Investors breathed a sigh of relief in 2025 as the global economy grew at rates consistent with historical trends despite US tariffs. Equity markets rose, and ACR strategies benefitted as portfolio company stock prices caught up with our fundamental value estimates. ACR results can be found at https://acr-invest.com/strategies/.
AI became the lead story again as tariff fears receded. In our 2025 year-end commentary, we will take stock of where we have landed three years into the AI LLM (Large Language Model) boom and three years after the 2022 bear market. We don’t know what the future holds, but valuations are resulting in higher portfolio cash levels. Our view is we are in the first few innings in the AI LLM tech layer buildout. Yet we remember well how the 2000-02 bear market occurred during the early innings of the Internet buildout. Financial market fluctuations often have their own seasons, and we believe the timing is practically impossible to predict. Thankfully, we have our valuation compass to see us through. When prices start to exceed values, we trim and sell.
Before getting to the financial and economic topics at hand, we would like to introduce you to the newest member of the ACR team. James Montier joined ACR at the end of last year as Managing Director of Fundamental Research and Performance. James is an expert in the fields of behavioral finance and value investing. He is a sought-after speaker and has authored many papers and four books on these and other economic topics. At ACR, James’ responsibilities will include publishing leading research, directing the firm’s fundamental performance initiatives, and supporting ACR’s objective of rational, unbiased investment decision-making. ACR investment team members have followed James’ work for many years and are excited to bring him into the ACR fold.
Our year-end commentaries tend to be longer than our quarterly updates. Mercifully, a high-level bullet point summary is provided below.
We begin by offering the following definition of a bubble: when returns implied by valuations are (i) historically low and (ii) significantly lower than the returns expected by investors.
The relationship between high cyclically adjusted P/Es and low future stock returns is robust based on historical data and the simple but powerful economic logic of the earnings yield (E/P).
ACR cyclically adjusted P/E and E/P calculations show the S&P P/E is at an all-time high of 46.6, and the earnings yield is at an all-time low of 2.1%.
AI LLMs are likely to be as revolutionary as the Internet. Corporate return-on-capital may be challenged due to large capital investments. The hope is higher productivity, but we cannot be sure. Consumers may ultimately benefit the most from AI LLMs.
We believe today’s greatest economic risk is a decline from current elevated P/Es. ACR protects against this risk by pruning portfolios of excessive valuations to assure very different valuation characteristics from the market.
Commentary Sections
I. Bubble Defined
II. The Double Bubble
III. The AI Boom
IV. Fundamental Reversion
I. Bubble Defined
James’ first piece will appear next month, and here we are, already stealing his thunder. He offers this concise dynamic definition of a bubble: “when returns implied by valuations are heading in a different direction to the returns expected by investors.” We will save evidence of rising returns expected by investors for James’ paper. Regarding returns implied by valuations, ACR has been reporting for many years how expected returns have been on a downward trend. The foundational premise begins with understanding the inverse relationship between the level of the S&P 500 cyclically adjusted P/E and future returns. The supporting data for this relationship has been well documented by us and others. We share this data again for reference below.
Note: S&P 500 Starting Cyclically Adjusted P/E is based on Real S&P 500 Price Per Share (PPS) divided by Ordinary Least-Squares Regression (OLS) trendline of S&P 500 Real Earnings Per Share (EPS) from 1926 to December 31, 2025. Each data point represents the forward S&P 500 Subsequent 10 Year Total Returns. Data points in the chart are displayed quarterly. Sources: S&P Dow Jones Indices; Robert Shiller; BLS- CPI Data; ACR Alpine Capital Research.
The implication is clear: the higher the P/E, the lower the return. Admittedly, the chart employs quarterly overlapping periods, and the hundred-year period includes just ten non-overlapping data points. The correlation is still compelling, but the economic logic upon which it rests solidifies the story. Higher P/Es are mathematically equivalent to lower E/Ps, and the earnings yield (E/P), like any investment yield, is a
measure of return. Similar to how interest income is the numerator in a bond yield calculation, the net earnings of a company is the numerator in the earnings yield. The bond yield, correctly calculated, is equivalent to the bond’s return when held to maturity. The stock earnings yield (E/P) is an excellent proxy for the long-term stock return. The three main differences are that company earnings are not fixed, may be re-invested rather than paid in cash, and are real rather than nominal if equities are a hedge against inflation. When “real”, inflation is added to the earnings yield to estimate total return. A more comprehensive explanation of stock return calculations is included in the Appendix.
Our calculation of the current cyclically adjusted earnings yield (E/P) is 2.1%. The following chart shows how the S&P 500 earnings yield (E/P) has been in a downward trend since the Financial Crisis of 2008 and has been in a downward megatrend back to the early 1980s.
Note: Earnings Yield (“EY”) is the inverse of the Price-to-Earnings Ratio (i.e., EPS/Price). EY represents an estimate of normalized earnings that could theoretically be paid out in dividends as a percentage of current price. Yields (income/price) are commonly used as a valuation metric to compare the margin of safety for an investment (e.g., Earnings Yield for Equities, Yield to Maturity for Bonds). Sources: S&P Dow Jones Indices; Robert Shiller; BLS CPI data; Compustat; Bloomberg; ACR Alpine Capital Research. S&P 500 EPS sourced from Shiller data through September 30, 2025. Most recent S&P 500 TTM EPS estimate as of December 31, 2025, sourced from Howard Silverblatt. Index data pulled on December 31, 2025.
The returns implied by valuations have clearly headed lower, satisfying at least one-half of our bubble definition. Spoiler alert. Returns expected by investors have not been declining with the returns implied by valuations. Most stock market investors unwittingly cheer as prices rise ever higher without considering whether they are rising faster than underlying earnings and by how much, or the consequences. This leads us to our next section.
II. The Double Bubble
The ACR investment team is mindful of a very different perspective on rising prices. We have been complaining about high prices (our specialty) for over a decade, and they continue rising! Indeed, we wrote in our 2019 year-end commentary: “The current cyclically adjusted P/E is 34.6 – alarmingly near the 1999 high of 35.5. What are the chances it rises significantly higher?” Odds aside, prices zoomed higher. We also posited: “… the ‘20s will not roar. Rather, they will be more like the ‘00s than the ‘10s for EQR and the market”. Six years into this decade, how wrong we were! The S&P 500’s roar could be heard across the globe at 15.1% annualized. Fortunately, our flagship EQR strategy returned 15.9% gross and 14.8% annualized net of a 1% fee1. EQR’s return was also unexpected as we normally do not keep up with runaway bull markets.
A lot has happened this decade so far. We never want turmoil to visit humanity, but the steep yet brief Pandemic declines, the rising interest rate bear market of 2022, and this year’s tariff tantrums were gifts. The volatility allowed us to put cash to work (in the case of the pandemic, very large sums) and to later take gains as prices rose toward our value estimates. The following P/E chart shows the “double bubble” of which we speak.
Note: a) ACR CAPE based on Real S&P 500 Price Per Share (PPS) divided by Ordinary Least-Squares Regression (OLS) trendline of S&P 500 Real Earnings Per Share (EPS) from 1926 to December 31, 2025. b) Arithmetic mean. Sources: S&P Dow Jones Indices; Robert Shiller; BLS- CPI Data; ACR Alpine Capital Research.
There are three periods in which the S&P 500 exceeded 30x earnings: the late 1920s, the late 1990s, and the current period. Only once has it exceeded 40x, the high water mark it sets today. 1929 and 1999 were single peaks: straight up and straight down. The aftermath of 1929 was of course especially catastrophic.
It was also avoidable given what we know today. While incomparable to the 1930s, the 2000-02 bear market was no slouch. The S&P 500 declined 49.1% peak to trough.
Given that we have been hovering over 30x earnings for over ten years, the question naturally arises, are we in a new normal? Until the mid-1950s, stock dividend yields were almost always higher than US treasury bond yields. That relationship reversed in the middle of the 20th century. US treasury bond yields have almost always exceeded dividend yields since late 1950s. Could this be another paradigm shift? Our answer is “no.” Logic and evidence show that stock returns revert to fundamental returns sooner or later. Conversely, the dividend payout is a choice. We address the economic phenomenon and math of fundamental reversion in Section IV. But first, we will opine on the ongoing topic du jour, AI.
III. The AI Boom
The Artificial Intelligence talked about today is a specific kind of AI, yet it has become what most are now simply calling AI. A few key terms which describe the technology and its beginnings are worthy of noting. AI LLMs can be traced back to a 2016 internal research project at Google called “Transformers: Iterative Self-Attention and Processing for Various Tasks” which led to the now famous 2017 paper, “Attention is All You Need.” Neural networks were once a discredited approach to AI. While LSTM (Long Short Term Memory) recurrent neural networks marked a comeback for the technology, the game changer turned out to be “self-attention” neural networks. They produced better translations of a word by referencing any other part of a passage simultaneously which helped clarify the word’s intent. The way self-attention models handled information was suited to the more powerful parallel processing chips found in gaming systems. The term transformer simply came from the architecture’s ability to efficiently transform input data into rich representations of meaning.
The ACR investment team’s goal is not to pick AI winners. Rather, it is part of our job to understand transformative technologies and to assess their impact on the economy and the companies we own or could own. Opinions vary on our team. All agree that AI is a useful technology. We are also implementing AI tools in each department at ACR. While these new tools are speeding up a few processes and creating efficiencies, we do not see AI helping us to become better investors yet.
Our thoughts on AI’s potential impact begins with a brief outline of possible outcomes, then we will discuss several economic issues.
In our 2024 1st quarter commentary, we laid out three general scenarios for thinking about AI’s impact:
Base case: AI is as revolutionary a technology as the Internet.
Positive tail outcomes:
Positive. AI proves to be the most profound technology in modern history, providing meaningful gains in productivity.
Pink swan. AGI (Artificial General Intelligence) ushers in an unparalleled period of human prosperity.
Negative tail outcomes:
Negative. AI proves useful but results in a weak jobs market or fizzles as a transformative technology.
Black swan. AGI/AI is not properly harnessed resulting in a dystopian, radically unequal world controlled by malicious humans and machines.
As Co-CIO of ACR, the above summary continues to reflect my views. The most crucial point is how widely drawn the tail outcomes are likely to be. The impact of AI may turn out to be somewhat better or worse than we think today, but we see no reason to forecast wildly optimistic or pessimistic outcomes at this time. The remainder of this section addresses more specific economic issues. We organize our discussion into five categories: capital investment, return-on-capital, employment, productivity, and asset prices.
The amount of capital being invested into AI is undeniably massive. Hyperscaler spending by Microsoft, Google, Meta, and Amazon alone was $228 billion in 2024, with $363 billion expected this year and $472 billion in 2026. A paper authored by two St. Louis Federal Reserve economists combined the BEA series for software, R&D, and information processing equipment with Census Bureau data on data-center construction to estimate AI-related capital investment growth.2 They showed the total rising from $2.083T in 2024 to $2.301T in 2025. From there they estimated the contribution to real GDP growth and concluded AI categories contributed 0.97% to real GDP growth in the first three quarters of 2025, higher than the Internet IT contribution of 0.81% in 2000, which represented 39% of total GDP growth for 2025 versus 28% in 2000.
What should we make of these large-scale investments and their contribution to growth? The first consideration is the impact of rising and falling capital investment on real GDP growth. For now, the economy is experiencing a tailwind from AI capital investment. The dynamic of continued rapidly rising investment and a sudden pullback could cause an AI bust. We will be monitoring the capital investment figures closely but are not overly concerned about these dynamics. It is the devil we know. The capital boom and bust cycle is familiar and avoidable at the investment level. Indeed, we had our best years during the Dot.com bubble crash. To the extent the bust is systemic, we prepare by adjusting the normalized earnings of economically sensitive companies and industries.
The other main issue with such significant investment is return-on-capital in the corporate sector. We performed an analysis, courtesy of James with input from Steve Fazzari, in which multiple capital payback period scenarios were estimated. They were based on varying levels of capital investment, productivity, and profit. The findings were not encouraging. Even optimistic scenarios struggled to produce breakeven profits. No doubt there will be big AI winners in the corporate sector. Yet there are likely to be more losers, and the overall corporate sector may or may not benefit. Consumers may ultimately benefit the most, as usefulness of the technology, in everything from entertainment to healthcare, accrues to their advantage.
The impact of AI on employment and productivity is complicated. Job loss is a major concern, especially in certain sectors of the economy. Yet rising living standards are synonymous with productivity gains which occur precisely because less work is required for the same amount of production. GDP per capita is a crude but useful measure of productivity. Our favorite chart shows undeniable long-term progress.
US GDP per Capita through 2024 Source: The Conference Board Total Economy Database, Angus Maddison, International Monetary Fund, ACR Alpine Capital Research
The replacement of handloom weavers, stagecoach makers, and back-office bank clerks by the power loom, automobile, and mainframe computer was painful for those who lost their jobs. Yet the long-term record shows job losses were absorbed, and general unemployment has never persisted. We are not aware of any innovation in history which produced systemic or mass unemployment and are skeptical this will happen with AI. The economic mechanics of this process, however, are anything but clear. Only the long-term data appears stable. We continue to watch developments in employment as they unfold. For now, we believe job losses will be episodic and related to select industries.
The promise of AI is the same as other general-purpose technologies, productivity gains and rising living standards. Yet it is notoriously difficult to connect specific technologies with rising productivity, even after the fact.
Source: FRED Business Sector: Labor Productivity (Output per Hour) for All Workers, Percent Change from Quarter One Year Ago, Quarterly, Seasonally Adjusted
The above chart shows how volatile the annual change in productivity is. Smoothing the annual changes is necessary to see the low single-digit gains consistent with long-term GDP per capita growth. The rolling periods also show how productivity has weakened in recent decades. If the figures can be trusted, the twenty-year average has been trending lower for some time. The more we delve into the details of productivity calculations, the less confident we become in being able to render the kind of assumptions required for a sensible forecast. The historical data ranges alone must serve as our guide. Our hope is that AI will result in productivity gains closer to the long-term trend of 2% rather than more recent gains nearer to 1½%, but we cannot be sure.
We are more confident in our understanding of asset prices. Unfortunately, we are also more pessimistic on that score. Our return-on-capital calculations suggest there is no reason to believe elevated profits relative to capital investment will support significantly higher asset prices. Conversely, we see no reason to expect our conservative cyclically adjusted earnings estimates will disappoint. Our asset price concern is not due to forecasts of a weak economy or disappointing corporate earnings. Rather, it is the price level relative to earnings, a topic which we take up in our final section.
IV. Fundamental Reversion
“Fundamental reversion” refers to the convergence of the stock return and fundamental return over time. The stock return is what the investor will earn ex-post. The fundamental return is established by the main economic components of the company. The essential premise is that the fundamental return in a functioning economy can be reasonably estimated based on the nature of a company’s economic components. The framework we use for the fundamental return in the tables below is the sum of the
dividend yield and earnings per share growth. The stock return is the fundamental return decreased or increased by the change in P/E multiple. The dividend yield on the current price is a fact, and we have spoken at length about the nature and stability of long-term GDP and corporate profit growth, most recently in our 2024 year-end commentary Mr. Market, Riding High, where the logic and data are included in the section “Corporate Profits and Mr. Market”.
The first table below shows ranges for the estimated fundamental return, and the second table what the S&P 500 stock return would be if the current cyclically adjusted P/E reverted to its historical average.
The annualized return table could show P/E ratios rising. However, given that the S&P 500 Cyclically Adjusted P/E ratio is at all-time highs, and our foremost aim is to protect capital, we are naturally inclined to reveal what happens if it declines. The reason we are pessimistic about stock returns is that even if today’s high P/Es stay where they are, the fundamental return is still uninspiring. And if today’s P/Es revert to their historical mean? Look out below.
Understanding if and when P/E ratios will decline is clearly important. ACR protects against the risk of declining multiples by maintaining portfolios with very different valuation characteristics from markets like today’s. A major market decline could cause recession as investors have less money to spend or feel as if they do. As mentioned previously, our antidote for general economic weakness is to adjust the normalized earnings of economically sensitive companies and industries.
While we cannot predict the timing, we believe some measure of fundamental reversion is inevitable for two reasons. The first relates to the equilibrium fundamental return. Under theories such as Tobin’s q (or Kaldor’s v), higher market values encourage new greenfield investment, where equilibrium is reached when replacement cost rises to equal market value (q = 1). There are many problems with this theory as applied to stock markets, and we look forward to exploring this topic further in future white papers. Unfortunately, findings from assessing the equilibrium fundamental return are unlikely to offer mean reversion to a P/E significantly higher than the historical average. The second reason is psychological. We have long discussed the fickle emotional state of “Mr. Market.” The three things in life we think can be counted upon are death, taxes, and market fluctuations. While the ACR investment team doesn’t wait around for market declines, and we are always busy turning over stones, we also welcome volatility to help shake loose a few extra opportunities.
Nick Tompras January 2026
APPENDIX
The humble yield is the simplest yet most fundamental of all metrics in finance. Debt investors naturally rely on it as their basic measure of return. Oddly, equity investors don’t pay much attention to the earnings yield, the equivalent measure for equities. The earnings yield under GAAP accounting is intended to represent what a company could pay in cash dividends while maintaining its current unit output (note that GAAP figures do not always portray the true economics of a company, which is why we typically make accounting adjustments and perform DCF (discounted cash flow) valuations especially in dynamic settings). If equities are a hedge against inflation, the investor’s return is the sum of its earnings yield and the inflation rate. Put another way, the company pays out all its earnings from selling the same unit volume of goods or services every year, while its economic value grows by raising product prices and holding costs at the rate of inflation.
A simplified example will help illustrate how this works. Let us say a company earns $1.00 and its stock price is $20. It therefore has a P/E of 20 ($20 / $1.00) and its earnings yield (E/P) is 5% ($1.00 / $20). Let us also assume the company pays all its earnings in dividends. In this case, the cash yield would be 5%. Assuming the company can raise its prices and hold its costs at the rate of inflation (a demanding assumption which requires validation!), the investor’s return would be 8% (5% earnings yield + 3% inflation). Most companies, however, do not pay all their earnings in dividends. The most important decision the board and management of a company can make is how much to pay shareholders in dividends (or share buybacks, the mechanics of which are explained in painful detail in Appendix 6 of our 2021 year-end commentary Pruning Risk, Creating Value). The concept is simple. The company should re-invest in organic growth or acquisitions if it can earn a higher return than its cost of capital. If it cannot, the company should pay dividends or buy back its shares. In our example, a company would re-invest if it could earn over 5% real or 8% nominal with 3% inflation.
Now let us say a company can earn 7% real or 10% nominal by investing in new production facilities. Management decides to reinvest 50% of its $1.00 in earnings ($0.50) and pay out the rest ($0.50) in dividends. The dividend yield would be 2.5% ($0.50 dividend / $20 stock price). The return from re-investment would produce $.035 in new earnings ($0.50 re-invested * 7% return from new production = $0.035) and earnings growth would be 3.5% ($1.035 / $1.00 -1 = 3.5%). The investor’s return is therefore 6% before inflation (2.5% dividend yield + 3.5% growth) and 9% after inflation of 3%. The 9% return from 50% re-investment is better than the 8% return when all earnings are paid in dividends. The reason is that the 10% return on re-investment is higher than the original 8% cost of capital. However, 9% is not as good as the 10% the company could have earned had it re-invested all its earnings in new production facilities ($1.00 * 7% = $0.07 in new earnings or 7% growth ($1.07 / $1.00 -1) + 3% inflation = 10% return).
One of the key considerations when investing in any company is to establish that the CEO is a solid capital allocator. All executives pay lip service to good capital allocation, but surprisingly few internalize and execute upon this critical principle. Some don’t have a clue. Others are more interested in capital allocation strategies beneficial to their own interests (unprofitably growing the empire, for example) rather than their shareholders’ interests.
A final caveat to our stock return calculation example is that for fast growing, high return companies, the earnings yield is not a good proxy for return. Let us say a company has a P/E of 50 or earnings yield (E/P) of 2% and is growing at 30% per year. Clearly the company’s expected return is higher than 2%, yet lower than 30% (assuming it cannot hold a P/E of 50 and grow at 30% forever). Investors need to take a different approach to valuing and setting expected returns for rapidly growing companies or portfolios. The earnings yield is a good proxy for the overall market and most ACR strategies.
END NOTES
1. EQR Returns
*Benchmark Inception Date: April 3, 2000 1Total return performance includes unrealized gains, realized gains, dividends, interest, and the reinvestment of all income. Returns greater than one year are annualized. 2Pure gross returns of the EQR Total Accounts composite are gross of all fees and do not reflect the deduction of transaction costs in wrap portfolios. Pure gross returns are supplemental information. 3Netof fees returns are pure gross returns of the EQR Total Accounts composite reduced by the highest wrap fee as required under GIPS; 3% is the highest tier fee under graduated fee schedules for smaller asset levels at some partner firms. Most ACR partners charge significantly less than 3%. 4Net of 1% fee returns are pure gross returns of the EQR Total Accounts composite reduced by a 1% fee.
ACR Alpine Capital Research LLC is an SEC-registered investment adviser. For more information, please refer to Form ADV on file with the SEC at www.adviserinfo.sec.gov. Registration with the SEC does not imply any particular level of skill or training.
Unless otherwise noted, all statistics highlighted in this research note are sourced from ACR’s analysis.
It should not be assumed that recommendations made in the future will be profitable or will equal the performance of the examples discussed. You should consider any strategy’s investment objectives, risks, charges, and expenses carefully before you invest.
This information should not be used as a general guide to investing or as a source of any specific investment recommendations and makes no implied or expressed recommendations concerning the manner in which an account should or would be handled, as appropriate investment strategies depend upon specific investment guidelines and objectives. This is not an offer to sell or a solicitation to invest.
This information is intended solely to report on investment strategies implemented by Alpine Capital Research (“ACR”). Opinions and estimates offered constitute our judgment as of the date set forth above and are subject to change without notice, as are statements of financial market trends, which are based on current market conditions. There are risks associated with purchasing and selling securities and options thereon, including the risk that you could lose money. All material presented is compiled from sources believed to be reliable, but no guarantee is given as to its accuracy.
The investment outlook represents ACR’s views on the economic factors that may affect the global capital markets. There can be no guarantee that these factors will necessarily occur as ACR anticipates, nor that if they do, they will lead to positive performance returns. There can be no assurance that any objective will be achieved.
The Equity Quality Return (EQR) Total Accounts Composite consists of equity portfolios managed for non-wrap fee and wrap fee clients according to the Firm’s published investment policy. The composite investment policy includes the objective of providing satisfactory absolute and relative results in the long run and preserving capital from permanent loss during periods of economic decline. EQR invests only in publicly traded marketable common stocks. Total Return performance includes unrealized gains, realized gains, dividends, interest, and the re-investment of all income. Pure Gross returns are gross of all fees and do not reflect the deduction of transaction costs in wrap portfolios. Pure Gross returns are supplemental information. Net of ACR Fee returns are Pure Gross returns reduced by 1.0% per annum, which is the standard management fee for the Equity Quality Return strategy. Please refer to our full composite performance presentation with disclosures published under the Strategies section of our website at https://acr-invest.com/strategies/eqr-equity-quality-return-performance/
The S&P 500 TR Index is a broad-based stock index that includes dividend reinvestment and has been presented as an indication of domestic stock market performance. It is unmanaged and cannot be purchased by investors. See EQR’s full composite presentation at https://acr-invest.com/strategies/eqr-equity-quality-return-performance/
Double Bubble or New Normal?
4Q 2025
Investors breathed a sigh of relief in 2025 as the global economy grew at rates consistent with historical trends despite US tariffs. Equity markets rose, and ACR strategies benefitted as portfolio company stock prices caught up with our fundamental value estimates. ACR results can be found at https://acr-invest.com/strategies/.
AI became the lead story again as tariff fears receded. In our 2025 year-end commentary, we will take stock of where we have landed three years into the AI LLM (Large Language Model) boom and three years after the 2022 bear market. We don’t know what the future holds, but valuations are resulting in higher portfolio cash levels. Our view is we are in the first few innings in the AI LLM tech layer buildout. Yet we remember well how the 2000-02 bear market occurred during the early innings of the Internet buildout. Financial market fluctuations often have their own seasons, and we believe the timing is practically impossible to predict. Thankfully, we have our valuation compass to see us through. When prices start to exceed values, we trim and sell.
Before getting to the financial and economic topics at hand, we would like to introduce you to the newest member of the ACR team. James Montier joined ACR at the end of last year as Managing Director of Fundamental Research and Performance. James is an expert in the fields of behavioral finance and value investing. He is a sought-after speaker and has authored many papers and four books on these and other economic topics. At ACR, James’ responsibilities will include publishing leading research, directing the firm’s fundamental performance initiatives, and supporting ACR’s objective of rational, unbiased investment decision-making. ACR investment team members have followed James’ work for many years and are excited to bring him into the ACR fold.
Our year-end commentaries tend to be longer than our quarterly updates. Mercifully, a high-level bullet point summary is provided below.
Commentary Sections
I. Bubble Defined
II. The Double Bubble
III. The AI Boom
IV. Fundamental Reversion
I. Bubble Defined
James’ first piece will appear next month, and here we are, already stealing his thunder. He offers this concise dynamic definition of a bubble: “when returns implied by valuations are heading in a different direction to the returns expected by investors.” We will save evidence of rising returns expected by investors for James’ paper. Regarding returns implied by valuations, ACR has been reporting for many years how expected returns have been on a downward trend. The foundational premise begins with understanding the inverse relationship between the level of the S&P 500 cyclically adjusted P/E and future returns. The supporting data for this relationship has been well documented by us and others. We share this data again for reference below.
Note:
S&P 500 Starting Cyclically Adjusted P/E is based on Real S&P 500 Price Per Share (PPS) divided by Ordinary Least-Squares Regression (OLS) trendline of S&P 500 Real Earnings Per Share (EPS) from 1926 to December 31, 2025. Each data point represents the forward S&P 500 Subsequent 10 Year Total Returns. Data points in the chart are displayed quarterly.
Sources: S&P Dow Jones Indices; Robert Shiller; BLS- CPI Data; ACR Alpine Capital Research.
The implication is clear: the higher the P/E, the lower the return. Admittedly, the chart employs quarterly overlapping periods, and the hundred-year period includes just ten non-overlapping data points. The correlation is still compelling, but the economic logic upon which it rests solidifies the story. Higher P/Es are mathematically equivalent to lower E/Ps, and the earnings yield (E/P), like any investment yield, is a
measure of return. Similar to how interest income is the numerator in a bond yield calculation, the net earnings of a company is the numerator in the earnings yield. The bond yield, correctly calculated, is equivalent to the bond’s return when held to maturity. The stock earnings yield (E/P) is an excellent proxy for the long-term stock return. The three main differences are that company earnings are not fixed, may be re-invested rather than paid in cash, and are real rather than nominal if equities are a hedge against inflation. When “real”, inflation is added to the earnings yield to estimate total return. A more comprehensive explanation of stock return calculations is included in the Appendix.
Our calculation of the current cyclically adjusted earnings yield (E/P) is 2.1%. The following chart shows how the S&P 500 earnings yield (E/P) has been in a downward trend since the Financial Crisis of 2008 and has been in a downward megatrend back to the early 1980s.
Note:
Earnings Yield (“EY”) is the inverse of the Price-to-Earnings Ratio (i.e., EPS/Price). EY represents an estimate of normalized earnings that could theoretically be paid out in dividends as a percentage of current price. Yields (income/price) are commonly used as a valuation metric to compare the margin of safety for an investment (e.g., Earnings Yield for Equities, Yield to Maturity for Bonds).
Sources: S&P Dow Jones Indices; Robert Shiller; BLS CPI data; Compustat; Bloomberg; ACR Alpine Capital Research. S&P 500 EPS sourced from Shiller data through September 30, 2025. Most recent S&P 500 TTM EPS estimate as of December 31, 2025, sourced from Howard Silverblatt. Index data pulled on December 31, 2025.
The returns implied by valuations have clearly headed lower, satisfying at least one-half of our bubble definition. Spoiler alert. Returns expected by investors have not been declining with the returns implied by valuations. Most stock market investors unwittingly cheer as prices rise ever higher without considering whether they are rising faster than underlying earnings and by how much, or the consequences. This leads us to our next section.
II. The Double Bubble
The ACR investment team is mindful of a very different perspective on rising prices. We have been complaining about high prices (our specialty) for over a decade, and they continue rising! Indeed, we wrote in our 2019 year-end commentary: “The current cyclically adjusted P/E is 34.6 – alarmingly near the 1999 high of 35.5. What are the chances it rises significantly higher?” Odds aside, prices zoomed higher. We also posited: “… the ‘20s will not roar. Rather, they will be more like the ‘00s than the ‘10s for EQR and the market”. Six years into this decade, how wrong we were! The S&P 500’s roar could be heard across the globe at 15.1% annualized. Fortunately, our flagship EQR strategy returned 15.9% gross and 14.8% annualized net of a 1% fee1. EQR’s return was also unexpected as we normally do not keep up with runaway bull markets.
A lot has happened this decade so far. We never want turmoil to visit humanity, but the steep yet brief Pandemic declines, the rising interest rate bear market of 2022, and this year’s tariff tantrums were gifts. The volatility allowed us to put cash to work (in the case of the pandemic, very large sums) and to later take gains as prices rose toward our value estimates. The following P/E chart shows the “double bubble” of which we speak.
Note:
a) ACR CAPE based on Real S&P 500 Price Per Share (PPS) divided by Ordinary Least-Squares Regression (OLS) trendline of S&P 500 Real Earnings Per Share (EPS) from 1926 to December 31, 2025.
b) Arithmetic mean.
Sources: S&P Dow Jones Indices; Robert Shiller; BLS- CPI Data; ACR Alpine Capital Research.
There are three periods in which the S&P 500 exceeded 30x earnings: the late 1920s, the late 1990s, and the current period. Only once has it exceeded 40x, the high water mark it sets today. 1929 and 1999 were single peaks: straight up and straight down. The aftermath of 1929 was of course especially catastrophic.
It was also avoidable given what we know today. While incomparable to the 1930s, the 2000-02 bear market was no slouch. The S&P 500 declined 49.1% peak to trough.
Given that we have been hovering over 30x earnings for over ten years, the question naturally arises, are we in a new normal? Until the mid-1950s, stock dividend yields were almost always higher than US treasury bond yields. That relationship reversed in the middle of the 20th century. US treasury bond yields have almost always exceeded dividend yields since late 1950s. Could this be another paradigm shift? Our answer is “no.” Logic and evidence show that stock returns revert to fundamental returns sooner or later. Conversely, the dividend payout is a choice. We address the economic phenomenon and math of fundamental reversion in Section IV. But first, we will opine on the ongoing topic du jour, AI.
III. The AI Boom
The Artificial Intelligence talked about today is a specific kind of AI, yet it has become what most are now simply calling AI. A few key terms which describe the technology and its beginnings are worthy of noting. AI LLMs can be traced back to a 2016 internal research project at Google called “Transformers: Iterative Self-Attention and Processing for Various Tasks” which led to the now famous 2017 paper, “Attention is All You Need.” Neural networks were once a discredited approach to AI. While LSTM (Long Short Term Memory) recurrent neural networks marked a comeback for the technology, the game changer turned out to be “self-attention” neural networks. They produced better translations of a word by referencing any other part of a passage simultaneously which helped clarify the word’s intent. The way self-attention models handled information was suited to the more powerful parallel processing chips found in gaming systems. The term transformer simply came from the architecture’s ability to efficiently transform input data into rich representations of meaning.
The ACR investment team’s goal is not to pick AI winners. Rather, it is part of our job to understand transformative technologies and to assess their impact on the economy and the companies we own or could own. Opinions vary on our team. All agree that AI is a useful technology. We are also implementing AI tools in each department at ACR. While these new tools are speeding up a few processes and creating efficiencies, we do not see AI helping us to become better investors yet.
Our thoughts on AI’s potential impact begins with a brief outline of possible outcomes, then we will discuss several economic issues.
In our 2024 1st quarter commentary, we laid out three general scenarios for thinking about AI’s impact:
As Co-CIO of ACR, the above summary continues to reflect my views. The most crucial point is how widely drawn the tail outcomes are likely to be. The impact of AI may turn out to be somewhat better or worse than we think today, but we see no reason to forecast wildly optimistic or pessimistic outcomes at this time. The remainder of this section addresses more specific economic issues. We organize our discussion into five categories: capital investment, return-on-capital, employment, productivity, and asset prices.
The amount of capital being invested into AI is undeniably massive. Hyperscaler spending by Microsoft, Google, Meta, and Amazon alone was $228 billion in 2024, with $363 billion expected this year and $472 billion in 2026. A paper authored by two St. Louis Federal Reserve economists combined the BEA series for software, R&D, and information processing equipment with Census Bureau data on data-center construction to estimate AI-related capital investment growth.2 They showed the total rising from $2.083T in 2024 to $2.301T in 2025. From there they estimated the contribution to real GDP growth and concluded AI categories contributed 0.97% to real GDP growth in the first three quarters of 2025, higher than the Internet IT contribution of 0.81% in 2000, which represented 39% of total GDP growth for 2025 versus 28% in 2000.
What should we make of these large-scale investments and their contribution to growth? The first consideration is the impact of rising and falling capital investment on real GDP growth. For now, the economy is experiencing a tailwind from AI capital investment. The dynamic of continued rapidly rising investment and a sudden pullback could cause an AI bust. We will be monitoring the capital investment figures closely but are not overly concerned about these dynamics. It is the devil we know. The capital boom and bust cycle is familiar and avoidable at the investment level. Indeed, we had our best years during the Dot.com bubble crash. To the extent the bust is systemic, we prepare by adjusting the normalized earnings of economically sensitive companies and industries.
The other main issue with such significant investment is return-on-capital in the corporate sector. We performed an analysis, courtesy of James with input from Steve Fazzari, in which multiple capital payback period scenarios were estimated. They were based on varying levels of capital investment, productivity, and profit. The findings were not encouraging. Even optimistic scenarios struggled to produce breakeven profits. No doubt there will be big AI winners in the corporate sector. Yet there are likely to be more losers, and the overall corporate sector may or may not benefit. Consumers may ultimately benefit the most, as usefulness of the technology, in everything from entertainment to healthcare, accrues to their advantage.
The impact of AI on employment and productivity is complicated. Job loss is a major concern, especially in certain sectors of the economy. Yet rising living standards are synonymous with productivity gains which occur precisely because less work is required for the same amount of production. GDP per capita is a crude but useful measure of productivity. Our favorite chart shows undeniable long-term progress.
US GDP per Capita through 2024
Source: The Conference Board Total Economy Database, Angus Maddison, International Monetary Fund, ACR Alpine Capital Research
The replacement of handloom weavers, stagecoach makers, and back-office bank clerks by the power loom, automobile, and mainframe computer was painful for those who lost their jobs. Yet the long-term record shows job losses were absorbed, and general unemployment has never persisted. We are not aware of any innovation in history which produced systemic or mass unemployment and are skeptical this will happen with AI. The economic mechanics of this process, however, are anything but clear. Only the long-term data appears stable. We continue to watch developments in employment as they unfold. For now, we believe job losses will be episodic and related to select industries.
The promise of AI is the same as other general-purpose technologies, productivity gains and rising living standards. Yet it is notoriously difficult to connect specific technologies with rising productivity, even after the fact.
Source: FRED
Business Sector: Labor Productivity (Output per Hour) for All Workers, Percent Change from Quarter One Year Ago, Quarterly, Seasonally Adjusted
The above chart shows how volatile the annual change in productivity is. Smoothing the annual changes is necessary to see the low single-digit gains consistent with long-term GDP per capita growth. The rolling periods also show how productivity has weakened in recent decades. If the figures can be trusted, the twenty-year average has been trending lower for some time. The more we delve into the details of productivity calculations, the less confident we become in being able to render the kind of assumptions required for a sensible forecast. The historical data ranges alone must serve as our guide. Our hope is that AI will result in productivity gains closer to the long-term trend of 2% rather than more recent gains nearer to 1½%, but we cannot be sure.
We are more confident in our understanding of asset prices. Unfortunately, we are also more pessimistic on that score. Our return-on-capital calculations suggest there is no reason to believe elevated profits relative to capital investment will support significantly higher asset prices. Conversely, we see no reason to expect our conservative cyclically adjusted earnings estimates will disappoint. Our asset price concern is not due to forecasts of a weak economy or disappointing corporate earnings. Rather, it is the price level relative to earnings, a topic which we take up in our final section.
IV. Fundamental Reversion
“Fundamental reversion” refers to the convergence of the stock return and fundamental return over time. The stock return is what the investor will earn ex-post. The fundamental return is established by the main economic components of the company. The essential premise is that the fundamental return in a functioning economy can be reasonably estimated based on the nature of a company’s economic components. The framework we use for the fundamental return in the tables below is the sum of the
dividend yield and earnings per share growth. The stock return is the fundamental return decreased or increased by the change in P/E multiple. The dividend yield on the current price is a fact, and we have spoken at length about the nature and stability of long-term GDP and corporate profit growth, most recently in our 2024 year-end commentary Mr. Market, Riding High, where the logic and data are included in the section “Corporate Profits and Mr. Market”.
The first table below shows ranges for the estimated fundamental return, and the second table what the S&P 500 stock return would be if the current cyclically adjusted P/E reverted to its historical average.
The annualized return table could show P/E ratios rising. However, given that the S&P 500 Cyclically Adjusted P/E ratio is at all-time highs, and our foremost aim is to protect capital, we are naturally inclined to reveal what happens if it declines. The reason we are pessimistic about stock returns is that even if today’s high P/Es stay where they are, the fundamental return is still uninspiring. And if today’s P/Es revert to their historical mean? Look out below.
Understanding if and when P/E ratios will decline is clearly important. ACR protects against the risk of declining multiples by maintaining portfolios with very different valuation characteristics from markets like today’s. A major market decline could cause recession as investors have less money to spend or feel as if they do. As mentioned previously, our antidote for general economic weakness is to adjust the normalized earnings of economically sensitive companies and industries.
While we cannot predict the timing, we believe some measure of fundamental reversion is inevitable for two reasons. The first relates to the equilibrium fundamental return. Under theories such as Tobin’s q (or Kaldor’s v), higher market values encourage new greenfield investment, where equilibrium is reached when replacement cost rises to equal market value (q = 1). There are many problems with this theory as applied to stock markets, and we look forward to exploring this topic further in future white papers. Unfortunately, findings from assessing the equilibrium fundamental return are unlikely to offer mean reversion to a P/E significantly higher than the historical average. The second reason is psychological. We have long discussed the fickle emotional state of “Mr. Market.” The three things in life we think can be counted upon are death, taxes, and market fluctuations. While the ACR investment team doesn’t wait around for market declines, and we are always busy turning over stones, we also welcome volatility to help shake loose a few extra opportunities.
Nick Tompras
January 2026
APPENDIX
The humble yield is the simplest yet most fundamental of all metrics in finance. Debt investors naturally rely on it as their basic measure of return. Oddly, equity investors don’t pay much attention to the earnings yield, the equivalent measure for equities. The earnings yield under GAAP accounting is intended to represent what a company could pay in cash dividends while maintaining its current unit output (note that GAAP figures do not always portray the true economics of a company, which is why we typically make accounting adjustments and perform DCF (discounted cash flow) valuations especially in dynamic settings). If equities are a hedge against inflation, the investor’s return is the sum of its earnings yield and the inflation rate. Put another way, the company pays out all its earnings from selling the same unit volume of goods or services every year, while its economic value grows by raising product prices and holding costs at the rate of inflation.
A simplified example will help illustrate how this works. Let us say a company earns $1.00 and its stock price is $20. It therefore has a P/E of 20 ($20 / $1.00) and its earnings yield (E/P) is 5% ($1.00 / $20). Let us also assume the company pays all its earnings in dividends. In this case, the cash yield would be 5%. Assuming the company can raise its prices and hold its costs at the rate of inflation (a demanding assumption which requires validation!), the investor’s return would be 8% (5% earnings yield + 3% inflation). Most companies, however, do not pay all their earnings in dividends. The most important decision the board and management of a company can make is how much to pay shareholders in dividends (or share buybacks, the mechanics of which are explained in painful detail in Appendix 6 of our 2021 year-end commentary Pruning Risk, Creating Value). The concept is simple. The company should re-invest in organic growth or acquisitions if it can earn a higher return than its cost of capital. If it cannot, the company should pay dividends or buy back its shares. In our example, a company would re-invest if it could earn over 5% real or 8% nominal with 3% inflation.
Now let us say a company can earn 7% real or 10% nominal by investing in new production facilities. Management decides to reinvest 50% of its $1.00 in earnings ($0.50) and pay out the rest ($0.50) in dividends. The dividend yield would be 2.5% ($0.50 dividend / $20 stock price). The return from re-investment would produce $.035 in new earnings ($0.50 re-invested * 7% return from new production = $0.035) and earnings growth would be 3.5% ($1.035 / $1.00 -1 = 3.5%). The investor’s return is therefore 6% before inflation (2.5% dividend yield + 3.5% growth) and 9% after inflation of 3%. The 9% return from 50% re-investment is better than the 8% return when all earnings are paid in dividends. The reason is that the 10% return on re-investment is higher than the original 8% cost of capital. However, 9% is not as good as the 10% the company could have earned had it re-invested all its earnings in new production facilities ($1.00 * 7% = $0.07 in new earnings or 7% growth ($1.07 / $1.00 -1) + 3% inflation = 10% return).
One of the key considerations when investing in any company is to establish that the CEO is a solid capital allocator. All executives pay lip service to good capital allocation, but surprisingly few internalize and execute upon this critical principle. Some don’t have a clue. Others are more interested in capital allocation strategies beneficial to their own interests (unprofitably growing the empire, for example) rather than their shareholders’ interests.
A final caveat to our stock return calculation example is that for fast growing, high return companies, the earnings yield is not a good proxy for return. Let us say a company has a P/E of 50 or earnings yield (E/P) of 2% and is growing at 30% per year. Clearly the company’s expected return is higher than 2%, yet lower than 30% (assuming it cannot hold a P/E of 50 and grow at 30% forever). Investors need to take a different approach to valuing and setting expected returns for rapidly growing companies or portfolios. The earnings yield is a good proxy for the overall market and most ACR strategies.
END NOTES
1. EQR Returns
*Benchmark
Inception Date: April 3, 2000
1Total return performance includes unrealized gains, realized gains, dividends, interest, and the reinvestment of all income. Returns greater than one year are annualized.
2Pure gross returns of the EQR Total Accounts composite are gross of all fees and do not reflect the deduction of transaction costs in wrap portfolios. Pure gross returns are supplemental information.
3Netof fees returns are pure gross returns of the EQR Total Accounts composite reduced by the highest wrap fee as required under GIPS; 3% is the highest tier fee under graduated fee schedules for smaller asset levels at some partner firms. Most ACR partners charge significantly less than 3%.
4Net of 1% fee returns are pure gross returns of the EQR Total Accounts composite reduced by a 1% fee.
2. Hannah Rubinton and Bontu Ankit Patro, “Tracking AI’s Contribution to GDP Growth,” St. Louis Fed On the Economy, Jan. 12, 2026.
IMPORTANT DISCLOSURES
ACR Alpine Capital Research LLC is an SEC-registered investment adviser. For more information, please refer to Form ADV on file with the SEC at www.adviserinfo.sec.gov. Registration with the SEC does not imply any particular level of skill or training.
Unless otherwise noted, all statistics highlighted in this research note are sourced from ACR’s analysis.
It should not be assumed that recommendations made in the future will be profitable or will equal the performance of the examples discussed. You should consider any strategy’s investment objectives, risks, charges, and expenses carefully before you invest.
This information should not be used as a general guide to investing or as a source of any specific investment recommendations and makes no implied or expressed recommendations concerning the manner in which an account should or would be handled, as appropriate investment strategies depend upon specific investment guidelines and objectives. This is not an offer to sell or a solicitation to invest.
This information is intended solely to report on investment strategies implemented by Alpine Capital Research (“ACR”). Opinions and estimates offered constitute our judgment as of the date set forth above and are subject to change without notice, as are statements of financial market trends, which are based on current market conditions. There are risks associated with purchasing and selling securities and options thereon, including the risk that you could lose money. All material presented is compiled from sources believed to be reliable, but no guarantee is given as to its accuracy.
The investment outlook represents ACR’s views on the economic factors that may affect the global capital markets. There can be no guarantee that these factors will necessarily occur as ACR anticipates, nor that if they do, they will lead to positive performance returns. There can be no assurance that any objective will be achieved.
The Equity Quality Return (EQR) Total Accounts Composite consists of equity portfolios managed for non-wrap fee and wrap fee clients according to the Firm’s published investment policy. The composite investment policy includes the objective of providing satisfactory absolute and relative results in the long run and preserving capital from permanent loss during periods of economic decline. EQR invests only in publicly traded marketable common stocks. Total Return performance includes unrealized gains, realized gains, dividends, interest, and the re-investment of all income. Pure Gross returns are gross of all fees and do not reflect the deduction of transaction costs in wrap portfolios. Pure Gross returns are supplemental information. Net of ACR Fee returns are Pure Gross returns reduced by 1.0% per annum, which is the standard management fee for the Equity Quality Return strategy. Please refer to our full composite performance presentation with disclosures published under the Strategies section of our website at https://acr-invest.com/strategies/eqr-equity-quality-return-performance/
The S&P 500 TR Index is a broad-based stock index that includes dividend reinvestment and has been presented as an indication of domestic stock market performance. It is unmanaged and cannot be purchased by investors. See EQR’s full composite presentation at https://acr-invest.com/strategies/eqr-equity-quality-return-performance/
Recent Commentary
Seeing Through the AI Haze2Q 2026 The AI Iran Rollercoaster
1Q 2026 Double Bubble or New Normal?
4Q 2025 Focusing on What Matters
3Q 2025 Capital Protection and Volatility
2Q 2025 Trump Tariffs, Business Chaos, and Valuation Adjustment
1Q 2025