Recessions, Expansions and Factor Performance: Not Much of a Factor-Timing Strategy
Given the recent inversion of the yield curve, and with companies like Apple guiding earnings estimates down, not to mention concerns about the impact of the US government shutdown and other issues, there’s more than just a whiff of a recession in the air. The Wall Street Journal headline on January 17 “Recession Tops CEO Fears”—was just one of several recent stories featuring the “R” word. The natural question on the minds of investors is, if things start to unravel, what do we do?
Recognizing that the popularity of factor investing is relatively new, and that many new factor-based investors may not be aware of the performance they might experience should the economy start to turn around, we decided to do a quick study.
Using the official definitions of the beginnings of recessions and subsequent expansions back to 1982 (“events”), we have calculated returns for the factors in the US4 model for the three months before and the three months following each official start date. We also calculated whether the average daily return during those periods was statistically different from the long-term average (p-value), as well as whether the pre-event return was statistically different from the post event. If we could expose some patterns in the data, it would suggest that investors may want to rein in or expand specific factor bets.
Alas, there was relatively little consistency in the differences in returns pre- and post-event across the three recessions and four expansions since the beginning of the model in 1982. There was also not much evidence that the averages were statistically different from what investors could expect over the long term—and when they were, the sign of the performance varied from one recession or expansion to the next. Short-Term Momentum (a component of the short-horizon model) saw returns related to both recessions and expansions that were statistically different from the long-term average three-month return in most of the periods under question. Unfortunately, that is probably the most difficult factor to implement, and the return was consistently in the same direction only in the period leading up to the official expansion date. Liquidity saw consistently higher-than-average returns in the period after an expansion started (so more-liquid stocks fared better than their less-liquid counterparts), but that is intuitive, as when investors start to come back into the market they often begin with the more-liquid and possibly safer names. Value, Volatility and Market Sensitivity all had consistently higher-than-average returns in the period following an expansion, but Volatility’s return was still negative in two of the four periods. Better returns for higher Market Sensitivity (higher beta) stocks as the economy recovers is not surprising, and high Value, particularly at the end of an economic contraction, tends to be populated with the most economically sensitive names.
The good news in this is that it does not appear to be necessary to change a factor-based investment strategy based on the economic environment. So, even if you can’t perfectly forecast the beginning and end of an economic cycle, sticking with your high-conviction strategy may be the best course of action.
Note: 3 months before, 3 months after and before-after rows all contain returns for the three-month period. Statistically-significant p-values are highlighted. The date stated is the month of the official beginning of the recession.
Note: 3 months before, 3 months after and before-after rows all contain returns for the three-month period. Statistically-significant p-values are highlighted. The date stated is the month of the official beginning of the expansion.
One final note. Perhaps the time periods we tested were poorly chosen—after all, markets tend to anticipate recessions well before they become official. Or the results could be inconclusive because even with more than 35 years of data there have been relatively few recessions and expansions. Observing returns in a certain direction in two out of three periods is not quite enough to conclude anything, although a 67% success rate with more observations could indeed drive a conclusion.
 Of course, many quants have been involved with factor investing for more years than we want to admit, but we refer here to the popularity of factor-based ETFs as the “relatively new” component.
The analysis is based on calendar months, but factor returns are daily. If the official date of the event is, for example, March 2001, we assume the three months prior start December 1, 2000 and end February 28, 2001 and the three months post start March 1, 2001 and end May 31, 2001.
 If you can, please give me a call.