Tuesday, 26 July 2011

A "new" indicator for possible inclusion in my Naive Bayesian Classifier

I have recently come across another interesting indicator on Ehler's website, information about which is available for download in the seminars section. It is the Trend Vigor indicator and below is my Octave .oct function implementation of it, first shown on the continuous, back-adjusted Natural Gas contract
and next the EURUSD spot forex market, both daily bars.
My implementation is slightly different from Ehler's in that there is no smoothing of the trend slope prior to calculation of the indicator. My reason for this is that smoothing will change the probability distribution of the indicator values, and since my Naive Bayes Classifier uses Kernel Density Estimation of a Mixture model, I prefer not to "corrupt" this density calculation with the subjective choice of a smoothing algorithm. As before, all parameters for this part of the classifier will be determined by using Monte Carlo techniques on "ideal," synthetically modelled market prices.

In the Natural Gas chart above it can be seen that the latter half of the chart is mostly sideways action, as determined by the Trend Vigor indicator, whilst my current version of the classifier ( the colour coded candlesticks ) does not give a corresponding similar determination for the whole of this later half of the chart. This is also the case in the second EURUSD chart. It is my hope that including the Trend Vigor indicator as an input to the classifier will improve the classifier's classification ability.

More on this in due course.

No comments: