ChatGPT was released by OpenAI on November 30, 2022, the S&P 500 index has risen close to 38 percent in total return terms (Chart II-1). By contrast, the S&P 500 equal weight index has risen by about 14 percent, which underscores how significantly optimistic expectations about AI have impacted the US equity market. While it is true that earnings have risen over the past year and a half because the economy has continued to expand, close to two thirds of the rise in the overall US equity market since the release of ChatGPT has come from multiple expansion. Chart II-1The "Generative AI Effect" Has Accounted For Over Half Of US Equity Market Returns Over The Past 18 Months In this report, we examine both estimates of AI’s potential to impact economic activity, and what those estimates might mean for the fair value of US stocks. We find that the US equity market is significantly overvalued unless the deployment of AI technology causes a 10-to-20 year productivity surge in line with what occurred during the IT revolution of the 1990s, with persistently high margins on the revenue generated from the improvement in growth. While that is technically a possible outcome, we regard it as the most realistic “best case scenario”, and we doubt that AI will end up truly boosting economic activity by this magnitude. It is possible that the AI bubble will burst before the next US recession, but that would likely depend on a major shift in sentiment about AI, which is extremely difficult to predict. A late-1990s-style “funding shortfall” will not be a catalyst this time around. As such, the US equity market may continue to rise over the coming few months, potentially strongly, until signs of a recession are unambiguous. But once the recession arrives, the extremely optimistic expectations about AI’s impact on growth underscore that the selloff in US stocks will likely be outsized relative to the impact of the recession on employment and GDP growth. Estimating AI’s Potential To Boost Growth: The View Of Others Investors who are optimistic about the potential for artificial intelligence (AI) to impact economic growth have several bullish private sector estimates to point to (Table II-1). For example, Goldman Sachs' baseline is that generative AI will boost aggregate labor productivity growth by 1.5 percentage points per year in the US over a ten-year period.1 The Brookings Institute has published estimates of a potential 1.8% annual improvement in labor productivity over either ten or twenty years.2 And McKinsey has argued that the automation of individual work activities enabled by AI and other technologies could provide the global economy with an annual productivity boost of between 0.5% and 3.4% until 2040, with generative AI contributing 0.1 to 0.6 percentage points of that growth.3 Table II-1Private Sector AI Growth Impact Estimates Are Very Bullish All of these estimates come with a variety of assumptions, but they are not the most optimistic estimates available. Goldman Sachs presents additional possible scenarios which could raise labor productivity growth by anywhere between 2.5 and 3 percentage points per year. On the other hand, a recent paper by Daron Acemoglu presented a much less optimistic view of the potential growth impact of AI.4 Acemoglu is a well-known economist whose research often focuses on the structural drivers of growth, as well as the impact of technological change. His paper concludes that predicted real output gains over the next 10 years from AI advancements are likely to be less than one percent in total over a ten-year period. The gap between Acemoglu’s estimates and those of other major organizations such as Brookings / McKinsey / Goldman Sachs lies, first, in the fact that Acemoglu only considered the potential for AI to impact existing tasks, rather than the growth impact of new tasks. Using Hulten’s theorem, which posits that the macro impact of microeconomic shocks can be estimated by what fraction of tasks are impacted and average task-level cost savings, Acemoglu arrives at his modest AI impact estimates through the following calculations: 20% of US labor tasks are exposed to AI, based on estimates from Eloundou et al. (2023)5 Among exposed tasks, 23% can be profitably performed by AI based on estimates from Svanberg et al (2024)6 Based on average time improvement studies, average labor costs savings from deploying AI are 27%, which are translated into overall cost savings of 14.4% using industry labor shares Applying this approach yields a total factor productivity boost of AI’s impact on existing tasks of no more than 0.66% (cumulative) over 10 years (20% times 23% times 14.4%). Acemoglu argues that even this is likely to be overstated because existing estimates of productivity gains and costs savings have been based on “easy-to-learn” tasks for AI, whereas “harder-to-learn” tasks would entail lower productivity enhancements. Acemoglu arrives at a final upper bound estimate of 0.53% total rise in total factor productivity (5.3 basis points per year), and a 0.9% total rise in real GDP (9 basis points per year). The GDP impact is higher than the TFP improvement because of an assumed improvement in the contribution of capital intensity to labor productivity. Assuming no change in Acemoglu’s average labor cost savings estimates and assuming an unchanged labor share estimate, Table II-2 shows how Acemoglu’s estimates of the impact of AI on annualized total factor productivity can be scaled up. The table highlights that the TFP impact implied by a Hulten’s Theorem approach falls short of the estimates cited by the major organizations noted above, even when we assume that most labor tasks are exposed to AI and that more than half of those tasks can be profitably employed by AI models. As such, investors should recognize that the AI growth potential estimates provided by several organizations appear to rely significantly on expectations that AI models will be able to conduct new tasks, and/or a much higher share of existing tasks that can be profitably performed by AI than existing studies show. Table II-2AI Growth Impact Estimates Based Only On Existing Tasks Are Nowhere Near As Bullish Estimating AI’s Potential To Boost Growth: Our Thoughts If Acemoglu’s estimates are too pessimistic because they don't account for AI creating new tasks, and if estimates from major organizations appear to be too optimistic, what should investors use as a benchmark for AI's economic impact? It is very difficult to determine how AI will boost economic activity if most of the gains are likely to come from new tasks, because it requires envisioning and quantifying the effect of tasks that do not yet exist. In our view, without more evidence of the utility and application of AI programs, the most reasonable approach is to identify a comparable episode of technological progress and measure the impact that it had on growth. Over the past 20-to-30 years, investors have witnessed numerous technological advancements, including the following: The rapid deployment of information and communication technology products and infrastructure in the 1990s The commercialization of the internet The development of smartphones The introduction of social media The advent of cryptocurrencies The impressive use of fracking technology to boost US oil output and, The growth of electric vehicles and alternative energy. Of these examples, the information technology revolution of the 1990s appears to be the most obvious comparable episode to the potential for generative AI. It is true that the acceleration in productivity growth that occurred in the 1990s was smaller than that of the last third of the nineteenth century or the 1930s-to-early-1970s period. Table II-3, however, shows that these periods benefitted from truly general-purpose technologies that occurred alongside a significant improvement in the average lifespan. The impact of that improvement on the effectiveness of human capital may not be fully captured in measures of labor composition. We would need to see much more evidence before believing that artificial intelligence could reproduce those kinds of effects. Table II-3The 1990s Was Not The Largest Productivity Surge In History, But Likely The Most Comparable To AI’s Potential Today What caused the IT-driven surge in productivity in the 1990s? Computers had been around for two decades prior to the 1990s, but significant cost reductions made the widespread deployment of IT equipment affordable. Advancements in office productivity software also significantly increased output per hour of service workers. Measuring exactly what the 1990s IT revolution did to economic growth is subject to a few assumptions. On the one hand, overall labor productivity accelerated from 1.6% on average between 1992 and 1995 to 3.2% on average between 2001 and 2005. Indeed, when looking at Chart II-2, it appears as if the acceleration in productivity lasted for almost a decade. Chart II-2Seemingly A Decade-Long Productivity Surge, But Not Really On the other hand, the period from 2001 to 2005 occurred during a recession and a subsequent period of weak corporate profits and employment growth. Chart II-3 shows that the labor productivity improvement that occurred during that period also happened alongside a significant reduction in hours worked, which is why real GDP growth was meaningfully weaker than it was from 1996 to 2000. Chart II-4 shows that the relationship between productivity growth and hours worked has been negative since the mid-1980s, even after accounting for the fact that this relationship appears to be especially negative during recessions. Chart II-3Strong Productivity In The 2000s Occurred For A Different Reason Than In The 1990s Chart II-4Over The Past Forty Years, Productivity Gains Have Not Fed Fully Through To Growth Given that generative AI has substantial potential to displace labor, productivity gains from this technology may not map directly to real GDP growth if there are negative consequences for the labor market. In our view, this reinforces our perspective that optimistic projections about AI’s potential to boost growth are overstated. All told, an analysis of the 1990s IT revolution shows that it boosted real potential growth by roughly a percentage point for a few years. In the case of AI, we think that a 1% annual growth improvement is somewhat reasonable, but we would characterize it as a very high-end estimate. We acknowledge, however, that its benefits may be longer lasting in such a scenario than what occurred in the 1990s. What Are Stocks Pricing In For AI? To gauge what the equity market is pricing in in terms of expectations for AI to boost economic activity and generate substantial corporate profits, we conduct a discounted cash flow (DCF) valuation exercise to determine the fair value of a sustained 1% boost to real output. We essentially regard this as the most realistic “best-case scenario” and we doubt that AI will end up truly boosting economic activity by this magnitude.7 But, if the rise in the equity market since late 2022 is meaningfully lower than the fair value implied by a 1% growth improvement, that could justify further gains for tech-related stocks. Unfortunately for investors who are optimistic about AI, it turns out that that is not the case. We make several assumptions when modeling the likely cash flows from a 1% growth improvement: We assume that risk-free interest rates are equal to nominal trend growth We assign a 2% equity risk premium to the risk-free rate when calculating the discount rate We assume that the corporate sector's share of the output generated is higher than is currently the case, varying between 40% and 60%. This compares to a 40% capital share of output in 2022, according to the BLS’ total factor productivity database. Strictly speaking, a full discounted cash flow approach that assumes an indefinite 1% annual growth improvement would value this development at between $11 and $16 trillion today. However, it is unrealistic to expect any technological development to boost growth every year in perpetuity. Thus, we conduct this valuation exercise using a different approach: a single-stage discounted cash flow model with either a 10- or 20-year time horizon. We calculate a terminal value at the end of the forecast period, employing a generous trailing P/E ratio of between 20 and 35 times profits. This approach values a 1% improvement in real GDP growth for the corporate sector at between $3 and $10 trillion, depending on the margin and multiple assumed. To determine what the equity market is pricing in regarding AI, we use a similar approach to that shown for Microsoft in last month's report, but applied to the US MSCI Growth Index and the overall US equity index. Our method calculates the change in market capitalization since November 2022 that is not attributable to an improvement in actual fundamentals, which we infer is the market cap increase associated with AI growth expectations. While it is true that expectations of a recession have been dialed back since late-2022, Chart II-5 highlights that there was no rise in our equity risk premium proxy that year. That strongly points to AI growth expectations as explaining most of or all of the subsequent rise in equity multiples given that bond yields have not declined meaningfully since. Chart II-5The Rise In Equity Multiples Since Late-2022 Has Occurred Due To Growth Expectations, Not Reduced Recession Risk Our approach highlights that US growth stocks have seen a $4.3 trillion increase in market capitalization since late 2022 from multiple expansion, and that the broad market has seen a $7 trillion increase. Chart II-6 contextualizes those numbers, by presenting them alongside the range of fair value estimates from our DCF approach. We also include a substantial scaling up of the growth impact estimates provided by Acemoglu, on the order of five and ten times the effect. Chart II-6The US Equity Market Is Pricing In A Substantial And Long-Lasting Boost To Growth From AI The chart vividly illustrates that the US equity market is currently priced roughly in line with the fair value of a 1% sustained improvement to real GDP growth and only slightly lower than what a ten-fold increase in Acemoglu’s existing-task estimates would imply. That may seem reasonable to investors with high hopes that AI will revolutionize the global economy but, from our perspective, it underscores a major source of risk to the US equity market. Chart II-6 may be framed in a different fashion: it suggests that the US equity market is significantly overvalued, unless the deployment of AI technology causes a 10-to-20 year productivity surge in line with what occurred during the IT revolution of the 1990s, with persistently high margins on the revenue generated from the improvement in growth. What Will Stop The Mania? If the US stock market is indeed grossly overvalued because of overly optimistic expectations about AI’s potential to boost economic growth, then it is important for investors to identify triggers that might pop the bubble. A starting point is to ask what popped the equity market bubble in the 1990s. The NASDAQ peaked on March 10, 2000, and rising interest rates were clearly a factor. Chart II-7 shows that the fed funds rate rose from 4.75% in the summer of 1999 to 6.5% in May 2000, a 175 basis point increase on what was already quite an elevated level. Chart II-8 shows, however, that funding was also an important element that had been flagging a warning sign well in advance of the peak in the stock market. The chart shows that the number of IPOs actually peaked in 1996, and fell to less than half their peak level in 1998. IPO activity rebounded in 1999, but the quality clearly diminished: 76% of the IPOs in that year had negative earnings, a metric that rose to 81% in 2000. Chart II-7Tighter Monetary Policy Clearly Played A Role In Bursting The Dotcom Bubble Chart II-8Dried-Up Funding Also Helped Pop The Dotcom Bubble Chart II-9Large Tech Companies Have A Lot Of Room To Continue Investing In AI On March 20, 2000, ten days after the peak in the equity market, Barron’s published a cover story, entitled “Burning Up.”8 The article focused on the fact that many internet companies were rapidly “burning through” their cash balances and would likely be out of money by the end of that year. This suggests that the narrative of “dried-up funding” was an important element in triggering the substantial selloff that followed. Today, while a “sudden-stop” shift in sentiment about AI is certainly a possible trigger that could pop the bubble in growth stocks, we doubt that funding will truly be a factor. NVIDIA’s surge in sales have gone largely to large tech companies, such as Google/Alphabet, Amazon, Microsoft, and Meta, and these firms have ample cash flow and cash holdings that can continue to be spent on AI chips and development (Chart II-9). It is true that the cash holdings of these firms has peaked and are trending lower, but their current level of $330 billion implies that large tech companies could fund another three years of NVIDIA’s most recent quarterly revenue without having to spend out of profits. If large tech firms significantly slow their pace of investment in AI, it is likely to be in response to one of two things: either a recession, or the belief / evidence that the pace of improvement of AI models is significantly slowing and that better quality data is the actual bottleneck rather than more “compute.” The latter has been hinted at in recent new stories in the financial press,9 but there is no sign yet from tech company executives that this is impacting their AI spending decisions. As such, we conclude that, while it is possible that the AI bubble will burst on its own, a recession remains the most likely trigger. Investment Conclusions Based on our analysis, the US equity market appears to be priced as if AI will deliver a late-1990s style productivity boom for 10-to-20 years. While possible, we doubt that this will occur, underscoring that US tech stocks are highly vulnerable to changes in the structural outlook for AI. It is possible that the AI bubble will burst before the next US recession, but that would likely depend on a major shift in sentiment about AI that will be extremely difficult to predict. A late-1990s-style “funding shortfall” will not be a catalyst this time around. A recession is likely to cause a major shock to AI-related stocks. Using Microsoft as a bellwether for the industry, Chart II-10 highlights that Microsoft’s capital expenditures as a share of EBITDA are quite elevated relative to history and are potentially subject to decline in a recession (which would also lower EBITDA itself). Chart II-11 shows that Microsoft’s capital expenditures have contracted in the past when the economy or corporate profits have been weak, suggesting that NVIDIA’s revenue will take a significant hit during a recession. That would be a cyclical effect that would not necessarily affect AI’s potential to impact economic activity over the longer term, but it would likely affect the “invulnerable AI” narrative for a time and would cause significant profit-taking among investors who are heavily overweight tech stocks. Chart II-10CAPEX Rates Are Elevated Relative To Cash Flow Chart II-11During A Recession, Tech Sector CAPEX Will Very Likely Decline The bottom line for investors is that the US equity market may continue to rise over the coming few months, potentially strongly, until signs of a recession are unambiguous. But the extremely optimistic expectations about AI’s impact on growth underscore that the US equity market selloff during the next recession is likely to be outsized relative to the impact on employment and GDP growth. Jonathan LaBerge, CFA Vice President The Bank Credit Analyst   Footnotes 1  Elder, Bryce. "Surrender Your Desk Job to the AI Productivity Miracle, Says Goldman Sachs." Financial Times Alphaville, 27 March 2023 2  Baily, Martin Neil, Erik Brynjolfsson, and Anton Korinek. "Machines of Mind: How Generative AI Will Power the Coming Productivity Boom." Brookings Institution, 5 May 2023. 3 Chui, Michael, et al. "The Economic Potential of Generative AI: The Next Productivity Frontier." McKinsey & Company, 14 June 2023. 4 Acemoglu, Daron. "The Simple Macroeconomics of AI." NBER Working Paper 32487, May 2024. 5 Eloundou, Tyna, Sam Manning, Pamela Mishkin, and Daniel Rock, “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models,” Technical Report, arXiv 2023. 6 Svanberg, Maja, Wensu Li, Martin Fleming, Brian Goehring, and Neil Thompson, “Beyond AI Exposure: Which Tasks are Cost-Effective to Automate with Computer Vision?,” 2024. MIT Working Paper. 7 Note to readers: I used ChatGPT to assist with elements of the production of this report. To put it mildly, the experience was not consistent with a 1% sustained boost to real GDP growth! 8 "Burning Up: Warning: Internet Companies Are Running Out of Cash - Fast." Barron's, 20 March 2000. 9 Mims, Christopher. "The AI Revolution Is Already Losing Steam." The Wall Street Journal, 31 May 2024.

In Section I, we examine some concerning signs of US economic weakness that emerged in June. We also discuss portfolio positioning in the face of falling interest rates and cross-check our recommended US equity overweight in the face of extremely optimistic expectations about AI’s impact on growth. We conclude that defensive positioning continues to be warranted. In Section II, we dig into those optimistic expectations for AI. We find that the US equity market is significantly overvalued unless the deployment of AI technology causes a 10-to-20 year productivity surge in line with what occurred during the IT revolution of the 1990s, with persistently high margins on the revenue generated from the improvement in growth. We doubt that AI will end up truly boosting economic activity by this magnitude.

Special Report

GAI technology has made tremendous gains over the past year. It has advanced from being a mere “curiosity” to becoming an everyday helper. While the promise of GAI is enormous, its effects are still limited: Companies are still struggling with monetization while productivity improvement is still at least a year away. In terms of evolution, the focus is shifting away from “picks and shovels” infrastructure companies toward model and application developers.

The soft landing and rate cuts narrative is being priced out, and the S&P 500 is overvalued and getting overbought. The Magnificent Seven are about to get a new moniker on the back of performance dispersion. However, without the cohort, S&P 500 earnings would have been even deeper in the red.

Outperformance of Growth sectors most likely has run its course. It is time to shift Growth vs. Value allocation to neutral, downgrade Semis, and upgrade Energy to overweight.

In Section I, we audit the market’s “soft landing” narrative in response to a meaningful challenge to our cautious stance from recent financial market developments. We acknowledge that US economic growth was stronger in the first half of the year than many investors expected, but we are unmoved by the recent uptick in “soft landing” hopes. A “soft landing” outcome very likely necessitates interest rate cuts before recessionary dynamics emerge, and it is far from clear that rate cuts or (especially) an easy monetary policy stance are likely to materialize over the coming year. As such, we continue to believe that conservative portfolio positioning is appropriate. In Section II, we discuss some simple approaches that we use when valuing the major asset classes that we cover. We conclude that global ex-US equities and ex-US developed market currencies are the main assets that can be considered “cheap” today.

Special Report Cyclically-speaking, the risk of global indebtedness does not appear to be acute. There are several pockets of sizeable private sector debt risk, and it is possible that the next US/global recession will cause a more pronounced economic downturn in some of these countries. Over the next one-to-three years, these risks are likely to be idiosyncratic. With the possible exception of France’s corporate sector, private sector debt risks appear to be manageable in the US, euro area, and China, the main drivers of global economic activity. However, over the longer-term, there are several problems with global indebtedness that will eventually “come home to roost.” US government debt is now excessive, and we expect meaningful net interest pressure for the US government in three-to-four years, even if the US does not experience elevated structural inflation. In China, the government’s strong desire to avoid aggravating structural imbalances will lead to the limited and finely balanced use of fiscal and monetary policy to boost growth, which is not good news for China-related financial assets. On balance, our conclusions are generally consistent with a structural bear market in the US dollar that is likely to begin after the next US recession. It also speaks to the possible structural outperformance of euro area stocks within a global equity portfolio, and possibly a continuation of the structural bull market in gold – which would benefit mightily from the development of any fiscal risk premia in US assets. The global financial crisis of 2008-2009, as well as the subpar economic recovery that followed, demonstrated to global investors the threat posed by elevated private sector and government debt. There has been a substantial improvement in the risk of indebtedness in some sectors of some countries over the past 15 years, but the risks of excessive indebtedness have increased in other areas of the global economy. In this special report, we check in on the indebtedness risk of a list of major economies using the BIS’ credit to the nonfinancial sector database and examine whether these risks exist primarily in the household, non-financial corporate, or government sectors. We contextualize the indebtedness data from the BIS into a risk score using several risk factors (by sector and by country), based on how elevated a given sector’s risk factor is relative not only to its own history but also the history of other countries. The sector risk scores are presented on pages 24 to 29, and we present a synthesis of our analysis below.1 We conclude that, while there are limited cyclical implications of recent trends in global indebtedness, there are several problems that will eventually “come home to roost” – particularly in the US and China. This would be consistent with a structural bear market in the dollar and a long-term uptrend in the price of gold, and could point to structural euro area outperformance within a global equity portfolio. A Global Indebtedness Report Card Table II-1 presents the aggregate risk score for each country by sector that we examined in our report. Several themes are evident from Table II-1 and the tables shown on pages 24 to 29. Table II-1A Summary Of Our Debt Risk Scores By Country/Region And Sector Shifting Household Sector Indebtedness Risk Chart II-1Shifting Household Sector Indebtedness The risk of household sector indebtedness has rotated from countries like the US and Spain to several other countries/regions, including Hong Kong SAR, Australia, Canada, and Sweden (Chart II-1). These are relatively smaller countries/regions and thus theoretically pose less of a risk to global financial stability than excessive household sector debt in the US and select euro area economies did in 2008. Mainland China remains one important wildcard for investors to watch. Ostensibly, the risk of China’s household sector indebtedness is only moderate according to our risk score methodology, given that its household debt-to-GDP ratio is lower than in many other countries. However, it has grown at a very significant rate over the past decade. In addition, household disposable income is lower as a share of GDP in China than in most advanced economies, and China’s housing sector has experienced a significant shock over the past two years. The fact that interest rates in China are likely to remain comparatively low versus the pace of economic growth, and that China’s property market is stabilizing, suggest that a major debt crisis in China’s household sector is unlikely over the coming year. The recent property market crisis, however, serves as a reminder of the potential structural vulnerability posed by Chinese household sector debt, which would almost certainly cause a global recession were a major deleveraging event to occur. Chart II-2Elevated Corporate Sector Indebtedness In Hong Kong SAR, China, Sweden, And France Some Surprises From The Trend In Corporate Debt Some countries with elevated nonfinancial corporate sector debt risk scores will not be surprising to investors. Chart II-2 highlights that Hong Kong's corporate sector indebtedness is massive and that mainland China's nonfinancial corporate sector debt risk is also very elevated. Mainland China's corporate sector debt risk is concentrated in state-owned enterprises, reflecting the significant quasi-fiscal spending (mainly in the form of infrastructure investment) that has occurred over the past decade in support of economic stability. However, Sweden and France also have very elevated nonfinancial corporate sector debt risk, whose corporate sector scores closely mirror their risk scores from the shadow banking sector. “Shadow credit” references credit that is not provided by domestic banks. A rise in shadow credit appears to be the source of the increase in nonfinancial corporate sector indebtedness in both Sweden and France. Shadow credit poses a risk to financial stability because credit availability from nonbank entities could tighten rapidly in a crisis; it thus points to potentially outsized economic weakness in Sweden and France in a bad economic scenario. Based on the IMF’s stress test results, we continue to regard Sweden’s nonfinancial private sector as one of the riskiest in the developed world. Real Long-Term Risks From US Government Indebtedness Investor concerns about the rise in US government debt have prevailed for over a decade following the surge in the debt-to-GDP ratio that occurred following the global financial crisis. However, with interest rates having fallen to extremely low levels during the last economic expansion, the debt servicing burden of US government debt was minimal. The COVID-19 pandemic changed that reality in two ways. First, the fiscal response to the pandemic resulted in another surge in the debt-to-GDP ratio. Second, the surge in inflation that occurred in the latter half of the pandemic has caused both short-term interest rates and expectations for future interest rates to rise. We expect interest rates to fall meaningfully during the next US recession, so a US government debt crisis is not imminent. However, we doubt that the fed funds rate over the coming decade will be as low as it has been over the past ten years. Higher average interest rates point to net interest costs exceeding their early-1990s levels later this decade (Chart II-3), which could cause financial market participants to force fiscal adjustment via a crisis. Chart II-3The US Will Likely Face A Fiscal Reckoning By The End Of The Decade The US is not the only country with elevated government debt risks. China, the euro area (excluding Germany) and the UK also rank highly according to our aggregate risk score methodology, as does Canada – although this reflects our use of gross rather than net debt to facilitate international comparability (see page 27 for details). The recent mini fiscal crisis in the UK is a preview of what may occur in the US and other countries on a grander scale in three-to-four years, given our view that the next US recession is likely to be mild and that the neutral rate of interest in the US and euro area is not as low as many investors believed prior to the pandemic. China’s relatively elevated government debt risk score reflects a significant rise in local rather than central government debt over the past decade, but that too carries risks for China’s economy given the way Chinese economic policy is carried out. Admittedly, these risks are much more likely to pertain to the risk of economic stagnation rather than an acute crisis. The Presence of Fiscal Space As A Buffer Against Private Sector Indebtedness In several of the countries identified with excessive indebtedness, the debt is concentrated in either the private nonfinancial or the government sector. For example, in the case of Sweden, its very concerning private sector debt load is somewhat offset by a very low government debt risk score, suggesting the presence of fiscal space in Sweden that could allow its government to respond to any private sector deleveraging event. However, in a few countries/regions, debt appears to be elevated in both the private and public sector: chiefly in Hong Kong, mainland China, and France (Chart II-4). France is a core member of the euro area; a corporate sector debt crisis in France would have a meaningful impact on European economic activity, but China’s very sizeable debt load is obviously more concerning given the importance of China as one of the three pillars of the global economy. Chart II-4Less Fiscal Space In Hong Kong SAR And Mainland China Than Before Investment Conclusions There are no real cyclical investment conclusions to be drawn from our analysis of global indebtedness. There are several pockets of sizeable private sector debt risk, and it is possible that the next US/global recession will cause a more pronounced economic downturn in some of these countries. However, with the possible exception of France’s corporate sector, private sector debt risks appear to be manageable in the US, euro area, and China, the main drivers of global economic activity. China’s nonfinancial corporate sector is indeed extremely leveraged, but much of this debt resides on the balance sheet of state-owned enterprises and thus is unlikely to pose a cyclical economic risk due to government support – especially given recent incremental easing in China. Tight monetary policy in the US and euro area is a much more proximate risk to the business cycle and, as described in Section I of our report, we expect a recession in the US to begin at some point over the coming six-to-twelve months. However, our analysis of global indebtedness highlights several problems that will eventually “come home to roost”. US government debt is now excessive. The likely future path for interest rates implies meaningful net interest pressure on the government in three-to-four years, even if the US does not experience elevated structural inflation. And in China, the government’s strong desire to avoid aggravating structural imbalances will lead to the limited and finely balanced use of fiscal and monetary policy to boost growth. As we noted in last month’s report,2 that is not good news for China-related financial assets, as it implies that Chinese policymakers will remain reactive and that China will become a more insular economy with even broader state influence or control. The Xi administration’s paradigm shift implies a very different China than many investors became accustomed to between 2008 and 2014, and one that is far less likely to stimulate global economic growth. In short, this is not, and likely will not be, the China that you have been hoping for. On balance, these conclusions are generally consistent with a structural bear market in the US dollar that is likely to begin following the next US recession. It also speaks to the possible structural outperformance of euro area stocks within a global equity portfolio, and possibly a continuation of a structural bull market in gold – which would benefit mightily from the development of any fiscal risk premia in US assets. Finally, once the next US administration is in place and a new high in the servicing costs of US government debt is within sight, investors should structurally monitor the spread between 10- and 30-year US Treasury yields for signs of an abnormally steep curve. An aggressive shift into short-duration positions will be warranted in response to any true signs of a budding fiscal crisis in the US. Jonathan LaBerge, CFA Vice President The Bank Credit Analyst Private Nonfinancial Sector The countries/regions most at risk from elevated private non-financial sector debt are Hong Kong SAR, Sweden, mainland China, France, Canada, and the Netherlands (Table II-2). Across all of the metrics shown in Table II-2 that measure the risk of indebtedness, Hong Kong consistently ranks as the riskiest market. This is particularly true based on debt service measures, which show an extremely large amount of income “lost” to repaying debt. Unlike the case of mainland China, Hong Kong’s sharp rise in private sector indebtedness over the past two decades (and especially since 2009) has not occurred due to government efforts to stabilize economic activity. Hong Kong’s pegged exchange rate effectively imports US monetary policy, which has been extraordinarily easy since the global financial crisis – particularly for an economy that did not suffer the same shock to household balance sheets that occurred in the US. The source of the risk from Sweden’s indebtedness is somewhat different than is the case in Hong Kong. Sweden’s private sector debt-to-GDP level is meaningfully below Hong Kong’s, although that is mainly indicative of how extreme the latter is. More importantly, the pace of leveraging in Sweden’s private sector indebtedness has been somewhat slower than in Hong Kong and indeed a few other countries/regions (such as Japan, France, and mainland China); it ranks third after Canada based on the first of our two debt service proxies. However, based on our second DSR that uses a measure of equilibrium interest rates, Sweden appears to be much riskier. Table II-2High Private Nonfinancial Sector Debt Risk In Hong Kong SAR, Sweden, China, France, And Canada The Household Sector The countries/regions most at risk from elevated household sector debt are Hong Kong SAR, Australia, Canada, Sweden, and the Netherlands (Table II-3). Relative to Hong Kong’s total private sector debt, the household sector is not the dominant contributor. When compared across countries/regions, however, Hong Kong’s household sector debt-to-GDP ratio is among the most extreme. Australia, Canada, and the Netherlands rank worse than Hong Kong in terms of household sector debt-to-GDP, but both economies have recently seen meaningfully slower household debt growth than has occurred in Hong Kong. Aside from the Netherlands, euro area economies rank quite low on the list of household sector indebtedness risk and nontrivially lower than in the UK. The risk of indebtedness posed by the household sector in mainland China may be understated in Table II-3. This is because China’s household disposable income is smaller as a share of GDP than most of the other countries/regions shown in the table, which causes artificially lower debt ratios when scaled relative to GDP. Relative to developed market economies, Chinese interest rates are meaningfully below the prevailing pace of income or GDP growth, so we still suspect that China’s household sector debt service ratio is not extremely high. Investors should acknowledge, however, that the risk posed by China’s household sector leverage is probably larger than conventional debt-to-GDP measures would indicate. Table II-3High Household Debt Risk In Hong Kong SAR, Australia, Canada, Sweden, And The Netherlands The Nonfinancial Corporate Sector The countries/regions most at risk from elevated nonfinancial corporate sector debt are Hong Kong SAR, Sweden, France, mainland China, and Canada (Table II-4). Unlike in mainland China, where most nonfinancial corporate sector debt is held on the balance sheets of state-owned enterprises, Hong Kong’s corporate debt does not have the same defacto state backing and is enormously concentrated in the real estate and financial sectors. Hong Kong’s real estate sector does enjoy significant structural policy support from the government. It is also true that the region has been highly indebted for some time. But Table II-4 highlights that Hong Kong’s nonfinancial corporate sector is massively leveraged and is thus vulnerable to a permanent rise in US policy rates and/or a property market crisis in the region. Commercial Real Estate (CRE) debt constitutes a large portion of Sweden’s corporate debt. IMF stress tests of Sweden’s CRE sector show that the median interest rate coverage would drop below one in a severe scenario, resulting in 75% of firms with debt-at-risk.3 We continue to regard Sweden’s nonfinancial private sector as one of the riskiest in the developed world. France ranks surprisingly high on the list of nonfinancial corporate sector indebtedness, the result of an M&A boom in the years prior to the COVID-19 pandemic. Our debt service ratio calculations suggest that the servicing burden of this debt may be lower than the BIS’ DSR would suggest, but it is still elevated even based on our measures. This suggests that the French nonfinancial corporate sector should be closely watched over the coming year, especially if the ECB were to keep its policy rate in restrictive territory. Table II-4High Corporate Sector Debt Risk In Hong Kong SAR, Sweden, France, China, And Canada The Government Sector The countries/regions most at risk from elevated government sector debt based on the BIS’ gross government debt data are Italy, the US, Canada, the UK, and Spain (Table II-5). If Canada were removed from the list, China would be the fifth most vulnerable country according to our methodology. We show gross debt-to-GDP in Table II-5 because of the lack of reliable net debt measures for China, but gross debt measures have many drawbacks. Canada is an example, as its gross debt-to-GDP ratio suffers from two international comparability problems. First, Canadian general government debt statistics include sizeable accounts payable (20% of GDP). In addition, the Canadian government holds significant financial assets; Canada’s net debt is very low compared to other developed economies. The gross/net debt issue also impacts the government indebtedness risk score for Japan, although Japan’s net government debt is still extremely elevated (160% of GDP). Very elevated debt levels in Italy, especially in net debt terms, underscore why the effective neutral rate of interest is likely lower in the euro area than would be the case if the euro area was one political and economic entity. The extraordinary US fiscal response to the COVID-19 pandemic underscores that the US will likely face a fiscal reckoning in the latter half of the decade as net interest costs eventually exceed their early-1990 levels. It is impossible to come up with a precise estimate of when the US will face market pressure for fiscal reform, but our best guess is that it will occur at the tail end of the next US administration. Table II-5High Government Debt Risk In Italy, The US, The UK, And Spain The Total Nonfinancial Sector (Private Plus Government) The countries/regions most at risk from total nonfinancial sector debt (private plus government) are Hong Kong SAR, mainland China, Sweden, Canada, and France (Table II-6). As noted above, Canada’s rank in Table II-6 is likely overstated due to the country’s much lower net debt ratio, although it would still rank relatively high given very elevated private nonfinancial sector debt. We agree that private sector debt is typically more of an economic risk than public sector debt. It is important to examine total debt, however, as it reflects the combined risk of a private sector deleveraging event that the government of that country will struggle to respond to because of a lack of fiscal space. The fact that Hong Kong and mainland China top this list underscores the risk of long-term economic stagnation in the region, and partially explains why the Xi administration is focused on improving China’s financial resiliency. Sweden’s government debt risk score is extremely low, but the country’s very elevated private nonfinancial sector debt is large enough for total nonfinancial sector debt to show up at an elevated level (similar to Canada). France’s comparatively high levels of government debt, even when measured in net debt terms, underscore the economic risks to the country were its highly leveraged nonfinancial corporate sector to experience a crisis following a period of meaningfully tight euro area monetary policy. Table II-6High Total Debt Risk In Hong Kong SAR, China, Sweden, Canada, And France Non-Domestic Bank Credit To The Private Nonfinancial Sector The countries/regions most at risk from excessive non-domestic bank credit (“shadow banking”) are Sweden, Hong Kong SAR, France, Japan, and Canada (Table II-7). The risk posed by shadow credit is that debt provided by non-bank entities is very rarely amortized, meaning that it needs to be periodically rolled over. The other risk is that lending standards or credit availability from these entities is more discretionary than is the case for banks and thus could tighten rapidly during a crisis. Combined with non-amortized loans/bonds that need to be rolled over, high levels of credit provided by the “shadow banking” sector could result in larger or more frequent credit “crunches.” Generally speaking, the list of countries with high shadow banking risk matches those that show up as high risk for the private nonfinancial sector. Japan is an exception. Global investors should be attuned to any potential credit availability issues that arise in Japan should JGB yields eventually rise, potentially in response to the end of the BOJ’s yield curve control policy. Table II-7High Shadow Bank Risk In Sweden, Hong Kong SAR, France, Japan, And Canada Appendix: Debt Risk Measures Our debt risk score tables present five measures of debt risk for three individual sectors and two aggregate sectors over fourteen countries/regions. The five sectors include: Households Nonfinancial corporations Government The private nonfinancial sector (aggregate of households and nonfinancial corporations) The total nonfinancial sector (aggregate of households, nonfinancial corporations, and the government) We also examine the private nonfinancial sector focusing on debt that is not provided by domestic banks (“shadow banking”). Our methodology scales each measure of debt vulnerability for each country across the matrix of histories of all fourteen[1] countries/regions for that debt vulnerability measure using a percentile rank. In that way, we compare each country’s measure to a range of country histories, rather than only its own history. We scale these measures as scores from 0 (best / lest vulnerable) to 10 (worst / most vulnerable) and present the most recent observations in the tables included in this report. Our five measures include: The BIS[2] Credit-to-GDP Ratio: Ratio of total credit provided to the sector to GDP The BIS Debt Service Ratio: Ratio of debt payment estimate to gross disposable income (GDI). This measure is not available for the government sector, the overall nonfinancial sector, as well as for nonfinancial corporations for China and Hong Kong SAR. The BCA Credit-to-GDP Gap: Measure of Credit-to-GDP relative to its 10-year moving average The BCA Debt Service Ratio (Proxy 1): Ratio of debt payment estimate 1 to gross domestic product (GDP) The BCA Debt Service Ratio (Proxy 2): Ratio of debt payment estimate 2 to gross domestic product (GDP) We also include an Aggregate Debt Risk Score, which aggregates the scores of all debt vulnerability measures available by sector for each country using an equal weight approach. Our BCA Debt Service Ratios are calculated in the following manner: We estimate principal payment schedules of 18 years for households and of 10 years for nonfinancial corporations. We then estimate a principal payment component of the total debt payment by dividing the stock of debt by the debt maturity. We do not consider a principal payment in cases where debt is exclusively not amortized, such as government debt. We then compute the measure of debt interest payment by multiplying the overall stock of debt by an interest rate proxy. For our DSR proxy 1, we use the 10-year government bond yield as a measure of effective interest rate plus a spread of 1.75% for household sector debt and 1% for nonfinancial corporate sector debt. One exception applies to Hong Kong SAR, where we use US 10-year Treasury yields given Hong Kong’s pegged exchange rate. For our DSR proxy 2, we use an estimate of the equilibrium interest rate instead of 10-year government bond yields with the same household/corporate sector spread estimates. Our estimate considers the median 10-year nominal GDP growth rate as the equilibrium interest rate, with exceptions for euro area members, Hong Kong SAR, and mainland China. For euro area economies, we use euro area GDP rather than the individual country GDPs due to the commonality of monetary policy. For Hong Kong SAR we use US GDP rather than Hong Kong GDP given its pegged exchange rate and its importation of US monetary policy. For mainland China we use half of the estimated equilibrium interest rate, given that China has consistently maintained a large gap between domestic interest rates and the prevailing rate of nominal GDP growth. We then add the interest payment estimate to the principal payment estimate (when applicable) to obtain total debt payment. We then express these debt payments as a percent of GDP. Gabriel Di Lullo Research Analyst   Footnotes 1 Please see the appendix on pages 30 and 31 for a description of our debt score methodology. 2 Please see The Bank Credit Analyst "April 2023," dated March 30, 2023, available at bca.bcaresearch.com 3 Sweden’s Corporate Vulnerabilities: A Focus on Commercial Real Estate, IMF Working Paper, Selected Issues Paper No. 2023/024, March 21, 2023