Conventional wisdom is that stocks are a safe investment over long holding periods, with long-term loss realizations in the U.S. being rare or nonexistent. 1 This leads to the common investment advice that young investors with long horizons should invest heavily in stocks and “set it and forget it.” However, a broader and longer examination2 of global stock market performance challenges this belief and suggests that stocks are not as reliable of a long term investment as many believe.
The U.S. stock market has shown strong historical performance, and many experts like Fama and French (2018a) estimate a low probability of loss and a high probability of substantial gains for investors with 20- or 30-year horizons. However, it’s worth noting that the U.S. return history is relatively short, spanning less than a century, which offers limited statistical information about what happens over 30-year horizons.
The U.S. stock market was the strongest performing market in the 20th century. As Stocks for the Long Run author Jeremy Siegel noted:
“The evidence of this study confirms the conclusion that U.S. stocks have been the best long-term investment over the past century. For the entire period from 1926 to 1996, the average real return on U.S. common stocks has been about 7 percent, far higher than the real return on any other investment. This long-run performance is unique not only in U.S. experience but also in that of other developed and emerging markets. The U.S. stock market has outperformed all other equity markets in the 20th century.”3
I believe that it was impossible to know this ex ante and if you had asked inventors circa 1900 what country offered the most promising investment returns for the coming century, Germany or the UK would have been far more likely candidates than the U.S.
In a more recent example, Japan seemed set to establish itself as an economic giant after tremendous economic growth in the post WWII period through the 1980s. At the end of 1989, Japan’s stock market was the largest globally in terms of aggregate market capitalization.
However, over the subsequent 30 years (from 1990 to 2019), a diversified investment in Japanese stocks produced returns of -9% in nominal terms and -21% in real terms, including dividends. Japan’s experience is not unique, as several developed countries have realized worse performance or even complete stock market failure (e.g., Jorion and Goetzmann, 1999).
In their paper Stocks for the Long Run? Evidence from a Broad Sample of Developed Markets, O’Doherty, Cederburg, and Anarkulova sought to take a broader view looking at developed markets around the world from 1841 to 2019. They noted that there is a survivorship bias in that continuous stock return data from successful markets are more readily available. The sample used in the study achieves substantially greater coverage of developed country periods compared to previous studies to minimize this bias.4
Consistent with findings in prior literature, the distribution based solely on U.S. data indicates a low 1.2% probability of a loss in buying power over a 30-year horizon. However, the full sample distribution, which includes data from other developed countries, reflects a much higher probability of loss at 12.1%.
Evidence from the developed country sample indicates a considerable risk of loss for long-term investors. The distribution suggests that the -21% real return realization in Japan over the past 30 years is not exceedingly rare. In fact, this observation lies in the 9th percentile of the wealth distribution meaning that over the sample studied, there was almost a 1 in 10 chance of a comparable or worse outcome.
An investor who learns about the distribution of 30-year returns using only the U.S. experience would assign a probability of just 0.5% that a return as extreme as the Japanese return realization could occur. The abundance of similar examples suggests that the U.S. distribution is overly optimistic.
One of the most prominent features of the results is the substantial uncertainty over real investment outcomes faced by long-horizon investors. For a ten-year investor, for example, the 1st percentile of real payoff is just $0.13, whereas the 99th percentile is $8.75. The dispersion in the payoff distribution is even more pronounced at the 30-year horizon, as the 1st and 99th percentiles are $0.14 and $53.45, respectively.
Their results show the 20-year and 30-year probabilities of a decline in wealth from stock market investments are still substantial. The loss probability is 15.5% at 20 years and 12.1% at 30 years. This means there is around a 1 in 6 chance of finishing with less than you began with over a 20 year period and 1 in 8 chance of finishing with less after 30 years. These results contrast those in Fama and French (2018a) and also contradict the conventional advice that stocks are safe investments at long horizons. The findings highlight the importance of guarding against biases related to survival and easy data in characterizing the risks faced by long-term investors. Even at long horizons in the world’s most developed markets, investors have had a considerable risk of loss.
The study also does not include the effects of commissions, fees, and expense ratios incurred to gain market exposure. This means that actual investor outcomes could be even worse than the results suggest.
One reasonable objection to these findings would be that markets change over time. The nature of financial markets in the 1850 and 1990s was very different in terms of the number of listed securities, concentration of firms across industries, trading technology, availability of pricing and financial information on listed firms, trading regulations, investor protections, among other things.
International equity markets in the late nineteenth and early twentieth centuries were also highly concentrated in railroad and mining-related firms.
To take this into account, they also looked at the distribution of real payoffs for samples starting roughly 40 years apart from the original 1841 sample: 1880, 1920, 1960, and 2000.
These alternative start dates show the original finding is pretty robust. In all but the post-1960 sample, there is still a greater than 1 in 10 chance of having lost money at a 30 year horizon and the 1960 sample results in just barely breaking even which I believe is a far worse outcome than most investors forecast.
We believe this is good evidence to suggest that investors should consider diversifying outside of equity focused portfolios. While historical evidence from the U.S. market alone might suggest a relatively safe investment environment for those with long investment horizons, this study shows that the experience in other developed countries is more nuanced.
Conclusion
In conclusion, this study challenges the conventional wisdom that stocks are a safe investment for long-horizon investors in developed countries. Using a sample that avoids biases related to survival and easy data, the study demonstrates that the risk of loss for long-term equity investors is significantly higher than most people believe. As a result, we believe investors should consider diversifying their portfolios outside of just stocks to mitigate the risk of significant losses.
Using the developed world sample, an investor who is comfortable with no more than a 2% chance of losing money over a 30 year period, should allocate no more than 35% to stocks, much less than is found in most portfolios today.
- see, e.g., Siegel, 2014
- Dimson, Elroy, et al. “Stocks for the Long Run? Evidence from a Broad Sample of Developed Markets.” Journal of Portfolio Management, vol. 28, no. 2, 2002, pp. 110-120. doi: 10.3905/jpm.2002.319361.
- Siegel, Jeremy J. “The Long-term Returns on the Original S&P 500 Firms.” The Journal of Finance, vol. 53, no. 6, 1998, pp. 2105-2128. doi: 10.1111/0022-1082.00075.
- Second, survivor bias can lead to an upward bias in performance relative to ex-ante expectations (Brown et al., 1995). To combat survivor bias, researchers used a classification of developed countries and treatment of return data that doesn’t condition on eventual market outcomes. Before 1948, countries entered the developed sample when their agricultural labor shares declined below 50%, drawing on evidence about labor patterns from the economics literature (e.g., Kuznets, 1973). After 1948, researchers used membership in the Organisation for Economic Co-operation and Development (OECD) and its European predecessor, the Organisation for European Economic Co-operation (OEEC). The treatment of return data in instances of market disruptions and failures (e.g., the temporary closure of stock exchanges or the permanent disappearance of the stock market in Czechoslovakia) reflects investor experiences to minimize survivor bias.
