There is a long history in financial economics of documenting some degree of predictability in asset returns. So why aren’t investors doing a better job of earning a risk premium? Is it because the prediction variables aren’t so useful after all? Or maybe the evidence showing support of earning higher risk premiums requires looking at longer periods than is the norm. Another possibility is that investors overall are incapable of mustering the emotional discipline required for exploiting forecasting opportunities.

Of all the possibilities—and there are many—the weakest seems to be dismissing the prediction factors as irrelevant. As one example, an inverted yield curve has a long history of preceding downturns in the economy. This is widely documented and recognized. As economist Bernard Baumohl advises in his book The Secrets of Economic Indicators,

…once the [yield] curve is inverted, the odds greatly increase that a recession is unavoidable. Just how certain can we be a recession will occur? Let history be a guide. Since 1960, all six U.S. recession have been preceded by an inverted yield curve months in advance. No other indicator has shown such consistency, not even the stock market.

Many other variables have shown varying degrees of usefulness in predicting returns. Indeed, a deep and broad array of published empirical literature since the 1980s in particular demonstrates that a range of predictors offer robust forecasting ability. There is an ongoing debate over the source of the predictability and the extent that the predictability translates into real-world profit opportunities. In the short term, for instance, there are few robust predictors. Meantime, there’s also an ongoing discussion about whether the predictability is evidence of a rational world where expected returns vary vs. seeing the predictability as evidence of irrational behavior. There is wide recognition, however, that expected returns vary with some degree of predictability. That’s no silver bullet, but it opens the door for thinking that we can enhance the equilibrium returns dispensed by the market portfolio.
A small sampling of the literature that supports the case for return predictability includes:
* Research on stock market dividend yield and other accounting-based metrics for equities
* Return-based volatility trends
* Correlation trends among asset classes
* Short-term return persistence (momentum)
* Relative return among asset classes
And, as we noted, the term structure of interest rates (i.e., the yield curve) is among the many predictors that have been productively analyzed through the years.
Critics argue that individual variables that perform well in historical tests often fail to deliver comparable results in out-of-sample tests, i.e., in periods outside the testing dates. For example, a recent academic study reports that individual predictors don’t outperform simple forecasts drawn from historical averages (“A Comprehensive Look at the Empirical Performance of Equity Premium Prediction,” by Amit Goyal and Ivo Welch, Review of Financial Studies, 2008).
On the surface, this looks fata for thinking that forecasting factors are useful in the real world. But the evidence that any one predictor fails at times is unsurprising. Even if a flawless predictor was able to deliver a constant stream of accurate return forecasts (a virtual impossibility, of course), its value would soon be arbitraged away. News of this predictor’s incessant forecasting strength in the past would quickly attract new investors, who would then embrace the predictor, thereby rendering its practical value nil by bidding up the relevant assets to prices that reduce expected return to zero or less. In addition, fluctuating economic conditions are likely to provide varying degrees of predictability power to any one predictor at different points in the business cycle.
The challenge for investors is interpreting the literature on predictors and deciding how to utilize the research while recognizing that the forecasts will be less than perfect at all times. A possible solution is combining predictors to minimize the potential for failure in any one predictor at times. In other words, by diversifying the set of predictors used to forecast returns, the reliability of the prediction may be enhanced, if only marginally.
The economic rationale for combining individual predictions as a means of raising the success rate of forecasts dates to research published more than 40 years ago (“The Combination of Forecasts,” by J.M. Bates and C.W.J. Granger, Operational Research Quarterly, 1969). This paper is widely cited as establishing the basis for showing that multiple forecasts are superior to individual forecasts.
Subsequent research over the years has strengthened the case for expecting pooled forecasts to outperform its components in isolation. A 1983 paper, for instance, observes that combined forecasts “can be quite accurate” (“The Combination of Forecasts,” by Robert Winkler and Spyros Makridakis, Journal of the Royal Statistical Society, 1983). And a 2004 study shows that combining forecasts of economic growth has a habit of outperforming individual predictions (“Combination Forecasts of Output Growth in a Seven-Country Data Set,” by James Stock and Mark Watson, Journal of Forecasting, 2004).
Applying the concept of combination forecasts to predicting the equity risk premium, a paper published this year demonstrates so-called out-of-sample predictive power for 15 economic variables when used in concert for predicting the excess return for stocks. Diversifying the set of predictors, in other words, minimizes forecast errors, according to “Out-of-Sample Equity Premium Prediction: Combination Forecasts and Links to the Real Economy,” by David Rapach, et al. (Review of Financial Studies, 2010).
Using multiple predictors to forecast risk premiums is a relatively recent line of research for financial economics, even though the underlying concept has been formally recognized for decades. Is this a short cut to big gains? Of course not. But as researchers continue to peel away the onion skin, the mysteries of asset pricing continue to be revealed, albeit slowly and unevenly.
The lessons from the evolving research with combination forecasts are intriguing. One of the implications is that forecasting risk premia requires a broad set of predictors. In effect, investors should diversify the sources of their forecasts. This reasoning is that each predictor’s value rises and falls through a business cycle. Different predictors harbor different information about what’s coming at different times. As such, drawing predictions from predictors whose information is highly correlated is subject to a higher failure rate compared to a more diverse and carefully chosen mix of predictors.
This may be a revelation to some, but it’s actually more of the same in financial research. “The revolutionary idea that defines the boundary between modern times and the past is the mastery of risk,” Peter Bernstein explained in Against the Gods: The Remarkable Story of Risk. This revolution is in fact a continuum of ideas and insights. That includes the growing evidence that we can’t blindly assume that one, or even a handful of randomly chosen predictors will suffice. Investing is hard work, and it’s destined to get harder. So it goes for investors as they continue to eat under the tree of financial knowledge.