Custom Search

Parallel functions for MTS

OMEGA WORLD ABSTRACT

Introduction

    This presentation is designed to detail an additional tool which can be used to make trading systems more self adaptive and therefore more responsive to current market conditions.

    If a system uses back data to any degree it can be regarded as being self adaptive to one degree or another. Moving averages, standard deviations, breakouts, neural networks, etc. all rely on some historical price movement to generate buy and sell signals.

    This programming technique takes the self adaptive concept one step farther by using the system itself to adjust its own trading parameters for each trade.

    I wish to emphasize at the beginning that this is a programming technique which must be applied in a different manner to each and every system on which it is used. It is not a canned function or add on program which can be applied to any system.

    Also, since the programming involved in the application of this technique can be quite involved and extensive, it should be emphasized that this is not a fix - all for mediocre or poor systems. In fact, it will probably worsen the results of a poor system since the variables will constantly be reset to extreme values, making the equity swings of the system even more pronounced.

    Systems which respond best to this technique are those which are considered robust in nature and remain profitable over a progressive set of input variables. Such a system should show a bell curve pattern when the results of an optimization over the critical inputs is performed. Systems which respond well to frequent optimization will find this technique useful in improving performance and smoothing out equity curves.

BackTest Regular Optimization

Traders and system developers regularly check variations of their system against recent back data in an attempt to discover if an underlying change in the market has effected the performance of their system. Done properly, this effort can be rewarded with improved system performance real - time. Improperly done, which is more often the case, frequent re - optimization leads to a system which is overly curve fitted and more prone to losses.

    The difficulty is knowing how often to optimize, over what system parameters and how much data should be used for each test. To come up with the correct testing parameters is a time consuming operation since there are so many variables to consider. Also, if the testing is to be accurate, a fairly large volume of past data should be considered.

    Through the use of parallel functions, one can set up a group of indicators which will graphically depict to the user when a significant change has occurred in the manner in which the system is responding to changes in market personality. Additionally, if warranted by indicator observation, the system can be altered to automatically change variable values when indicated by changes in the market.

    In this manner one is able to observe what the results of the system would have been had re-optimization occurred at regular, defined intervals over specified input values.