MarketTrak Question/Comment Message


Posted By: Peter   Date: Tue Jan 26, 2016

Title: Does system learn

Message:
 Does your system learn from past market movements verses the forecast?
 Is it more likely to turn negative at the next down turn?
 
 

Response:
  The answer to your first question is yes. There are two learning or training methods. The first occurs when new neural networks are created. Input data about previous forecasts and what actually happened in the market following the forecasts affects the placement of neurons in the network and the values of their connecting weights. Forecast errors are often corrected in this training process. The second method involves the manner in which the ANO is computed. This is a rather complex calculation but I can say that forecast errors are immediately taken into account by networks previously trained.

 The answer to your second question is I don’t know. It would depend on the market behavior leading up to the decline. If the behavior was similar to previous declines then I would expect the model to forecast the decline correctly. However, if the decline is a result of some shock, like China announces something rather negative after the close, then you can't expect the model to know this and the forecast could be wrong.

Hope this helps,
Rich

Return to message board




 

Copyright © 1996-2017 MarketTrak LLC.  All Rights Reserved.