Thursday, February 23, 2017

When Technical Analysis Works and When It Doesn't



Above we see a chart of the ES futures going back to January 23rd (blue line) drawn from early this morning.  A new data point is plotted every time we see 500 price changes in the contract.  This means that the X axis is denominated in price movement (volatility) units, not in time units.  When markets slow down (such as during overnight hours or at midday), we draw fewer "bars".  When we see an upswing in movement, we draw a greater number of bars.  Thus, when nothing is happening in the market, nothing is really happening in the chart.

The lookback period going to January 23rd is one that I identified as a stable market regime.  In statistical terms, the distribution of prices over that period was stationary.  I run simple tests in Excel to compare volume and buying/selling distributions within that lookback period to identify when we have a stable regime.  Within stable regimes, we can use simple technical indicators, such as overbought/oversold measures, to help us identify candidate buy and sell areas.  The overbought/oversold measure in red looks at how price deviates from its 50-bar average in standard deviation units.

As a rule, in a stable regime, I want to be a buyer of higher price lows (oversold areas where price remains higher than at the prior oversold levels) and a seller of lower price highs (overbought areas occurring at successively lower price highs).  When the recent market is not stable (significant differences in participation and in the behavior of the participants), there is no a priori reason for believing that technical indicator readings drawn from the recent past will be relevant to the immediate future.  

What that means in practice is that using standard preset levels on standard technical measures to derive trading signals in all markets is a very inefficient process.  Much of the time, we'll be inappropriately extrapolating the past into the future.  When those strategies yield (predictably) random results, traders become frustrated and then look to trading psychology to cure their woes. Clueless coaches are apt to provide those traders with less than helpful advice to "follow your process" and stay "disciplined" in trading.  Slavish adherence to a random process will only yield consistently random results.

Technical analysis is like card counting in blackjack.  It works if there is a constant number of decks from which cards are drawn.  If the number of decks in the shoe changes randomly, knowing the number of face cards played in the recent past will not provide information about the number likely to show up in the future.  If there is a relatively constant set of participants in the marketplace and their buying and selling activity falls within stable parameters, we can make a reasonable inference as to the probability of forthcoming buying or selling.

The smart trader is not looking for where to buy or sell.  The smart trader is looking to see if the current market activity is stable relative to the activity of the recent past.  The smart trader watches the dealer and figures out when card counting truly yields a betting edge.

Further Reading:  A Dynamic Approach to Technical Analysis
.