That's a really good question.

A caveat is that large price movements don't necessarily are bad for LPs. In a sandwich attack you'd get two large price moves but they almost cancel out. And >10% of all volume in the 0.05% WETH/USDC pool is sandwich attacks. In another study someone showed/guessed that the largest swaps come from hackers/exploiters who don't care about slippage, and the swaps are profitable for LPs because the price eventually reverts.

The bottom line is that we need more empirical data on the best way how to model the price evolution. Panoptic did study this a bit, see the thread and my discussion with Guillame Lambert in comments. If we don't want to look at the individual tx but multi-block price patters then GBM in fact isn't that bad.

https://twitter.com/Panoptic_xyz/status/1652463028674387968?s=20