@Larry I don't understand why future options pricing is necessarily different from the past. Certainly back-testing does not guarantee future results, but if something performs badly in back test there it would seem very unwise to use the strategy in the future. In the same way, if the strategy performs well in back-test you can have some confidence about its performance in the future.

I had some experience with this recently. I had a theory that option implied volatility was a predictor for future returns (see http://www.bearcave.com/finance/options_iv/index.html). My idea was based on the speculation that higher volatility and option demand meant that the "market" had a view about the future that was reflected in the option volatility. As it turned out, at least for the data I was using, this theory appears to be wrong. If I had not tested the theory via back-testing, but traded real money on the basis of the model I probably would not have done well.

Of course one might say that given superior wisdom and experience such theories would not be entertained. Since I don't have such wisdom, I'm going to keep back-testing.

Back testing as commonly discussed is fraught with problems. Options are absolutely and categorically impossible. Past data offers nothing useful since there is no possibility for setting a program that yields executable transactions. Thus the entire exercise is corrupted beyond redemption. Back testing in the traditional provokes questions and often exposes shortcomings of any system. The flaws in traditional and commonly practiced back testing forced the development of a program that captures executable transactions. However, the program can not be applied to options or derivative securities. Regarding implied volatility, it was rejected here in favor of measuring actual volatility and price behavior. Risk tails were established and hedges or other transactions set against the tails. In this manner the risk exposure and probability was clearly evident and implied volatility irrelevant. In the hedged or paired transaction program transactions were based upon return and the ability to cover tail risk. Meeting the objective rendered implied volatility irrelevant since the payment for putting on the position was sufficient for the risk. By the way, no one truly possesses superior wisdom and there are no experts. The quest here was to define the purpose and determine the acceptable risk factor. That is all that truly matters.

I actually concur with Ian's perspective. If a theory performs badly when back tested, it is suspect by definition and the theory and hypothesis likely requires revision. My own experience is that most back testing has limited utility since the favorable testing results may be flawed due to the real life execution issue. Concerning implied volatility, I salute your research effort. Truth be known, the entire Greek series (except theta) and implied volatility was discarded here as a data set without much utility or value within the context of system design. In any event, options are absolutely hopeless as to producing any meaningful back testing results. Component pricing is asynchronous albeit by fractions of seconds for many issues during the session and there is the issue of spreads. It was possible to build an observation system embedding executable transactions for "look backs" at individual issues and pooled funds. As for options, it was live or nothing at all.

@Larry Thanks for the interesting post.

I was working from several research papers that had reported implied volatility and various options price factors as predictors. But as you note, it was pretty much a bust. In effect I built a simple factor model and the regression result was terrible from a prediction point of view.

I have never traded options, but for the last two quarters in graduate school I have been studying options and options modeling (I'm finishing a Monte Carlo class now). My view of options, admittedly academic, is that option prices are a function of the underlying behavior and supply and demand in the market. Perhaps you point is that you cannot predict supply and demand, which governs the actual price of an option in the market. But in theory this price will not be that far from price you can arrive at through simulation (since American options cannot be analytically priced).

While I don't have your experience, I remain, for now, in the back test camp. I would be reluctant to trade any model that I could not back test.

Ian Kaplan wrote "My view of options, admittedly academic, is that option prices are a function of the underlying behavior and supply and demand in the market."

To which "supply and demand" and "market" are you referring? Remember that options are created out of thin air -- at the instant that bid and ask meet. ("Yo! Open interest!") Demand is not so hard to model, but Supply? Absent the omnipresent commissions, all costs are opportunity costs (of foregone margin-consuming alternatives).

Please enjoy yourself. Play this game right, and you'll enjoy it for life.