I am working on using AI to build models to decide whether or not you
should be long/short/out of the market. I am playing with logistic
models and neural networks at this time, but there are a number of other
tree based or Bayesian models I could try in due course.
However, I was just wondering if anyone had any comments on which data
you are best using to build your models on. Say one is interested in
the SPI (or SPY): are you better using the index data even though you
will end up dealing in the future or some other derivative? What I have
in mind is that these things don't move in lock-step so there may be
different information in the data concerning the index and that
concerning the future of CFD, etc.
If anyone is interested in making a comment I would love to hear it.
Stephen Choularton Ph.D., FIoD
Hi Stephen, I think it actually depends on the purpose of your analysis. If I am into fair pricing and want to look the strength of deviation in the traded price of some derivative as compared to that fair price then I would go for modelling the underlying. Also for Risk management I would rather model the underlying. However if my goal is just to understand the market moves (hence prediction) for some derivative then I think directly modelling those historical derivative prices would be better as in actual trading lot of other noises generally involved in the traded price. However in this approach you may occasionally find less data points especially near to the inception of that derivative.