Optimizing the forex trading system parameters: USD/JPY

User Rating: / 2
PoorBest 
Written by Forex Automaton   
Friday, 01 May 2009 12:11
Article Index
Optimizing the forex trading system parameters: USD/JPY
Optimizing the forecasting parameter
Optimizing the stop-loss parameter
Optimizing the trade entry parameter
Optimizing the trade exit parameter
Summary of progress
All Pages

I continue with the series of forex trading system optimization reports for the individual currency pairs, traded on the day scale, which began with AUD/JPY. In the algorithm, the forecast signal whose nature will not be disclosed is fed into the money management framework driven by three adjustable parameters. In the current test setting, the parameters form a multi-dimensional space populated by figures of merit, as obtained by Monte Carlo simulation of independent trading histories. The set of parameter combinations constitutes the totality of possible trading strategies under study. The goal is to optimize the trading strategy by finding the best values of parameters on the basis of the simulated trading data.

My present plan is to go through a similar study for every major forex exchange rate involving USD. After that, certain things related to money management in the context of such an algorithm should become clear, and I should be able to zoom down on a "golden" patch in the money management parameter space.

In the production regime, all major forex exchange rates will be analyzed simultaneously, therefore the dimensionality of the picture and the computational complexity of the "AI" procedure will grow considerably. So will the CPU demands of the optimization procedure. But the money management insights extracted from optimizing the individual currency pairs will hopefully prove helpful in limiting the multitude of virtual traders to simulate, which will speed up the optimization.

Analysis approach and the data set

It can not be overemphasized that for the analysis to be of any value, the algorithm may not trade the data used to train its decision making. A run of the program included simulations of trading histories of over 13,000 independent "virtual traders" (forex robots), each of them being an incarnation of the same algorithm, differing by the setting of the adjustable knobs. This report uses the USD/JPY day scale data covering the time interval from August 20, 2002 to March 23, 2009. The key concepts of conditional projection distributions and profile histograms have been explained before.

There are a few points worth recording here, related to the code development. This is a newer version of the code as compared to the one used to obtain the data for the previous two reports (AUD/JPY and EUR/USD). Most changes had to do with introducing data serialization (object persistency), an issue important for production running, but not for the optimization. The main change that can affect the results of these studies: in this version, the last business day of the week is used exclusively to close all positions; the positions are not kept through week-ends.



Last Updated ( Monday, 04 January 2010 12:36 )