Skip to content

Creating the Adaptive Time Machine

September 15, 2009

The “Adaptive Time Machine ” project was developed in collaboration with Corey Rittenhouse of Catallactic Analysis. For the purposes of this series we will demonstrate a variety of simple learning algorithms that  function well and incorporate some of the concepts underlying our proprietary algorithm.  The technology is called “ARO” which stands for Adaptive, Robust, and Optimal. These are the qualities that were emphasized in creating the machine to make it stand apart from neural networks or conventional machine learning. In a sense, it was designed to straddle the line between being robust, yet optimal, and capable of functioning in new environments using an adaptive mechanism.

There are a lot of factors to consider when creating a learning algorithm, but one of the most important is that it is “bias free.” What I mean is that the algorithm should not have any preconceived trading  style such as a bias toward trend following or mean reversion. It should also consider all possible combinations of  a given strategy within reason. To be “bias free” the ” adaptive time machine” should be just as likely to go long or short given strategy under the right circumstances. In this example we will focus on a short term strategy algorithm to keep things simple, but our testing reveals that it is effective using intermediate and long-term strategies as well.

To start our little experiment, we will look for runs within the last 5 trading days, which reduces bias in the machine so that it is not “discovering” daily follow through -a well-documented effect in the literature.   When you use runs over the last 5 trading days the best entry/exits change very frequently over time, forcing the learning algorithm to adapt to be able to survive new environments. The algorithm is able to go long or short after a series of “n” down or up runs and exit after a series of “m” up or down runs. Thus there are 50 possible different combinations of entries and exits. The algorithm  investigates all of the possible strategies and determines: 1) whether a strategy should be traded either long or short 2) whether a strategy is statistically significant from zero 3) whether a strategy beats buy and hold 4) the “optimal” strategy

To be Continued!

3 Comments leave one →
  1. Craig permalink
    November 7, 2009 4:49 pm

    Am I the only one wondering why there is a auto-generated link relating to ‘bagging and boosting’?

Trackbacks

  1. The Adaptive Time Machine: The Importance of Statistical Filters « CSS Analytics
  2. Trade Using Market Regimes? « Quantivity

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: