macrocephalopod

This is pretty much my point. Not that all the dealer hedging stuff doesn’t work at all. Just that it’s often presented in a way which massively overstates how significant it is (eg chart above, or plots of dealer gamma exposure vs return). These charts show a big effect but

98%+ of the effect you see in those charts is straightforward contemporaneous mechanical relationships in the underlier => options direction. Measuring the strength of the predictive options => underlier relationship (what you actually care about) is hardly ever discussed.

I would split the problem in 2 parts - 1.estimating dealer(delta hedged) positioning and 2. predicting performance of options given the positioning. 2. is pretty pretty good grounding, ad can even be modelled (Loepr). 1. is super hard

1. is a fool's errand. Statistics works a lot better here in my biased opinion.

What makes you say that? I haven’t implemented the method in the recent-ish @HauVolatility report but it seems a priori plausible, and the only good test of whether the effect everyone says exists really does exist.

Confess need to re-read Hau's work, but when I pinged, my understanding is it mostly looks to understand whether OI is bought/sold, no? Not the actual dealer net Greek exposure.

Tracking inside the OI--probabilistically assigning buys / sells / tied orders / etc-- to see where residual risk lies on a market maker's book who chooses to hedge delta is one in the same as tracking net greek exposure (just calc the greeks for the inventory assignments!).

Well, it’s the same if you assume spread-payers don’t hedge deltas and spread-receivers do.