If I Knew Then What I Know Now
I love cars (actually I find engineering in general fascinating). When I was about eight I was watching a Formula One race and the commentators mentioned that the current car was a big improvement over that of the year before. I remember vividly that this struck me as odd. If the same people made the improved car, why didn’t they just make that version first? Whether this thought was unusually astute or just stupid isn’t the point here. The point is that we can’t improve until we have something to improve on. We don’t just learn from mistakes, we also improve on our successes.
The first issue of “Volatility Trading” came out in 2008 and the much expanded second edition was released in 2013. I still firmly believe in the process I wrote about: find an edge, size appropriately, execute, evaluate, repeat. I also (typos aside) believe in the individual theories about volatility measurement, forecasting, hedging and sizing. But I have drastically changed my mind about the relative importance of some of these things.
In particular, I think model based option trading is nowhere as effective as it was and that situational option trading is now the place to focus.
This classification is quite general. With model driven trading, we create a model that encapsulates the world we are concerned with. We always have a fair value. An example would be in sports gambling where we simulate an entire baseball game in advance, using our estimates of each player’s batting abilities. This would give us win probabilities for every game. Alternatively, situational trading just looks for special cases. What happens when a team lost the previous game by eight runs?
All trades can be classified like this. Card counting is model driven. Ace tracking is situational. Playing ranges in poker is model driven. Reading tells is situational. Value investing is model driven. Momentum investing is situational. Each type of trade has pros and cons but that isn’t my point here.
When it comes to option trading, the idea of forecasting realized volatility and comparing this to implied volatility is model driven. Collecting the variance premium is situational. And over the last 10 or so years, the first approach has become far less profitable. In 2007 this was a nice way to make money. Now it is a grind that only just beats costs. But the variance premium is still solid, and because of the solid economic and psychological reasons for it, I am confident it will continue.
So that is what I have most changed my mind about.