We observe the following from our analysis, based on the US market data:

• In 83% of realized drawdowns, the magnitude of drawdown is more severe than would have been expected if returns were normally distributed. In the shorter international samples, realized drawdowns were worse in 70% to 79% of cases. Realized drawdowns were, on average, worse than simulated drawdowns for all factors without exception.
• For the value factor, the duration of the drawdown (from performance peak to trough) tends to be on par with what we expect under normality. The recovery, however, tends to be quite sharp, usually as the market bubble bursts, which differs from our expectation under normality.
• When the momentum factor is defined within the small-cap universe, we observe some of the most extreme differences between realized and simulated performance. Under normality, we would expect to observe a drawdown of 27.8% only once every 17,000 years, a period so long that, to go back that far, much of the planet would have been in the midst of the Ice Age. For the momentum factor, in general, the time from previous peak to trough is substantially shorter than what we should expect under normality because of momentum’s tendency to experience sharp crashes.
• The low-beta factor, defined within both the large- and small-cap universes, exhibits the greatest magnitude of underperformance of all factors. As we noted earlier, the low-beta factor does not earn a positive risk premium before we control for market risk. This means that low-beta strategies can underperform the market portfolio for long periods of time. The shrewd reader will recognize we are dealing with a benchmark-mismatch problem; that is, the market portfolio is not the appropriate benchmark for the low-beta portfolio, because the benefit of the low-beta portfolio is in risk reduction and not in delivering value-add over the market portfolio. A low-beta portfolio earns an average return comparable to that of the market, and its benefit comes from the fact that it assumes less market risk.
• Diversification does reduce risk, but far from completely. The realized drawdown characteristics of the six-factor portfolio are disappointing compared to the simulations. In the US market simulation, the worst drawdown was 13.2%, with the period from peak to recovery extending a little over five years. The realized worst drawdown in the United States over the last 55 years was 18.7%—and still we have not reached the last peak observed almost a decade ago. Even the second-worst realized drawdown of 14.2% is more severe than the 13.2% worst drawdown under the assumption of normality. This observation drives home the point that if we assume correlations are constant, and that we can diversify away almost all factor risk by investing in multiple factors, we may be in for an unpleasant surprise. Investors in the last quant crash, who were assuming low correlations among investment assets, can bear witness to the cost of such an assumption.

In summary, realized periods of underperformance when compared to underperformance estimated by theoretical normally distributed data show 1) more severe drawdowns; 2) for value factors, quite prolonged periods from peak to trough often followed by speedy recoveries; 3) sharp momentum crashes, and 4) limited diversification benefits from combining factors into portfolios.