Really interesting, Antony! My development environment also allows for "drag-and-drop" style factor research, although I don't explicitly do anything with risk yet. A couple questions.
Congrats! These look amazing.
What's your methodology for discovering alpha factors? Do you start with alpha lens? Do you start with an economic thesis or do you just mine the data for correlations?
Ah, I see. Interesting that the "validation" runs all look fantastic compared to what you'd expect for OOS, but then the "test" runs have quite a drop-off, including the pronounced alpha decay.
It's always possible your alpha has been discovered?
Another thing I noticed is that your strategies here tend to have extended periods of "consolidation" and then periods where they "run." Could there be some sort of underlying market regime shifts to explain this? I've used this notebook to discover that one of my strategies was entirely dependent on volatility regime, for example:
Perhaps that notebook could be adapted to asses strategies vs interest rate regime, etc. or whatever hunch you might have.
Then when the model is set to run in the main contest, slippage is turned on and the orders aren't clearing fully --- which is especially problematic for fast signals.
That makes sense! Maybe set rebalance 1-2 hours before close for the version you submit to the contest. The tearsheet will ignore the slippage/commissions dings, but incomplete fills will be an issue.
What do you mean by "fast signals"? Presumably for the Q Fund, no fast signal will work, since it operates on a 1-5 day delay.