Discussion about this post

User's avatar
Overcoming Bias Commenter's avatar

My personal experience is very different than the advice of this post might indicate. For instance, after coming to the asset management business from computer science, I was shocked at the poor statistical inference practices I saw. Everything is based on t-statistics and arbitrary cut-offs for significance. Many models are evaluated and chosen using only "first principles" and econ intuition. And there's an open culture about the fact that this intuitive style that lacks rigor is clearly bad, clearly suboptimal, clearly could be improved with minimal effort, clearly not cost effective. But incentive schemes make it very lucrative for those who can wheel and deal in the low-quality inference jargon, who can data mine for significance but put spin and marketing on it to appear very credible. I've seen many successful modeling techniques from Bayesian inference and machine learning just tossed aside, almost laughably, because how would anyone ever explain the "first principles" or "intuitive" reason why it did not work in some period?

I may only have limited exposure, but my take away has been this: economics seems to be a special case of a signal processing problem, and the modeling motivations that underlie much of the math and computer science literature on signal processing, statistical learning, and Bayesian inference seem to utterly subsume and supersede any of the econ or market first principles that are widely trumped up. But those are not as easy to sell as simplistic stories about overreaction/underreaction, micro to macro aggregation stories, etc., so they collect dust on the shelf despite being more accurate tools with strictly better info about what they seek to model.

Expand full comment
Overcoming Bias Commenter's avatar

The kind of sources that usually feature in American "debates", like the deep sea oil fields in the American part of the Gulf of Mexico.

Expand full comment
35 more comments...

No posts