Serving the Quantitative Finance Community

 
User avatar
MartinGale7
Topic Author
Posts: 70
Joined: April 1st, 2011, 7:42 am

Popular Methods for Non-ARCH Forecasting of Volatility and Option Prices?

September 7th, 2018, 1:32 pm

In the CQF, we learned about GARCH etc. I assume this is somewhat 'entry level' and that there are a wide variety of other mechanisms for pricing volatility and options based (ONLY) on historical price data.

I have been doing quite a bit of investigation on using machine learning to price puts and calls based only on past data. This seems to hold up well.

I am curious to know what other families of solutions quants have pursued in this area? Obviously I'm not asking for anything proprietary, but more a feel as to where the research is being focused and what types of algorithms are currently popular (beyond the ARCHs).
 
User avatar
Alan
Posts: 2958
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: Popular Methods for Non-ARCH Forecasting of Volatility and Option Prices?

September 7th, 2018, 5:41 pm

Two comments.

1. Using ONLY historical data is sub-optimal in vol prediction (i.e., a mistake) given the existence of forward looking data embedded in options, VIXes, etc. 

(Source: "Sidebar: Volatility prediction and prediction in general", pg 257 of "Option Valuation under Stochastic Volatility II").

2. Having said that, you might want to consider the Hidden Markov Model (HMM) techniques discussed in the same cited chapter. Unlike GARCH, in many models volatility is a latent (unobserved) variable. HMM methods give tractable ways to approach the problems of smoothing, filtering, and prediction with latent states.
 
User avatar
MartinGale7
Topic Author
Posts: 70
Joined: April 1st, 2011, 7:42 am

Re: Popular Methods for Non-ARCH Forecasting of Volatility and Option Prices?

September 8th, 2018, 11:59 am

Thank you. There is very important information which isn't included in the historical price, such as upcoming Non-Farm Payrolls, Rate Decisions, Dividend Payments and Company Reporting. From that perspective, a model based on past data will always be missing a trick.

I'll look up the book.
 
User avatar
Alan
Posts: 2958
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: Popular Methods for Non-ARCH Forecasting of Volatility and Option Prices?

September 8th, 2018, 2:29 pm

Thanks on the book.

Yes, you give excellent examples of things that will likely be missing in a GARCH analysis. Of course, in principle, one can add seasonals when a historical time series contains repetitive things. But, to add to the list, sometimes the (volatility-inducing) scheduled future event is *not* a seasonal. For example, I recall a shift -- actually a 'turn-up' -- in the upside SPX smirk  just prior to the 2003 coalition invasion of Iraq.