Thanks for Your interesting post.

We trade some kind of gap strategy (so in its core is simple rule: if open is lower than yesterday close by more than... ) and some time ago try to improve it results by using binary classifiers as a filters. Our strategy is short term, it trades in average about 20 minutes from open.

In our research we use about 40 thousand variables/features as inputs, in next step try to select important feature and than use it with binary classifiers as a filter to say us - trade or not. Filters are independent form strategy rules itself. We get quite impressive results. There was many interesting points in this research and one of them was use of some kind of market environment indices.

We use price of asset, many indicators using this price but also market indices (like different stock market indices RUT,S&P, VIX etc) and bonds, gold and oil prices and indicators using it.

We try to learn different kind of binary classifiers, input variables was determined by historical transactions of strategy and decision variable was simple transaction gain (discretized - 1 if gain 0 if loss or even).

When we was selected important variables/features we found that majority of them is not price of asset itself and its derivatives (indicators) but various indicators showing market environment.

So I suggest to improve Your result You could try to use some kind of indicators and/or indices showing broad market environment.

If You wish You could find our research : http://ssrn.com/abstract=2462169

Regards,

Andrzej

Jonathan,

Thank you for sharing results of the analysis. Technical Analysis, and trading patterns and rules formulated in terms of prices, volumes, open interests (in futures), quantified news supported by historical data were always under investigation by traders. It is a matter of definitions what to include in Technical Analysis.

These subjects did not escape academic interest either. As an example of the latter I would like to attract your attention to the paper: Neftci, Salih. Naive Trading Rules in Financial Markets and Wiener-Kolmogorov Prediction Theory: A Study of "Technical Analysis", The Journal of Business, Volume 64, No. 4, October 1991, pp. 549 - 571. While the paper was written more than 20 years ago, Neftci names the condition under, which a practitioner may expect profits. I am citing it as is:

"I showed that a few of these rules generate well-defined techniques of forecasting. Under the hypothesis, economic time series are Gaussian, and even well-defined rules were shown to be useless in prediction. At the same time, the discussion indicated that if the processes under consideration were nonlinear, then the rules of technical analysis might capture some information ignored by Wiener-Kolmogorov prediction theory. Tests done using the Dow-Jones industrials for 1911-76 suggested that this may indeed be the case for the moving average rule."

Therefore, many researchers "decompose" the problem and study the effects "separately" checking whether price changes are Gaussian or not, are there dependencies between the financial variables, or linear correlations. This "puts water" on the wheel of linear or non-linear properties of financial time series. As an example of recent findings in futures I can reference to my recent research "Optimal Trading Strategies As Measures of Market Disequilibrium" http://arxiv.org/abs/1312.2004 (direct PDF download http://arxiv.org/pdf/1312.2004v1.pdf ). Such properties are not permanent making tools monitoring them valuable. What was useless can become useful and vice versa several times during the life of an individual trader.

Best Regards,

Valerii