This might seem very elementary or even silly, but all comments are much appreciated. As science major I do not have deep technical knowledge in statistics, so any guidance is much appreciated.

Say we have this hypothetical problem:

Say I would like to come up with a model to predict returns of some financial instrument $I$ for a next day. I have the return values $R_n$ of the $I$ for the past $n$ days.

Additionally for each day $d$ I have some independent data $P_d$ (say $m$ numbers) that could be used to predict return value $R_d$ of today, but I have no idea what this data means and where it came from.

Are there any standard ways of finding how data $P_d$ relates to $R_d$, for example some kind of software package that loops through many possible models and check how well they fit?

Or what would be a way to tackle this type of problems, because it seems to me there could be infinite number of ways to use $P_d$ to predict $R_d$?


Actually, I was thinking of posting a similar question, though coming from a different angle.

Firstly, your data $R_d$ and $P_d$ form a time series with potential interactions and you might end up using a vector autoregreesion methon (as in https://www.otexts.org/fpp/9/2)

In that case, you have to analyse all the auto-correlations and so on. Additionally, with a high dimensional $P_d$ you will get time series in a very high dimensional space.

So far I have no experience in the time-series regime, but this seems not very feasible.

Hence, one could also use a regression task that is fed with your data. You still might so enrich you data $P_d$ with additional features.

An advantage might be to use a nonlinear model, since they might find some structure in the data you don't see directly.

For example a neural network can be used as is pointed out in here or in here

You can also use a random forest as is pointed out here Note that here an optimization is performed to obtain the best parameters for moving averages that contain the data from previous events/time-steps


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.