What is the state of modeling at top options market making firms?

[Incster]

I've been working on a model, and I don't have much knowledge about what the state of theo modeling is at various firms to know whether or not it's useful. I have some connections at a few top-15 prop firms (by volume) to know how it stacks up against them, but these firms aren't ones like SIG or CTC who're known for their top notch models.

The firms I'm familiar with don't really use a model per se at all. They usually stick to modifying components of a black-scholes style pricer to hack around the deficiencies of the black-scholes model. I'm not really interesting in knowing how good these types of models are, but instead the types of models like Heston that represent a full description of a stochastic process. Can anyone here shed some light into how good these models are at top trading firms? For example, do any firms run a model that can run theos inside the BBO in SPX for 99% of the options? If not, what is the best they can do?

I'm asking because I'm going to make a pitch to some people about buying/using/investing in my model at some point, and I'd like to know if I have something that is at all interesting to the players in these markets. My model can price inside the NBBO in 90% of options, and inside a "double width" NBBO 99% of the time while running in 10ms, but I really have no clue if this would be useful or impressive. Any insights would be welcome!

 

[CrossGamma]

Look at Vola Dynamics LLC - they are former OMM quants. Their library is used by some 2nd tier firms and the general approach (Black-Scholes++) is similar to all places I've heard of or seen so far.

 

[CrossGamma]

In OMM it's not about complex pricing dynamics like stochastic volatility and compound Poisson jumps but about models that are flexible enough to fit the market and have parameters that are intuitive for the traders to interact with. More complex dynamics can be interesting for risk management as they imply correlation structures between instruments, but I've typically seen more ad-hoc (e.g. parameterized) or data-driven approaches to this as well. The devil is often in getting all the details right, especially for American exercise. This includes things like cash dividends, event jumps, rate compounding, settlement effects, ...

 

[Incster]

Look at Vola Dynamics LLC - they are former OMM quants. Their library is used by some 2nd tier firms and the general approach (Black-Scholes++) is similar to all places I've heard of or seen so far.

Thanks for the link. Is it fair to assume that their models C8, C10 and C12 have 8, 10 and 12 parameters respectively? If I had to really guess, C is "cubic" and the number is number of nodes.

 

[Incster]

In OMM it's not about complex pricing dynamics like stochastic volatility and compound Poisson jumps but about models that are flexible enough to fit the market and have parameters that are intuitive for the traders to interact with. More complex dynamics can be interesting for risk management as they imply correlation structures between instruments, but I've typically seen more ad-hoc (e.g. parameterized) or data-driven approaches to this as well. The devil is often in getting all the details right, especially for American exercise. This includes things like cash dividends, event jumps, rate compounding, settlement effects, ...

So, if I had a model that was similarly intuitive, flexible and fast, but also actually a model, this would hypothetically be interesting?

 

[CrossGamma]

Thanks for the link. Is it fair to assume that their models C8, C10 and C12 have 8, 10 and 12 parameters respectively? If I had to really guess, C is "cubic" and the number is number of nodes.

Yes, I think the numbers indicate the number of parameters. The actual parametrization is only available to their customers. I don’t think it’s cubic since they claim absence of arbitrage. Some of their curves are based on extensions of SSVI.

 

[CrossGamma]

So, if I had a model that was similarly intuitive, flexible and fast, but also actually a model, this would hypothetically be interesting?

Maybe. 90% within NBBO doesn’t sound amazing for OMM and it’s not clear to me yet what the selling point of your model is beyond being a model. I have a few more detailed follow-ups and will DM you.

 

[vanillaice]

In OMM it's not about complex pricing dynamics like stochastic volatility and compound Poisson jumps but about models that are flexible enough to fit the market and have parameters that are intuitive for the traders to interact with. More complex dynamics can be interesting for risk management as they imply correlation structures between instruments, but I've typically seen more ad-hoc (e.g. parameterized) or data-driven approaches to this as well. The devil is often in getting all the details right, especially for American exercise. This includes things like cash dividends, event jumps, rate compounding, settlement effects, ...

Hi CrossGamma, can you elaborate on the data driven approach? especially in event volatility modeling distinct from BS basevol