Friday, August 07, 2009

VIX % Change Advantage

Soggy bottom in the Qs today after first hour should have warned you we were going up!!!
And now for something completely different. . . . . .

One thing that's caught my attention for a while is the daily % change in the VIX relative to the major indices, including the Dow, SPX and NDX.
I've noticed a tendency for exaggerated VIX moves to foreshadow subsequent short term trends in the indices and here's a simple little study to validate my intuition.
I choose to make the indices the target of the trade rather than the VIX since, by it's very nature, there's considerably more daily noise in the VIX than in the underlying indices.
That format almost makes it easy to test alternate indices performance relative to the VIX.
Simply put, the system sells the SPY if the daily % change in the SPY is greater than the daily % change in the VIX, and then exits the position after a fixed period. Vice versa for the sell side.
.
A 65% percent profitable return may not seem very impressive, but this is a very simple system and easily improved by adding entry filters such as a % change threshold, an MA trend, etc., as well as exits conditions, trailing stops, etc.
Those improvements could also improve the drawdown and consecutive winners/losers ratio. . . 2 of the most important metrics that I consider when evaluating a system.


And, as always, keep in mind that these studies are not meant to be canned solutions but rather jumping off points for your further exploration. That's one of the reasons they're free. The equity curve really begins to take off around trade # 20, which was initiated on 12/27/07 which is reflective of the beginning of the bear paradigm. Were we to rerun the data backtest with this as out start date, the results would be improved considerably without any further code modications.


Below is the TS 2000i code with the fixed bar exits optimized for the SPY. In future posts I'll look at some of the other indices, add a few filters and some other tweaks. So far this simple approach looks very encouraging for both capturing short term gains and creating a probability model for short term direction.

9 comments:

Bill Luby said...

Another interesting piece of work, Bob.

Thanks for posting,

-Bill

Michael S said...

haven't tested specifically on this strategy, but i suspect this might be a factor: http://marketsci.wordpress.com/2009/07/22/a-flaw-in-logic-trading-the-relative-strength-of-assets/

michael

bzbtrader said...

Michael,
As I said in the post . . this should be regarded as a jumping off point.
I respect your efforts to question the validity of asset correlations, but . .
For my own part I have found my 3 finger lead and reverse systems, which utilize relative strength in several high beta components of the Qs to forecast short term Qs momentum, to be extremely reliable on both a daily and 10 minute bar basis.
Without parsing words, I don't consider the VIX to be an asset but rather a reflection of market stability.
The VIXEN studies that I've posted extensively have shown that following the VIX on 1 and 2 minute bars on a Qs chart will produce consistently low risk daytrades and I consider this current post a continuation in the spirit of that research.
I'll investigate your alternate approach further. Maybe you could provide the code language you're using, which would speed up the comparison testing considerably.

Michael S said...

Hello Bob - not a code issue as much as a logic issue - the question is does the strategy perform better with the VIX component or is that extraneous data. For instance, which performs better: SPY % chg less than VIX, or SPY % chg less than zero? In the post I showed that a lot of strategies our there involving daily relative strength of two assets, are in fact really trading just the tendency of one of those assets (in this case, the SPY to reverse some of the previous day's direction). Not saying your study falls under that because I haven't tested it, but just a thought...

michael

Toptick said...

Bob, Michael: Of course I would rather have tested this to make sure, but I think I see what's going on and I'll make the argument. This isn't a case of the relative performance problem, but is a demonstration of daily mean reversion. We know that VIX and SPX are strongly anti-correlated, so much so that when they both move in the same direction it recently has been a strong signal.

Going to the code, if SPX is up, VIX is almost always down, so SPX change is greater than VIX change, and you take the Sell. Conversely, if SPX is down, VIX is likely up, so you take the Buy. You should get nearly the same results by not considering VIX at all.

Thanks to all!

bzbtrader said...

Toptick,
I afraid I'm not ready to throw in the towel just yet on the VIX advantage. Being in my sixth decade and somehwat dim and slow, my main interests are feeding my hummingbirds, http://bzbtraderdownloads.blogspot.com/2009/08/tending-my-flock.html
paddling around in my swimming pool and playing golf so this whole trading thing is a real distraction. Nevertheless, I'm not quite grasping your mean reversion strategy on the SPX. Monday's double 3s post is kind of an attempt in that direction although that is based on higher highs and lower lows to define mean reversion thresholds and to generate trade triggers.
If you just outline the precise rules of the mean reversion trade you are imagining then I'll code it up and post the results while I'm waiting for my and chicken noodle soup and Jell-O lunch and prepare for my afternoon nap.

Toptick said...

I'm certainly not saying it doesn't work, only that it looks to me like it shows mean reversion instead of a VIX effect.

My idea is to compare the code as above with code where the SPX change is compared with 0 instead of compared with the VIX change:

"If PercentChange(Price,Length) > 0 and MarketPosition = 0 then Sell ..."


[BTW, I'm absolutely not trying to give you a hard time -- you are creative and very generous to share your work.]

bzbtrader said...

Toptick,
OK, now I'm about ready to test your idea. What the exit signal?

Toptick said...

I would leave the exit unchanged (hypothesis is that taking comparison to VIX away gives similar results, so apples-to-apples test would be the same exit).

Thx!