How reliable and variable are the most widely accepted long-short factor premiums across asset classes? Can investors time factor premium? In their June 2019 paper entitled “Factor Premia and Factor Timing: A Century of Evidence”, Antti Ilmanen, Ronen Israel, Tobias Moskowitz, Ashwin Thapar and Franklin Wang examine multi-class robustness of and variation in four prominent factor premiums:
- Value – book-to-market ratio for individual stocks; value-weighted aggregate cyclically-adjusted price-to-earnings ratio (P/E10) for stock indexes; 10-year real yield for bonds; deviation from purchasing power parity for currencies; and, negative 5-year change in spot price for commodities.
- Momentum – past excess (relative to cash) return from 13 months ago to one month ago.
- Carry – front-month futures-to-spot ratio for equity indexes since 1990 and excess dividend yield before 1990; difference in short-term interest rates for currencies; 10-year minus 3-month yields for bonds; and, percentage difference in prices between the nearest and next-nearest contracts for commodities.
- Defensive – for equity indexes and bonds, betas from 36-month rolling regressions of asset returns versus equal-weighted returns of all countries; and, no defensive strategies for currencies and commodities because market returns are difficult to define.
They each month rank each asset (with a 1-month lag for conservative execution) on each factor and form a portfolio that is long (short) assets with the highest (lowest) expected returns, weighted according to zero-sum rank. When combining factor portfolios across factors or asset classes, they weight them by inverse portfolio standard deviation of returns over the past 36 months. To assess both overfitting and market adaptation, they split each factor sample into pre-discovery subperiod, original discovery subperiod and post-publication subperiod. They consider factor premium interactions with economic variables (business cycles, growth and interest rates), political risk, volatility, downside risk, tail risk, crashes, market liquidity and investment sentiment. Finally, they test factor timing strategies based on 12 timing signals based on 19 methodologies across six asset classes and four factors. Using data as available from as far back as February 1877 for 43 country equity indexes, 26 government bonds, 44 exchange rates and 40 commodities, all through 2017, they find that:
- All four factor premiums are robust and significant for every asset class over the last century. Specifically, annualized gross Sharpe ratios for value, momentum, carry and defensive portfolios are 0.62, 0.67, 0.84, and 0.78, respectively. For a multifactor portfolio across all asset classes, annualized gross Sharpe ratio is 1.59, indicating a large diversification benefit.
- Factor portfolio performances for pre-discovery and post-publication subperiods are similar, offering little support for belief that markets changed after publication of the four factors.
- However, factor portfolio performances are typically 30% lower during pre-discovery and post-publication subperiods than during discovery subperiods, suggesting that discovery samples are lucky for factor specifications.
- Value and momentum returns exhibit fairly strong negative correlations across asset classes. Momentum and defensive returns have consistently positive correlations. Value and defensive returns have consistently negative but small correlations.
- There is little evidence that factor premiums relate to economic variables, liquidity, investor sentiment or crash risk.
- Factor premiums vary significantly over time in a mildly predictable way. However:
- Predictability is inconsistent across forecasting methods, asset classes and factors.
- Incremental trading frictions largely confound factor timing based on this mild predictability.
In summary, evidence indicates that premiums for four widely accepted asset return factors are: (1) persistent over time, but smaller than originally reported; (2) mutually diversifying (especially value and momentum); and, (3) likely unprofitable to time.
Cautions regarding findings include:
- Although there is a discussion of breakeven trading frictions for factor premium timing, reported results are gross. Accounting for monthly portfolio reformation and shorting costs/constraints would lower returns and Sharpe ratios. As noted by the authors, estimating these costs over long periods across asset classes is very difficult.
- Factor portfolio formation and maintenance as described is beyond the reach of most investors, who would bear fees for delegating to a fund manager.
See results of this search for other perspectives on factor timing.
