A Comprehensive Look at the Empirical Performance of Equity Premium Prediction

Here's the abstract:

Among many other variables and variations of various strategies, they investigated PE10, which is the strategy being discussed in this thread. They report their results in section 4.3 on page 18:Given the historically high equity premium, is it now a good time to invest in the stock market? Economists have suggested a whole range of variables that investors could or should use to predict: dividend price ratios, dividend yields, earnings-price ratios, dividend payout ratios, net issuing ratios, book-market ratios, interest rates (in various guises), and consumption- based macroeconomic ratios (cay). The typical paper reports that the variable predicted well in an in-sample regression, implying forecasting ability.

Our paper explores the out-of-sample performance of these variables, and finds that not a single one would have helped a real-world investor outpredicting the then-prevailing historical equity premium mean. Most would have outright hurt. Therefore, we find that, for all practical purposes, the equity premium has not been predictable, and any belief about whether the stock market is now too high or too low has to be based on theoretical prior, not on the empirically variables we have explored.

Note that the "earnings-price" ratio (EP) used by Goyal and Welch is simply the inverse of the "price-to-earnings" ratio (PE). So the results they found for EP10 apply equally well to PE10. See the paper for the tables and charts referenced in the quote above, as well as for details on methodology, other results, etc.Lamont (1998) explores variations of the E/P ratio and the payout ratio. Table 8 thus explores variations on the computation of earnings and dividend ratios. For example, Earning(10Y) are the moving average 10-year earnings. We explore two different horizons: one in which the forecast begins in 1902, another in which the forecast begins in 1964.

Panel A shows that there is both statistically significant outperformance (both in-sample and out-of-sample!) the price variables (e/p and d/p) if we use longer-term moving average price ratios and if we begin our forecasts in 1902. The economic significance reached 33 basis points for the Earnings(10Y)/Price ratio, though it is below 10 basis points for all other variations. The payout ratio (d/e) does not work. Unfortunately, if we begin our forecasts in 1964, it is not just that our variables are no longer statistically significant, they outright underperform.

Panels B through D explore 3-year, 5-year, and 10-year horizons, respectively. Panel B shows that the 3-year horizon predictions look like the one-year horizons: significant outperformance of long-memory price ratios if we begin in 1902, but underperformance if we begin in 1964. Panel C and D show that the 5-year and 10-year horizon predictions become progressively worse. The 10-year horizon predictions, however, show statistically insignificant overperformance when we begin predictions in 1964, but none if we begin predictions in 1902.

In sum, there is a tiny hint that long-memory earnings-price ratios might have better in-sample and out-of-sample performance than the prevailing mean; but the empirical evidence is so modest that it is better interpreted as not speaking against e/p, instead of speaking in favor for e/p. It looks decent primarily because the other predictive variables look so incredibly bad.

So Goyal and Welch are not as enthusiastic about using PE10 as are the authors of the paper discussed here. Make of this what you will. I believe that it remains a rather contentious topic in the academic world, with respected researchers on both sides of the debate. I'm not competent to judge all of this, but it is rather interesting, and if anyone is still following this long thread, I thought you might like to see a contrary point of view.

John Norstad