Posted By: HS.   Date: Sun Mar 29, 2009

Title: Automatically adapting model versions ?

Message:
  I feel that the switch to version 25 was quite reactive, as opposed to proactive, which contradicts the aim of the MarketTrak tool: to attempt to predict markets ahead of time. The result is that the old model was short/cash for the past ~3 weeks, when the markets jumped up ~10%.
 
  I realize v25 would have eliminated this lost opportunity over the last 3 weeks, however in the future this will happen again and again - as we're waiting for the markets to evolve, and then after that they evolved and shifted, we will be applying another trading version that would have better performed (and this is key) in HINDSIGHT.
 
  Isn't there a way to develop a model that adapts to market situations automatically and ahead of time?? Otherwise, we would be consistently using the wrong model/strategy in real-time, and correcting it in hindsight.
 

Response:
  First let me say that version 21 did a very good job. It beat a buy and hold strategy by 20 percent in the three months that we used it. My expectation is that this model will continue to do well.

 Version 25 is based on a different technology. The input data variables are the same but the processing is different. I expect this version to beat version 21 in making money for us in the coming months.

 To answer your question, our models are designed to adapt to changing market conditions. However, there is a limit to how many different conditions a model can classify and still have a predictive capability. Said another way, I don't think that you can build a model that will classify a market state far from any it has seen in the past, although I don't have proof of this.