/biz/ - Business & Finance

>>10420004

scoops, I read that whitepaper. interesting stuff.
things I like
>nonlinear over linear models
this is essentially reducing the rationality in the models by drawing meme curves instead of lines
>careful attention to avoiding forward looking bias
>attention to meta parameters
>frequent retains
though this is a double edged sword
>breaking up by sector
a decent proxy for controlling for macro

things I don't like
>length of tail for training data
>meta parameter choices
they essentially hand tuned these values to maximize their results. also these were both static. much better if they were adaptive. I particularly also don't like 3 years tail for quarterly positions. I think it should be a longer tail but then that's based only on instinct
>no proper incorporation of macro changes
though to be fair this is really hard, but I bet weighting in a couple of indexes could do it
>training set 1983 to 2010
there's a lot of past bias here, and I worry that it doesn't account for the modern explosion of algo trading

>>10257965
Invest in rape. Africa a best sector

Oil finally on a downward breakout

>>10225337
3dpd get out reeeeee
>>10225330
mfw