9 Comments
User's avatar
vehctor's avatar

One thing that is interesting about hummingbot's implementation is that it seems to be modeled for spot markets

HangukQuant's avatar

the difference would be...that balance is non negative? I don't know, I just saw it was abit funky and I did not follow through on the logic actually...

Thierry Henkinet's avatar

I recommend you read the paper from Olivier Guéant on Optimal market making (model B). It’s a more accurate resolution of the same equations. For example, your B (base spread) is just the inverse of kappa and q is the sum of “deltas” the unit order quoted.

HangukQuant's avatar

definitely will! I always seen references around for that but not yet gotten to it. thanks for the reco

Thierry Henkinet's avatar

There are also limits on +/-q, which is critical in how you quote. These limits were absent from AnS.

Thierry Henkinet's avatar

You are welcome. It’s going to fix a couple of your assumptions on q for instance. The take away is very similar though: calibrating 2 parameters. Liquidity being quite simple because you can indeed estimate it visually. The other one is a bit more tricky.

Thierry Henkinet's avatar

and just focus on model “B”, it is the simpler one and the one which is used by professionals.

Thierry Henkinet's avatar

finally, this paper does use the “reservation price” model but just give the optimal spread on the bid and in the ask, but you can compute a reservation price with a bit of maths and both approaches are exactly the same at the end.

Thierry Henkinet's avatar

*not use