To dig a little deeper, I downloaded markets and macroeconomic data to see what energy spike episodes looked like in the past. Before turning to that analysis, a brief aside on Claude, Anthropic’s AI model. At QuantStreet, we’ve been avid users of machine learning and AI tools for some time. For example, we have a nightly news agent scrape the day’s main headlines and email us a summary of how these impact our portfolio positions. We’ve recently started using Claude Code to help us run our analytics processes, as well as to help us streamline and extend our codebase. With the caveat that no AI model is perfect—and certainly Claude makes mistakes and needs human oversight—the capabilities of Claude are astounding. The analysis in this piece owes a lot to Claude, which must have saved between 10-20 hours of human effort. Now back to our analysis.

Subscribe now

Using month-end spot WTI prices from FRED, the next chart shows all identified price spikes in WTI since the 1970s. Here is how Claude explains the spike-detection algorithm (with some minor edits from me):

We scan monthly WTI prices from 1973 to the present, looking for episodes that follow a simple pattern: a sharp run-up followed by a meaningful pullback. The algorithm tracks a rolling price trough, and once prices rise at least 30% from that trough, it locks in a qualifying spike. It then waits for a pullback of at least 20% from the peak to mark the episode’s completion—recording the start, peak, and end dates along with the size of the run-up and drawdown.

Each past price spike is indicated by the pink, shaded area, with the vertical dashed line indicating the peak WTI level in each spike. The shaded region’s edges indicate the start and end of each spike episode. Because the data used to construct the chart are monthly, the current energy spike (so far, entirely intramonth) is not included in the sample. With this caveat, there have been 12 prior energy price spikes since the 1970s.

The next chart zooms in on the first of these. This one encompasses two price shocks, though according to our definition, this counts as a single spike: a low, followed by successive highs, and a selloff of more than 20%. The first spike is the Arab oil embargo following the Yom Kippur war in 1973 and the second is the aftershock of the Iranian revolution in 1979. Oil prices went from $3.56 per barrel to around $10 after the embargo in 1974, then rose to a peak of $39.50 in 1980 (indicated by the vertical dashed line), with the episode ending in late-February of 1982 after oil fell to $31.56. Not counting dividends, this was a weak decade for the S&P 500, with the index starting and ending at the $120 level. Year-over-year CPI inflation experienced two spikes, peaking at 12.5% after the oil embargo and again at close to 15% following the Iran revolution. The oil embargo was highly contractionary, leading to a more than 10% drop in industrial production; the Iran revolution price spike was also contractionary, though not as severe.

The next table provides a summary of each episode. Each row of the table shows the start, peak, and end dates of each price spike. The number in parentheses below the end date is the length of the episode in years. The WTI (up) cell shows the percent price appreciation in the WTI spot price from the episode start to the peak, and the WTI (dn) cell shows the price fall from the peak to the episode end. The SPX (up) and SPX (dn) cells show the analogous measures for the S&P 500 index, e.g., SPX (up) is the percent rise from the start of the energy peak episode to the month of the episode’s peak oil price. And the CPI (up) and CPI (dn) cells show the change in CPI inflation, i.e., the percent change in the CPI index, over these intervals. The final column shows a brief description of important energy news events in this window (summaries and research done by Claude); when this is blank, it means Claude did not identify a major energy news event that took place in this window.

A few observations:

• By far the largest and longest price shock was the over 1000% uninterrupted price increase oil experienced from the early 1970s to the early 1980s. This lasted over 9 years. During this time the S&P 500 was flat in price terms and the price level (CPI) rose over 100%.

• During the First Gulf War of 1990 to 1991, there were two price spikes. One lasted for 1.7 years, and saw a 64.1% WTI price rise, while the second took place in mid-1990, but lasted only 5 months with a price rise of 92.7%. During the first WTI price spike, the S&P 500 rose over 26%, but the S&P 500 fell during the second price spike by around 6%.

• From the late 1990s until mid-2008, oil prices experienced a series of sharp spikes, followed by mild corrections. The net effect was to take prices from the mid-teens to close to $140 per barrel, before oil prices suffered a major correction as the economy entered a recession during the Global Financial Crisis (GFC).

• The post-GFC rebound and Arab Spring, which took 1.5mm barrels of Libyan oil off the market, coincided to lead to another large oil price increase. Stocks, though, did very well during this episode with a price increase (not counting dividends) of roughly 75%.

• Finally, the economic rebound following COVID, combined with Russia’s invasion of Ukraine in 2022, precipitated a 594% oil price increase, over a 2.5-year period, though stocks were up roughly 25% in price terms in this window.

Interestingly, the oil price / stock price relationship is not always negative during large price spikes. For example, in the 1993 to 1997 price spike (next figure), WTI prices had a peak-to-trough rise of 75%, but the S&P 500 was up over 65% in price terms.

However, other energy price spikes are associated with falling stocks. For example, in the 2001 to 2003 price spike, which was associated with OPEC production cuts and the start of the Second Iraq War, energy prices went up 85.6% while stocks fell close to 20% in price terms.

The takeaway is that oil price spikes are sometimes, but not always, associated with negative economic outcomes and negative stock market returns. A lot depends on the reason for the shock, with supply-driven price spikes—especially those caused by prolonged geopolitical events in the Middle East—generally being negative for economic growth and financial markets, while unattributed price spikes—presumably caused by increased demand due to strong growth—are much more easily absorbed by financial markets and the economy.

Two other wildcards are the length of the spike, with shorter spikes presumably having less impact, and the oil sensitivity of the economy overall.

Looking ahead

WTI futures are contracts that lock in a delivery price for a barrel of oil at some future date at a price that is agreed on today. Futures prices consist of two components. First, they reflect the market’s expectation of what WTI spot prices will look like on future dates. Second, they reflect oil producers’ demand to forward sell future production at a prespecified price and oil consumers’ demand to procure future oil, also at a prespecified price. The group with the stronger hedging demand will push futures prices either below (if producers hedge more) or above (if consumers hedge more) the market’s expectation of future spot prices. This tug of war makes futures prices noisy predictors for future spot prices; nevertheless, futures prices still contain a great deal of information about the market’s expectation of the path of WTI prices over the coming years.

As of the afternoon of March 12, 2026, the WTI futures curve, given by the white curve below, is forecasting a reversion to pre-war energy prices in two years, and to $70 oil in one year. For reference, the yellow curve shows the same set of futures prices from one month ago, which was prior to the start of the Iran conflict and before the recent WTI price spike. By year five, market expectations today are equal to the market’s expectation—or more correctly, hedge-demand adjusted expectations—from one month ago. Interestingly, by 2037, the market now believes in lower oil prices than it did one month ago, perhaps reflecting the expectation that ultimately a regime change in Iran will take place—though it may take a long time—which will result in long-term lower energy prices as the Iran risk premium gets priced out of the market.

There are, of course, those who disagree with this. See, for example, Paul Krugman’s piece “Oil Prices Could Easily Go Much Higher.” But keep in mind that those who really believe in $150 oil in a few months can easily buy September 2026 oil futures. The fact that these currently trade at $80.74 indicates that most do not believe this dire scenario. One reason for this is that the US is keenly aware of the problems in the Strait of Hormuz, and is directing great resources to reopen the Strait to international shipping.

According to Apollo’s Torsten Slok, the US and global economies are now much less dependent on oil than they once were. This, together with the futures market prediction that the price spike will be short-lived, bodes well for the stock market’s and economy’s ability to absorb and move past the present oil price shock.

Only time will tell what actually happens, but the historical evidence—not all energy price spikes are scary for markets and the economy—and the market’s view that the present price spike will last one year, but not more, give reason for cautious optimism.

Subscribe now

Working with QuantStreet

QuantStreet is a registered investment advisor. It offers financial planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic, data-driven, and shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our email list.

Since QuantStreet launched in December 2021, we’ve had a bullish view on US assets and on the dollar. The former is still true. We are believers in the AI revolution and don’t buy the AI bubble argument. But the latter is starting to change. Based on the behavior of the US stock market and the dollar after last year’s tariff selloff, this view seems to be the consensus.

Let’s first take a look at the dollar positives.

Subscribe now

Positives

US growth exceeds that of the rest of the world. This raises US interest rates and attracts global capital, both of which are dollar positive.

Second, US rates are high relative to other countries.

By swapping out of 10-year US Treasuries into 10-year Italian or French government bonds, investors give up over 0.60% in yield, and arguably switch into lower quality credits to boot. The terms of that swap also make the dollar attractive.

But the negatives are starting to pile up.

Politics

The aha moment for me came after reading the WSJ article from Dec 27, 2025 titled “Trump wants a weaker dollar. Some Chinese say he has a point.” To quote from the article:

“You make a helluva lot more money with a weaker dollar,” the president said in July. When the dollar is strong, “you don’t do any tourism, you can’t sell tractors, you can’t sell trucks, you can’t sell anything.”

Nothing new here. President Trump is a mercantilist at heart and wants the US to have balanced trade with the rest of the world. Though it will come with many knock-on effects, one way to unwind our trade deficit is to have dollar depreciation: foreign goods become more expensive and US exports become cheaper.

The article continues:

Liu Shijin, a longtime top economic adviser to the (Chinese) government, said in a speech this month that the U.K. and U.S. also started out as primarily manufacturing powers with puny currencies but eventually matured beyond that stage.

Liu cast an envious eye toward Trump’s ability to make the world shudder over U.S. tariffs, which he said was the result of dollar-wielding American consumers’ buying power. Liu said that the whip hand rightfully belongs to China, since its population gives it a potential consumer market “far exceeding the U.S.”

“We should aim for a basic balance between imports and exports” and push for a strong, globally used currency, Liu said.

If influential thinkers in China are pushing for a stronger yuan and weaker dollar, a large part of the trade mechanism supporting dollar weakness disappears. I don’t think a strong yuan represents the CCP’s official policy, but markets will move long before the official policy does.

Add to this President Trump’s desire for lower rates and his search for a new Fed chair who will deliver them—which would diminish both the dollar’s carry advantage and the Fed’s inflation-fighting credibility—and the political ingredients for dollar weakness are in place.

Real effective exchange rate

Another pillar supporting the weaker dollar thesis is the dollar’s high real effective exchange rate. Consider a consumption basket of US goods, with a price of $P. Say the price of one dollar in terms of euros is €S (right now, the euro price of a dollar is around €0.85) and the price of a similar consumption bundle in Europe is €P_E. The real dollar-euro exchange rate is then

dollar/euro real exchange rate = S x P / P_E.

This is the ratio of the euro price of buying the consumption bundle in the US to the euro price of buying the consumption bundle in Europe. The higher this is, the more “expensive” is the dollar. A trade-weighted average of this ratio across all global currencies yields the dollar’s real effective exchange rate (REER).

Expensive currencies tend to depreciate in nominal terms, though this is not always the case (e.g., the yen has been “cheap” on a REER basis for a long time and continues to get “cheaper”). This happens because countries with expensive currencies run trade deficits (since foreign stuff is cheaper than domestic stuff), so their currencies go out into the world to buy other countries’ goods. As the global supply of the currency grows, assuming demand stays fixed, prices adjust downward.

The next chart shows the dollar’s REER (in black) and nominal effective exchange rate (NEER, red, dashed line). The dollar’s REER is close to an all-time high, despite the dollar’s nearly 10% sell-off in 2025. The bottom panel of the chart plots the one-year ahead change in the dollar’s nominal exchange rate against the dollar’s current REER. For every point above its steady-state level (of around 96, where the regression fit line crosses the x-axis), the dollar tends to depreciate by 0.329 points over the next year. This point has been made in the academic literature, e.g., Cashin and McDermott (2004), so the result stands up to statistical scrutiny.

Sources: QuantStreet, BIS, FRED

Given its current valuation level, the dollar is forecasted to depreciate by 4.27% over the next year, and by twice that amount over the next two years.

The next figure applies the same analysis to the euro. The euro’s valuation appears neutral based on its historical REER, and the euro forecasted change for the year ahead is close to zero.

Sources: QuantStreet, BIS, FRED

Flows

An unintended consequence of sanctioning foreign countries by freezing their dollar assets has been to push them into other stores of value. The Jan 23, 2026 WSJ article “Five reasons gold is surging toward $5,000 an ounce,” carries the following chart using data from the World Gold Council which shows extreme buying of gold by global central banks.

These central bank flows into gold have been augmented with investor flows into exchange traded funds like GLD and IAU. Together with gold’s inelastic supply, this has pushed up the price of gold to near $5,000 per troy ounce, a dramatic appreciation both in nominal and inflation-adjusted terms.

The interpretations of the above chart vary. One is that gold is in a bubble. After such an extreme price run-up—both in speed and magnitude—it’s hard to definitively rule out a bit of froth in the gold price. But the other interpretation is that, while the dollar and Treasuries still represent the majority of global reserve holdings, the dollar’s position as the sole global reserve currency is slowly deteriorating. This shift to other stores of values is most readily seen in assets whose supply cannot be adjusted in the short-run, like gold.

Not helping matters is the general frustration among America’s allies with the Trump administration’s foreign policy. For example, while I think the Greenland situation will be resolved to the satisfaction of both the Trump administration and our European allies, the way it has been handled over the last week has diminished the desire of European to hold US assets, with some going so far as to call for a concerted effort to reduce exposures. I don’t think this is a plausible path forward (neither does the FT’s Robin Wiggleworth), but even a marginal move in this direction will continue to chip away at the dollar’s reserve currency status. This FT article explains further.

Portfolio implications

To summarize the dollar thesis, here are the two positive:

• US interest rates are high relative to other developed market economies; and

• US economic growth is relatively strong.

The negatives are:

• The Trump administration wants a weaker dollar;

• High dollar valuations forecast future nominal depreciation;

• Political interference with the Fed is dollar negative;

• The signal from the gold market is bad;

• Widespread frustration with the Trump administration’s foreign policy.

Whatever one thinks of the prospects of US versus international stocks, a 4.27% annual headwind will be difficult for US assets to overcome. Based on QuantStreet’s analysis of long-term valuations trends, our forecast for US and international stock returns are close to each other. Adding in the dollar headwind for US assets, and conversely its tailwind for international ones, suggests investors with light international allocations should rethink their portfolio positions. If you’d like our advice on how to do that, please reach out.

Subscribe now

Working with QuantStreet

QuantStreet is a registered investment advisor. It offers financial planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic, data-driven, and shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our email list.

The most recent cycle began with the market trough of September 2022. Despite the significant consternation engendered by it, this recent bull market is neither anomalously long (when measured in months) nor anomalously large (when measured in percent return).

Subscribe now

To get a better sense of the duration and magnitude of each bull cycle, the left panel of the next figure stacks all eight cycles on top of one another, each starting at a value of $1 invested in the S&P 500 index. The legend shows the dates of each bull cycle, starting with the longest one (measured in months) down to the shortest one; the colors are the same as in the above chart.

The current bull cycle (highlighted by the shaded blue line) has run for 40 months. Since 1960, only two bull cycles have been shorter (one lasted for only 7 months), while five have been longer. Of these, the shortest one lasted 60 months, while the longest three lasted roughly 13 years. The duration of the current bull cycle is not anomalous.

Through the first 40 months, the current bull cycle has been the best performing one, though the difference in returns is not large. A $1 investment in the S&P 500 back in September of 2022 would have grown to roughly $2 now, pretty much exactly in line with the 2009-02:2021-12 bull market. A $1 investment in each of the previous 13-year bull markets ultimately grew to a value above $8. If history repeats—more on this below—there is more return left in the current bull cycle.

The message from CAPE

As the right panel of the above chart shows, the cyclically adjusted price-to-earnings (CAPE) ratio has risen during this cycle (highlighted in blue) from a starting value in the high-20s to its current value of approximately 40 (an increase of 11.3 points). A rise in the CAPE ratio at the start of a bull cycle is not unusual, though the increase in the current cycle has been faster than typical. CAPE ratios rise because stock prices anticipate future, presumably higher earnings, whereas the E in CAPE is a reflection of the last ten years of earnings (adjusted for inflation). What gives investors more pause, however, is the high level of CAPE in this bull cycle.

While the CAPE ratio has historically had an inverse relationship with future returns (higher CAPE, lower future returns), the measure has several well-known problems:

1. CAPE has a long-term upward trend, partially due to the inability of historical accounting measures to properly reflect modern business practices (like marketing spend to acquire clients). I wrote about this about a year ago.

2. The CAPE-to-future-return relationship is much flatter over the last several decades than it has been historically. So a higher CAPE ratio predicts less dire returns than it once did.

3. The companies that make up past earnings of the S&P 500, i.e., what is reflected in the E of CAPE, are not the same as the current index constituents. Making this adjustment renders the current CAPE ratio less extreme.

4. Related to the above, past earnings may not be a good proxy for future earnings. Based on the price-to-forward-earnings ratio, market valuations also look less extreme.

To get a better handle on the trend component of CAPE, the blue stars in the left panel of the above figure show the starting CAPE level of each of the eight bull cycles since 1960. The orange dashed line shows the time trend in these starting CAPE levels. It is clear that over the last 65 years, on average, successive bull cycles have started from progressively higher CAPE levels. This is consistent with the longer-term trend in CAPE I mentioned earlier. The CAPE ratio at the start of the most recent bull cycle is shown circled in red.

The middle panel of the above chart shows the relationship between the starting CAPE level of each cycle (x-axis) and the length in months of the subsequent bull run (y-axis). Again the most recent starting level is circled in red. There is a clear inverse relationship between the starting CAPE level of each cycle and the cycle’s length. The right panel shows the same analysis but rather than showing the cycle’s starting CAPE ratio on the x-axis, this plot shows the starting level of the CAPE ratio relative to its trend. Interestingly, the right panel provides a much closer fit to the data (a 76.8% R-squared versus the trend-unadjusted R-squared of 57.2%).

While the starting CAPE level in the current cycle is high, the current cycle’s starting CAPE value is only 3 points above its trend, a relatively small deviation. Based on the historical relationship between cycle length and the trend-adjusted starting CAPE ratio, the current cycle is projected to last 54.7 months. Despite all of its flaws as a forecasting variable, a simple trend adjustment to CAPE suggests the current bull cycle has a while yet to run.

Conclusion

It would certainly be preferable, from a historical perspective, if current market valuation ratios were lower. The high level of prices relative to past earnings clearly reflects investor expectations for high future earnings growth. The CAPE chart above showed both the change in the CAPE ratio in each bull cycle, as well as the compounded annual growth rate (CAGR) of earnings (the E used in CAPE) over each cycle. Some cycles are characterized by mid-single digit earnings growth, while others had earnings CAGRs in the double digits. Having double digit earnings growth turns out to be very important.

Say that the current CAPE ratio is 40, and we expect stocks to return 9% per year for the next ten years. Since the dividend yield of the S&P 500 is 1.1% per year, this means price appreciation of 7.9% per year. Say that the E in CAPE growth at 10% per year for the next ten years (this would happen if every earnings number that falls out of CAPE is replaced with a new one which reflects a 10% CAGR over the ensuing ten years). Under these assumptions, in ten years the CAPE ratio will fall to 40 x 1.079¹⁰ / 1.1¹⁰ = 33, still a relatively high historical level but better than 40. To put the 10% earnings growth assumption into context, according to FactSet “[f]or CY 2026 analysts are projecting (year-over-year) earnings growth of 14.5%.”

My main point is that a continuation of the current bull cycle for the next several years would not be unprecedented (in fact it would be quite typical) and with sufficient earnings growth, market valuations could well fall during that time.

As one caveat, I’ve argued in the past against the type of line-them-up analysis I am doing in this note (see the Nvidia and Enron piece from early 2024). But I feel this analysis may be more salient for individual macro series, like market-wide returns, than single stocks (less room to data mine). Furthermore, as with any market forecast, please regard this one with a great deal of caution. Markets may do considerably worse than I estimate. And the historical relationships I rely on for this study may not hold in the future. Please consult a financial professional before making investment decisions and make sure your portfolio properly reflects your risk tolerance and liquidity needs.

Subscribe now

Working with QuantStreet

QuantStreet is a registered investment advisor. It offers financial planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic, data-driven, and shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our email list.

Appendix

The next charts repeat this note’s analysis using a 15% threshold to define a market selloff. The qualitative features of the results remain.

In light of the start of the AI revolution—which arguably began on November 30, 2022 with the introduction of ChatGPT—almost exactly 50 years after the start of the information and telecommunications revolution, her book, published in 2002, appears especially prescient.

Subscribe now

Perez argues that each revolution proceeds in four phases:

• The irruption phase sees the introduction of the new technology to the world.

• This is followed by the frenzy stage when society begins to understand the power of the new technology, and invests resources in its development. She argues that the interaction of financial capital and technological change ultimately leads to a market crash at the end of the frenzy stage.

• At this point, the synergy stage begins, where the best uses of the technology are discovered and resources are efficiently funneled into supporting these.

• Finally, the revolution enters its maturity phase, at which point most investments have already been made, and the technology settles into the fabric of ordinary life. At this point, a new “great surge” begins as another revolutionary technology enters the public consciousness.

Against this backdrop, Perez produces timelines of the five great technological revolutions. In her estimation, the Industrial Revolution started in 1771 and lasted through 1829. At this point, the Age of Steam and Railways began, lasting until 1873. In 1875, steel, electricity, and heavy engineering carried the baton until 1918. The Age of Oil, Automobiles, and Mass Production began in 1908, overlapping for a decade with the end of the prior technological epoch. This lasted until 1974. Finally, the Information and Telecommunications Age began in 1971 and had not yet run its course when the book was published in 2002.

Financial markets and technological revolutions

Our contribution is to connect Perez’s timelines with stock return data, which we obtain from the macrohistory.net website. This data set covers annual economic and market statistics about 18 developed market economies, starting in 1870.

The next chart shows the value of 1 unit of local currency invested into the U.S., U.K., and German markets starting at the end of 1874. The left panel zooms in on the first ten years of data. The legend shows the compounded annual growth rate (CAGR) of each market, as well as the annual volatility and the worst one-year performance in the sample period.

In the ten years following the 1875 start of the Age of Steel, Electricity, and Heavy Engineering, world stock markets were up between 5.4% and 8.3% per year. The worst one-year performance was -12.9% in the U.S. in 1884. The selloff in the U.S. market in 1876 and 1877 represents the tail-end of the Long Depression of 1873-1877, which began in response to overinvestment in railroads, the focus of the prior (1829) technological revolution. The right panel extends the analysis window to 25 years, through 1899. Over this time, the compounded annual growth rate was between 6.1% and 7.4%. The worst one-year return was -15.4% in the U.S. in 1893. Overall, the Age of Steel, Electricity, and Heavy Engineering saw reasonably robust stock market performance for the next 25 years.

The next chart shows the ten- and 40-year performance after the 1908 start of the Age of Oil, Automobiles, and Mass Production. The left panel shows that in the ten-year period after the end of 1907, the U.S. market was up 6% per year, with a worst one-year return of -23.6% in 1917, ten years after the start of this technological epoch.

The right panel shows that extending the analysis to the end of World War II captures the market crash that happened during the Great Depression. Over 1930 and 1931, the U.S. market was down 54%. The U.S. market would not sustainably reach a new high until the early 1940s, over ten years after the start of crash. The market crash associated with the Great Depression, even if it was partially caused by technological overinvestment (which is not the common narrative), still happened 20 years after the start of the oil and automobile age.

Finally, the start of the information and telecommunications revolution saw strong global market returns from the end of 1970 until 1980, with CAGRs of 8.5% to 14.3% per year, as seen in the left panel of the next graph. However, both the U.S. and U.K. markets experienced severe strains in 1973 and 1974, down 38% and 64% respectively, due to the 1973 oil price shock. This shock was caused by conflict in the Middle East and was not directly related to the technological revolution itself.

Expanding the analysis to the 35-year period after 1971 shows annualized returns between 7.5% and 13.9%. There were two more large shocks in this extended time frame. First, the Japanese market was down 46% in between 1990 and 1992, largely due to the bursting of the Japanese asset bubble. The U.S. and U.K. markets would continue to set new highs until the bursting of the dot-com bubble from 2000 to 2002, a period over which the U.S. stock market was down 34% when measured using end-of-year prices. While the dot-com selloff was attributable to overly exuberant tech company valuations, the bubble did not materialize until almost 30 years after this tech epoch began in 1971.

The Age of AI

Arguably the Age of AI began with the November 30, 2022 introduction of ChatGPT. This set in motion an unprecedented rush to develop the software and hardware infrastructure needed to support the AI revolution.

Taking 2022 as the starting point, thus far the annualized rates of return across global markets have been from 7.4% to 11%. This is higher than the CAGRs in the ten-year windows following 1875 and 1908, but in line with the ten-year window following 1971.

History shows that the starting points of technological revolutions are not invariably followed by the onset of large stock market selloffs, though certainly the latter do occur in the subsequent years and decades. What comes next this time around is a topic for another day, but the historical precedent we obtain from Perez’s book coupled with the macrohistory.net data provides a bit of guidance.

Working with QuantStreet

QuantStreet is a registered investment advisor. It offers financial planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic, data-driven, and shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our email list.

Subscribe now

With interest in gold hitting a fever pitch, we now revisit our gold investment thesis.

A tail of gold in nine charts

As the next chart shows, gold has rocketed upward since 2020, doubling over the last two years, and rising by over 50% so far in 2025.

As of the time of this writing, gold is trading north of $4,000 a troy ounce.

Subscribe now

If gold were the perfect hedge for inflation—it isn’t—then gold prices divided by the overall price level would be constant over time. However, real gold prices (adjusted for inflation by dividing by the CPI price level) are currently at multi-decade highs.

Any time an asset price runs up this much this quickly, one wonders whether bubble dynamics are at play. More on this below.

The next two charts examine the price of gold relative to two other valuation benchmarks. The top chart shows that the price of gold, when measured in units of bitcoin, is close to its all-time lows. Without getting into the debate about the merits and flaws of crypto investing, anyone holding bitcoin as a hedge against all manner of bad things might well consider switching part of that hedge into gold.

Similarly, the price of gold, when measured in units of the S&P 500 index (including reinvested dividends), is also near multi-decade lows. There is no particularly compelling reason to think that stocks—which entitle their owners to the earnings power of the underlying firms—should not outperform gold over long stretches of time, but at least we take some comfort in knowing that gold, even after its recent large price runup, has lagged equity returns over the last several decades.

ETFs, because they enable investors to hold gold as a financial asset in their brokerage accounts and not as a physical commodity in their bank vaults, greatly increased access to the gold market.

The two largest gold ETFs, GLD and IAU, together hold over $200 billion of physical gold at current market prices. The $200 billion assets under management (AUM) figure consists of two components: new inflows of investor dollars into the ETFs and appreciation in the value of existing holdings due to the runup of gold prices.

The next chart isolates the investor flow component of ETF gold holdings by dividing their AUM by the gold price level, normalizing the adjusted figure to start at one in early 2005, which is when the ETF holdings data become available.

An important feature of this year’s gold price runup is that it has been associated with large investor inflows into the two largest gold ETFs. Coupled with the multi-year trend of gold buying by global central banks, gold prices are not rising in a vacuum, but are doing so in response to strong investor demand. As the demand curve for gold shifts outward, unless the supply curve can shift outward equally quickly—which it cannot—gold prices will rise.

A similar observation about investor inflows into the gold market appears in a recent article by Nir Kaissar, of Bloomberg News, though he concludes that the gold trade in now long in the tooth and is best avoided.

A natural question is the degree to which investor inflows into gold line up with gold price appreciation. The next chart—which first appeared in a very interesting LinkedIn post by Cam Harvey of Duke University—shows the real price of gold (i.e., after adjusting by inflation) plotted against a scaled version of the above investor inflows measure.

As Cam Harvey points out, the two series closely track each other up until the last few years, at which point the real gold price begins to diverge from the price level implied by investor inflows into gold ETFs. He refers to the underlying cause of this divergence as “de-dollarization,” with a promise to write more about this phenomenon soon.

Our own take is that the dollar’s status as a reserve currency has come into question over the last few years. Some of the well-known reasons for this—geopolitics, high deficit and debt loads, questions about the Fed’s independence, government shutdowns—are detailed in our recent investor letters (here and here). It is likely that central bank buying of gold is due to exactly these factors.

Looking at the part of the real gold price not explained by investor flows (the difference between the yellow and green lines in the above chart) and plotting this against the (negative) U.S. dollar price index (in red), reveals a striking pattern.

The gold price runup coincides almost exactly with the dollar’s peak (or trough of the negative dollar index) of September 2022. The gold price appreciates more than the dollar weakens, but the timing of the move lower in the dollar and move higher in gold is unlikely to be a coincidence.

A key driver of investor demand is that gold often acts as a safe-haven asset in times of economic and market turmoil. One way to see this is to calculate gold’s average return in 12-month periods of large S&P 500 selloffs. We find that gold returns in such periods tend to be slightly positive. Another take on this question is the correlation between gold and stock returns.

The above chart shows the last 12-month correlation of monthly gold and S&P 500 returns, plotted over time. This correlation is often negative, which means that at times when the stock market sells off, gold frequently acts as a portfolio hedge. In recent years, the correlation of monthly stock and gold returns has been particularly negative, suggesting that gold has served as an effective hedge in investor portfolios.

Subscribe now

Looking ahead

The currently high interest rate differential between U.S. rates and those of other countries (U.S. rates are higher) is dollar positive, as is America’s faster growth rate and lead in the AI revolution. If the dollar selloff were to reverse, and if questions around the dollar’s reserve currency status were to subside, gold may retrace parts of its recent rally. On the other hand, gold’s positive price momentum and tendency to appreciate during Fed easing episodes argues for continued strength. And with (geo)political uncertainty—the Russia-Ukraine war, China-U.S. jockeying for global influence, questions about Fed independence—still high, investor demand for gold is unlikely to subside overnight.

Whether or not gold makes sense for an individual investor depends on many considerations outside the scope of the present analysis. At QuantStreet, we will continue to evaluate gold in the context of our systematic portfolio allocation process, while considering the qualitative factors raised in this analysis.

Working with QuantStreet

QuantStreet is a registered investment advisor. It offers financial planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic, data-driven, and shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our email list.

Subscribe now

If the investor does nothing today (Case 2), she will earn an 8% return for the next 20 years, which will generate $466 in the account at the end of year 20. At this point, we assume the investor liquidates the entire portfolio at a future capital gains tax rate of 20%. Relative to an $80 cost basis, this leave $389 in the account at the end of year 20 after taxes have been paid.

The alternative for the investor is to sell the entire portfolio today (in reality, the investor may rebalance only a portion of the portfolio), pay a $4 capital gains tax ($20 gain x 20% tax rate), and reinvest the proceeds from the sale into the new, higher returning portfolio. The cost basis of the new portfolio will be $96. This then gets invested at 9% for the next 20 years, generating a pre-tax value in year 20 of $538. After paying a 20% capital gains tax relative to a $96 cost basis, the post-tax value in the account is $450.

The portfolio rebalance, despite its immediate capital gains cost, generates $61 additional post-tax dollars in the portfolio in year 20. If the investor was hesitant to do this rebalance, she should not have been.

Analysis

There are two factors at play. First, how much does the investor pay in taxes now if she sells? The cost of paying higher taxes now is that the money paid in taxes cannot earn interest for the next 20 years in the investment account. (A benefit of rebalancing is that the portfolio’s cost basis goes up, which means lower taxes in the future.) Second, how much extra return does she earn from rebalancing the portfolio today? If the extra anticipated return is sufficiently high, it outweighs the return drag caused by paying taxes earlier.

While this was the case in the above example, it is not always so. The next table shows a scenario analysis which varies the portfolio cost basis (y-axis) and the anticipated return gain from rebalancing (x-axis, in basis points). The scenario just analyzed, which generates a $61 gain from rebalancing, is shown in the cell labeled 80 (cost basis) and 100 (annual basis point gain from rebalancing).

But the gains from rebalancing are not always positive. The lower the cost basis, the higher the drag from paying taxes now and foregoing the returns earned on the amount paid in taxes. The lower the anticipated future return differential, the less offset there is to paying taxes earlier. For example, if a $100 portfolio has a cost basis of $50 and the anticipated return gain from rebalancing is only 30 basis points per year (0.3%), then the investor loses $10 in the year 20 post-tax portfolio value from rebalancing today.

As the above scenario analysis shows, if the anticipated return gain is close to one percent per year (100 basis points)—as would often happen from moving from high fee to low fee mutual funds—then almost regardless of the portfolio’s cost basis, the investor is better off rebalancing into the lower fee funds immediately, if her planning horizon is 20 years out.

Impact of step-up cost basis

In many cases, when assets go to loved ones after someone’s passing, they are subject to a step up in cost basis to the market prices prevailing at the time of the inheritance. This is typically advantageous because the cost basis is usually considerably lower than the market value of the inheritance. Depending on the current cost basis of the portfolio, this step-up feature can either advantage or disadvantage the rebalance-now scenario. When the current cost basis is relatively high compared to the portfolio’s market value, a step-up basis at inheritance can make the rebalancing option look more attractive.

This is what happens in Cases 3 and 4 below. This scenario is the same as Cases 1 and 2, but we now assume that the capital gains tax rate in 20 years is zero (due to the step up in basis). The post-tax value in Cases 3 and 4 is the same as the pre-tax value, and the edge from an early rebalance increases from $61 to $72.

However, a step-up basis can also make the rebalancing less attractive in cases where the current tax basis is low relative to the market value of the assets. In such cases, illustrated in the example below, the benefit of never paying capital gains on current assets (with high unrealized gains) outweighs the benefit of higher anticipated returns, even over the 20 year life of the investment.

Cases A and B below show the analysis with a very low cost basis portfolio ($20 cost basis per $100 of market value) with a 20% capital gains tax rate in year 20 and a 0.8% annual return differential from rebalancing. The rebalancing option in this scenario generates a benefit of only $3 because the foregone interest from the high initial capital gains tax payment is sizable. In fact, the pre-tax year 20 value of the rebalancing option (Case A) is lower than the pre-tax value of the no-rebalance option (Case B), but because of the latter’s extremely low cost basis, the future capital gains tax is very punitive, giving the rebalancing option a higher post-tax account value.

Cases C and D show the same scenario, but with a step-up basis and thus an effective future capital gains tax rate of 0%. This generates a very large benefit for the no-rebalance scenario (Case D) because the very low $20 cost basis of the portfolio does not generate a large future tax bill. In this case, not rebalancing yields a $12 benefit over the rebalance option.

The next table performs a scenario analysis by varying the current portfolio cost basis (y-axis) and anticipated return differential (x-axis), while assuming a step-up basis upon inheritance, and thus an effective future capital gains tax rate of 0%.

The rebalancing option is still almost always beneficial when the anticipated return differential is close to 1% (100 basis points) per year, even with a step-up cost basis. The benefit of rebalancing is now lower for the low cost basis portfolios (because having zero capital gains tax rates in the future is so beneficial for portfolios with a low future cost basis, i.e., those the result from not rebalancing today), but it is sometimes higher for today’s high cost basis portfolios (because all of the anticipated higher pre-tax account value after rebalancing stays in the account, and does not have to get paid out in taxes).

Portfolio level analysis

The above analysis can be nicely explained in the context of modern portfolio theory. In the chart below, the blue curve—known as the efficient frontier—shows, for every level of expected return (y-axis), the minimum volatility (x-axis) at which that expected return can be achieved.

The investor’s high unrealized gain portfolio can be thought of as the red X in the figure. It has an anticipated return of 8% per year and a return volatility (which measures how much the portfolio can change in value in a given year) of 18%. This portfolio is not optimal. For example, the investor would be better off rebalancing to the green triangle because that portfolio has the same volatility as the red X (18%), but a higher anticipated return (8.9%). The 0.9% gain in annual anticipated return can be used in the above scenario analysis to determine the dollar benefit of paying capital gains taxes now and rebalancing.

Another portfolio that is preferable to the red X is the green square. It has the same anticipated return (8%) but a much lower volatility (15.7% per year). (In fact, any portfolio in the quasi-triangle demarcated by the dashed lines and the blue curve is better than the red X.) Using the tools of modern portfolio theory, we can translate this 2.3% per year volatility differential into an implied difference in anticipated returns of around 1.55%.2 Again, an inspection of the above scenario analysis tables shows that with this anticipated return differential, rebalancing immediately is almost always beneficial (in fact, the tables only show return differential up to 1%).

Caveats and conclusion

Several real-world considerations may impact the conclusions:

• In the first part of the analysis, we ignored uncertainty. Assuming a deterministic annual return of 8% or 9%, while simplifying the analysis, may overstate the benefit of immediately rebalancing by not considering possible adverse return paths. This can be evaluated using Monte Carlo studies, but is beyond the scope of the present article.

• Related to the above concern, we are not explicitly considering that portfolios consist of multiple securities with correlated return streams and different risk characteristics (like the propensity to experience large drawdowns), which may not be well represented by modeling the portfolio as a single return stream. Again, Monte Carlo studies can be used to analyze this scenario.

• We did not consider 351 exchanges, a new type of financial structure which may allow for efficient diversification (e.g., moving from the red X to the green triangle) without paying any capital gains taxes in the short term.

• The tax code may change in the future. For example, capital gains tax rates may be higher or lower, or there may be a state capital gains tax in currently exempt states. Tax code changes can increase or decrease the anticipated benefit of an immediate portfolio rebalance.

These, and potentially other, considerations impact the benefit of portfolio rebalancing in the presence of capital gains taxes. But when the anticipated gain in returns is of the order of 1% per year or more, or the drop in portfolio volatility is meaningful, and the planning horizon is in the 20+ year range, it often makes sense for investors to rebalance legacy portfolios. This is true even if doing so generates capital gains taxes in the short run.

This article is for educational purposes only and is not investment advice. Please consult with a tax professional and your investment advisor before deciding on a portfolio rebalance or making other investment decisions.

Subscribe now

Working with QuantStreet

QuantStreet is a registered investment advisor. It offers financial planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic, data-driven, and shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our email list.

I’ve now spent the last ten years working on applying NLP tools to the study of financial markets (you can check out this work here), on teaching this material to our MS and PhD students at Columbia Business School, and the last four years on setting up QuantStreet, an asset allocation and wealth management firm.

Subscribe now

Since the release of ChatGPT on November 30, 2022, the pace of technological innovation in AI has grown exponentially. Today’s AI models can ingest and intelligently reason about hundreds of pages of information without any additional training. This creates an opportunity to develop what truly are super-human news summarization engines: Feed several hundred pages of news into an AI model, provide it with an appropriate set of questions—for example, about how the news impact portfolio positions—and let the model generate in a few minutes output that would take dozens of man-hours, if not more.

QuantStreet’s approach

Our approach to the news summarization problem involves three steps.

1. Collect daily news headlines and article snippets. Keep these in a central storage repository for later use.

2. Access a state-of-the-art large language AI model (LLM), ideally switching to the newest model version when it becomes available.

3. Periodically collate the last several days or weeks worth of news, combine this with information about portfolio positions, and feed the combined text data (potentially hundreds of pages of text) to the LLM to summarize the news flow and explain how each portfolio position is impacted by the news.

Because this is not a process that can be implemented on one’s home PC, we use Google Cloud’s compute infrastructure, for scheduling and running jobs, storing the captured data, and then feeding the collated text file into a Gemini LLM using the Vertex AI platform. This process took us a few days to set up, but without access to the cloud compute, storage, and AI architectures, it would have taken a team of engineers months (if not longer) to create something similar.

Caveats

Setting up this monitoring process highlighted something that dedicated readers of financial news already knew from experience: News headlines predominantly skew negative. (This is something the academic literature has also pointed out.) Without providing the model proper context, the LLM’s interpretation of recent news flow also skews highly negative. In fact, asking the naive question of “How does this set of news impact portfolio positions?” often leads the model to come up rather cataclysmic interpretations, which are not useful for making actual portfolio decisions. Prompt engineering—i.e., asking the LLM model the right set of questions—is a crucial component of designing a proper AI-driven workflow, and it took us some time to develop a prompt that leads to reasonable answers in our use case.

Below is the model’s analysis of news flow, as of the evening of Saturday, June 21st, 2025. The news fetch did not include the U.S. attack on Iran’s nuclear infrastructure, so you will not see this specific event covered in the analysis. Also, despite our careful prompt design, the model’s analysis still skews negative for many portfolio positions. The human decision maker ultimately has to account for this bias in his or her interpretation of the model output, but the LLM’s ability to synthesize large amounts of news data and explain succinctly how these impact multiple portfolio positions is stunning.

Gemini’s analysis as of June 21, 2025

The following section was generated using an artificial intelligence model based on publicly available news sources. It is provided for informational and educational purposes only and does not constitute investment advice or a recommendation to buy, sell, or hold any security. AI-generated content may be incomplete, inaccurate, or reflect automated interpretations that differ from QuantStreet’s views. QuantStreet does not independently verify the accuracy or completeness of the analysis below and uses this type of analysis as only one of many reference points. Actual market outcomes may differ from any interpretations expressed. 

1. Summary of News Events

Recent news is dominated by escalating geopolitical tensions, primarily the intensifying conflict between Israel and Iran, with increasing discussion of potential U.S. involvement and its impact on global markets, particularly oil prices. Alongside this, the ongoing war in Ukraine and U.S. domestic political actions, including presidential directives on immigration and military deployments, contribute to a climate of uncertainty.

Technologically, news highlights the continued advancement of AI, with new models and applications emerging from major tech companies, though concerns about AI’s societal impact (e.g., job displacement, misinformation, data privacy) and operational risks (e.g., mistaken bans, bad advice) are also present. The successful launch of a new gaming console and major data breaches underscore the dual nature of the tech sector.

In health, a resurgence of measles cases in the U.S. and the identification of a new, severe COVID subvariant indicate ongoing public health challenges. However, positive developments in medical research, such as an experimental cholesterol pill, gene therapy advancements (despite an associated death), and promising results for psilocybin in treating depression, suggest progress in healthcare innovation.

Economically, the Federal Reserve’s projections hint at stagflation concerns, adding to market caution. Discussions around potential changes to U.S. tax incentives for electric vehicles and renewable energy sources signal possible shifts in industrial policy. Consumer-focused news includes significant box office success for a new film, major sporting events, and labor policy changes affecting hospitality and agriculture sectors.

Overall, the news reflects a mix of significant geopolitical risks, continued technological innovation with associated challenges, persistent public health issues alongside medical breakthroughs, and a cautious economic outlook.

2. Summary of Impact on Each Portfolio Position

Here is an analysis of how the recent news impacts the operating fundamentals and risks of each portfolio position, sorted alphabetically:

• IAU (iShares Gold Trust)

  • Impact: Moderately Positive

  • Confidence: High

  • Reasoning: The escalating conflict between Israel and Iran, with discussions of potential U.S. involvement, significantly increases global geopolitical uncertainty. News of the Fed’s economic projections hinting at stagflation (slow growth with high inflation) further contributes to a risk-off sentiment and inflation hedging. Gold (IAU) is traditionally considered a safe-haven asset, benefiting from increased geopolitical risk and concerns about economic instability or inflation. These headlines are likely to increase demand for gold, supporting its price.

• QQQ (Invesco QQQ Trust)

  • Impact: Mixed to Slightly Negative

  • Confidence: Moderate to High

  • Reasoning: QQQ is heavily weighted towards large-cap technology and growth companies. While there are positive developments in AI (e.g., Google Gemini 2.5 Pro, Midjourney AI video model, AI for brain understanding) which are tailwinds for many of its core holdings (e.g., Google, Meta), several headlines present significant risks. The massive data breach affecting Google, Apple, and Facebook poses material cybersecurity, reputational, and regulatory risks for these key components. Reports of Instagram’s AI-related mass bans point to potential operational and reputational challenges for Meta. Furthermore, Trump’s proposed bill to end EV subsidies is a direct negative for Tesla, a prominent QQQ holding, by potentially impacting demand for electric vehicles. The Amazon CEO’s comments on AI reducing corporate jobs could signal efficiency gains but also broader economic disruption concerns.

• VGK (Vanguard FTSE Developed Markets ETF)

  • Impact: Slightly Negative

  • Confidence: Moderate

  • Reasoning: VGK has significant exposure to European and other developed non-U.S. markets. The heightened geopolitical tensions, particularly the Israel-Iran conflict and its potential for broader regional instability, pose a direct risk to global economic sentiment and supply chains that impact these international markets. While specific company news (like Estee Lauder founder’s death) or localized events (like the Air India crash) have limited broad impact, the overall increase in international political risk is a headwind. The potential for expanded U.S. travel bans could also indirectly affect international tourism and business, impacting some companies within VGK’s holdings.

• VGSH (Vanguard Short-Term Government Bond ETF)

  • Impact: Neutral to Slightly Positive

  • Confidence: High

  • Reasoning: VGSH invests in short-term U.S. government bonds, known for their safety and liquidity. In times of increased market uncertainty stemming from geopolitical conflicts (Israel-Iran) or domestic political tensions, there can be a “flight to quality” that benefits government bonds. While the Fed’s hint at stagflation could imply some interest rate uncertainty, short-term bonds are less sensitive to interest rate fluctuations than longer-term bonds. The primary operating fundamental for VGSH is the stability and backing of the U.S. Treasury, which remains strong despite political headlines.

• VOO (Vanguard S&P 500 ETF)

  • Impact: Mixed to Slightly Negative

  • Confidence: High

  • Reasoning: VOO represents broad U.S. large-cap equities. The escalating Israel-Iran conflict and the associated rise in oil prices are direct negatives, increasing market volatility and investor risk aversion (“Stock futures fall as investors eye Israel-Iran conflict”). The Fed’s stagflation concerns also signal potential headwinds for corporate earnings growth. While the underlying positive trends of AI adoption continue to benefit many S&P 500 companies through efficiency gains, these are counterbalanced by the immediate and significant geopolitical and macroeconomic risks highlighted in the headlines. The large data breach also represents a broad risk to consumer and business trust, impacting numerous companies.

• VXUS (Vanguard Total International Stock Index Fund ETF Shares)

  • Impact: Negative

  • Confidence: Moderate

  • Reasoning: VXUS offers broad international equity exposure, including both developed and emerging markets. The intensifying Israel-Iran conflict has direct negative implications for Middle Eastern markets and significant potential for broader global economic disruption and increased risk premiums for international investments. The potential for expanded U.S. travel bans adds further uncertainty to international trade and tourism. While there was news of a peace deal in Rwanda/DR Congo, its impact is too localized to significantly offset the broader geopolitical risks affecting a diversified international fund.

• XLC (Communication Services Select Sector SPDR Fund)

  • Impact: Mixed

  • Confidence: Moderate to High

  • Reasoning: XLC includes major technology and media companies (e.g., Google, Meta). The continued advancements in AI (Google Gemini 2.5 Pro, Midjourney AI video model) are fundamentally positive for these companies’ innovation and future growth. WhatsApp’s move to a subscription model and ads could also represent new revenue streams for Meta. However, these positives are significantly counteracted by negative news, particularly the massive data breach affecting Google, Apple, and Facebook, which creates substantial operational and reputational risk. Reports of Instagram’s AI-related mass bans also highlight potential platform instability and user dissatisfaction.

• XLF (Financial Select Sector SPDR Fund)

  • Impact: Mixed to Slightly Positive

  • Confidence: Moderate

  • Reasoning: The Senate’s passage of a stablecoin bill is a significant positive for the financial sector, providing regulatory clarity and potentially opening new avenues for growth in digital assets for banks and financial services firms. The private equity firm financing Harvard research also suggests continued robust activity in capital markets. However, the Fed’s hint at stagflation concerns could pose headwinds for traditional banking activities like lending, potentially impacting loan growth and asset quality. Geopolitical tensions could also contribute to overall market volatility, which can affect trading and investment banking revenues.

• XLRE (Real Estate Select Sector SPDR Fund)

  • Impact: Slightly Positive

  • Confidence: Moderate

  • Reasoning: Trump’s directive to curb immigration enforcement at hotels and restaurants is a positive development for the hospitality sector, a significant component of real estate through hotel REITs and commercial properties. This policy could alleviate labor shortages and support demand for related commercial real estate. While there were reports of deadly flash floods, these are localized events that do not typically impact the diversified national portfolio of XLRE. Potential Medicaid cuts could negatively affect real estate related to rural healthcare facilities, but the broader impact seems limited compared to the positive for hospitality.

• XLU (Utilities Select Sector SPDR Fund)

  • Impact: Mixed to Slightly Negative

  • Confidence: High

  • Reasoning: The forecast for a significant heat wave across the U.S. is generally positive for utilities, as it drives increased demand for electricity for cooling, boosting revenues. However, the news that the Senate wants to end wind and solar tax credits is a significant negative for the utilities sector, especially for companies heavily invested in or planning to invest in renewable energy projects. This could increase the cost of capital for green initiatives or slow the transition to renewables, impacting future growth and profitability for a segment of the sector.

• XLY (Consumer Discretionary Select Sector SPDR Fund)

  • Impact: Mixed to Slightly Negative

  • Confidence: High

  • Reasoning: Trump’s move to curb immigration enforcement at hotels and restaurants is a positive for the hospitality and restaurant sub-sectors within XLY, by ensuring labor supply and supporting operational stability. The strong box office performance of “How to Train Your Dragon” and the FIFA Club World Cup indicate healthy consumer spending on entertainment and leisure. However, Trump’s proposed bill to end EV subsidies is a significant negative for Tesla, a major holding within XLY, which could dampen EV sales and impact the company’s fundamentals. The broader economic concerns, including hints of stagflation, could also lead to reduced consumer spending on non-essential goods and services, posing a risk to the entire sector. The massive data breach could also erode consumer trust and impact online spending habits.

QuantStreet is a registered investment advisor. It offers financial planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic, data-driven, and shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our email list.

The dollar sell-off has been accompanied by a large underperformance of U.S. stocks relative to international peers. The next chart shows that the S&P 500 (SPX) has underperformed global stocks ex-U.S. (MXWOU) by over 13% year-to-date. Part of this underperformance is attributable to dollar weakness (since MXWOU is the U.S. dollar price of foreign stocks).

Subscribe now

While this year’s dollar weakness and U.S. stock market underperformance are notable, neither is particularly anomalous in a historical context. The next chart shows the worst three-month dollar return and U.S. stock market underperformance (relative to MXWOU) in each year going back to the 1980s.

The blue bars show that this year’s dollar sell-off of around 10% has happened many times in the last several decades, most recently in 2022 and again in 2023. Both of those episodes were associated with U.S. stock underperformance, shown by the red bars, though this year’s trough underperformance is worse by a few percentage points. The last time the U.S. stock market was this much weaker than its global peers was during the Global Financial Crisis of 2009.

Nevertheless, in the grand scheme of things, this year’s dollar weakness and stock market underperformance are not particularly anomalous.

Why did it happen?

In a late-2024 research note on how to restructure the global trade system, Stephen Miran, the chairman of the President’s Council of Economic Advisors, wrote that:

“The classic reason currencies offset changes in tariffs is that tariffs improve the trade balance, which then puts upward pressure on the currency for traditional reasons.”

That is, it was Miran’s belief (and the belief of many others) that once the U.S. began to impose tariffs on goods imports to the U.S., the dollar would strengthen, thus offsetting some of the tariff pain for U.S. consumers.

The “traditional reasons” cited by Miran reference a paper by Jeanne and Son (2024), who analyze the impact of a permanent increase in tariffs on the home country’s currency. In Jeanne and Son’s words: “In all cases, the exchange rate moves in a direction that tends to offset the impact of the tariff: a tariff on imports appreciates the home currency and a tariff on exports depreciates it.” The mechanism works something like this: A permanent and unanticipated increase in tariffs on imports causes the price of imports to become high relative the price of the home good. Because (in the model) it is costly for firms to adjust prices, the adjustment to bring the relative price of the goods back into equilibrium happens through an appreciation of the home currency (thus making the import good cheaper in home currency terms).

The model is complex, and involves both supply and demand components. A simpler story is that an improved trade balance means there are fewer dollars leaving the U.S. to buy foreign goods, which effectively shifts the dollar supply curve to the left, thus increasing the price of the dollar in units of foreign currency. But the end results is the same: higher tariffs, stronger dollar.

Unfortunately, this is not (yet) how this particular tariff cycle has played out. Rather than dollar strength, we have seen pronounced dollar weakness, and alongside that, an underperformance of U.S. assets relative to foreign peers. Goldman Sachs attributes this to a decline in international dollar demand:

“The increase in trade tensions and other uncertainty-raising policies are eroding consumer and business confidence, Goldman Sachs Research senior currency strategist Michael Cahill writes in the team’s report. The team finds that changing perceptions of US governance and institutions are also affecting the appeal of US assets for foreign investors, and the rapid back-and-forth on policy decisions makes it difficult for investors to price outcomes other than high uncertainty.”

This explanation dovetails with an emerging media narrative of a decay in the dollar’s status as a reserve currency due to “erratic policymaking and rising trade barriers” (FT April 12, 2025). The same article mentions the market’s concern that “the Trump administration [may] take further steps to depress the value of the dollar.” Though this particular concern has recently been rebuffed by members of the administration themselves: “Bessent has sought to dispel global angst that the US is actively seeking a weaker foreign exchange rate. Since February, he’s repeatedly said that the strong dollar policy remains ‘intact,’ including after Trump’s April 2 tariff announcement triggered a broad selloff in American assets.” (Bloomberg May 14, 2025).

What comes next?

The initial belief that tariffs will result in a stronger dollar have been replaced by the fear that the dollar’s status as the global reserve currency is being undermined by international angst about the Trump administration’s economic and trade policies. So what comes next?

The next chart provides a historical take on this question. The x-axis in the graph shows decile buckets for dollar returns over the prior three months. The -10% trough return experienced by the dollar this year places the current situation into the 1st (worst) decile, which contains three-month dollar outcomes between -11.8% and -5.29%.

The y-axis shows the next three-month return starting from each decile bucket. The box in each decile shows the median (solid line) and the interquartile range (IQR) for future outcomes (between the 25th and 75th percentile). The top horizontal line (top whisker) represents the 75th percentile plus 1.5 * IQR and the bottom horizontal line (bottom whisker) represents the bottom 25th percentile minus 1.5 * IQR. The circles in the plot show next three-month outcomes outside of the whiskers. Most extreme future moves following a three-month period of dollar weakness tend to be to the upside, with very few (only three) extreme downside moves. From a purely statistical viewpoint, the dollar has the potential to see strong next three-month returns.

Turning to the specifics of the present situation, the Trump administration has backed away from the more extreme aspects of the initial April 2nd tariff announcement. Since that time we’ve signed a trade deal with the UK and had fruitful negotiations with China, which led to a temporary (and hopefully ultimately permanent) reduction in reciprocal tariffs. There are active negotiations with many other countries.

In addition to this, the Trump administration has secured an agreement with Saudi Arabia in which they pledge to invest $600 bn into the U.S. In a gigantic order, Qatar has committed to buy 210 planes from Boeing, propelling Boeing’s stock to a 50% gain from its early-April lows. This is in addition to the hundreds of billions of dollars major corporations, e.g., Apple, Nvidia, and TSMC, have pledged to invest in high-tech manufacturing capacity in the U.S.

One may be forgiven for thinking that the Trump administration’s economic policy decisions are not quite as erratic as advertised. If market perception comes around to this point of view and as long as the administration does not embark on a weak-dollar policy—which they have said they will not do—the dollar may yet begin to rally and U.S. assets may reverse some (or all) of their recent underperformance relative to the rest of the world.

Working with QuantStreet

QuantStreet is a registered investment advisor. It offers financial planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic, data-driven, and shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our email list.

Subscribe now

Our view has been that the substantial outperformance of U.S. versus international markets has been driven (to a large extent) by faster earnings growth. The underpinnings of that growth are likely structural in nature: the U.S. features a more pro-shareholder culture, more research and innovation, and more freedom and better legal safeguards for corporations. These factors argue for continued U.S. outperformance.

However, Sharma does have a point. A lot of the good news about U.S. stocks has been priced in. The next chart shows the price-to-forward-earnings ratio—the price relative to one-year ahead analyst earnings expectations—of the U.S. market (SPX), relative to Europe (SXXP), Asia (MXAP), and world stocks excluding the U.S. (MXWOU).

Subscribe now

In the early years of the millennium, U.S. stocks traded at a price-to-earnings (P/E) discount to the rest of the world. By the time of the global financial crisis (GFC) of 2007-2009, the valuation of U.S. stocks had caught up with that of international peers, and since then U.S. valuations have raced ahead of the pack. As of the end of 2024, the U.S. P/E ratio of 24.88 was approximately 10 turns higher than the P/Es of European and Asian stock markets.

Finance 101 teaches us that stock price multiples reflect two factors: discount rates (higher discount rates mean lower prices) and expectations about future earnings growth (higher growth expectations mean higher prices). Unquestionably, part of the valuation differential reflects the belief that U.S. companies will have higher earnings growth in the future, either because of higher top-line growth, margin expansion, or greater share buybacks which increase earnings-per-share (EPS) by decreasing share count.

As the above chart shows, over the last 20 years, U.S. per-share earnings have grown at close to 7% per year, while European and Asian corporate earnings growth has been around 5%. With the tail wind of the AI revolution, the market is betting that this earnings growth differential will persist, and perhaps even widen.

Another important consideration for projecting relative stock performance is that non-U.S. stock markets now pay considerably higher dividends. The next chart shows the last-12-month dividend yield of U.S. and global stock markets. The U.S.’s 1.27% dividend yield is swamped by the 3.41% dividend yield of European stocks and the 2.42% dividend yield of Asian stocks.

The U.S.’s relatively low dividend yields and relatively high P/Es mean that U.S. earnings growth has to be that much higher than in the rest of the world for U.S. stock market outperformance to continue.

Reversion-to-the-mean returns

To understand how future returns may be impacted by these three forces—P/Es, dividend yields, and earnings growth—we perform a simple simulation. In the case of the U.S., we assume its current dividend yield will approach its historical dividend yield in a straight-line fashion over the next 10 years. Similarly, we assume the U.S. price-to-forward-earnings ratio will go from its current level to its historical average level over the next 10 years. We assume an initial earnings per share expectation of $1 (or 1 unit of local currency for non-U.S. markets) and that this grows at the compounded annual EPS growth rate observed over the last 20 years (i.e., our assumption is that earnings expectations grow at the same historical rate as actual earnings). At the end of each simulation year, the price is the end-of-year expected earnings multiplied by the end-of-year P/E ratio.

The return on invested capital (assumed to start at $100 or 100 local currency units) in a given year equals the price appreciation from that year plus the dividend yield that prevailed at the end of the prior year. For U.S. stocks, should dividend yields and P/Es revert back to their historical averages over the next 10 years, and should earnings growth equal its 20-year average, the annual rate of return would be 4.72%.

The next table summarizes this analysis for all four global markets. All returns are calculated in local currency terms. We argue in the Appendix that comparing returns while ignoring exchange rate effects is an acceptable approximation. The top row of the table replicates the results of the above chart and shows a prospective annual U.S. equity return of 4.72%. For Europe, despite its lower historical earnings growth rate, the high dividend yield and low P/E ratio combine to generate a forecasted reversion-to-the-mean annual return of 9.22% over the next 10 years. For Asia and global non-U.S. stocks, the analogous projected annual returns are 7.98% and 8.47% respectively.

Subscribe now

In the reversion-to-the-mean scenario, the anticipated U.S. stock return is low because the projected decline in P/Es and low dividend yield offset the relative high U.S. earnings growth rate. European and Asian stock returns are higher, despite lower earnings growth, because of projected P/E expansion and higher dividend yields in the mean-reversion scenario.

Analyzing more scenarios

Thus far, we’ve analyzed the single scenario of everything returning to its historical average over the next 10 years. It is possible, likely in fact, that this is not what the market believes. The market may think that U.S. P/E ratios are now at a permanently higher steady state level than in the past (due to, for example, higher expected earnings growth) and that U.S. earnings growth over the next 10 years may be higher than its historical 20-year average.

What would have to happen to 10-year ahead P/Es and 10-year ahead earnings growth rates for the U.S. market to outperform the European reversion-to-the-mean returns shown above. The next table shows the projected 10-year ahead U.S. stock returns at different assumed ending points for the P/E ratio and for different assumed earnings growth rates. To make the analysis easier, we assume the same dividend yield progression—from the current level to the historical average—over the 10 years of the simulation.

The cells shown in green (and marked with an asterisk) show scenarios where U.S. stocks outperform the European stock base case return of 9.22%. Future winning-path P/Es can be as low as 19 times, as long as earnings growth comes in at 11% per year. If earnings growth comes in at only 9%, then future U.S. P/Es would have to be at least 22 times. And even if U.S. P/Es don’t fall, i.e., end at 25 times, earnings growth would still have to come in at a robust 8% for U.S. stocks to do better than the European mean-reversion base case. (The hurdle rates for U.S. outperformance against the mean-reversion base case Asian stock return are a bit less demanding, and are shown in the Appendix.)

Is the high level of earnings growth needed for sustained U.S. outperformance reasonable? The next chart shows the annualized earnings growth over the prior 10 years for each of the four asset classes. The one episode of very high recent earnings growth was the 10-year period that began in late-2009, at the earnings trough following the global financial crisis, and which ended in late-2019, just prior to the onset of COVID-19.

In this case, the S&P 500 10-year growth rate peaked as of December 31, 2019 at a annualized rate of 10.95%. The growth rate of Asian earnings was even higher, at right around 20%. But both of these high growth rates came off of very depressed post-GFC earnings levels. While a 10-year ahead earnings growth rate of 9 or 10% is certainly possible, it is a tall order, even accounting for the productivity gains likely to accrue from AI adoption.

Next, we flip the tables and ask what sort of forward P/E ratio and earnings growth would be needed for European stocks to outperform the 4.72% return in the mean-reversion scenario for the U.S. market. This picture is exceedingly benign. Short of a lost decade of growth with disastrously low earnings coupled with a downward repricing of European equities, it is hard to see how European stocks—with their relatively low valuations, decent earnings growth, and high dividend yields—don’t eke out a higher-than 4.72% gain. (The table below has the same P/E and EPS scale as the U.S. versus Europe table shown above.)

The Appendix shows that the outperformance path for Asian stocks relative the U.S. base case is a bit more challenging than for the Europe-versus-U.S. comparison. Similarly, the Appendix shows that in most scenarios, European stocks are likely to outperform the Asian mean-reversion scenario stock returns.

The mother of all bubbles?

Are U.S. stocks in a massive valuation bubble? We don’t think so. Will U.S. stocks outperform their European and Asian counterparts over the next 10 years? Maybe. But it will require a combination of a historically elevated future P/E ratio and very robust earnings growth over the next 10 years. A more likely scenario is a bit of P/E mean reversion, say to 20 times, with strong earnings growth, say 8%, which would result in an annualized 10-year ahead U.S. return of 7.23%. This would be a perfectly good return outcome for the U.S., but it is lower than the mean-reversion base case for Europe, Asia, and the rest of the world excluding the U.S.

Of course, it is possible the U.S. really will chalk up 10%+ earnings growth (or, more precisely, growth of earnings expectations) on the back of an AI productivity boom. In that case, the U.S. market will likely again be the winning global stock market over the next 10 years. Only time will tell, but the valuation case for non-U.S. global markets is as compelling as it’s been in a while.

Looking ahead

Our philosophy at QuantStreet is to combine tactical and strategic thinking. The above analysis falls into the strategic category: Based on a long-run analysis of P/E progression and earnings growth, the bar for U.S. outperformance is very high. Europe and Asia have an easier path—just reversion to the mean—to achieve future outperformance versus the U.S. But long-term thinking and medium-term thinking in markets often don’t align. In the medium-term, other dynamics matter besides valuations and earnings growth. These latter considerations are captured by our machine learning one-year ahead return forecasting model blended with a one-year (ex-last-month) trend signal. Portfolios should reflect both strategic and tactical considerations.

QuantStreet is a registered investment advisor. It offers wealth planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services. The firm’s approach is systematic and data-driven, but also shaped by years of investing experience. To work with or learn more about QuantStreet, contact us at hello@quantstreetcapital.com or sign up for our firm mailing list.

Subscribe now

Appendix

Our analysis thus far has ignored currency effects. Fortunately, the relative level of the euro versus the dollar has not shown any evidence of a trend over the last 50 years (the data prior to 1999 are from a euro-like currency basket).

The same is true of the dollar against a broad currency basket over an even longer time frame.

The next table shows the ending P/E and earnings growth assumptions needed for U.S. outperformance relative to the Asian mean-reversion base case.

The next table shows the ending P/E and earnings growth assumptions needed for Asian outperformance relative to the U.S. mean-reversion base case.

The next table shows the ending P/E and earnings growth assumptions needed for European outperformance relative to the Asian mean-reversion base case.

The next table shows the top companies in Vanguard’s VGK ETF, which tracks the European market. The top companies are global leaders in semiconductors, software, healthcare, food and drink, energy, fashion, and finance. The composition of the MXAP index is shown here.

In a recent LinkedIn post that references CAPE, AQR writes that “Equity valuations matter (eventually). While equity valuations have continued to rise from already-stretched levels and recent strong stock returns have bolstered many investor portfolios, the future may not be as bright.” James Mackintosh — generally not prone to bouts of optimism about markets — also wrote about the currently high CAPE ratio in a recent WSJ article titled “Is This Wildly Overvalued Stock Market Doomed? Yes, but Maybe Not Yet.” He went on to explain that “History shows no link between nosebleed valuations like today’s and next year’s returns. Expensive stocks can always get pricier.”

Subscribe now

We agree with the first part of his explanation: Over the last few decades, elevated levels of CAPE have had no bearing on year-ahead stock market returns. As for the second part, if high levels of CAPE are not associated with future low returns (at least in the past several decades), then one has to question what exactly is the meaning of stocks being “expensive.”

Trend in CAPE

One interpretation of the above figure is that the expected return on stocks is now lower than it has been historically. Applied to market returns over the next ten years, this is the conclusion reached by Goldman Sachs and Bank of America according to the above James Mackintosh article. But there are two alternative explanations that should be considered:

1. Investors may expect faster future earnings growth, which could justify a higher price-to-earnings ratio without implying lower future returns.

2. Corporate earnings today — thanks to things like the non-capitalization of R&D and marketing expenses — may not be directly comparable to corporate earnings of yore due to disparate accounting treatment.

We wrote about both phenomena in an earlier piece and point the interested reader there. (We plan to explore the first point in more detail in a future research piece.)

If either of these factors is at work — and especially if the accounting treatment of earnings is structurally different now than in the past — then the average level of CAPE may simply have risen over the years. The next picture shows clearly that CAPE has had a pronounced time trend over the last 140 years.

Yet, as the bottom panel of the figure shows, even relative to its pronounced upward trend, the current detrended CAPE ratio is still elevated (though not quite as elevated as the unadjusted CAPE ratio). In light of this, isn’t the market truly overvalued?

What does CAPE forecast?

The overvaluation question is typically addressed by bucketing months based on their ending levels of CAPE and then checking what the forward return looks like in each CAPE bucket. (This is what the AQR piece from earlier did.)

Our version of this analysis is shown in the next table. Each row in the table corresponds to one of nine CAPE buckets, from lowest to highest. [1] The columns corresponds to the average year-ahead return following months that fall into each CAPE bucket, as well as the minimum, maximum, and standard deviation of these year-ahead returns.

Since 1881, higher detrended CAPE values, i.e., the deviation of the CAPE ratio from trend, have indeed been associated with lower future returns. The lowest detrended CAPE bucket is associated with a 19.7% average year-ahead return while the highest bucket is associated with a 2.4% average year-ahead return. The current detrended CAPE falls into this highest bucket, leading some to conclude that future market returns are likely to be low. However, the full sample averages shown in the table — taken using 140+ years of data — hide important time series variation in how CAPE ratios are related to future returns.

When do the forecasts take place?

The next figure shows each month that falls into one of the nine detrended CAPE buckets from the above table and the associated one-year ahead return. For example, the upper left panel in the figure shows the lowest detrended CAPE bucket which contains months with end-of-month detrended CAPE values between -13.4 and -7.9. The average one-year ahead return following months that fall in this bucket is 19.7%. However, the majority of these data points are from prior decades. The only relatively recent months in this bucket are from the time period of the global financial crisis.

The lower right panel shows the highest detrended CAPE bucket, with CAPE values from 6.7 to 22.7. Conditional on being in this high valuation bucket, year-ahead returns are very low, with an average of 2.4%. However, many of these low returns, and certainly the most negative ones, came in the aftermath of the Great Depression (the local maximum of CAPE happened in September of 1929 at a value of 32.6.). It is hard to argue that the Great Depression stock market experience was caused by high market valuations, rather than by a cascade of poor macroeconomic conditions and bad policy decisions.

The next figure keeps the same bucket classifications, which use the full data sample, but zooms in on the post-1990 time period. Here, the most frequent detrended CAPE bucket was the high-valuation one, with 108 months falling in this bucket, versus 59 for the second most frequent bucket (which has detrended CAPE values of -1 to 1). In particular for those waiting to invest in the stock market until detrended CAPE falls into one of the three low valuation buckets, you might find yourself out of luck because these buckets contain months that are predominantly from the pre-1990 time period. Almost the entirety of the strong stock market performance over the last three decades took place from relatively high detrended CAPE buckets.

Since 2003, average year-ahead returns in the highest detrended CAPE valuation bucket are 7.7%, and they would be even higher if not for the 2020 market selloff associated with the onset of COVID. It is hard to argue that the COVID selloff was caused by high stock market valuations, rather than factors associated with the pandemic.

For those concerned that detrended CAPE is forward looking — since it uses the entire data sample to take out the time trend — the Appendix repeats the analysis using raw, non-detrended CAPE. The results from using raw CAPE are even more supportive of the thesis that high values of CAPE in the recent few decades had no bearing on future stock market returns.

Is the current market just like the late 1990s?

Repeating the above analysis using three-year forward returns and detrended CAPE since 1990 shows that the highest detrended CAPE bucket had an annualized three-year ahead return of 5.2%. Excluding the dot-com bubble would make the three-year ahead return in the highest (and current) detrended CAPE bucket look downright rosy. (Note that the global financial crisis selloff did not happen from highly elevated CAPE levels.) Concerns about current valuations therefore critically depend on whether now is just like the late 1990s.

Current valuations are nowhere near as stretched as market valuations were in the late 1990s and our view is that current tech euphoria is much more grounded in earnings than the late-1990s market bubble. (Some market commentators agree with this, but others do not.) Taking Nvidia as a case in point, while its stock price is up massively over the last few years, its price to expected earnings is flat suggesting the entirety of the runup is due to investors’ upward reassessment of the company’s earnings.

Another data point is the price-to-expected-earnings ratio for the broad technology sector using the MSCI USA Information Technology Index (the Nasdaq price-to-expected-earnings ratio doesn’t go back to the early 1990s). The recent sharp runup in technology stock prices has been supported by increasing earnings expectations. The price-to-expected-earnings ratio of the tech index hit a 63.5 multiple prior to the dot-com bubble crash, whereas today the index is trading at around 35 times forward earnings. While this is a historically high valuation multiple, it is nowhere near the market excesses of the dot-com bubble. The recent runup in prices has been much more fundamentally driven than the dot-com bubble market rally.

Takeaways

The market-is-overvalued argument is predicated almost entirely on the analogy of today to either the dot-com bubble or to the Great Depression. In both of those cases a high CAPE level was followed by a market selloff. But those are just two data points, and neither is particularly apropos to the state of markets today. Over the last 20 years — outside of the COVID selloff which was not driven by high valuations — there is not a strong relationship between CAPE and future returns.

It is, of course, impossible to know what the market has in store for investors in the coming months and years. But using CAPE would not have been an effective forecasting variable for market returns over the past 20 years. Perhaps CAPE will prove more prescient in the years ahead. Our bet is the next market selloff will happen for an as-of-yet unknown reason and the current high historical levels of detrended CAPE are not particularly worrisome based on the last several decades of data.

Subscribe now

[1] We use nine CAPE buckets because 3x3 graphs use space more efficiently than 5x2 ones.

Appendix

The next table repeats the analysis from the main part of the article but using raw, not detrended, CAPE.

Since 2003, raw CAPE has spent the vast majority of its time in the upper tercile of its historical range. Either the market has been extraordinarily imprudent over the last 20 years, or the level of CAPE from 1890 may not have much to say about the current level of CAPE.

Large selloffs

Not surprisingly, we find that the S&P 500 index is prone to large selloffs. We count 13 since the 1960s. Connecting potential forecasting variables to the incidence of past selloffs using two different neural network models shows that, ex-post (i.e., after the fact), these selloffs might have been forecastable using commonly available economic and market data. Unfortunately, we doubt that this feat could have been accomplished nearly as accurately in real time. However, even if we don’t believe future selloffs are (fully) forecastable, it does not mean that this modeling exercise is for naught.

Subscribe now

Our analysis reveals that the state of markets and the economy today does not resemble the state preceding prior large selloffs. Of course, we can have a selloff for a whole new set of reasons that were not observed historically, but at least we can rule out some of the reasons for why we had selloffs in the past.

To make matters more concrete, let’s define that we mean by a market selloff.

How do you know when a selloff ends?

Our definition of selloff is a market drop of 15% or more from a prior peak. The selloff ends once a 10% retracement has taken place. Any subsequent selloff constitutes a brand-new selloff episode. The blue line in the top panel of the next chart shows the cumulative return on a dollar invested in the S&P 500 index in 1960. The orange line shows the life-to-date maximum return earned by the initial $1 investment. When the blue line dips below the orange line, we call that a market drawdown. This drawdown, in percent, is shown in the bottom panel of the figure.

Selloff starts are indicated by the right green triangles. Selloff troughs are identified by down purple triangles. Keep in mind that you would not have been aware of a selloff trough in the month in which it took place because you would not have known whether the market would sell off in the next month or not. The left red arrows indicate the ends of selloff episodes, which occur following a 10% or higher retracement from the trough. Because of this 10% retracement rule you would have known about the end of the selloff in the month in which it took place. When we analyze post-selloff returns, we only start counting after the first 10% of the post-trough bounce has already taken place.

AI and large market selloffs

To find the association of current economic and market conditions with future selloffs, we first identify 12-month periods with a start-to-trough selloff of at least 20% (this is a slightly different definition of selloff than in the prior section, but it generates many more selloff episodes to train the model).

The blue line in the above chart shows past occurrences of 20% or larger selloffs in the subsequent 12 months, and the red and orange series show the model probabilities of such selloffs. The orange line shows the forecast from a simple, one-layer neural network designed to predict such selloffs. The model takes as inputs several market and macroeconomic variables, and identifies an association between these and future large selloffs. While the simple neural network captures several of the selloff events, it does a poor job with the 1987 market crash. It also predicts some historical selloffs which never took place, such as its large selloff forecast just prior to 2000.

One of the key advances in machine learning over the last several decades has been the advent of deep learning, which involves training neural networks with multiple layers, to better capture complex, real-world relationships. The red line in the above figure shows the large selloff forecast resulting from a multilayer neural network model. While this also is not perfect, it does a (slightly) better job with the 1987 crash, and shows considerably fewer false positive, i.e., forecasts of large selloffs that never materialized.

Of course, neither the simple neural network model, nor its deep learning cousin, is perfect. All these models can do is find a relationship between forecasting variables and past selloff episodes. It is possible – indeed likely – that future selloffs will happen for very different reasons, and that our market and macroeconomic forecasting variables may not be rich enough to capture the relevant state of the world for forecasting future selloffs. That being said, the models can answer the question of how likely it is for a large selloff to occur, if it were to occur for a reason that was observed in our training sample.

With these caveats in mind, both of our forecasting models are currently generating benign large selloff probabilities. The simple model forecasts an 11.4% percent chance of a large selloff, while the deep neural network model — our preferred model — forecasts only a 0.9% percent chance of a large selloff. The model is sanguine because there are offsetting effects from the forecasting variables. Despite the fact that some dimensions of the market environment are suggestive of large future selloffs, other dimensions are not.

For example, we find that high 2-year stock returns are associated with an elevated probability of large future selloffs. And while lagged 2-year stock returns are indeed high, there are offsetting effects. For example, the 2s10s curve (the difference in yields between 10-year and 2-year Treasury bonds) was often very steep (i.e., 10-year rates high relative to 2-year rates) prior to past selloffs, and this is not currently the case. Similarly, economic news sentiment prior to past selloffs has typically been negative, and economic sentiment today is benign. The net result of these offsetting effects is that the model’s large selloff forecast is currently quite low.

What happens next?

The next table shows the return on U.S. stocks in 12-month periods which followed the ends of prior large selloffs. Keep in mind that large selloffs end strictly after the first 10% of the post-trough bounce has already taken place. Thus, the ends of large selloffs are identifiable in real time. Of the 13 large selloffs seen in U.S. markets in the prior 60 years, 12 of them were followed by strongly positive returns over the subsequent 12 months. Only once – during the bursting of the dot-com bubble – was a large selloff followed by another large selloff. In 12 of the 13 cases, large selloffs were followed by large rebounds. The average post-selloff 12-month return is 21.5%. And this is after leaving the first 10% of the rally on the table. Those finding themselves in possession of investable assets in the aftermath of large selloffs are likely to be investing into a good future environment for stocks.

Our conclusion is that it is better to manage the risk in your investment portfolio in a way so that, when the inevitable selloff does come, you are not left scrambling to liquidate your positions and being forced to move to cash. Instead, if you were able to properly size your risk taking prior to the selloff, then rather than liquidating you may be able to add to your stock exposure at a time that is historically associated with high future returns.

For our part, we will continue to regularly update our large selloff forecasting model and will share the results with our clients. No model is guaranteed to be able to forecast future selloffs in real time, because these often happen for reasons that have never before been seen. Nevertheless, it is useful to have a mapping from market states to future selloffs, and to use this lens to monitor the current state of markets.

Subscribe now

Working with QuantStreet

QuantStreet is a registered investment advisor. We offer wealth planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services to our clients. Our approach is systematic and data-driven, but also shaped by years of investing experience. If you are an existing client or if you are thinking about working with us, we'd love to hear from you. Please reach out to us at hello@quantstreetcapital.com.

Over the last 20 years U.S. stocks and bonds have meaningfully outperformed their global peers, both in absolute terms, and on a risk-adjusted basis. The next chart shows the average annualized excess returns [1] of different asset classes, plotted against their annualized volatility. The returns of international asset classes, highlighted in red, are shown in U.S. dollar terms. Points to the upper left of the chart are preferable, because they have higher excess returns (y-axis) and lower risk (x-axis). It is clear that the international asset classes are in all cases far removed from the upper-left-most points of the figure.

Subscribe now

World non-U.S. government bonds (WBNDXUS) are clearly dominated by U.S. bonds (USIG, USTIPS, USTREAS) which all have higher average excess returns and lower risk. World stocks ex-U.S. (MXWOU) is similarly dominated by the S&P 500 (SPX), as well as by several sector indexes (healthcare, utilities, and REITs) all of which have offered higher excess returns with lower risk over the last 20 years. Finally, emerging market (EM) stocks (MXEF) and EM ex-China (EMXC) are similarly dominated by the aforementioned U.S. indexes, as well as by Nasdaq, midcaps (MID), and small caps (R2000), all of which have higher excess returns and lower volatility than EM stocks.

The Sharpe ratio (SR), defined as the average excess return of an asset class divided by its volatility, quantifies the amount of return investors earn per unit of risk taken in a given asset class. Higher SRs are desirable since they indicate higher return to risk ratios. As the next chart shows, over the last 20 years virtually all U.S. stock sectors, except financials (FINS), had higher Sharpe ratios than their international peers. Similarly, U.S. bonds had higher Sharpe ratios than international non-U.S. government bonds.

In the last 20 years, U.S. stocks and bonds had considerably higher absolute and risk-adjusted returns relative to their international peers. In the remainder of this article, we explore why U.S. stocks have outperformed and ask whether this outperformance is likely to continue.

Drivers

Certainly valuation multiple expansion is part of the story. The next chart shows the price-to-expected-earnings ratio for U.S., European, and Asian stocks. After trading roughly in line for many years, the valuation gap between U.S. and international stock markets has recently grown quite a bit wider. If the U.S. outperformance story is all about multiple expansion, then it may not persist into the future (as has been argued by Cliff Asness of AQR, for example). Fortunately, the story is not just about the valuations gap.

It is important to note that the sector compositions of U.S., European, and Asian markets are quite different. The U.S. stock market is much more tech-heavy, and the relatively higher valuation of tech firms — which partially reflects the sector’s faster earnings growth — accounts for part of the valuation gap. One manifestation of the difference in sector compositions is the growth rate of earnings, and therefore dividends, in the U.S. relative to other stock markets.

As the above chart shows, tech (NASDAQ) has seen the fastest earnings-per-share (EPS) growth over the last 20+ years. Value stocks’ (USVAL) EPS has grown slowly, relative to the S&P 500 (SPX), while small caps (R2000) and especially midcaps (MID) have seen relatively faster growth. International ex-U.S. stocks (MXWOU) have lagged U.S. EPS growth by 1.9% per year, while emerging market stocks (MXEF) have lagged U.S. EPS growth by 3.5% per year.

To get a sense of what might come next, it is important to disentangle the valuation and growth drivers of U.S. outperformance.

What happens next?

To get a sense of what future returns for U.S. and international stocks might look like, consider the basic Gordon growth model for stock prices: P = D/(R-g), where P is the stock price, D is next year’s expected dividend, R is the expected return from owning the stock, and g is the dividend growth rate. This formula says that dividends grow at g percent per year, and the price equals present values of all future dividends when discounting at the rate R. This formula can be inverted to show that the expected return is equal to the dividend yield plus the anticipated dividend growth rate, or R = D/P + g.

Since higher earnings multiples (like in the U.S.) imply lower dividend yields (D/P), the only way the U.S. might be expected to have higher future returns relative to international markets is if the U.S. ends up having much higher dividend growth rates. We already saw that the U.S. had historically higher EPS growth rates, and assuming that dividend payouts are a relatively fixed portion of earnings, the U.S. might therefore be expected to have higher dividend growth rates as well.

The next table shows that, based on the last year of data, the U.S. indeed has a lower dividend yield (1.29%) than that of European (3.25%) and Asian (2.32%) stocks. To reflect the different dividend yields, in order for all three stock markets to have 8% per year expected returns, the market must believe that U.S. stocks will have annual dividend growth rates of 6.71% (i.e., 8% – 1.29%), versus 4.75% dividend growth rates in Europe and 5.68% dividend growth rates in Asia. This implies a U.S. dividend growth differential of 1.96% versus European stocks and 1.03% versus Asian stocks. If U.S. dividend growth comes in this much higher, then U.S. stocks will earn the same 8% returns as European and Asian stocks, despite having higher valuation multiples. The higher valuation multiples may thus simply proxy for higher expected growth rates.

Another take on this question is to consider the actual historical dividend growth rates for U.S. and international stocks. These are given in blue in the above table. Interestingly, historical dividend growth rates are, in all cases, quite close to the implied dividend growth rates that would be needed to achieve 8% anticipated returns, shown in the “Implied g” row of the above table. Adding the historical dividend growth rate to the current dividend yield produces an expected return forecast for all three markets. The U.S. expected return (8.25%) is 0.15% higher than the European expected return and 0.85% higher than the Asian expected return. [2]

Given current valuation levels, as reflected in dividend yields, in order for U.S. stocks to meaningfully outperform European equities in the future, the dividend growth differential between U.S. and European stocks would need to increase from its historical level of 2.11% over the last 20 years. Similarly, to do better than the forecasted 0.85% return advantage vs Asia, U.S. dividend growth rates would need to exceed those of Asian markets by more than the historical 1.88% differential. The next section discusses some reasons why this might happen.

Why U.S. outperformance may continue

Our prior take on why U.S. stocks may continue to outperform their global peers focused on the CATO Human Freedom index, which tracks things like rule of law, property rights, court independence, personal freedom, among many other categories. By these metrics, the gap between the U.S. and the rest of the world, and especially emerging markets, is increasing in favor of the U.S. Other factors favoring the U.S. include: a culture of shareholder activism where underperforming companies are forced to change by market forces; a retirement system heavily reliant on equity market performance which aligns political and shareholder incentives; more reasonable regulation than in other countries; and world leadership in technological innovation.

Some representative headlines from recent articles highlight the issues:

• “Xi Unleashes a Crisis for Millions of China’s Best-Paid Workers,” Bloomberg, September 18, 2024.

• “Meta to European Union: Your Tech Rules Threaten to Squelch the AI Boom,” Wall Street Journal, September 19, 2024.

• “Europe Regulates Its Way to Last Place,” Wall Street Journal, January 31, 2024.

• “Is Red Tape Strangling Europe’s Growth?” Financial Times, September 9, 2024.

• “Glenview Capital Plans Push for Changes at CVS,” Wall Street Journal, September 30, 2024.

There are many other reasons for U.S. dividend outperformance as well, and a full analysis of why the U.S. has and may continue to outperform other global stock markets is beyond the scope of this article. Of course, the future may be quite different from the past and future dividend growth differential might tilt in the favor of European and Asian markets going forward, though that is not our base case.

Long-term perspective

The analysis thus far looks at the last 20 years of data, and converts all asset returns to dollar terms. The next two tables extend the analysis back to 1948, and then back to 1870, using a wonderful data set on country-level stock and bond returns made available at macrohistory.net. Returns are now presented in local currency terms and thus represents the performance available to investors in their home countries. In the post-WWII period, the U.S. and Sweden have been the two best-performing global markets, as measured by their Sharpe ratios (though of course, the size of the U.S. market is vastly larger). The next closest markets are Switzerland, the Netherlands, and Japan.

Going back even further to 1870, the U.S. is now the second best market by Sharpe ratio, trailing only Australia, though Japan is a close third. However, the outperformance of the Australian market appears to be concentrated in the 1870-1945 period because in the post-WWII sample, Australia is a relative laggard.

A key to the performance tables: Stk_ret is the average annualized stock return; Bnd_ret is the average annualized bond return; Stk_xr is the average annualized stock return above a short-term investing rate (e.g., U.S. T-bills); Stk_xr_sd is the annualized volatility of stock excess returns; SR is the annualized Sharpe ratio equal to Stk_xr divided by Stk_xr_sd; Dvd_growth is the annualized rate of dividend growth.

Summary

Based on recent, less recent, and even very old data sets, the U.S. stock market has been the global leader for the last 150 years. Our view is that U.S. market leadership will continue.

It is important to note that U.S. investors may still want to hold international assets in their portfolios even if U.S. stocks are expected to have higher future returns and Sharpe ratios, as long as international assets are diversifying enough and the expected return differential is not too great. We leave this analysis for future research, but note that given historical and potential future underperformance, the bar for including international assets in the portfolios of U.S. investors is quite high.

This article provides educational content only, and is not financial advice. Investing in U.S. and global stocks entails many risks, including the risk of large losses. Your investments should reflect your own risk preferences and liquidity needs. Investment in U.S. or global stocks may not be appropriate for you. Please carefully consider your own circumstances and/or consult a financial professional before making any investment decisions.

Subscribe now

Working with QuantStreet

QuantStreet is a registered investment advisor. We offer wealth planning, separately managed accounts, model portfolios and portfolio analytics, as well as consulting services to our clients. Our approach is systematic and data-driven, but also shaped by years of investing experience. If you are an existing client or if you are thinking about working with us, we’d love to hear from you. Please reach out to us at hello@quantstreetcapital.com.

Appendix

The next table explains the security abbreviations used in the main text.

Say you are an investor with $120 to invest. Should you allocate the full $120 to stocks on day one? Or, should you slowly allocate that money to stocks over the course of a year — say $10 per month — while keeping the remainder sitting in a savings account or in a money market fund? We refer to the latter, incremental investment strategy as dollar cost averaging.1

Subscribe now

With respect to the U.S. market hitting all-time highs, it should be noted that the S&P 500 in the U.S. and many markets globally have been hitting successive all-time highs for the last 100 years or longer. The next chart shows the S&P 500 index, in log terms to make the trend visible, going back to 1926. Historically, long-term investors would have been poorly served by not buying S&P at the highs, as new highs were perpetually on offer.

Timing market entry

For this analysis, we take dollar cost averaging (DCA) to mean that an investor with $X to invest will allocate a fixed amount of $X into stocks over time, while keeping the remainder in cash. Importantly, DCA may have psychological benefits for investors: less sense of regret; less anchoring to a fixed price level; a feeling of doing something active. We will not examine these important aspects of the strategy, but instead we focus only on economic outcomes.

To understand the investing implications of DCA, we simulate two daily trading strategies, both of which receive a $1 deposit, inflation adjusted at an annual rate of 2.5%, at the start of each calendar year. So there is $1 to be invested at the start of year one, then $1.025 at the start of year two, then $1.051 ($1 x 1.025 x 1.025) at the start of year three, and so on.

• In the first variant, investors puts the inflation-adjusted $1 into the U.S. stock market at the beginning of each calendar year. We refer to this as the start-of-year strategy.

• In the second variant, investors puts 1/252 (there are 252 trading days in a year) of the inflation-adjusted $1 into the U.S. stock market in every trading day of the year. Alternatively, investors allocate to stocks 1/252 of the value of their cash account, if this is greater.

We add two additional allocation strategies to serve as benchmarks:

• Invest the full cash amount into stocks at the high point of each year. This is not an implementable strategy because it requires knowing the high point of stocks in every year, but it provides a benchmark for a really bad timing strategy.

• Invest the full cash amount into stocks at the low point of each year. Ditto about implementability. But this serves as a benchmark for a really great timing rule.

In all cases, any uninvested cash earns interest at the short-term T-bill rate. Once allocated to stocks, capital stays in the stock account (so no going back to cash). The only timing being tested is when to use cash (or a portion of cash) to buy into the U.S. stock market.

Performance

The next chart shows the evolution of the cash account under the four strategies in the early years of the simulation. The solid blue line in the upper left shows the start-of-year fully-invested strategy. The cash account is always at zero because, as soon as money comes into the portfolio at the beginning of every year, it gets immediately allocated to the stock market. The upper right panel of the figure shows the percent of the portfolio value represented by the cash account under the dollar cost averaging plan. The saw tooth pattern indicates the receipt of an inflation-adjusted $1 at the start of each year, which then gets slowly allocated to the U.S. stock market over the remainder of the year. Successive jumps in the fraction of wealth sitting in the cash account are smaller because the value of the portfolio grows more quickly than the assumed inflation rate of 2.5%.

The lower two panels show what happens to the cash accounts (as a fraction of the portfolio value) in the max-of-year and min-of-year strategies. In both cases, cash sits in the account until a given year’s maximum or minimum stock market level is attained, at which point the cash is invested into the stock market.

The next chart shows the cumulative return, in log terms, of each strategy. Keep in mind that, in each of the four cases, an inflation-adjusted $1 deposit hits the account at the start of each calendar year. From an initial value of $1, all accounts end up in the vicinity of $300,000 at the end of the simulation period (shown in the TR column of the table below). This just shows what consistent savings combined with compounded interest can achieve.

The average return of the start-of-year strategy is 10.15% per year. Since this version of the strategy is always fully invested in the stock market, this just represents the average annualized return of the U.S. stock market over the time period of the analysis. The annual volatility of the start-of-year strategy is 17.14%. The reward-to-risk ratio, as measured by the Sharpe ratio, is 0.447. The Sharpe ratio is the ratio of the average historical return of an investment above the risk-free (e.g., T-bill) rate divided by the volatility of such excess returns.

The dollar cost averaging strategy has a lower annual return, of 9.91%, it also has a lower volatility, at 16.7%. The net effect of these two is that the Sharpe ratio of the DCA strategy is 0.442, just slightly lower than the Sharpe ratio of the start-of-year investment strategy. DCA lowers average returns but it also lower risk, while largely maintaining the same reward-to-risk tradeoff as the invest-right-away variant.

The max-of-year and min-of-year strategies

Not surprisingly, the worst strategy is the max-of-year allocation plan, which allocates the cash account into stocks on the day of the year which represents the high point of that year’s stock prices. What is much more surprising is that the almost 100-year performance of this atrocious timing strategy is still as respectable 9.25% annualized return with a 0.409 Sharpe ratio! So if, in every single year going back to 1926, you had picked the absolutely worst day on which to invest in the stock market, you would still have done very well.

On the other hand, if in every year you had a crystal ball telling you to invest at the very lowest price of that year, your annualized return would have been 10.83% with a Sharpe ratio of 0.49. So even this full-prescience, optimal timing strategy only does a little bit better than the plain invest-on-the-first-day-of-the-year variant.

The main takeaway from this analysis is that being invested in the market matters more than the exact day on which you choose to make the investment. At least this has been true in the U.S. over the past almost 100 years.

Rolling five-year windows

So far we’ve analyzed an almost 100-year investing strategy, which committed an inflation-adjusted $1 into the market in each calendar year. To examine a more practical investment horizon, we now analyze how dollar cost averaging works in rolling five-year windows, starting in 1988. In each window, an investor can allocate only $1 into the stock market. She can either do this at the start of the entire period, which we again refer to as the start-of-year strategy, or allocate daily 1/252 of the $1 into stocks starting on the first day of each five-year window.

The next chart shows the average annual returns earned by the start-of-year (SoY) and DCA strategies in each five-year window, which are labeled with their ending year (so the 1988-1992 window is labeled with “1992”). The bottom portion of the figure shows the annual difference between the start-of-year and the DCA strategy in each five-year window. DCA did better in the five-year windows ending in 2004 (i.e., 2000-2004), 2005, and 2006 because these contain the dot-com bubble, as well as in the five-year window ending in 2012 because this window begins with the 2008 market selloff associated with the global financial crisis. On average, the start-of-year investment strategies outperform the DCA ones by 0.69% per year.

The next chart shows that, while they have lower average returns, the DCA strategies are, in all cases, less volatile than the ones where the investment is made at the beginning of the five-year period. Particularly large volatility gaps occur in the five-year periods ending in 2006, 2012, 2013, and especially 2024. The latter is associated with using dollar cost averaging at the start of the 2020 COVID-19 pandemic, which would have greatly reduced the volatility of the DCA strategy relative to the full capital commitment at the beginning of 2020. Despite this large volatility gap, as of now, there is almost zero return differential in the 2020-2024 time period between the start-of-year and DCA strategies.

The next chart reports the annualized Sharpe ratios of the start-of-year allocation and DCA strategies in each of the rolling five-year windows. While there is some time variation in the relative level of the Sharpe ratios, the average Sharpe ratio difference between the two strategies is exactly zero, suggesting that the difference in returns and volatilities between allocating to stocks immediately and dollar cost averaging leave the overall return-to-risk tradeoffs of the two plans exactly unchanged.

Subscribe now

Summary

Relative to committing capital immediately to the stock market, dollar cost averaging is associated with lower returns and lower risk levels. However, the return-to-risk tradeoff, as measured by the Sharpe ratio, of the two strategies turns out to be the same. Interestingly, full-prescience strategies don’t help that much on the good side (min-of-year allocation) or hurt that much on the bad side (max-of-year allocation). Just being invested in the stock market is the key dynamic. Timing entry using DCA lowers returns by exactly the amount warranted by the drop in volatility.

Outside of behavioral issues, like experiencing regret or anchoring of price levels, investors appear to be better off just setting a targeted risk level for their portfolios, e.g., the appropriate stock-bond mix, and trading to that risk level right away. Waiting to invest simply has the effect of implicitly modifying the targeted risk level.

The caveat is that, while these observations have been true in U.S. markets since 1926, they may not continue to hold in the future. All investors must ensure that their portfolios are constructed in line with their own risk preferences and liquidity needs.

• Looking at a 150-year history of the cyclically-adjusted price-to-earnings (CAPE) ratio of the S&P 500 index suggests that the stock market is very overvalued.

• However, this analysis ignores that earnings today are not the same as earnings from 150 years ago. Adjusting for these differences makes the market overvaluation seem much less dramatic.

• Furthermore, interest rates, while considerably higher than last year, are still relatively low by historical standards. Adjusting for this makes stocks look very close to fairly valued.

Subscribe now

The Cyclically-Adjusted Price-to-Earnings Ratio

There are many perspectives on the question of whether the stock market is overvalued. In this article, we look at one measure in particular, the Campbell and Shiller cyclically-adjusted price-to-earnings ratio (CAPE). The CAPE ratio looks at the inflation adjusted price of the S&P 500 index relative to the last 10-year average of inflation-adjusted S&P 500 earnings. The benefit of averaging earnings is that this makes the CAPE ratio much less sensitive to cyclical fluctuations in corporate earnings, which tend to occur right around every economic downturn. You can download the CAPE ratio and see explicitly how it is calculated from Robert Shiller's website, as well as from other sources, like the NASDAQ data site for example.

Based on the current value of the CAPE ratio, many argue that the U.S. stock market is very overvalued. Looking at the CAPE ratio going back to the 1880s shows that the CAPE ratio is, indeed, high relative to its long-term average level of 17.6 (the orange line below).

The bottom chart shows the CAPE ratio relative to its long-term average value. Currently, CAPE is 12.8 points above its long-term average, which is one of the main data points underlying the "stocks are expensive" narrative.

Are the 1880s a good benchmark for 2023?

One of the implicit assumptions in the above analysis is that the investing world of the late-1800s is a good proxy for the investing world today. But many things have changed in the last 150 years. Three things in particular may contribute to today's price to earnings (P/E) ratios being higher, on average, than P/E ratios from the 1880s.

1. First, it is possible that the expected returns on the overall stock market have decreased over the last 150 years. This one is hard to refute definitively, but looking at realized stock returns over 20-year periods does not suggest that stock returns are lower now than in the past (in fact, they appear to be higher).

2. The earnings growth of S&P 500 stocks may have accelerated since the 1880s. This would result in higher P/E ratios through the simple algebra of stock prices being equal to expected future dividends divided by discount rates minus anticipated growth rates (i.e., D/(r-g) or the dividend discount model). Looking at the graph below, it certainly looks like the per-share earnings growth of the S&P 500 index has accelerated in the last several decades relative to the early part of the sample.

3. The other major difference, while a bit of a wonky accounting issue, is actually really important. Tangible investment by companies (e.g., buying a printing press) shows up as an asset on the balance sheet, and does not represent a hit to earnings. But intangible investment, like a firm's repository of computer programs, does not show up as an asset, and is charged against earnings. As the economy moves away from tangible to intangible capital, corporate earnings mechanically shrink because of the differences in this accounting treatment. Morgan Stanley has a good piece on this. So today's earnings may just be systematically understated relative to the earnings of 150 years ago.

Adjusting for differences

It is, of course, difficult to model all of the disparate influences that might render 1880s earnings and P/Es different from those of 2023. One "trick" is to check whether the time series in question (i.e., the CAPE ratio) has what economists call a time trend. A time trend reflects the possibility that a particular series may be steadily increasing or declining over a long period of time. The following chart repeats the one we saw earlier, but rather than assuming a constant CAPE ratio over 150 years, the graph shows the time trend in the CAPE ratio. The time trend is the best fitting straight line to the CAPE ratio so that the peaks above the line and those below exactly offset.

The first observation is that there is a distinct time trend in the CAPE ratio over the last 150 years. This effect is highly statistically and economically significant. The average level of CAPE in the 1880s was just over 10, and it is now over 20 (actually 23.9). Will this average level of CAPE fall back to its low-teens level from 100+ years ago? For the reasons already discussed, to us this seems unlikely.

Second, the bottom panel of the chart shows that the current CAPE ratio is still high relative to the long-term trend, but the difference is now 6.4 points, exactly half of what we saw in the first version of this analysis.

What about interest rates?

Recalling that a very simple model for stock prices is that they equal D/(r-g), that is the expected future dividend divided by the discount rate minus the expected growth rate, changes in interest rates, via their effect on the discount rate in the above formula, may also impact the P/E ratio of the stock market. Lower interest rates should be associated with higher P/E ratios. We now modify the above analysis to extend our trend model to also take into account the current level of 10-year Treasury rates. As the next chart shows, while interest rates are now higher than a year ago, they are still quite low by historical standards.

Taking into account the current level of interest rates produces our third and final analysis of the CAPE ratio. The orange line below, superimposed on the blue CAPE ratio, shows the level of the CAPE ratio implied by both the time trend in the CAPE ratio and the current, relatively low, level of interest rates.

The bottom panel of the chart shows the current CAPE ratio minus the model-implied CAPE ratio, where the model takes into consideration the time trend and the current level of interest rates. The CAPE ratio is still higher than the fair-value CAPE ratio level (of 26.3) but the difference is now down to 4 points. This represents a little bit of overvaluation of the U.S. stock market relative to its 150-year history, but the overpricing is much less dramatic than the raw CAPE ratio would suggest.

How is this possible?

When you are bombarded by warnings of the imminent demise of U.S. stocks, just keep in mind that comparing today's CAPEs to those of 150 years ago is really an apples-to-oranges comparison. There are two competing hypotheses here. One -- the doomsday hypothesis -- argues that the market is skating on an irrational plateau of overvaluation, a bubble sure to be imminently pricked. The other hypothesis is that things change over time, and that these changes have led to permanently higher average CAPE ratios than those that prevailed in the past.

Which view is correct? Only time will tell. Our money is on the permanently higher CAPE hypothesis.

As always, investors should make sure to position their portfolios in line with their own risk tolerances and liquidity needs.

Subscribe now

The story begins with the following set of charts which documents the public debt-to-GDP ratios of the 18 countries tracked in the Macrohistory data set. As a percent of GDP, several countries — notably Japan and the U.S. — are hovering close to their 150-year high points of national leverage.

Subscribe now

As of the end of 2020, in addition to having very high levels of debt relative to GDP, many of the Macrohistory data set countries also had interest rates close to or at their all-time lows.

Keeping in mind that the level of debt issued by governments is an endogenous outcome — in the sense that government can’t issue voluminous amounts of debt without strong investor demand to buy that debt issuance — the low interest rate levels prevalent as of 2020 suggest that investor demand for such debt was quite strong.

For an issuer’s debt load to be sustainable, two factors must coincide. First, when the debt comes due, the issuer must be able to either fully repay the debt or to refinance it by issuing new debt. Second, the interest expense associated with the borrowing must be low relative to the issuer’s income. For governments, interest expense can be measured by multiplying countries’ debt-to-GDP ratios by the average interest rate owed on that debt. The average interest rate depends on the maturities of the sovereign borrowing and on the shape of the yield curve at the time of bond issuance.

While this information is not available in the Macrohistory data set, to get a rough proxy for the interest rate expense, I average the short-term and long-term interest rates (roughly 2- and 10-year rates) reported in the Macrohistory data set and multiply each country’s debt-to-GDP ratio by this blended rate. This gives a very rough estimate of each country’s annual interest rate expense relative to its GDP (under the assumption that the interest rates at the time of issuance were similar to the current blended rate).

I define the debt service-to-GDP ratio as a country’s debt-to-GDP ratio multiplied by its blended rate. As can be seen in the above charts, for most countries in the sample, as of the end of 2020, their debt service-to-GDP ratio, i.e., the estimated ratio of their annual interest expense to GDP, was actually at all-time lows. So while many countries had very high debt levels, their typical annual interest expense was actually quite low.

Debt and interest rate levels

To dig deeper into the relationship between debt and interest rate levels, the next chart plots both series for the U.S. The data start in 1870 and end in 2024. The blue and orange lines are annual series obtained from the Macrohistory data set, and the green and red lines extend these from 2020 to today using data from the FRED website. It is visually clear that, at least for the U.S., there appears to be an inverse relationship between overall debt levels and the levels of interest rates. High debt levels are associated with low interest rates, and low debt levels are associated with high interest rates. This strongly suggests that the equilibrium outcome in global markets is that countries issue debt at those times when investors are most willing to buy that debt issuance.

To test this relationship more broadly, I regress the Debt/GDP level of all countries in all years of the sample on the countries’ blended interest rates (the average of short- and long-term interest rates) in those years. This regression should not be interpreted in a causal way — i.e., it does not imply that lower interest rates cause countries to have more debt — but it simply points out the general association of higher debt levels with lower levels of interest rates.

The regression model is extremely simple, and ignores important statistical issues like country-level variation in the level of debt and interest rates. But in very broad terms, a typical country with a blended interest rate of 5% would have an average debt-to-GDP ratio of 68.6% - 2.86 x 5% = 54.3%. For countries with blended interest rates of 0%, their typical debt-to-GDP ratio would be 68.6% and for countries with blended interest rates of 10%, their typical debt-to-GDP ratio would be 40%. The R-squared of the regression — a measure of how well the model is able to explain the data — is only 5.3%, suggesting that this very simple model does not capture a great deal of the variation in debt-to-GDP ratios across countries and across years.

Zooming in

I now zoom in on two countries: Japan and the U.S. Japan has famously had an extremely high debt load for many decades. The next chart shows the debt-to-GDP ratio for Japan, as well as a blue, shaded region. The blue, shaded region shows the range of debt-to-GDP ratios across all countries in a given year. That Japan’s debt-to-GDP ratio is at the top of the blue shaded region indicates that Japan has had the highest sovereign leverage ratio among the 18 Macrohistory countries for the last 20 years. (Japan effectively defaulted on its debt post World War 2, which explains the missing observations in the 1940s and the drop in its debt load to near zero by the 50s and 60s.)

The debt vigilantes have been so enamored of this high debt load that they have been shorting Japanese government bonds for the last two decades, alas with very little success. In fact, the trade of shorting Japanese government bonds has been so spectacularly unsuccessful, that it has been dubbed the widow-maker.

The widow maker’s ensnarement of unsuspecting bond vigilantes can be explained by the next chart.

Though it has had a high debt load, Japan’s debt service-to-GDP ratio has tracked near the bottom of the range across the 18 countries in the Macrohistory data set. The shaded region in the above chart shows the minimum and maximum range of debt service ratios across all 18 countries in each year. Apparently, the state of having high debt-to-GDP ratios and low interest rates is persistent and self-sustaining. The low interest rates make servicing even high debt loads relatively inexpensive, which removes credit risk from government bonds, and allows their yields to stay low.

Turning attention to the U.S., the next chart shows that our debt-to-GDP ratio has, indeed, risen since its local trough in the 1970s. But despite the recent rise, U.S. debt-to-GDP sits squarely in the middle of the range of other countries’ leverage ratios, both today and historically. The orange line shows the debt-to-GDP ratio calculated using FRED data, which extends the Macrohistory series to the present day. Compared to prior peaks — for example the one which followed the allies’ heavy borrowing during World War 2 — the current U.S. debt-to-GDP ratio is not at historically extreme levels.

In fact, as of the end of the Macrohistory data in 2020, the U.S. debt service ratio hovered near all-time lows. However, the rise in rates over the last few years has pushed my measure of debt service to a higher, though not unprecedented, level. Because U.S. government bonds have not all been issues in the last few years, the actual interest expense on U.S. debt relative to GDP is lower than this figure indicates (the Peterson Foundation estimate is 2.4%).

What comes next?

The above figure should be interpreted as offering a preview of what U.S. debt service would look like were the U.S. to refinance all of its current debt at the present level of interest rates. If this happens, the U.S. debt burden will become considerably more costly and nearly 6% of the U.S. GDP will be used to service our national debt.

Of course, it highly unlikely that this is the scenario will actually play out. Why?

First, with such high debt loads, an even small increase in interest rates will have a strong contractionary effect on the U.S. economy. Given the jobs numbers over the last few weeks, high interest rates are already starting to slow economic growth. As growth, and inflation, come down, interest rates will fall, and the above debt service ratio is likely to retrace a good deal of its recent run-up.

Second, should rates indeed remain at these new high levels for the foreseeable future, it is unlikely — at least one hopes it is unlikely — that the U.S. government will maintain its current spending habits and the current budget deficit will need to come down. At some point, if rates remain high, the White House and Congress will need to get their act together and reduce our national debt level until the debt service ratio falls to more moderate levels.

It is, of course, possible that the U.S. will not get its act together, and our ballooning budget deficit and high interest rates will lock the country into a negative feedback loop along the lines of what happened in the eurozone following the Global Financial Crisis. Should this happen it will be a crisis solely brought about by our own poor choices.

For me, the odds are that either scenario one (falling rates) or two (fiscal prudence), or some combination of the two, plays out and the U.S. debt service ratio will stabilize as it has over the last 150 years. Given the dollar’s relative strength, the rest of the world largely buys into this thesis. Investors have to wait and hope that policymakers do the right thing.

Subscribe now

TIPS are U.S. government bonds whose principal and coupon payments are tied to the rate of inflation. TIPS, like regular Treasury bonds, are backed by the full faith and credit of the United States. When inflation goes up, TIPS principal and coupon payments rise with the percent change in inflation, and when inflation falls, TIPS principal and coupon payments fall alongside. According to the TreasuryDirect website:

The principal (called par value or face value) of a TIPS goes up with inflation and down with deflation.

When a TIPS matures, you get either the increased (inflation-adjusted) price or the original principal, whichever is greater. You never get less than the original principal.

Subscribe now

The version of inflation which drives TIPS principal values comes from the Consumer Price Index for Urban Consumers (CPI-U), available from this page at the Bureau of Labor Statistics. The principal adjustment due to CPI-U fluctuations is called the index ratio. The index ratio is a number like 1.01, which means that the principal of your TIPS investment is 1.01 times higher than the original principal balance at the time of issue. The index ratio can be located at the TreasuryDirect website.

While the index ratio can rise in periods of high inflation, it can also fall if there is deflation. For example, it can drop from 1.01 down to 1.005 in a time period where there was -0.50% inflation. For an investor who owns $1,000 principal amount of a TIPS bond, when the Index Ratio is 1.01, the TIPS coupon rate will be paid on a principal of $1,010 (i.e., $1,000 x 1.01) and then the index ratio is 1.005, the coupon rate will apply to a principal amount of $1,005. A more detailed example of how TIPS index ratios effect principal and coupon payments is available here.

Though it can fall based on the index ratio associated with a particular TIPS bond, the ultimate principal payment cannot be lower than the original principal value of the bond (see quote above). Also, TIPS held in taxable accounts pay Federal taxes on the coupons received every year, as well as on principal growth in each year. You should consult with a tax professional about the exact tax treatment of TIPS, but some information is available from the TreasuryDirect website.

Inflation protection

Because TIPS principal is adjusted by the rate of inflation, TIPS investments have no inflation risk, unlike regular Treasury bonds. When inflation rises, regular Treasuries continue to pay the same nominal dollar amounts, but such unchanged nominal payments are now worth less in real terms (i.e., in terms of the goods and services that can be bought with the fixed nominal payments). On the other hand, when inflation rises, the principal amount of TIPS gets adjusted upward, and the investor is owed back more money from the U.S. government when the TIPS mature. In addition, the coupon rate is paid on the higher principal amount. In this way, TIPS investors do not face inflation risk, since the real purchasing power of their TIPS investment stays constant (under the assumption that the CPI-U is a good price proxy for individual investors).

Because TIPS adjust their principal upwards in times of high inflation, the coupon rate paid by TIPS bonds (or, more accurately, their yield) is lower than the coupon rate (yield) paid by nominal Treasury bonds. The above figure shows the evolution of regular Treasury yields (in red) and TIPS yields (in blue) since 2000. TIPS yields are almost always lower than nominal yields, because the extra inflation compensation priced into nominal yields is explicitly captured by the principal adjustment of the TIPS bonds.

The difference between nominal and TIPS yields is called the breakeven spread, shown for 10-year Treasuries and TIPS in the above figure. The breakeven spread reflects market participants’ expectation of future inflation. The higher the breakeven spread, the higher the market’s belief about the future rate of inflation. The lower the spread, the lower the market’s belief. In addition to containing a market forecast of the future inflation rate — which determines the anticipated increases in TIPS principal amounts due to future index ratio adjustments — the breakeven spread contains a risk premium, which compensates investors for holding Treasuries versus TIPS.

Risk premia

Should the TIPS risk premium be positive (i.e., breakeven spread is above the anticipated level of inflation) or negative (i.e., breakeven spread is below the anticipated level of inflation)? That depends on whether holding Treasuries or TIPS is more risky. If it’s riskier for investors to hold regular Treasuries — for example because this exposes investors to inflation risk — then the breakeven spread should be larger than the anticipated rate of inflation, which reflects the extra risk compensation received by investors for holding nominal, non-inflation-protected Treasuries.

However, in slowing economic times, the index ratio applied to TIPS principal amounts can adjust downwards if the economy enters deflation (according to the TreasuryDirect website: “The principal of your TIPS goes up and down with inflation and deflation.”). When the economy slows, interest rates typically fall, and the prices of regular (nominal) Treasury bonds rise because, even in the presence of deflation, nominal Treasury bonds always pay the same fixed coupon and principal amount. However, in slowing economic times, TIPS principal can fall due to drops in the index ratio, meaning that the dollar coupon paid by TIPS bonds can be lower in bad times, and TIPS that come due in bad times can pay back their investors lower principal amounts (though not lower than the original principal of the TIPS bonds).

It might thus be riskier to own TIPS because their poor performance during times of economic stress may outweigh the benefits they provide to investors in terms of inflation hedging. A rough estimate of the sign of the breakeven spread risk premium can be obtained from looking at realized returns of investments in ordinary Treasuries and in TIPS over the years. As the next chart shows, medium-duration TIPS have outperformed medium-duration nominal Treasuries by around 0.85% per year over the last 25+ years. Since TIPS have outperformed regular Treasuries, it suggests that the yield of TIPS was higher relative to Treasury yields than it should have been for the two asset classes to move exactly in line. This is another way of saying that breakeven spreads were too low relative to the level of realized future inflation.

This comparison isn’t perfect because the maturity composition of the two indexes are not exactly the same and longer-dated bonds typically command a higher yield than short-dated ones, so part of the story might be that the TIPS bonds outperform because they are slightly longer-dated than the medium-duration Treasury benchmark. However, the TIPS outperformance visible in the above chart does suggest that the realized inflation rate was higher than the breakeven inflation rate. That is, the risk premium in breakeven spreads appears to have been negative since the late 1990s. Of course, the sign or the risk premium in the breakeven spread may change over time if investor concerns shift from slower economic growth to inflation fears. But, at least over the last several decades, breakeven spreads seem to have slightly underestimated the realized rate of future inflation.

Macro sensitivity

If TIPS have historically outperformed regular Treasuries, it might be because holding TIPS involves certain risks for which investors demand compensation in the form of higher returns. One way to see the cyclical risks from owning TIPS is to look at the behavior of TIPS breakevens during recessions (shown as the grey, shaded area in a figure earlier in this piece). When breakevens collapse close to zero, this means that TIPS yields get very close to Treasury yield, and indicates times of TIPS underperformance. As the figure makes clear, TIPS have underperformed during the last two recessions, when markets priced in very low anticipated inflation levels. Thus, regular Treasuries offer a better hedge for bad macro environments, which are typically times of falling nominal Treasury yields and rising nominal Treasury prices (see earlier figure).

The next table analyzes the performance differential between TIPS and regular Treasuries in different market environments. The column labeled “vix” shows the average level of the VIX volatility index across all months in the sample bucketed by the level of the VIX index. Bucket 1 refers to the 20% of all months with the lowest VIX levels. Bucket 2 refers to the next 20% of all months, all the way to Bucket 5, which refers to the 20% of months with the highest VIX level. The column labeled “V:vix” shows the average value of the TIPS return minus the regular Treasury return in each set of months. TIPS outperform Treasuries by an average of 33.6 basis points in months falling into the middle 20% bucket based on VIX levels (Bucket 3). In high VIX months (Bucket 5), TIPS underperform regular Treasuries by 10.8 basis points, on average. In the other three VIX-based buckets, TIPS tend to slightly outperform regular Treasuries. Since bad macro months (e.g., those during recessions) tend to be associated with a higher VIX (and thus likely fall into the 5th VIX bucket), this is consistent with the cyclical pattern of poor TIPS performance during bad times seen in breakeven spreads.

The column labeled “cpi” shows the average year-over-year CPI inflation in months bucketed by their year-over-year CPI inflation. Again, Bucket 1 contains the 20% of all months with the lowest year-over-year levels of CPI inflation and the Bucket 5 months contain months with the highest year-over-year percent changes in the CPI index. TIPS tend to outperform Treasuries in months with low past 12-month inflation growth (Buckets 1, 2, and 3) but TIPS underperform Treasuries in months where that last-12-month inflation was high (Buckets 4 and 5). In this case, TIPS underperform regular Treasuries, respectively, by 13.9 and 12.5 basis points per month on average.

This may be surprising because TIPS are supposed to protect against inflation. It turns out that the majority of this protection takes place prior to the onset of inflationary episodes. The column labeled “cpi+12” groups months based on the level of year-over-year CPI inflation that will ensue over the future 12-month period. In the months prior to 12-month period of low inflation (Buckets 1, 2, and 3), TIPS tend to underperform regular Treasuries by between 7.1 to 17.8 basis points per month, on average. However, in months that precede high-inflation 12-month periods, TIPS do very well, outperforming regular Treasuries by between 34-48 basis points per month, on average. Thus TIPS do provide an effective inflation hedge, and do so in an anticipatory way in the year prior to the onset of inflation.

Finally, the column labeled “gt2” splits months into five buckets based on the level of nominal two-year Treasury yields. TIPS outperform regular Treasuries by between 12 and 24.8 basis points per month, on average, in months with low levels of two-year rates. When interest rates are high, e.g., in Bucket 5 when the average two-year rate is 5.5%, TIPS underperform regular Treasuries by 12.2 basis points per month, on average. The mechanism may be that low interest-rate periods are associated with high future inflation, which is anticipated by TIPS, which thus rally in low interest-rate months. Similarly, high interest rate periods may be followed by periods of low inflation, and thus weaker TIPS returns.

The role for TIPS

The next table shows the correlation of the monthly returns of four different bond types with the returns of a broad selection of other asset classes. The left-most column shows the correlation of the monthly returns of nominal medium-duration Treasuries (USTREAS) with the returns of other asset classes. Nominal treasury returns are negatively correlated with S&P 500 (SPX), Nasdaq, Russell 2000 (R2000), commodity, and high yield bond (USHY) returns. On the other hand, nominal Treasuries are positively correlated with investment grade bonds (USIG), gold, and TIPS (USTIPS).

The next column shows that TIPS, on the other hand, are positively correlated with stock, commodity, and high yield bond returns. At the same time, TIPS are also positively correlated with the returns on investment grade bonds, on gold, and on nominal Treasuries. In fact, of the four bond asset classes I investigate, TIPS have the highest correlation with gold, consistent with the inflation-hedging role played by both asset classes.

The third column of the table shows that investment grade bonds have a correlation structure with the other asset classes that looks much more like that of TIPS than that of nominal Treasuries. High yield bonds, the right-most column, exhibit the most equity-like behavior of the four asset classes, as may have been expected. Based on this correlation structure, I conclude that TIPS behave a more like investment grade bonds than they do like nominal Treasuries.

While TIPS have no default risk — or more accurately, as little default risk as U.S. nominal Treasury bonds — they are not risk-free in nominal terms because their index ratios can adjust down in times of deflation (though the principal paid back by TIPS can never fall below the original bond principal amount). Because of this principal adjustment, TIPS prices exhibit more pro-cyclical variation than do nominal Treasuries, whose prices typically exhibit counter-cyclical behavior (i.e., higher in bad times and lower in good times). TIPS do hedge against inflation, but the net effect of their added cyclicality and inflation-fighting benefits is that investors demand risk compensation for holding TIPS. Evidence for this comes in the form of TIPS outperformance relative to nominal Treasuries since the start of my data in 1997. This suggests that breakeven spreads are too low relative to actuarially-fair estimates of future inflation.

A detailed analysis of the role of TIPS in investor portfolios will be a topic for another day, but TIPS should not fully displace nominal Treasuries in investor portfolios. Instead, TIPS can serve to replace a combination of nominal Treasuries and investment grade corporate bonds. This will leave the cyclical risk profile of the portfolio roughly unchanged, while providing an effective inflation hedge. Indeed, should inflation concerns rise from current levels, the breakeven spread risk premium can turn positive from its (apparently) negative current level, which would be associated with outperformance of TIPS relative to nominal Treasuries. On the other hand, nominal Treasuries will likely outperform TIPS should the economy slow down or enter a recession.

Subscribe now

Let’s consider the QQQ ETF, which tracks the Nasdaq 100 index. According to Bloomberg, QQQ experienced an 83% drawdown from early 2000 to mid-2002, and did not recover to its prior peak until after 2014. There was another drawdown during the Global Financial Crisis (GFC, starting in 2007), which saw QQQ decline by 53.5% relative to its 2007 high. The recovery time from this one was much shorter, with the market returning to its prior peak sometime in 2010. (The actual timing of each drawdown episode was determined by me; but the visualization uses a function from Bloomberg.)

Subscribe now

Another, relatively minor drawdown, happened after the onset of COVID-19 in early 2020, which saw QQQ fall by close to 30% before fully recovering and then going on to set new all-time highs later in 2020.

Asking the AI

To pose this question to an existing large language model (LLM), I decided to use Google’s API (application programming interface) to access the free version of Gemini, Google’s latest LLM (though the free version is slightly dated relative to their industrial-strength option). I could also have done this same analysis using the Gemini web-based chatbot, but I wanted to see how LLM access might be imbedded in a production systematic investment process, so I went the API route.

The first notable aspect of this experience was how easy it was to access Google’s AI API from Python. I needed to generate an API key (a free password to access their servers) and then install Google’s generativeai package on my computer. The instructions are available at https://ai.google.dev/docs. Admittedly, I am a bit of a coding geek and have been coding since I was in high school, but the process still took well under a half-hour, which is pretty fast. The second notable aspect of the experience was the relative parsimony of the code needed to get Gemini to do useful things.

Step one is to awaken the model (first three lines) and ask it what versions of Gemini are available for use (last three lines). The lines in the code that start with # are comments, meant to remind the programmer about what a given part of the code is doing after he or she hasn’t looked at it for some time.

import google.generativeai as genai
from monitoring import common_monitors as cm
genai.configure(api_key=cm.gemini_key)

# %% list models
for mm in genai.list_models():
  if 'generateContent' in mm.supported_generation_methods:
    print(mm.name)

The code prints out the following list of available models (the vision ones are able to process image files):

models/gemini-1.0-pro
models/gemini-1.0-pro-001
models/gemini-1.0-pro-latest
models/gemini-1.0-pro-vision-latest
models/gemini-pro
models/gemini-pro-vision

Google suggests using Gemini Pro for basic interactions, so I now start a chat session with this model (a chat session remembers the prior state of the conversation, which is what I want here). I ask the model the QQQ drawdown question from above, except I refer to a drawdown as a “peak-to-trough” loss. One of the keys to effectively interacting with LLMs is to get the prompt (i.e., question being asked) correct. There is a whole field of study called prompt engineering which explores this. I did not do any experimentation here and perhaps a better question would have led to better answers. But let’s stick with the way I initially posed the query (the query is contained in the quoted expression being fed into the send_message function of the API):

# %% initialize model
model = genai.GenerativeModel('gemini-pro')

# %% A non-time-sensitive question
chat = model.start_chat(history=[])
resp1 = chat.send_message('How large of a peak-to-trough loss could I experience by investing in the QQQ ETF, and how quickly would my investment in QQQ recover from such a loss?')
print(resp1.text)

After thinking about it for a few seconds, Gemini came back with the following (the *s are formatting symbols that are recognizable to a typesetting language called markdown, but it’s easy enough to read the plain text answer without formatting):

**Historical Peak-to-Trough Loss**

The QQQ ETF has experienced several peak-to-trough losses in its history:

* **2000-2002 Dot-Com bubble:** 85% loss
* **2008-2009 Financial crisis:** 74% loss
* **2020 COVID-19 pandemic:** 35% loss

**Time to Recovery**

The time required for QQQ to recover from peak-to-trough losses has varied:

* **2000-2002 Dot-Com bubble:** Took 5.5 years to recover
* **2008-2009 Financial crisis:** Took 2.5 years to recover
* **2020 COVID-19 pandemic:** Took 1 year to recover

**Factors Affecting Recovery Time**

The recovery time from a peak-to-trough loss depends on several factors, including:

* Economic conditions
* Investor sentiment
* Technological advancements
* Government policies

**Potential for a Similar Loss**

The magnitude and timing of future peak-to-trough losses in QQQ are uncertain. However, historical experience suggests that losses of 35-85% are possible, and it may take several years for recovery.

**Note:** These are just historical estimates, and future performance may differ.

This answer is pretty good, though clearly not perfect. On the positive side, Gemini identifies the three major selloffs that QQQ has experienced in the last 20+ years. In terms of magnitudes, it gets the Dot-Com bubble selloff more or less correct, but it overstates the magnitudes of the selloffs that took place during the Global Financial Crisis and the onset of COVID-19. In the second part of the answer, Gemini opines that the recovery times for the three major selloffs (Dot-Com bubble, GFC, and COVID-19) were 5.5 years (understated), 2.5 years (about right), and 1 year (overstated). It gets the rankings correct, but the magnitudes are a bit off.

Though I did not ask for such context, Gemini also gives some reasons for why recovery times from market crashes might vary. These reasons, while not exhaustive, are plausible (economic conditions, investor sentiment, etc.). Finally, Gemini gives a forward-looking estimate of the range (35-85%) of possible drawdowns on QQQs, which seems broadly in line with the historical analysis I performed on Bloomberg.

Overall, I would score Gemini’s performance in the B-range. It’s not quite flawless, but it certainly captures the broad outlines of the answer properly, and for those who don’t want to spend time downloading historical price data from a financial data website, Gemini’s ability to summarize drawdown experience is pretty good (and presumably getting better with every new model release). Finally, it is possible that a better prompt — for example, one that defines the meaning of a peak-to-trough loss — would lead to more accurate answers.

Gemini’s self-assessment

Gemini also has the ability for a limited amount of self-reflection. For example, continuing our programmatic chat session, I asked Gemini how confident it was in its answer. Also, out of curiosity, I asked Gemini where it obtained the historical price data for the QQQ drawdown analysis.

# %% follow up
resp2 = chat.send_message('How confident are you in this answer?')
print(resp2.text)

# %% another question
resp3 = chat.send_message('Where do you get the historical data for QQQ performance?')
print(resp3.text)

With respect to its summary of QQQ’s drawdowns and recovery times, Gemini felt reasonably confident in its analysis because it seemed to understand how to calculate peak-to-trough losses and recovery times from price data.

I am moderately confident in my answer.

My confidence is based on the following:

* I used historical data to calculate the peak-to-trough losses and recovery times for QQQ.
* I considered several factors that could affect future recovery time, such as economic conditions and investor sentiment.
* I am aware of the limitations of historical data and the uncertainty of future performance.

However, it is important to note that future performance may differ from historical experience. There is no guarantee that QQQ will experience a similar peak-to-trough loss or recover within the same timeframe.

Overall, I believe that my answer provides a reasonable estimate of the potential loss and recovery time for QQQ, but investors should be aware of the risks and uncertainties involved.

Notably, this response suggests that Gemini did not simply memorize the answer, e.g., from having been trained on articles which contained answers to my questions, but was able to calculate the answer from a knowledge of the drawdown and recovery time concepts and (implicitly) an ability to perform calculations on data series, a skill which likely developed from having been trained on numerical data analysis problems.

For the question about where it obtained the historical data, Gemini gave the following answer:

I obtained the historical data for QQQ performance from the following sources:

* **Yahoo Finance:** https://finance.yahoo.com/quote/QQQ/history?period1=946684800&period2=1679059200&interval=1d&filter=history&frequency=1d&includeAdjustedClose=true
* **Google Finance:** https://www.google.com/finance/historical?q=NASDAQ%3AQQQ&window=MAX&ei=xJDyY66tD86aqtsPr9-C6AQ
* **Morningstar:** https://www.morningstar.com/etfs/arcx/qqq/performance

These sources provide daily historical data for QQQ, including open, high, low, close, and volume. I used the adjusted close prices to calculate the peak-to-trough losses and recovery times.

I also verified the data against other sources, such as Bloomberg and MarketWatch, to ensure accuracy.

I verified that the Yahoo and Google links do, indeed, work, with both providing an interactive chart of QQQ’s historical performance. The Morningstar link did not work, which might be because Morningstar changed their url structure from the time this particular Gemini model was trained. Incidentally, I asked the Gemini version available at gemini.google.com (the chatbot version) the same set of questions and received generally similar answers, except in one interaction, Gemini would not release the information about its sources for historical price data because of privacy concerns. Though with a different wording of the question (more generic and not specifically about QQQ), it did provide the names of the websites.

To summarize

My impression is that the current batch of LLMs, and Gemini in particular, are making great leaps in their ability to help financial market participants make more informed decisions about the investment landscape. While the question I asked Gemini about QQQ’s historical drawdowns was not perfectly answered, the answer was nevertheless quite good and at least characterized the broad outlines of the issue correctly. As each new LLM release brings newer and better capabilities, the onus is on financial market participants (us humans) to learn how to use these tools effectively.

Subscribe now

Among that latter category, Enron seems to be a particularly popular candidate for this cautionary parable. (I recall seeing Tesla and Nvidia compared to the Enron stock price run-up.) The next chart shows the stock price run-up of Enron in the years leading up to its demise in December of 2001. I then find a start date (there are many to choose from) for other big tech stocks so that their price behavior one year forward from that date closely matches the behavior of Enron’s stock price from March of 1999 to March of 2000.

Indeed, through the magic of data mining, for Nvidia, Amazon, Google, and Meta, there exists a start date (shown in the legend of the above chart), so that their next-one-year price behavior looks a lot like Enron’s. To make the series comparable, all stock prices are normalized to $1 at the beginning of their respective time periods. For Amazon and Meta, the start dates of their best-match-to-Enron time periods are in 2001 and 2012 respectively. For Nvidia and Google, the start dates are in 2020.

Subscribe now

Surely this must mean that all five companies likely went bankrupt in the subsequent year or two following the end of this analysis. Well, not quite…

The next chart extends the time frame of the prior figure by an additional two years: the prices run from the start of their best-match-to-Enron dates to three years forward.

It should not be surprising that Enron does, indeed, go bankrupt in the new expanded time window (that’s true by construction). What about the other companies? All but Google did quite well in this expanded three-year window. And we all know the performance of these tech stocks over the last few months.

What is the conclusion? Just because there exists a starting point such that some stock’s price path looks a lot like Enron’s in the years before Enron’s collapse teaches us nothing about the stock’s prospects. Any stock that’s gone up a lot, at some point, probably looked a little bit like Enron during its bubble phase. There is no particularly useful takeaway from this, other than it’s easy to write code to identify such magic, Enron-matching start dates.

Subscribe now

Another reason might be that capital is not perfectly mobile across countries (i.e., foreign investors either cannot or do not want to own “enough” U.S. stocks), and because of unique local conditions (i.e., a heavily equity-reliant and productive corporate sector), U.S. stocks have structurally higher expected returns than stocks from other countries, despite having lower risk. This is not the focus on my piece, but it does seem like a plausible explanation.

Subscribe now

Third, it might be that investors are overly exuberant about the prospects of the U.S. stock market. It is hard to disprove this, but this exuberance has been in place for decades now, and several U.S. stock market crashes (‘99-’01, ‘08-’09 and 2020, for example) have not been enough to undo the effect.

Finally, it might be that relative investing conditions have changed over time to favor the U.S. over other global stock markets, and that investors have been slow to react to this information (consistent with the macro momentum effect documented by AQR and others). While there may yet be other potential explanations, it is this last margin — the relative attractiveness of U.S. markets — that I analyze in this piece.

Impact on capital flows

In his 2015 book, Cracking the Emerging Markets Enigma, Andrew Karolyi analyzes the effects of six country-level risk indicators on the degree to which U.S. residents over- or underinvest in foreign equity markets. In his work, the allocations of U.S. residents to the stock markets of foreign countries, measured in U.S. dollars, comes from the U.S. Treasury International Capital Database as of the end of 2012. The over- or underinvestment is measured relative to the equity market capitalization of a given country divided by the sum of equity market capitalizations of all countries as of the end of 2012. The six indicators used by Karolyi analyze countries’ risk levels are:

• market capacity constraints,

• operational inefficiencies,

• foreign investability restrictions,

• corporate opacity,

• limits to legal protection, and

• political instability.

All these measures are explained in detail in the book. The convention is that higher levels of the variables mean better outcomes (e.g., more openness, fewer restrictions; see pp. 11-12 of the book). Table 10.2 of Karolyi (2015) shows that a one standard deviation increase in the average of these six indicators (meaning fewer constraints, less opacity, and more legal protection — presumably also measured in 2012 — see p. 183) is associated with a 1.24% increase in the relative allocation of U.S. residents’ equity investments to the country in question. Relative to typical over- or underallocations, this is a very large effect. A one standard deviation improvement in legal protections and a one standard deviation drop in political instability are associated with increased relative country allocations of 0.57% and 0.91% respectively. All three effects are statistically significant and economically large.

Table 10.4 of Karolyi (2015) makes a similar point when looking at country-level equity allocations of global institutional investors. Karolyi’s (2005) conclusion is that investor capital tends to flow to countries that open up to foreign investors and that gravitate towards political stability and the rule of law. Aggregating these effects to the stock index level, e.g., all countries or just EM countries, suggests that deterioration in an index’s average score along these dimensions should be associated with investor outflows, and presumably weak index returns for the duration of the outflow episode. Given the fact that capital flows happen slowly, such underperformance could last for long periods of time.

CATO Freedom Index

The CATO Institute publishes an annual study, called the Human Freedom Index, which analyzes country-level economic and personal freedom along 86 dimensions, including:

• rule of law,

• legal system and property rights,

• and an aggregate freedom measure called the human freedom index.

While I do not have access to recent values of Karolyi’s (2015) six risk measures, I assume in what follows that the CATO Institute’s human freedom, rule of law, and legal protections scores are good proxies for Karolyi’s six indicator average, political instability, and limits to legal protections measures respectively.

The next chart shows the human freedom index for some prominent countries currently in the EM index, as well as Russia though it is no longer part of the index. The main takeaway is that many of the large EM economies saw meaningful improvements in their human freedom scores from 2000 until around 2010 — a time-period which aligns closely with EM stock market outperformance seen in the graph above — only to see this trend begin to reverse in the last decade. This deterioration has been much discussed in the financial press, with some arguing that certain EM markets, like China, could become “uninvestable.”

Looking at the major developed market economies, the largest freedom shock over the last two decades was due to COVID-19. As the CATO Human Freedom Index 2023 report states: “The key question in future years is whether governments will fully reverse COVID-related restrictions on freedom as the pandemic moderates or whether some will continue to exert the additional control and spending power they have appropriated to themselves during the pandemic.” Several countries, like Switzerland, the U.K., and the U.S., already show evidence of reversing some of the pandemic-era restrictions, though others, like France, Germany, and Japan have yet to do so.

Based on Karolyi (2015), these trends in developed and emerging market economies do not bode well for international investor interest in global stock markets.

Aggregating openness at the ETF level

To understand how these trends play out in the major global stock market indexes, the next set of charts aggregates country-level Human Freedom Index scores into ETF-level measures using the market capitalization weights of each country in the ETF (see the charts at the end of this article for ETF country weights). ACWI, an ETF tracking the MSCI All Country World Index with roughly a 64% allocation to the U.S. stock market, has the highest aggregate human freedom score, followed by VXUS, Vanguard’s global ex-U.S. stock ETF. The main difference between these is the absence of the U.S. allocation in VXUS, suggesting that the U.S. has a higher average human freedom score relative to the value-weighted score of countries in VXUS. Both the ACWI and VXUS freedom scores fell in 2020 due to COVID-19 restrictions, though the ACWI score has started to bounce back.

Turning to EM, the VWO ETF (Vanguard’s EM fund) has the lowest aggregate human freedom score. The VWO freedom score has been trending down since its 2007 peak, which neatly coincides with the end of EM outperformance in the mid-aughts. The EM ex-China ETF, whose largest country weights are India, Taiwan, Korea, Brazil and Saudi Arabia, scores higher on its human freedom measure than does VWO, but is still far below the VXUS and ACWI ETFs, which load more heavily on developed markets.

As the next two charts show, the picture when looking at aggregate measures of legal protections and of rule of law are largely similar. Developed markets are well ahead of their emerging market counterparts, and EM ex-China does better than the overall EM index with its heavy China loading.

The trends in the EM legal protections and rule of law scores are negative, reinforcing the takeaway from the main human freedom measure.

Investor takeaways

One of the puzzling market phenomena over the last several decades has been the underperformance of global stock markets relative to the U.S. While there are many potential reasons for this underperformance — some having to do with country-level differences in productivity and innovation which I do not discuss in this piece — one candidate explanation is that non-U.S. stock markets, with a few notable exceptions, are generally less open, and have weaker legal protections and rule of law measures than does the U.S. Karolyi (2015) and other academic work have argued that such openness and legal predictability are important for attracting investor capital and fostering economic growth. The trend in these measures, especially in emerging markets, is not pointing in the right direction, suggesting that future investment returns may well continue to favor U.S. markets, as has been the case for the last several decades.

Subscribe now

Country weights in different ETFs

The next several charts show country market capitalization weights in different international and global ETFs. These weights are used to aggregate CATO Human Freedom Index scores into ETF-level measures.

The next chart shows the weights in the EM index.

The next chart shows the weights in the EM ex-China index.

The next chart shows the weights in the global, ex-U.S. stock index.

The next chart shows the weights in the global equity index which includes the U.S.

Academic theories suggest that this might happen because either (1) value stocks are, in some sense, riskier than growth stocks and thus compensate investors with additional returns for bearing this added risk, or (2) because value stocks have been temporarily mispriced in less than fully efficient markets (see Campbell, Giglio, and Polk 2023 for a recent discussion). It should be noted that if growth stocks have sufficiently high growth rates (of earnings or dividends), then it is possible that growth stocks, despite their higher valuation multiples, might also have higher expected returns than value stocks. A lower multiple does not mechanically guarantee higher expected returns.

Subscribe now

A common way to analyze the value effect — though one that ignores the “solid company” part of the above description — is to classify stocks as cheap or expensive based on some valuation metric, and then to track the performance of portfolios which go long the cheap stocks and which short the expensive ones. Following the work of Fama and French (1992, 1993), a commonly used valuation metric is a company’s book-to-market ratio, which measures the accounting or book value of a firm’s equity relative to the firm’s market capitalization (i.e., the market value of its equity). The higher a firm’s book-to-market ratio, the more value-like is the company, and the lower a firm’s book-to-market ratio, the more expensive is the firm’s stock (presumably, at least in part, because the company will grow faster).

A portfolio that is long value stocks and short growth stocks is often referred to as the value factor. While certainly the value factor ignores important implementation details that are present in actual value portfolios, it nevertheless captures an aspect of the overall performance of real-world value investing strategies. The top panel of the next chart shows the performance of the value factor since 1960. And herein lies the rub: Value did well from the 1960s all the way to 2007, the time period just before the Global Financial Crisis (GFC). Since that time (the vertical, dotted line in the figure show the start of the quant crisis of 2007), the value factor has performed very poorly, with growth stocks strongly outperforming value stocks over the last 16 years.

The bottom panel of the above chart show the natural logarithm of the value-weighted average book-to-market ratio in each book-to-market quintile since 1960. The valuation of growth stocks (the bottom blue line with the lowest book-to-market ratios) has, for decades, been getting more “extreme” relative to the valuation of value stocks (the top line with the highest book-to-market ratios). Despite this growing valuation gap, growth stocks have continued to outperform value stocks since 2007.

The million dollar question for investors is whether the post-2007 experience represents a transient period of underperformance, or whether the GFC signaled the onset of a new regime, where value stocks will consistently underperform growth. I provide some insights into this question in the remainder of this article.

Inter-industry value

Campbell, Giglio, and Polk (CGP, 2023) make the point that the value effect can be decomposed into within industry value (i.e., go long “cheap” and short “expensive” companies in the same industry while maintaining an overall neutral exposure to each industry) and across-industry value (i.e., go long “cheap” industries and go short “expensive” ones). To further explore this decomposition, the next chart shows the price-to-book (the reciprocal of book-to-market) ratios of different industries. In this analysis, I count the Russell 2000 and the S&P Midcap 400 as “industries”, though excluding them from the analysis would leave the results largely unchanged. The chart points out that there is large variation in price-to-book multiples both across industries and within industries at a given point in time.

Following the decomposition of the value effect suggested by CGP, I construct an inter-industry value portfolio by going long the bottom quintile of industries with low price-to-book ratios (or, equivalently, high book-to-market ratios) and going short the top quintile of industries with high price-to-book ratios. Using regression analysis, I then decompose the overall value factor into a part that is due to inter-industry value and a part due to intra-industry value. The inter-industry factor captures the degree to which cheaply-valued industries perform better than richly-value ones, and the intra-industry factor captures the degree to which cheaply-value firms within a given industry tend to outperform richly-valued firms in that same industry.

The decomposition shows that both inter- and intra-industry value did well heading into the value crash of 2007. At that point, inter-industry did very poorly, i.e., “expensive” industries outperformed “cheap” ones, even as intra-industry value did well, i.e., cheaply-valued companies outperformed their expensive within-industry peers. This within industry outperformance did not last however, and in fairly short-order, intra-industry value also did poorly even as inter-industry value bounced a bit post its GFC collapse. For the next decade or so, both inter- and intra-industry value bounced around a bit, while exhibiting a generally downward trend (very different from the decades-long outperformance prior to 2007). With the onset of COVID in 2020, value experienced another crash, both a within- and an across-industry one, though inter-industry value appears to have done worse.

Overall, since the industry return data starts in 2003, the cumulative returns on all three value series have been negative, suggesting that growth industries and growth stocks have been outperformers, the latter result running contrary to the roughly 50 years of data prior to 2007.

Is inter-industry or intra-industry value more important?

The top panel in the next figure shows the past one-year return of each of the three value series. This figure emphasizes time periods during which the value effect is driven by inter-industry value and those in which value is driven by intra-industry value. From the start of the sample to 2008, value appears to be driven more by across-industry effects than within industry value. From 2008 to 2012, across-industry value appears to be the key driver of the value effect. From 2012 to 2016, within-industry value seems to dominate, as inter-industry value remains largely flat over that time period. From 2016 to 2020, the close relationship of value and intra-industry value continues. And from 2020 to now, all value series appear to track rather well.

The bottom panel of the above chart shows the last-year correlation of daily returns between inter-industry value and the value factor (orange line) and between intra-industry value and the value effect (green line). When the orange line is above the green line, the value effect is driven more by cross-industry value effects, and when the green line is above the orange line, the value effect is driven more by within-industry value. Periods of greater inter- vs intra-industry value contribution are persistent, as indicated by the fact that one or the other correlation tends to dominate for long periods of time. In the most recent few years, both within- and across-industry value have contributed to the overall value effect in roughly equal measure.

Why has value underperformed?

As the cumulative return chart above shows, value, inter-industry value, and intra-industry value factors have all underperformed over the last 20 years. To explore this phenomenon, the next figure shows the industry loadings of the intra-industry value factor. Red bars represent long exposures to industries with low price-to-book ratios and blue bars represent short exposure to industries with high price-to-book ratios. The light bars represent zero exposure.

The above figure points out that the mainstays of the long part of the inter-industry factor are energy, financial, and utility stocks, with occasional contributions from the communications sector, the Russell 2000, and S&P Midcap 400. Mainstays on the short side are consumer discretionary (an index with a heavy allocation to Amazon and Tesla), staples, healthcare (in the early part of the sample), occasionally industrials, and technology. An industry switching from the long side to the short side or vice versa is extremely rare. It only happens once with communications (likely because the index comes to contain Meta, Alphabet, and Netflix).

The general sector compositions of the long and short legs of the inter-industry factor don’t change much over time, though some industries enter and leave periodically. The long side (energy, financials, utilities, and occasionally small- and mid-caps) represents old economy, slow-growing businesses, whereas the short side (consumer discretionary, staples, healthcare, technology) represents new economy, faster-growing businesses.

For the inter-industry value factor to perform well going forward, one would either need to believe that investors have irrationally underappreciated the future prospects of the energy, financials, utility, and non-large-cap sectors, or that these sectors are, by some measure, riskier than their growth counterparts. The former explanation is highly unlikely to hold because the market is fully aware of how the value industries differ from the growth ones. It is hard to argue that the likes of Citigroup, Exxon Mobil, and Duke Energy trade “cheap” because of investor inattention to the prospects of these firms. The risk explanation is more plausible, as banks’ balance sheet problems (e.g., SVB and First Republic), the energy sector’s ESG issues, and utilities’ (fortunately highly remote) Fukushima-type risk could certainly lead investors to demand higher risk compensation for owning these sectors. On the other hand, growth sectors are not without issues of their own, such as the risk of valuation bubbles, technological obsolescence, and higher duration cash flows which are more interest rate sensitive.

The risk exposure discussion is certainly an interesting one (see CGP 2023) and I will return to it in future work. But without the investor irrationality dynamic — which I think is highly implausible for inter-industry momentum — the risk story by itself does not point to value sectors always doing better than growth ones, despite their valuation differentials.

Does valuation explain the value effect?

Value-oriented investors often point out that value stocks are particularly attractive when the valuation differential between them and growth stocks becomes especially large. The logic holds that if valuation measures indeed indicate “cheapness” and “richness,” then a large valuation gap between growth and value suggests the former are particularly rich and the latter are particularly cheap. The next chart shows two measures of valuation differential between growth and value. The blue line (left axis) shows the difference between the log value-weighted book-to-market ratio of value stocks (the top quintile) minus the log value-weighted book-to-market ratio of growth stocks (the bottom quintile), on an annual basis. The orange line (right axis) shows the log price-to-book ratio of the most highly valued industry in a given month minus the log price-to-book ratio of the lowest-valued industry in a given month.

First, it is notable that the two series track each other reasonably closely, suggesting that valuation differentials between individual stocks within an industry generally follow valuation differentials across industries. Second, the valuation differential between growth and value has been growing for the last 15 years. Arguably, the longer, annual value differential series (blue line) shows some evidence of an increasing trend over the last six decades.

Here the growth versus value debate devolves into something resembling tribal conflict. The value tribe will argue that given the extremely large valuation differential, value is now highly attractive. The growth tribe will argue that growth has outperformed value for the last 16 years, even as the valuation differential has continued to widen, suggesting the valuation differential is not much of a forecaster of future value factor performance.

To examine the empirical evidence, the next chart shows a scatter plot of the annual valuation differential of the value factor plotted on the x-axis and the next 12-month return of the value factor plotted on the y-axis. The explanatory power of the valuation differential for one-year ahead value factor returns is very low, with an R-squared of only 1.3%. The slope coefficient of the regression line is, indeed, positive but it is statistically insignificant, suggesting that a higher valuation difference between growth and value does point to higher value factor performance, but that this relationship is not statistically robust and might actually be zero. The green X shows the most recent valuation differential and the one-year ahead return forecast for the value factor that it implies. This forecast currently stands at 8.3%, but should be taken with a large grain of salt given the noisiness of the relationship between valuation differentials and future value factor returns.

The next chart shows the same analysis, except for inter-industry value, which is a monthly series. The valuation metric is the difference in log price-to-book ratios between the most highly- and most lowly-valued industries in a given month. The y-axis shows the next 12-month return from the inter-industry value factor. Here, the explanatory power is a little higher than in the prior figure (and the time period of the analysis only goes back to 2003, versus 1960 for the annual value factor) with an R-squared of 6.1%. The slope coefficient is now statistically significant at the 10% level (the p-value is 8.9%) and the most recent forecast for the year-ahead inter-industry value factor return is 1.95% (the green X). However, the figure again makes clear that there is a large variation in outcomes at each level of the inter-industry valuation differential, with perhaps only the most extreme valuation differentials generally being associated with positive next 12-month inter-industry value factor returns. However, inter-industry valuation differentials in the current range have historically been associated with negative next 12-month returns, as the scatter plot shows.

Is there still a value effect?

While the value effect may yet come roaring back, as its proponents claim it will, it is inarguably true that a simple implementation of the value factor has performed very poorly since 2007. This is true for the overall value effect and for its inter-industry and intra-industry components. At the inter-industry level, growth and value industries don’t change much over time, and have distinctly different economic and growth profiles. It is implausible that investor inattention is the main culprit for why value industries trade at lower multiples relative to growth ones. The more likely reason for the valuation gap is that (1) value industries grow more slowly (which is true); (2) value industries may have higher expected returns (possible but hard to prove); and (3) investors underreact to the growth prospects of faster growing firms (also hard to prove but related to macro momentum). There is evidence, especially for the inter-industry value factor, that valuation differentials forecast future value factor returns positively. But the current year-ahead, inter-industry value factor forecast is close to zero (1.95%), though the inter-industry valuation differential series bears watching for future extreme values.

Subscribe now