Friday, July 11, 2014

Understanding the Markets You're Trying to Predict

The recent post highlighted the notion of event time and the potential value in redefining the X-axis of charts.  Once we have a fresh X-axis that normalizes intrasession changes of volume and volatility, cyclical patterns in markets become clearer.  It's yet another example of how quant processes can potentially benefit discretionary traders.

The chart above runs from May 6th, 2014 to yesterday morning:  it's one of the things I was tracking closely at the start of Thursday's trade.  Each point on the chart represents 500 price changes in the ES futures.  So, in other words, every time we get 500 ticks higher or lower in the ES front month contract, we draw a new bar.  So time is measured in units of market movement, not in movement of the clock.  

What that does is create many bars during busy, volatile market periods and fewer ones during slow, non-volatile ones.

Suppose we get very little price movement during the course of 500 ticks in the index.  What that tells us is that price change is occurring within a very narrow band:  there is a high degree of consensus in the market at that moment regarding the location of value.

If we get a great deal of price movement during the course of 500 ticks, it means that price change is occurring within a much wider band.  That suggests a higher degree of uncertainty in the market regarding the location of value.

The above chart measures current value uncertainty versus its longer-term moving average.  When we have values above 1.0, there is relative uncertainty regarding the location of value.  When we have values below 1.0, there is relative certainty.  

As with traditional measures of volatility, you can see the tendency for there to be greater levels of uncertainty at relative market bottoms and greater levels of uncertainty at relative market peaks.  The normalization of the X-axis and the construction of the indicator around its recent moving average--creating a measure that is relative, rather than absolute--makes that relationship easier to identify.

One certainly could test this measure in- and out-of-sample and use it as part of a trading system.  Indeed, I have conducted such tests and have found the value uncertainty tool to be helpful.  My use of it, however, is more qualitative and discretionary:  I'm using it simply to identify whether the market environment is becoming more or less uncertain.

So, for example, I closed out a long position late Wednesday because, although we had a bounce that day, the market uncertainty was rising, not falling.  At the same time, I could see that some sectors of the market--most notably small caps--were not participating meaningfully in the bounce.  Indeed, all told we had only about 500 more advancing stocks on the day than declines.  Stock sentiment was bullish--the equity put/call ratio was a relatively low .76--but that bullish tide was not lifting all boats.  Nor was market action suggesting greater certainty in the location of value.

Notice in this example that I am using tested market measures to anchor a reasoning process.  My goal is as much understanding as prediction.  (Indeed, I did not predict Thursday's sharp overnight decline; I simply identified Wednesday's absence of strength).  I want to understand what is happening in the market and why; not just blindly predict future movement on the basis of an algorithm linking variables that happen to fit a given lookback period.  A major turning point in my trading occurred when I stopped predicating trades on predictions and instead used predictive inputs to inform a process of understanding.  Trading systems and tested indicators are most useful when they reflect an underlying understanding of markets; they cannot substitute for such understanding.

Further Reading:  Putting Historical Odds on Your Side
.