22 December 2006

Another Look At Historical Volatilty

Adam Warner said something yesterday that has re-ignited a thought process I've had bubbling away in the back of my head for some time.
Historical volatility is not a perfect proxy for the volatility of a stock. It factors in the daily range but DOES NOT tell you how it got there.

In other words, a $50 stock with a 50 cent range that doesn't fluctuate at all intraday would spit out the same historical volatility as a similar price and range stock that has many intraday moves up and back within the range. But clearly the second stock affords better trading opportunities.

So when I use historical volatility as below to guestimate the ability to make money flipping a stock, it is just an approximation.
Historical volatility a.k.a. statistical volatility is after all, a theoretical statistical measure and says little about the direction or trendiness of price movements. As such, it may not accurately reflect the the time period it is actually measuring.

For instance, a stock can put on a great move over the period of 20 days, yet statistical volatility can be at lows. We need look no further than the S&P500 in the last few months

This can have implications for option sellers trying to select appropriate strikes. Of course the big problem is not being able to see in the future, as short strike smashing trends can develop from absolutely nowhere. Condor traders on the S&P are getting crunched, yet statistical volatility is suggesting that a condor ws a good trade.

This has all got me thinking about a more appropriate measure of volatility that can account for trendiness of price movement. But will it be useful?

Just something to think about as I zone out from the familial bickering that is inevitable during Christmas dinner.

Any thoughts?

1 comment:

Anonymous said...

there is a whole lot of googlable stuff on ''realized volatility'' or ''realized variance'' which amounts to looking at 1,2,5 minute squared price changes

there is a similar idea of just looking at the sum of the absolute tick for tick price changes across the day

but a lot of this ''intraday movement'' stuff can be aproximated by using parkinsons high/low/close volatility estimator, especially useful if you are looking to reduce the computational intesity of many constructing long timeseries of implied volatilities