Historical financial data can be a useful tool. It can confirm common sense ideas such as the concept that stocks should usually earn more than bonds, and that bonds should usually earn more than cash.
And it can be used to find flaws in plans — as William Bengen did in his famous 1994 study, which found that you’re setting yourself up for trouble if you spend from your retirement portfolio at a rate equal to its historical average return.
But it’s easy to get into trouble by using specific historical figures as a tool for predicting the future.
Which Figures Should We Use?
According to my 2012 edition of the Ibbotson SBBI Classic Yearbook, the annualized after-inflation return for U.S. stocks from 1926-2011 was roughly 6.6%.* And the inflation-adjusted return for U.S. Treasury bills over the same period was roughly 0.6%.
So, if we’re trying to pick a number to use for average stock returns for the future, should we use the 6.6% real return figure? Or should we expect the more stable figure to be the 6% equity risk premium (that is, the difference between stock returns and Treasury bill returns)?
If it’s the risk premium that we expect to be more stable, given how low interest rates are right now, that would give us an expected real return for stocks of just 4.1%.**
And according to a paper from the Credit Suisse Research Institute***, from 1900-2012, the global equity risk premium was just 4.1%. So going forward, should we be projecting based on the historical equity risk premium in the U.S.? Or should we be using a global average? Using that 4.1% figure would give us expected inflation-adjusted stock returns of just 2.2%.
The World Changes
In statistics, you learn about a given population by studying a sample of the population. And the larger your sample size, the more confident you can be in your conclusions about the underlying population.
With investing, the only way we can increase our sample size is to wait. For instance, if we’re concerned about annual U.S. stock returns, we have no choice but to collect that data at the glacial pace of one data point per year.
And a decent case can be made that the underlying population from which we’re drawing our sample is in fact changing over time, thereby reducing the usefulness of our older data points.
For example, in the 1930s, most U.S. households did not own stocks, placing a trade took several minutes and required talking to an actual person, trades were much more expensive, there were hardly any mutual funds (and no index funds or ETFs), nobody had up-to-the-second news, and the regulatory environment was entirely different.
So, if the purpose of our statistical analysis is to draw conclusions about what we can expect in the future, should we really be including results from the 1930s in our analysis? What about the 1940s? It’s not really clear where to draw the line.
Should We Ignore History?
My point here isn’t that history is useless. Rather, my point is that any time you encounter projections or conclusions based on historical figures, you would be well served to maintain your skepticism. In many cases you will find that there are alternative ways to interpret and apply the historical data that would lead to different conclusions.
*”Stocks” meaning the S&P 500 and the S&P 90 prior to the creation of the S&P 500.
**Calculated as 0.1% Treasury bill yield, plus 6% risk premium, minus 2% inflation (based on the market’s apparent expectation of roughly 2%, calculated as the spread between yields on short-term TIPS and nominal Treasuries).
***The study in question has since been taken offline. The 2014 version, however, can be found here.