Glimmers Of A Return To Normalcy In Quantitative Strategies?
Things seem to be settling down in the quant space these days and after the ride we’ve all experienced over the past 12 months, this is truly welcome news. We could use the relief. The correlations across factors and quantitative investing themes seem to be headed down, back to their historical levels. Implied correlation across stocks (the common correlation element) is subsiding. And quantitative factor volatility has fallen markedly since May 18, dropping to approximately 1/3 of where it was only several months ago.
This return to greater normalcy has us feeling optimistic about the prospects for quantitative investing. We believe that in this “regular” regime there is hope that once again our alpha models will work, that our risk models will be calibrated correctly now and that active stock picking may have a shot at being successful. This would all be good news for fundamental based quantitative investing.
As we have written extensively, quantitative investing faced five specific challenges over the past year. First, we had seen a number of dramatic shifts in the style driving the market. We had seen periods when value was working quite well and times when it failed resoundingly. We have seen the same for Sentiment: marked outperformance and striking underperformance. These periods changed sharply and suddenly. Moreover, the performance of the style that was working was utterly dominant: value worked and sentiment failed. Or vice versa. But there was nothing gray about style performance.
Second, we saw factor volatility elevate to historically high levels. From June 2008 through May 2009, we saw the greatest level of volatility in quant factors that we have ever seen, measured back to 1950. For example, April saw the 2nd worst, 1st best and 7th worst days ever for our Sentiment index. It also saw the 5 best, 5 worst and 10 best day ever for the Value index. This environment was unpleasant for any investor who does not enjoy experiencing radical daily swings in the value of their investments.
Third, there have been a series of significant rotations in the correlation of quantitative factors. Whereas historically Sentiment and Value were moderately negatively correlated (-0.2 correlation), in recent months the two themes became nearly perfectly negatively correlated (-0.95 correlation). Where previously these were two factors, they now became one factor. There was no diversification. However, within the Value theme a number of factors became decoupled. Free Cash Flow to Price, Sales to Price, Total Yield, Book to Price, and Earnings to Price started deviating from historical levels of correlation. Where previously the denominator of these ratios was the driving and unifying force, today the numerator became the key factor in explaining performance.
Fourth, risk models did a poor job of predicting tracking error. Realized tracking error deviated substantially from predicted tracking error. Where we targeted tracking error of 2.85%, we consistently realized tracking errors of two to three times that level. On the basis of numerous conversations with clients, we do not believe this is solely a problem with the risk model we use but a problem with all risk models recently. Nearly everyone reported a similar story. Today, realized tracking error is much more closely approximating target tracking error. We welcome this change.
Finally, fifth, the average systematic correlation across stocks reached all-time highs, exceeded only by the 1987 crash and a brief period in 1954. We measure this “implied correlation” as the correlation among a portfolio of stocks, where we assume the correlation is constant for each pair of stocks. In other words, implied correlation is the value one gets from doing portfolio math and ascertaining what the dispersion is among stocks within that grouping.
In recent months there has been very little dispersion in stocks across the market. Systematic risk drove everything. Stock specific news was irrelevant. Getting the individual names right in your portfolio was never less important. Getting your systematic risk exposures right was never more important.
This trend is reversing at the moment with implied correlations beginning to come down to levels last seen prior to the Lehman bankruptcy. However, there is still substantial room to go before implied correlations return to their historical levels.
When examining the characteristics of the Index today, what is most remarkable is how similar both the long and short baskets appear on a wide range of metrics. Previously the median market cap in the long basket was approximately 3 times as large as the median market cap in the short basket. Today the market cap in the two baskets is, for all practical purposes, identical. Previously, the short interest in the short basket was notably higher than the short interest in the long basket. Today, the short interest in the low Sentiment names is identical to the short interest in the high Sentiment names. Previously, the short basket was heavily tilted to names with a stock price under $5. Today, low price stocks are not an important part of either portfolio. Previously, the long basket was a relatively low beta portfolio and the short basket was a relatively high beta portfolio. Today, while the long portfolio is still has a lower beta than the market, the relative beta differences between the two portfolio has shrunk considerably.