Does the earnings yield (inverse of price-to-earnings ratio, or E/P) usefully predict returns for individual stocks? In their April 2011 paper entitled “Reexamination of the Earnings-Price Anomaly by the Buy-Sell Strategy”, Hsin-Yi Yu and Li-Wen Chen test a long-only strategy that forms monthly value-weighted portfolios based on time-series sorting rather than cross-sectional sorting. Time-series sorting ranks stocks according to current E/P of each relative to its range over the prior decade. The strategy tested buy stocks near the top of their respective ten-year ranges and subsequently sells them when they move to the bottom. Intuitively, stocks near the top (bottom) of their respective historic E/P ranges are likely to be undervalued (overvalued). For reference, they also test a strategy that forms portfolios based on cross-sectional sorting by current E/P and held for a fixed interval, while noting that such sorts make little sense because average E/P varies considerably by industry. Using earnings and price data for all common stocks listed on NYSE, AMEX and NASDAQ from January 1962 to December 2010, they find that:
- Portfolios formed via monthly cross-sectional sorts of stocks on current E/P with a one-month holding period, whether long-only the highest E/P stocks or long-short the highest-lowest E/P stocks, generate no three-factor (adjusted for market, size, book-to-market) alpha over the sample period. The book-to-market factor essentially explains returns for this strategy.
- Portfolios formed via monthly time-series E/P ranking with a one-month holding period, whether long-only the highest ranked stocks or long-short the highest-lowest ranked stocks, also generate no alpha over the sample period.
- In contrast, portfolios formed monthly by buying stocks whose time-series E/P ranking is in the top 20% and holding them until they drop into the bottom 30% generate an average gross three-factor monthly alpha of 0.17% (approximately 2% annually) over the sample period. Furthermore:
- Buying and holding the same stocks indefinitely generates no alpha; the sell signals are crucial.
- Shifting the buy threshold to a higher E/P ranking and shifting the sell threshold to a lower E/P ranking generally increase alpha.
- The average stock holding period for the strategy based on time-series E/P ranking is five to six years, suggestion exploitation of a long-term reversion effect. Controlling for intermediate-term price momentum does not affect strategy alpha.
- Results for the strategy based on time-series E/P ranking are not sensitive to delays of up to six months in acting on buy or sell signals.
- However, alpha for the strategy based on time-series E/P ranking concentrates in the last two decades of the sample (1990-2010) and is not significant in prior subsamples.
In summary, evidence suggests that investors may be able to identify value based on individual stock E/P relative to its long-term history, but the modest alpha of this approach and the absence of alpha during extended subperiods undermine confidence in it exploitability.
It would be interesting to know the raw returns from the time-series sorts (not presented in the paper).
Cautions regarding findings include:
- Requiring ten years of price and earnings history to calculate E/P time-series rankings excludes many stocks from consideration.
- The sample period (49 years) is not long in terms of independent decade-long lagged E/P ranking intervals.
- Given the number of buy-sell criteria tested, there may be substantial data snooping bias in results. In other words, the best-performing combinations may incorporate a material amount of luck.
- Reported portfolio returns are apparently gross, not net. However, the dent in alpha from trading frictions should be modest because of the long average holding period.
- Statistical significance tests assume stock returns have normal distributions.
