Instead of "mean reversion" in the classical sense, you should expect to see negative autocorrelation in the next price move: You would typically see that if it upticks (i.e. midprice goes up on next best bid or best offer level formation), it is more likely to downtick on the next price move. The next midprice change should be negatively autocorrelated.
Many still refer to this as a "mean-everting behavior", so it's really what you want to call it.
I think what you're looking for is literature on how to capture this as an alpha and the truth is that there's no free lunch because the stylized observation here is too simple:
- The most naive reaction to this microstructural observation is to join the best bid after a downtick and vice versa.
- However, most MMs are aware of this, so you should also expect your "desired" fill rate to be low on the best bid after a downtick, and low on the best offer after an uptick.
- Moreover most aggressive participants are aware of this as well, so you should expect your "adverse" fill rate to be high on the best bid after a downtick, and high on the best offer after an uptick.
- This partly shifts the focus to how you can improve your fill rate, which is a hard open problem that there's no good academic literature on: the most obvious follow-up to increase your fill rate is to layer more levels ahead of time, but then this also means your open order risk is higher and the strategy state space becomes much more complex.