Trading Signals

I’m putting together a framework to evolve, test, and optimize signals using a genetic algorithm approach. Signals with statistically significant results will be further combined into bayseian networks and fed back into this testing framework.

Determining the conditionality of one signal against another requires insight and guesswork, or evaluating permutations of networks. With a GA optimizer and enough computing power should be able to determine networks that successfully amplify the combination of signals to one that is correct more often than not (or at least is more successful in the profitable situations than the losing ones).

The universe of events and indicators that have potential to be significant is large, as are their parameterizations. Choosing the search space will be a challenge, as is sometimes having access to the required data.

I plan to overlay our tick UI with trading signal indicators indicating a probability weighting when a signal reaches a non-neutral threshold. Should be interesting to visualize the results.



Filed under bayesian networks, genetic algorithms, indicators

2 responses to “Trading Signals

  1. bbc

    I have one confusion, that we all know that market is not static as human decisions step in, the price is no more than the combination/censor of different opinions on a large set of information on every moment. We convert high dimension data into price, volume and time. And from these we get indicators, which are correlated, then we reduce the dimension of the data into one dimension to make a binary decision. I really doubt it will work better than pure empirical decision making. Consistency is another issue, GA might be adaptive to the market condition, but you need to make two assumptions on top of it: market changes gradually/smoothly, your adaptive algorithm adapts to the change much faster than the frequency of your trading. GA is not a fast algorithm for sure.

    I personally prefer Boosting with Weighted Majority Algorithm.

  2. tr8dr

    I wrote this a long time ago 😉 Although I do use GAs and other optimisation techniques to search high-dimensional sets from time to time, I use other machine learning techniques to adapt a strategy through the day.

    A GA does not naturally present a gradual evolution to changes, rather in a brute force manner tries to find an optimum in a specific circumstance. Using a local optimum is, very often, not a good long-run approach. Rather evolving gradually I find works much better.

    So I suspect we agree on this.

Leave a Reply

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

You are commenting using your 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