This morning I saw a thread about filters which quickly turned into a flame of mathmatical methods used in finance. So I wanted to post my latest idea to solve a problem I have. This has to do with guessing where implied volatility will open the day after an earnings announcment or an event. A quick run down for those unfamiliar; a way to play to an earnings announcment is to sell front month and buy a back month against it. Edge comes from difference in the IVs, so that this only works when there is a significant difference between them. The day after an event the front month IV collapses and more than likely the outer month falls as well. This is profitable when former falls more than the latter even though you may be net long vega. Of course, a large move after the event may wipe out the profit and could cause a significant loss. The problem is the estimation of where the IVs will open the next day. If it is a stock I've traded before, say CSCO, I have a pretty good idea. However, I want to find a better way than I'm using now for stock I'm unfamiliar with. Right now, I just go to Ivolatility and look at their historical IV chart(the quick way) or I do a HV calculation and look back at where they open the day after earnings. My thought on how to come up with a good estimate is to filter out the jumps and then use the new IV as my next day number. The process would go something like this: remove returns larger than 3 standard deviations from the mean return and replace it with the mean return, recalculate the STD to see if it is significantly different form the first. The process is repeated until the STDs converge. The remaining volaility number is an estimate of where the IV will be the next morning in the front months. I used this in my time series class last semester, but I haven't applied it to anything yet as I didn't figure this out until after earnings season was over. So, I was wondering what people thought about this or if anybody has tried something similar. I'll have to get historical data to test it, but it seems like a quick and easy way to get a number to go by.
Here is the paper from where I got the idea. Note, it talks about mean reversion mostly, but thats not what I'm talking about(I did use mean reversion in my class project). I'm interest the jump process and incorporating that into my trading. Also, I could estimate the expected jump post event using this methodology as well. The part I'm talking about is on page 41(or 43 in Acrobat Reader).