Testing Size, Value and Momentum Return Predictors

Do commonly used indicators reliably predict stock size, value and momentum strategy returns? In the June 2014 version of his paper entitled “A Comprehensive Look at Size, Value and Momentum Return Predictability”, Afonso Januario examines the abilities of 17 fundamental and technical indicators and indicator combinations to anticipate returns for these three factors. He defines factor portfolios based on market capitalization (size), book-to-market ratio (value) and return from 12 months ago to one month ago (momentum), reformed monthly, as follows:

1. Size = (SmallValue+SmallNeutral+SmallGrowth)/3 – (BigValue+BigNeutral+BigGrowth)/3
2. Value = (SmallValue+BigValue)/2 – (SmallGrowth+BigGrowth)/2
3. Momentum = (SmallWinners+BigWinners)/2 – (SmallLosers+BigLosers)/2

He selects the 17 indicators (such as book-to-market ratio, dividend yield, earnings-price ratio, return on equity, lagged return, short interest and implied volatility) from prior published research on predictive variables. He measures indicator values each month as the averages only for stocks in long or short sides (and the spread between them) of each of the above three factor portfolios. He applies linear regressions at monthly and annual frequencies to determine whether an indicator is more effective than the historical average factor portfolio return in predicting future factor portfolio returns. Using relevant sets of data for a broad sample of relatively liquid U.S. stocks from initial set availability (ranging from 1950 to 1995) through 2012, he finds that:

• Over the entire sample period, long-short portfolios based on size, value and momentum factors generate gross annualized returns 2.1%, 4.5%, and 9.0%, respectively, with gross annualized Sharpe ratios of 0.21, 0.47 and 0.66. However:
  • Over recent years, the size anomaly disappears.
  • While most stable, the value performs poorly during the 2007-2009 financial crisis.
  • While most successful by many measures, momentum suffers crashes (as during the 2009 recovery).
• Regarding the size effect:
  • At a monthly forecast horizon, prior-month return is the most consistently effective predictor.
  • At an annual forecast horizon, dividend yield/earnings-price ratio growth and book-to-market ratio/earning-price ratio growth combinations are most consistently effective.
• Regarding the value premium:
  • At a monthly forecast horizon, book-to-market ratio and prior-month return are most consistently effective, but little better than the historical average return.
  • At an annual forecast horizon, book-to-market ratio and the book-to-market ratio/earnings-price ratio growth combination are most consistently effective.
• Regarding the momentum effect:
  • At a monthly forecast horizon, exit rate (fraction of institutions completely exiting the stock relative to the number of institutions holding the stock in the previous interval) is most consistently effective, but little better than the historical average return.
  • At an annual forecast horizon, stock variance (average of stock return variances in a portfolio), correlation (average correlation of the excess returns of each stock in a portfolio with the excess return of the remaining stocks in the same portfolio) and the book-to-market ratio/earnings-price ratio growth combination are most consistently effective.
• Constraining portfolios to long-only with no leverage generally reduces/eliminates predictive power of all indicators for all three factor portfolios.

In summary, evidence shows that few indicators used in prior research have consistently significant power to predict gross size, value and momentum factor returns.

Cautions regarding findings include:

• Reported returns are gross, not net. Accounting for the costs of monthly portfolio reformation and shorting costs would reduce returns. Since different factor portfolios generate different turnovers, net findings may differ from gross findings. Shorting may not be consistently feasible.
• Testing of multiple factors and many indicators on the same set of data introduces snooping bias, such that the performances of the best indicators tend to overstate expected results. Testing across two subperiods somewhat mitigates this bias.
• The study generally does not address the directions of indicator-future return relationships (whether higher or lower is better).
• Sample periods are generally short for non-overlapping annual forecasts.
• The methodology is complex and beyond the reach of many investors (or costly if delegated), but conveys a sense of which indicators may be most effective. Less elaborate methods applying findings to more extreme factor portfolios may be useful.