Sunday, June 15, 2014

Toward a Dynamic Approach to Technical Analysis

In my recent post, I suggested that a dynamic technical analysis would not rely upon a fixed set of indicators and chart patterns to be interpreted in a uniform manner.  Rather, fresh predictors would arise from a study of drivers impacting the most recent market regimes. Traders would follow the indicators that have demonstrated predictive accuracy; not untested ones presumed to possess universal validity.

Let's take an example from the current market.  Using tests of stationarity, I have identified a stable period of market behavior embracing the recent past.  That stability means that market price changes during that period can be assumed to proceed from a single process.  If the time period under consideration possessed wildly different statistical properties, such as 2008 and 2014 for stocks, then there would be no assurance that price changes were resulting from a stable process.  That would give us no reason to extrapolate patterns from one period to the other.

A major problem with traditional use of technical indicators is that they are employed uniformly across non-stationary periods--especially intraday.

It turns out that during the most recent stationary period (i.e., "regime") a major driver of short-term price has been correlation.  How SPX is correlated with other equity instruments has been significantly predictive of short-term forward price change.  

The catch is that single measures of correlation were not particularly informative in my research.  Rather, aggregated correlations across a large number of stocks and sectors and over intraday horizons ended up being an excellent measure for short-term trading signals.  In other words, the most valuable indicator of correlation was not a standard technical indicator, but rather a measure that required research.  Moreover, I had to research it in such a way as to not overfit the data relationship.

But there it is:  when the correlation measure exceeds .40, the next day's price change has averaged +.35%.  When the correlation has been below .40, the next day's price change has averaged -.37%.  As long as we stay in the current regime--a key assumption--I expect market strength when correlations are high and rising and weakness when they are low and falling.

That becomes a potential trading "setup" for the current regime.  In a future regime, correlation may be much more modestly associated with forward price change--or it may be predictive over longer time frames.  Setups become dynamic, because they adapt to changing markets.  

But correlation is but one factor that sets up in the current regime.  There are others, from options ratios to intraday volatility.

This is how many hours of preparation go into a single hour of trading.  It takes time to create and test large correlation matrices.  It also takes time to test the many other variables that are associated with the factors of sentiment and positioning that are key drivers in the current regime.  Once the research is done, however, the trader knows the indicator levels that are significant and is able to automate alerts.  That way, the trader can seamlessly incorporate multiple tested technical signals with his or her discretionary judgment about the market.

The challenge is not that technical analysis doesn't work.  The problem is that technical analysis works the way that parenting works:  powerful when dynamically adapted to situations; diluted when applied uncritically.