Sunday, May 17, 2015

Volatility, Correlation, and What Makes For Good Trading

I'll be talking more about this topic at Thursday's NYC trader gathering:  how market volatility and correlation are related to forward price movement in the broad indexes.  The above chart tracks an index that I created that combines volatility and correlation across the major stock market sectors.  The bars represent average forward 10-day price change since 2012 when the volatility/correlation index has been in its various quartiles.  In general, we see subnormal returns when volatility and correlation are low and unusually strong returns when the two are high.  It is in the latter situation that we have seen selloffs, with everything taken down in a risk-off mode.  When volume and volatility are low and some sectors are showing strength and others not (weak breadth), that's when we've seen punk near-term returns.

We currently reside in the lowest quartile:  low volatility and low correlation.

The division of forward returns into quartiles based upon a candidate predictor is the start of market analysis, not an end point.  At least two further issues remain:

1)  What is the variability around the returns in the quartiles?  We're looking at averages, but perhaps these are skewed by a relative handful of very low or very high values.  Strong differences of means are far more significant when there is little variability around the means than when individual values are all over the place.  In practical terms, a given mean difference in returns may not be helpful if pursuing the difference exposes you to large drawdowns.

2)  Is your candidate predictor truly unique?  Maybe volatility and correlation are low and high simply because the prior X days have demonstrated a strong or weak return.  Just because you have a promising variable doesn't mean that the variable is *uniquely* predictive of forward returns.  A simple procedure is to construct a stepwise multiple regression where past price change is the first variable entered to see if it is predictive of the next X-day change in index prices.  Then add your candidate variable in the second step and see if this accounts for significant further variance in predicting the next X-day change.

Even if we pass the above two criteria, what we have left is a hypothesis, not a conclusion.  Any historical analysis assumes that the patterns of the past will play themselves out in the immediate future.  As a quantitatively informed discretionary trader, I will then look for evidence in the day's session that the predicted pattern is or isn't playing itself out.  When price action and market behavior lines up with what the model is predicting, there's the possibility of a trade.  When they don't line up, it means that something unique is happening in today's trading session, such that a historical pattern is not playing out.

That, too, is information to a flexibly minded discretionary trader.

Good trading occurs at the intersection of rigorous analysis and keen pattern recognition.

To be continued...

Further Reading:  My Recent Trading Experiment