Quant Cofounder for Options Arbitrage

I've got a lot of experience but bottom line is I'm not making money yet.

EOD prices are just fine for me at the moment. That's one snapshot per underlier per day but with 10-20 strikes and 3-4 maturities and some 5000 underliers, that's a lot of data to go through in backtests. And in a live scenario I'm anyways limited by Interactive's Brokers 100 simultaneous marketdata subscriptions (or up to 1000 of I pay for them), even with a round-robin rotation algorithm I can't cover that many quotes.

About the option pricing models, they aren't worth much... until they're worth a lot

 

Any reason why you are limiting yourself to options strategies? What about intra-day directional futures strategies? If the end goal is a good return on capital, why not go after opportunities with lower cost of acquisition (cost as in obtaining and cleaning data, research, etc.).

 

WTF is this abortion of a thread? Reproductible? My take is that mommy shouldn't have.

 

Same! If you worked on strats and "quant" as the OP states you just haven't found the multiples of combinations that can work with high probability of success, now you want someone else to bring it to you for only a generous 50% split and all of $500 a month? And it has only been 11 or 12 years?

 

Experience. Took a long time to acquire the knowledge and I prefer to keep the direction.

 

Actually this looks like a good conversation to have, as we’re all working on similar problems. So let me explain why models seem worthless to me:
In terms of my own arbitrage, I had to keep placing 10,000 orders (maximum allowed by IB) on two accounts, just to observe which orders may get fills and at what prices. In such case models didn’t matter as I could use basic math (only sometimes basic BS) to validate that I’m not going to lose money, as arbitrage had to work somewhat similarly to buying debit spreads or butterflies at credit where you wouldn’t need any models. I’d use more precise model (still BS) only when taking some risk.
While the issue with any type of truly riskless arbitrage is that someone else instantly loses money, usually market makers, and they quickly analyze why/how they’re losing, and then plug leaks in their system. I was still able to catch some arbitraged deals last week, worth around $2k but that’s not much unless I could do this every day (and even then not scalable, unless possibly I'd gain an additional advantage by becoming an actual market maker). While I could observe how those leaks stopped couple days later.
But if you do something longer term that isn’t instantly profitable (and therefore likely not riskless arbitrage) then you have a chance of getting away with it.
Though even then, once someone loses money then they’ll adjust their models too. Also, if you do anything at scale then your own trades will affect options pricing, so if you buy some options and sell others then the ones you’ll start buying will quickly get more expensive, and vice-versa for the ones you’re selling. So the more you want to scale the more risk you’d need to take, while affecting your own and everyone else's models, and basically trade like everyone else.

However, I never understood how people may use various models. I saw some gigs where clients wanted to use different option/vol models to project future options prices, but I didn’t understand how that could work.
For example if I use an exotic model to predict that a combo should be priced $X at time A, then my model won’t really matter if everyone else will simply decide that the same combo should cost $Y at time A. Even just for kicks, if I imagine that the whole world will conspire and agree that from now on everyone uses a model that prices all options exactly $10 at time A. Then if at time A everything worked as planned, I may come in and say “hey, I’ve used a better model and this option is worth $20”. But everyone else will say “so what, our models say $10 and that’s how much it is, go play somewhere else.” This actually already happens as I see some people arguing how much their flys should be worth while nobody else cares because they’re worth whatever the market decides.
That’s why all models are worthless to me, except for the actual model used in the market. And I would even love to know what is the actual model used by the market at the current time and that’s difficult enough. I can’t even buy/sell too many combos with better than $0.05-$0.20 slippage, unless I place tons of actual orders and see what I can “catch”.
But I simply don't know if I'm not missing something by not looking into using various models.

Other than that I can consistently make money “arbitraging” volatility smile (not riskless), but margin requirements are too high to make much more than 5%-10% per year, while taking more risk than I’d like.

So just my conclusion would be to change focus from doing something too complex and trying to crack the market, while going back to basics. For example I heard quite a few times about pro traders making 20%-40% per year using iron condors like in the video I posted earlier. I haven’t focused on such basic strats yet, but it seems possible, maybe with some aid of TA. Have you considered that and would 30%/year be enough for you?

 

i'd do it.

 

It's "volatility arbitrage" I'm talking about. You can replace "volatility" with "probability" (or more generically with "market dynamics") and it's more clear what the technique is about.

If the market dynamics is a normal distribution of constant daily variance, that's a random process so there's no way to predict direction or the exact prices each day (TA is useless), but you have a descriptive statistics of those prices expressed in the so called "volatility", like you have a description of the probability of a coin falling head or tail in the 50% probability. So in this case there's an "average value" of the option prices and in the long run (or large number of trades) you get close to that average. If the bid-ask spread is small and the market is misinterpreting the volatility (say market says the coin probability is 50% and bid/ask trades at 49.5% / 50.5% but you notice some wear on a side of the coin and compute that the probability is actually 51%) then you are right and they are wrong and you can arbitrage that probability / volatility difference away.

It's still "directional" in that you need to guess something (the volatility / the model) and it's only arbitrage in the statistical way and if you're right in your guess. If today were 1950 and you had the Black-Scholes model up your sleeve, it'd be more than enough to make tons of money. The problem today is that you need far more precise models than Black-Scholes and it gets increasingly as hard to find them as say, curing cancer.

 

Assuming you could "crack the code", what sort of returns would justify the effort?

 

OK i have a strategy for you. Its written in c++ but v simple . It sells 6w 15 delta index puts (dax,dxy,spx, whatever) low strikes ,very stable high vol. We use an ITM call instead of OTM put though. It delta hedges the position to expiry by selling ITM 75 delta calls or puts (instead of forwards, same expiry, solves for hedge notional) when it hits predefined delta limits. Its very safe if the world explodes the call goes to 0 so you don't need to worry about short "vol" style drawdowns. More volatile spot paths are marginally more profitable because of the extra theta from the delta hedges. The edge in the strategy is the index vol is expensive in this region, using a call not a put so safe in size and then the improve using the short in in the money options to delta hedge which doubles the theta you earn instead of forwards. v simple to code, blacksholes everything and only requires one option chain. Do you have money though i don't?