Stocks for the Long Run? Evidence from a Broad Sample of Developed Markets

Aizhan Anarkulova is a Ph.D. candidate in Finance at the University of Arizona; Scott Cederburg is an Associate Professor of Finance and the Sheafe/Neill/Estes Fellow in Finance at the University of Arizona; and Michael S. O’Doherty is Associate Professor of Finance and the Charles Jones Russell Distinguished Professor of Finance at the University of Missouri. This post is based on their recent paper, forthcoming in the Journal of Financial Economics.

Prior research suggests that investing in stocks is an extremely attractive option for investors and institutions with long holding periods. For example, Mehra and Prescott (1985) demonstrate that the historical equity premium in the United States (U.S.) is quite large, and Siegel (2014) documents that long-term losses from diversified stock market investments in the U.S. are infrequent. In a recent study, Fama and French (2018) use simulation methods based on U.S. data to show that equity investors with 20- or 30-year horizons face a very small chance of losing money and a large chance of realizing a substantial gain.

Although these conclusions based on the historical record of the U.S. equity market are reassuring, there are reasons to be skeptical. First, the U.S. return history is short from a statistical perspective in that even a 100-year sample period contains few independent observations of long-horizon (e.g., 30-year) investment outcomes. Second, conclusions from U.S. data likely suffer from both easy data bias (Dimson, Marsh and Staunton, 2002) and survival bias (Brown, Goetzmann, and Ross, 1995). Easy data bias follows from researchers’ tendency to consider uninterrupted return data from successful markets which are more readily available, and survival bias follows from conditioning on end-of-sample outcomes. Third, long-term investors should be concerned not only with average investment outcomes, but also with extreme tail events. There are several notable cases of developed stock markets experiencing losses over long holding periods, in contrast with the historical U.S. experience. As a classic example, Japan had the largest equity market in the world at the end of 1989, but experienced a stock market return of -9% in nominal terms and -21% in real terms over the subsequent 30 years.

Given these concerns, our study develops a quantitative approach to estimating the distribution of long-horizon outcomes from diversified stock market investments in developed markets. Relative to prior work, our empirical design incorporates three important features to produce a more definitive estimate of the return distribution. First, we use a long time series of returns and extend the sample beyond the U.S. by including an expansive set of 39 developed countries. This approach allows us to provide a much better statistical assessment of long-horizon returns by considering a wide range of historical investor experiences. Second, we address easy data bias by carefully reflecting in our data the experiences of investors throughout periods of war, hyperinflation, political revolution, financial crisis, and stock exchange closure. Many of these episodes are omitted from prior research. Third, we combat survival bias by classifying countries as developed (and including them in our sample) based on measures of economic development that would plausibly have been available in real time.

Our final sample contains approximately 2,700 years of stock market return data across 39 developed countries. We use bootstrap simulation methods to estimate the distributions of real (i.e., adjusted for inflation) buy-and-hold returns for horizons ranging from one month to 30 years. Much of our analysis focuses on an assessment of long-horizon returns, and we also contrast our results with those based only on U.S. data. The observed differences across the two samples allow us to quantify the impact of easy data and survival biases.

The analysis yields three primary findings. First, long-horizon return outcomes from diversified equity investments are highly uncertain. Based on our broad sample of 39 developed markets, we construct a distribution of 30-year buy-and-hold returns, which reflect the real end-of-period value (measured in local currency) of a $1.00 initial investment. We estimate the 5th percentile of this distribution to be $0.47, whereas the 95th percentile is $23.30. This disparity in investment results stands in stark contrast to the conventional view that stocks are safe investments over long holding periods. Second, catastrophic investment outcomes are not rare. We estimate a 12.1% chance that a 30-year investor will lose relative to inflation from her stock market investment, and the 1st percentile real investment outcome over a 30-year holding period from an initial $1.00 investment is just $0.14. Third, our findings based on the historical record of stock market performance across dozens of developed markets are drastically different from those based on U.S. data. The U.S. sample, for example, would lead investors to believe that mean and median investment outcomes are considerably better than the corresponding outcomes from the full developed sample, and the 30-year real loss probability estimated from U.S. data is just 1.2%. An investor learning solely from the U.S. data would also be extraordinarily surprised by the recent -21% real 30-year return in Japan, as the estimated probability from U.S. data of realizing such a poor outcome is only 0.5%. These results suggest that the historical U.S. stock market experience leads to an overly optimistic characterization (ex ante) of future investment outcomes.

Our evidence on long-horizon stock market performance has implications for several academic and practical applications. The most notable applications are those requiring a quantitative characterization of the stock market return distribution, including the choice of optimal portfolio allocation across asset classes, the design of pension systems, and the construction of target-date mutual funds.

For readers interested in this work, we have a follow-up project that extends our evidence on domestic stocks to a broader set of asset classes (i.e., international stocks, government bonds, and government bills).

The complete paper is available here.

Both comments and trackbacks are currently closed.