Posts
VIX, the volatility index is meant to tracks stock market volatility. I have written before that in recent years it is more of a "bearishness" indicator. This is proven by looking at the correlations between VIX and SPX or the DJIA. The VIX and the SPX are highly (negatively) correlated. I December, the correlation was -0.92. In Q4 2009, it was -0.85. Please see our numerous studies in correlations.
Today I performed a study to see whether the VIX could be used as a predictor of price movements on an intra-day basis. For this, I compiled a list of intra-day historical prices for today and looked at the correlations.
There are over 300 data points, practically one price per minute. Because VIX does not actually trade (like a stock), and only trades options (and European style at that), I used the VXX, which is an ETF that tracks the VIX, and does trade as a stock.
The data is available as a Google document.
The correlation on a minute by minute basis is -0.41. The correlation is negative, which is correct as if the VIX goes up the SPX goes down, but that is about all that can be said. 0.41 is a very bad correlation, it is closer to non-correlation (value of zero).
So the conclusions from the above is that VIX cannot be used as an intra-day predictor.
We then shifted the prices in time, for one series at a time. The column Vix Ahead indicates whether the correlation of the VIX being from 1 minute ahead to 10 minutes ahead. The more it is shifted, the worse the correlations gets.
Similarly, we repeated, this time with the SPX being ahead by the same amounts of time.
The chart below shows the correlations for all cases.
The closer it gets to zero, the worse it is.
Clearly, the VIX cannot be used as a predictor of prices for the given above intra-day time frames.
Since they are fully correlated in the longer term, looking at SPX is a lot easier. VIX is also a horrible vehicle to trade options on as often in-the-money options trade for less then intrinsic value.
Perhaps it could be used on seconds-advance basis (i.e, on an intra-minute basis). If it did, maybe only Goldman Sachs and the likes will have the tools to act as fast!
Caveats: VXX was used instead of VIX; The last traded price in each trading minute was used. We only looked at today's data, from 9:30:00 to 14:52:00
4 comments:
However, VIX vs. VXX themselves are not correlated well, so the usage in this case in questionable.
http://bigcharts.marketwatch.com/charts/big.chart?symb=vix&compidx=aaaaa%3A0&comp=vxx&ma=0&maval=10&uf=0&lf=4194304&lf2=8&lf3=4&type=64&size=2&state=8&sid=1704273&style=320&time=18&freq=7&nosettings=1&rand=5304&mocktick=1
Thanks for the comment, that is an excellent point. The correlation between VXX and VIX in Q4 was 0.81, in Nov-Dec09 was 0.87, and in Dec was 0.82. So that indicates they are correlated (as they should, and we'd hope!). The correlation is not perfect however.
Correlation should not be verified visually, it needs to be computed from data. It could be computed from a chart, which amounts in the end to using the data from the chart. The right thing to do would be to use the VIX intra-minute data, which unfortunately my software cannot obtain as VIX is not a real stock that trades. So, yes, there are caveats in this approach and it is not exhaustive by any means. Thank you!
Thanks for the work and for hte explaination. If we can test VIX data directly (not VXX) that would be more useful, as VXX is mixed with back months pricing, like UNG which is far different from gas futures.
Post a Comment