5 May 2009

"Systemic Correlation Is At October 1987 Levels"

From Zero Hedge. quants see instability.....

First, there has been yet another dramatic shift in the style with the recent rally that began on March 10th. Value has markedly outperformed while Sentiment has strikingly underperformed. This marks the fourth such dramatic change in the past 10 months. From early May 2008 through July 14, 2008, Sentiment dramatically outperformed, returning over 50% during this period as Financial and Consumer Discretionary stocks fell and Energy and Material names rose. In mid July, these trends sharply reversed with the bursting of the energy bubble. Through the Fall of 2008, we saw Sentiment re-emerge strongly, again, while Quality also out-performed. Styles reversed yet again starting around November 21st, 2008 with the catalyst being the announcement by the Treasury of the TALF program and a general reduction in risk-aversion. Deep value names strongly outperformed, led by Book-to-Price and other normalized earnings factors, while Sentiment and Momentum factors manifestly underperformed. This trend yet again reversed itself starting on January 7th. Our Sentiment Index went on another tear and our Value Index declined sharply. And then on March 10th, yes, you guessed it, it all switched once again and reversed gears very strongly. Picking the right style has been critically important to generating outperformance while the Style that has been working has shifted repeatedly and sharply. Thus, this has been a very difficult environment in which to generate consistent outperformance.

Second, we continue to see factor volatility remain at historically high levels. Not only have there been a number of significant longer trends but we have seen the daily volatility of factor returns rise to levels with scant precedence. This month saw a number of days with extraordinary positive and negative returns to our theme indices as well as individual factors. Specifically, April 9th was the 2nd worst day ever for Sentiment and the 5th best day for Value, measured back to July 1950. April 20th was the single best day ever for Sentiment and the 5th worst ever for Value. And April 23rd was the 7th worst day ever for Sentiment. Just within this past month, we have seen some of the best and worst days ever for our quantitative indices, when measured over approximately 15,000 days. This environment certainly presents a challenge to both investors and portfolio managers who might 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 Historically, the correlation between Valuation factors and Sentiment factors has been mildly negative. Presently, the correlation is almost perfectly negative. Moreover, within the factor theme portfolios, there have been significant rotations among factors. For example, free-cash-flow based variables and forward earnings variables have decoupled from more traditional normalized earnings variables. Dividend Yield and Total Yield are now almost perfectly negatively correlated with Book-to-Price and Sales-to-Price as distressed cheap companies have cut dividends.

In turn, this has caused the natural balance built into quant models to become unglued. In the context of a multi-factor model, one cannot take for granted that the valuation and momentum are offsetting each other. Of course, this depends upon the specific Valuation and Sentiment factors one is using in one’s models. As described above, a number of Valuation metrics have become negatively correlated, thereby tempering the overall performance of the Valuation Theme. On the other hand, Sentiment variables have increased their correlation, thereby amplifying the negative performance of the theme. Hence, for many quantitative models, the natural offset of Valuation and Sentiment has broken down with the result that the negative underperformance of Sentiment is dominating overall performance.

Fourth, risk models continue to do a poor job of predicting tracking error. As shown in Figure 14, since August 2007, the risk model we use has consistently been underestimating risk. We are presently targeting tracking error of 2.85% but have consistently been realizing tracking errors two to three times that level. This month was actually slightly better as realized tracking error was only 150% of the predicted level. From our conversations with clients, we do not believe this is solely a problem with our risk model but with all risk models. Nearly everyone is reporting similar stories.

Finally, fifth, the average systemic correlation across stocks is at near all times highs, exceeded only by the days following the October 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 correlations are the values one gets from doing portfolio math and ascertaining what the dispersion is among stocks within that grouping. A high correlation means that there is very little dispersion among the stocks and a low correlation means that there is high dispersion in performance. Today there is very little dispersion when measured in stocks across the market as a whole. Systemic factors are driving stock returns across the market. Stock specific news is largely irrelevant and that this is the case in the middle of earnings season, when stock specific news should be at its height, is truly remarkable. Getting the individual names right in the portfolio has never been less important. Getting your systematic risk exposures (e.g. your style tilt) correct has never been more important.


No comments: