Testing the Buffet Indicator: Market Cap-to-GDP
Introduction
Initially popularized by the legendary investor Warren Buffet, the Buffet Indicator is the ratio of US stock market capitalization to Nominal Gross Domestic Product (GDP). As a valuation metric the Buffet Indicator asks a simple question: “how valuable are public stocks compared to 1-year of economic output?”.
As of July 2021, the Buffet Indicator is as follows:
Total US Stock Market Capitalization: $47.13T
Gross Domestic Product: $22.72T
Buffet Indicator = $47.13T ÷ $22.72T = 207.4%
By this measure the Buffet Indicator is at an all-time high which suggests the market is currently significantly overvalued compared to underlying economic production. As the below chart demonstrates, we are truly in unprecedented territory.
The Buffet Indicator is popular amongst investors and one of the valuation metrics we track closely at Light Finance, but as any good analyst knows one metric cannot alone tell the full story. The analysis to follow has four objectives:
- Describe the history of the Buffet Indicator and what it purports to measure
- Describe the data and construction used for the analysis
- Quantify what the Buffet Indicator says about future returns
- Offer critiques and counterpoints
History
Warren Buffet first discussed what has become the Buffet Indicator in 2001 in an article for Fortune magazine titled “Warren Buffett On The Stock Market”. Even after 20 years, the article is worth reading in full as it offers an excellent summary of the history of the US stock market in the 20 thcentury, investor psychology, and how to invest for the long run. It isn’t until the end of the article that Warren lays out the rationale behind the Buffet Indicator saying:
“Below is a chart, starting almost 80 years ago and really quite fundamental in what it says. The chart shows the market value of all publicly traded securities as a percentage of the country’s business-that is, as a percentage of GNP. The ratio has certain limitations in telling you what you need to know. Still, it is probably the best single measure of where valuations stand at any given moment. And as you can see, nearly two years ago the ratio rose to an unprecedented level. That should have been a very strong warning signal.”
Indeed, had you heeded the Oracle’s advice in 1999 when the Tech Bubble reached its zenith you would have been quite pleased when it popped. This prophetic call solidified the Buffet Indicator as a key measure for investors that is closely tracked to this day.
Theory
Market Capitalization
The broadest and most widely accepted measure of the aggregate value of the US stock market is the Wilshire 5000. The Wilshire 5000 is a market capitalization weighted index which as of March 2021 contains 3,500 publicly traded stocks. Thinly traded stocks, those without readily available prices, or those worth less than $.01 are excluded from the index; in practice this makes almost no difference.
Data for the Wilshire is available beginning in January 1970. Data for the period 1970–1979 is available quarterly and daily data begins in 1980. The Wilshire was designed in such a way when it was first introduced such that a one-point change in the index was equal to a $1B change in total market capitalization. However, index divisor adjustments due to corporate actions and index composition has changed the relationship over time. By 2020 each index point reflected a change of about $1.05B in the total market capitalization. For more information, you can access the Wilshire guidebook here.
While this is a good starting point, ideally, we would go back further with the hope that a longer time series might yield more insight, but index data is hard to come by. For data prior to 1970 I use the Z.1 Financial Account — Nonfinancial corporate business; corporate equities; liability, Level, available via by the Federal Reserve and the FRED database. The Z.1 account provides a quarterly, aggregated estimate of market capitalization by adding up the value of outstanding stock as reported by companies in their 10-Q filings. The data technically goes back to 1945 but is spotty until 1950.
For this analysis I have combined the Z.1 Financial Account and Wilshire into a single index. Since the Wilshire’s index adjustment has drifted over time, I have adjusted the data back to inception (and projected it to present day) using linear interpolation. Moreover, since we’ll be working with monthly data starting in 1980, I have taken an average of the daily closing values throughout the month in order to avoid the distortion of a given endpoint. This allows us to create an estimate for market capitalization. For example, as of July 2021 the Wilshire index value was 45,838 which translates to a market capitalization of ~$47.13T.
GDP
Gross Domestic Product (GDP) represents the total value of goods and services produced by the US economy. GDP is calculated and released quarterly by the US Bureau of Economic Analysis (BEA) and typically goes through several revisions before the final figure is produced at the end of the following quarter. As such, GDP is a static measurement of prior economic. One major drawback of using GDP in the denominator of the Buffet Indicator is that we don’t observe it in real time. Contrast this with the market capitalization of the Wilshire which we can observe up to the minute (if not more frequently!).
To overcome this deficiency some analyses incorporate the GDPNow estimate available from the Federal Reserve of Atlanta. GDPNow is not so much a forecast as an estimate of GDP in real time. Using monthly data the “now-cast” attempts to derive production throughout the quarter as opposed to only measuring economic activity that has already taken place. This approach is compelling, but not without its limitations. The Federal Reserve of Atlanta states:
“Overall, these accuracy metrics do not give compelling evidence that the model is more accurate than professional forecasters. The model does appear to fare well compared to other conventional statistical models.”
Moreover, the error rate (particularly early in the quarter) suggests that incorporating GDPNow into the Buffet Indicator may introduce more problems than it solves.
For the purposes of computing the indicator and the analysis to follow I’ve chosen to use only the officially published figures for GDP as of the end of the most recent quarter. The reasons for this are twofold:
- This formulation of the indicator is the same as that originally proposed by Buffet.
- The objective of the Buffet Indicator is to give the “big picture” of market valuation. A difference of several months is probably not going to change the conclusion significantly. Furthermore, considering the error inherent in any forecast it’s difficult to say whether incorporating it into a formal model increases accuracy.
Putting the Two Together
The Buffet Indicator is attractive because it is simple. It only has two inputs
Given that stock market capitalization represents the expectation of future economic activity and corporate profitability, and GDP is a measure of the most recent actual economic activity, the ratio of these two gives expected future returns relative to current performance. In this way the Buffet indicator is similar to a Price-to-Sales Ratio of a specific stock; where GDP is analogous to the “sales” of the US.
Warren puts it best:
“What’s needed is an antidote, and in my opinion that’s quantification. If you quantify, you won’t necessarily rise to brilliance, but neither will you sink into craziness. On a macro basis, quantification doesn’t have to be complicated at all.”
Current Analysis
The most current plot of the Buffet Indicator can be seen below:
This is where it gets tricky. There is some visual evidence that the Buffet Indicator has been trending upward over time. There are some plausible explanations for this apparent trend:
- The past several decades of US economic life have witnessed a preference for capital over labor. At a basic level GDP can be broken down into wages or profits; the “sales” either get paid out to workers or retained as profits. GDP going to profits has increased over time hence the indicator has trended up.
- Technology is much more important to the US economy today compared to the 1960’s and 70’s which favored manufacturing. Technology stocks typically have higher operating margins and command higher valuations hence the overall market valuation is structurally higher.
- Trend economic growth in the US is significantly lower today than it was in the 1960’s and 70’s. In a “slow growth” economy capital in place becomes more valuable (see Piketty’s Capital in the Twenty-First Century). Valuations have increased as a result.
It’s difficult to say which of these explanations’ accounts for the apparent upward trend in the Buffet Indicator. For this analysis, let’s accept the trend as fact and examine the consequences.
It’s important to say up front that when working with economic or financial time series data it is crucial that the data be stationary or cointegrated. It is very uncommon to detrend financial time series. Most financial time series are difference stationary rather than trend stationary and substituting one for another can lead to spurious results and misleading conclusions.
Alas, that is not the promise of the Buffet Indicator. The idea of the Buffet Indicator is to determine if the ratio of market cap to GDP is significantly higher or lower than the historical average. The hope is that observing the indicator through time will grant us insight into whether the market is over or undervalued. Hence it is necessary for us to work with level data regardless of possible statistical malpractice. Nonetheless we can try our best to follow sound principles of econometric research.
Detrending
In the plot below, I present the “detrended” time series which depicts the Buffet Indicator as a percentage deviation from historical trend. Having now removed the trend we are implicitly assuming that the Buffet Indicator will mean revert toward a fair valuation. Moreover, we rule out the possibility that the detrended indicator “gets stuck” or remains at an extreme range compared to historical values.
Also presented are standard deviation bands. Again, we need to make a crucial assumption, namely that the detrended indicator is normally distributed which (if I’m being honest) is unlikely. BUT if we make this assumption, then generally speaking ~68% of the data should lie within 1 standard deviation of the mean (which here is 0 i.e., fairly valued) and ~95% of the data should lie within 2 standard deviations. A value for the Buffet Indicator greater than/less than the 2 standard deviations should be regarded as significantly over/under valued.
Summary
The current market cap-to-GDP ratio is 78% above its long-term trend which should be considered significantly over-valued.
Predictive Value of the Buffet Indicator
As I mentioned in the introduction, the Buffet Indicator first gained notoriety following the Tech Bubble when valuations reached historically unprecedented levels. The most relevant question for investors when assessing valuation criteria is: “what does this model say about the prospects of future returns?”. While predicting the stock market is notoriously difficult, it is easy to look at historical data and see how the market performed following periods of exceptionally high or low valuations.
The below plot depicts monthly datapoints from January 1980 to August 2016. As discussed in the Data section, prior to 1970 we only have quarterly estimates of market capitalization taken and aggregated from corporate 10-Q filings. Furthermore, Wilshire data, which begins in 1970, is only available quarterly until 1980. Therefore, to build the model using monthly data we need to work with the abbreviated period which contains 440 observations.
The explanatory variable (x-axis) is the Buffet Indicator’s deviation from historical trend while the dependent variable (y-axis) is the 5-year forward annualized return. The objective is to determine if a relationship exists between the Buffet Indicator and subsequent returns. A regression line, confidence and prediction interval are also depicted.
The results of the accompanying regression model follow. Standard Errors are corrected for heteroscedasticity and autocorrelation.
The results are quite strong and consistent with what we anticipated a priori. The coefficient of the Buffet Indicator is negative and highly statistically significant. This suggests that market valuation as measured by the market capitalization-to-GDP ratio (detrended) is important in explaining future expected return. Crucially, the coefficient of the Buffet Indicator is negative which tells that (on average) higher valuation is associated with lower future return.
The Adjusted R 2of the regression is ~.47 which tells us that 47% of variation in the 5-year future annualized return is explained by current valuation. This is a remarkably high figure for such a simple model and lends credibility to the Buffet Indicator as a tool for investors.
Some other casual observations stand out. The data is clearly heteroscedastic (I’ll spare you the formal tests), but in laymen’s terms this means that the variability of ‘y’ depends on ‘x’. For example, future returns are much more dispersed when the Buffet Indicator is close to 0 v. at the extremes. Around 0, the mean annualized return is ~10% as we might have guessed. This is right in line with the long run historical average annual return for large cap stocks. However, around 0, annualized returns have historically ranged from approximately -10% to 25%. This is a wide range and not very helpful information if you’re an investor.
The extremes tell a different story. For “high” valuations (those greater than .25), the data is clustered quite tightly to the regression line. At high valuations, forward annualized returns have quite reliably been around 0 or slightly negative. For “low” valuations (those less than -.25) annualized returns have consistently been quite good; ranging from ~9% to 20% bearing in mind that for “low” valuations we don’t have a lot of data points from which to draw conclusions.
The below plot labels the extreme data points so we see which years are associated with these returns.
On the low end the points correspond to 2008 and 2009. Not surprisingly this was the depth of the Financial Crisis when stocks were at some of the lowest levels on record. Subsequent years generated impressive returns as the economy snapped back. On the high side the points correspond to 1999 and 2000 and the dismal returns that followed the popping of the Tech Bubble.
So, what does this mean for investors today? Using the regression model, we can develop a forecast for the expected 5-year forward annualized return. The Buffet Indicator is currently 78% (.78) above trend. Expected return is computed as follows:
E(5 Year Forward Annualized Rtn.)= .089- .2512(.78)= -.104 or -10.4%
An expected return of -10.4% is…pretty rough and should give any investor pause.
Summary and Critiques of the Buffet Indicator
In this article we have gone on a deep dive into the Buffet Indicator. We have reviewed its history, theoretical construction, and technical aspects of the data. We also developed a model to assess the indicator’s predictive capabilities and built a forecast to help investors better understand what today’s valuations imply about the likelihood of future returns. Along the way I’ve tried to highlight possible errors or flaws in the Buffet Indicator, so investors have a well-rounded understanding of what the market cap-to-GDP ratio can tell us and where it may lead us astray.
The Buffet Indicator is not my preferred valuation metric. In fact, I don’t pay any attention to it at all. From a theoretical and practical perspective, I think it has too many issues to be reliable. Many investors purport to use it, but I do not. I offer some criticisms below.
- It is not grounded in theory. Finance 101 tells us that we should value an investment based on its expected cash flows. For a firm this may be dividends, earnings, or free-cash flow (FCF, in my opinion, being the most complete). Corporate earnings are distinct from GDP. I’m confident this distinction is not lost on Warren Buffet, but if we are to develop a valuation criterion then it should be specific to its application. A metric like GDP is much too macro.
- It lacks a well-researched literature. The Buffet Indicator has not been widely studied. Most of the literature I have seen comes in the form of articles and blog posts. While not all knowledge needs to be peer reviewed (far from it) it does help to have a rich library of research to draw from to inform decision making. Other valuation metrics are available that have been studied and their usefulness (or lack thereof) is well documented.
- Trend in the Buffet Indicator has been added after the fact. As originally proposed, the Buffet Indicator was simply market cap divided by GDP; Warren did not mention anything about trend in his article. As the indicator continued to rise following the GFC and the market didn’t collapse the utility of the indicator was called into question. It wasn’t until later that people (not Warren) began to suggest that the indicator was trending over time and that we need to be looking at deviation from trend rather than raw values. This could be the case, but it also smacks of data mining and trying to force data to fit a narrative.
- Detrending financial time series is VERY uncommon. I mentioned this before, but it bears repeating that if you see someone detrended a financial time series then this should immediately raise a red flag.
- It’s not stationary. When working with financial time series it is very important to ensure that the data is stationary. If it is not and you are proposing a model, then you should argue that it is cointegrated with your variable of interest. The Buffet Indicator is not stationary. Moreover, the detrended time series failed tests for stationarity (which I omitted). So rather than having a sound statistical foundation we had to invoke normative arguments that the indicator is probably stationary. I believe the Buffet Indicator cannot rise indefinitely. Hence in a crude way the time series is stationary. What I don’t believe is that it may/has changed structurally and that the mean we observed in the past represents the mean we will observe in the future.
The Buffet Indicator is popular because it is simple, but we should not substitute simplicity for coherence. The Buffet Indicator is interesting to consider and shows promise based on historical data. The point I want to drive home is that it should be used prudently.
Refer back to this page periodically for an up to date calculation and analysis of the Buffet Indicator! Thanks for reading!
-Aric Lux.
Originally published at https://lightfinance.blog.
