Automated Options Market Making

Jurassic · Post by **Jurassic** » Sat Sep 08, 2018 7:19 pm

@loltrading are you saying that as delta move fast you sweep stale orders in spot or vol?

also, how is it a tradeoff between vol mols and infra?

rickyvic · Post by **rickyvic** » Tue Dec 04, 2018 9:30 pm

I read this too.... Quite some time ago.

It should work, I think the point is not looking at greeks which are model dependant.... Assuming that they are active and liquid enough....

bullero · Post by **bullero** » Tue Dec 04, 2018 11:04 pm

My current state of understanding is that both methods exist - at least in the stories.

But just out of curiosity, EL could you elaborate a bit what you mean (exactly or in a vague sense) when you say souped-up vol model? I am not positive if this has any relevance but few weeks ago I was just browsing that side of the internet where you prefer to have your tinfoil hat and found a post from this one ex-HFT trader who linked some semi-random (but based on the appearance of the LaTeX seemingly legit) publication where these authors were basically doing some vol calibration using NNs. Basically the feeling that I was able to decode from the comment in that post was that the some OMMs were doing something similar.

I have to say that I am not in the position to comment this approach from a theoretical point of view - if the approach makes sense or not. Now, lets us assume that you would do that. What did I gain from this exercise? I find it hard to understand what the 'beef' would be here. My naive estimate would be that doing all the matrix multiplications and mapping intermediate nodes using non-linear activation functions all the way to the result would be at least as expensive in time as naive Black-Scholes, maybe many times more expensive (?).

So the gain here should be higher than the cost I pay for waiting my computations. I could always just go with naive pre-cached BS. Am I missing something?

nikol · Post by **nikol** » Wed Dec 05, 2018 12:40 am

EL: "just use order book dynamics to set quotes"

During recent BTC collapse I managed to increase initial investment by buying at bid and selling at ask deep OTM june BTC calls. Notional was little, but anyway I was surprised. On the move to automate this trading. I see it can be done using model-independent tricks and straight application of ANN to price different strikes. Maybe I am wrong, but so curious, that it drives me forward.

ronin · Post by **ronin** » Wed Dec 05, 2018 10:26 am

> But what happens if you throw out the vol models completely, and just use order book dynamics to set quotes?

You couldn't manage the risks.

The problem is that you have next to no liquidity in any single option contract. You get one, maybe two trades per day. Maybe zero. If you got in, you probably won't get out - not by eod, and not cheaply.

So you offset your risks where you can and where it is cheap.

If you are thinking equities, it is more similar to pair trading than single stock market making.

Having said that, my understanding of the current options market making space is that 99% of pnl comes from taking priority on all levels as soon as the book opens, and maybe 1% from filling stale quotes when the market moves. Which is roughly what the pnl looks like in equities market making anyway. So on the grand scale of things, it is all the same thing. But the details are different.

nikol · Post by **nikol** » Wed Dec 05, 2018 11:27 am

"If you got in, you probably won't get out - not by eod, and not cheaply"

why? you can almost always zero (net) exposure without closing it at reasonable cost. keep it till expiration or till door opens again.

Jurassic · Post by **Jurassic** » Wed Dec 05, 2018 12:02 pm

> Having said that, my understanding of the current options market making space is that 99% of pnl comes from taking priority on all levels as soon as the book opens, and maybe 1% from filling stale quotes when the market moves. Which is roughly what the pnl looks like in equities market making anyway. So on the grand scale of things, it is all the same thing. But the details are different.

Why is being the first in the queue key in market making (99% of pnl)?

eeng · Post by **eeng** » Wed Dec 05, 2018 12:33 pm

Alpha wise this is a very leveled playing field (just read Baird book or even easier, simply plug your vol estimator of choice into a BS calculator and update your quotes) so the only point where you make a difference is in arriving soon than everyone else doing the same stuff.

bullero · Post by **bullero** » Wed Dec 05, 2018 12:36 pm

@Jurassic, If I paint the levels at the beginning of the day and assume that each level, on average, receive fills with same lambda* my queue position rises evenly until I am the top of the queue on both sides. If I receive fill on my bid/ask I still have high probability of getting out making the spread before move.

*Of course this is not realistic.

ronin · Post by **ronin** » Wed Dec 05, 2018 12:45 pm

Guys, honestly...

>why? you can almost always zero (net) exposure without closing it at reasonable cost. keep it till expiration or till door opens again.

Well, that is kind of my point. You manage the risks by netting them out over the entire surface. You can't manage the risks on the level of a single contract.

>Why is being the first in the queue key in market making (99% of pnl)?

It's called adverse selection. This was discussed many times in this forum and elsewhere - look it up. Basically, if you are last in the queue, after you are filled the best price is at the next level and you lost money on your fill. Having people behind you in the queue gives you an opportunity to get out of your fill before the price moves to the next level.