The Predictive Power of VIX

So, I wanted to look a little deeper into the question of whether CBOE Volatility Index (VIX) outperformance vs. expectations is at all predictive of future market performance. My short answer -- from anecdotal experience and from the post I had last week on the topic -- was "no." My longer answer after looking at it in greater depth is still no, but I'd leave open the possibility that different methodology might lead to a different conclusion.

Let me explain how I went about it. I defined VIX expectations in an adjustable linear fashion. In other words, X% move in the SPDR S&P 500 ETF Trust (SPY) should lead to Y% move in VIX. I'll call the multiplier Z. But I keep Z a variable; specifically, it's the 20-day rolling average mutliplier. For example, right now the 20-day linear multiplier is about negative 6. If SPY lifts 1%, it suggests VIX should drop 6%. But let's say VIX only drops 3%. I'd score that as a +3% outperformance for VIX.

Over time, that negative 6x multiplier varies. I always use the current multiplier to put VIX in context. I also chose to use addition and subtraction to define VIX overperformance and underperformance. There's probably a good case to use standard deviation instead. My concern was it would overstate VIX deviations on tiny market moves.

Yada yada yada. I came up a "VIX vs. expected VIX" number for every day going back to the start of SPY in 1993, and then correlated it to SPY performance going one month forward from each day, just as a basic way to see if there was any sort of relationship.

And the answer was -- well, I gave you the answer already. The correlation is about negative 0.015, or about as random and uncorrelated as possible. If I filter it down to only the biggest 10% of VIX deviations, the results don't change all that much. The correlation goes all the way up to negative 0.049. VIX overperformance is ever-so-slightly bullish, but it's not statistically significant.

But alas, there's an oddity that's not particular surprising. Many of the days of VIX outperformance were days that saw big up moves in the market. The top four days are March 8, 1993 (SPY up 2.23%), Aug. 26, 2015 (SPY up 3.86%), Oct. 28, 1997 (SPY up 5.77%), and May 5, 1997 (SPY up 2.37%). What's really happening on those days is that VIX is "outperforming" by simply not dropping as much as the market pop would suggest. And that makes some sense. Remember, it's a volatility index, not a contra-market index. These are days of high volatility and likely many a short squeeze. There's some big downside days thrown in, as well, which is probably more of a case of VIX going "exponential" on very ugly days.

So I ran the numbers again, but deleted days where SPY rose or dropped 2% or greater. That lopped almost 10% of all trading days out, which was more than I expected, honestly.

And the results? Still scant correlation. It measures negative 0.0103. Even if I just look at the 100 biggest VIX outperformance days on sessions with SPY moving under 2%, correlation goes all the way up to ... negative 0.0779.

Long story short, I'm not looking further into this for now. Lots of people have opinions on what VIX "should" do on a given day or week or month. They define "should" very subjectively. I attempted to at least put some objective context to it, and came up with really nothing to suggest contextualizing VIX to its expectations will predict much in the way of future market moves.

Disclaimer: Mr. Warner's opinions expressed above do not necessarily represent the views of Schaeffer's Investment Research.