Hello All,
I’ve decided to build a market timing signal into my system.
To start, I have decided on two criteria for the research process:
-
Must make maximum use of native P123 capabilities. The use of Excel and a scripting language with stats/machine-learning libraries are also acceptable.
-
Must make minimal use of back-testing.
Given these constraints, what are your thoughts on research best practices?
My initial idea is to build a variance-minimized regression for the n-period logarithmic change of an index/ETF (or the spread of indices/ETFs). My reasoning here is to avoid doing manual optimizations and to be able to use the resulting signal in Kelly betting framework. But that is neither here nor there. I am far more interested in your thoughts on where to begin as far as variable selection (both in terms of where to look for alpha and how best to evaluate significance and dependence), optimal regressions types, and how best to implement the signal within a broader strategy.
Thx!
//dpa