Friday, August 1, 2008

Another Large, Low Volume Drop - Massive Edge or Statistical Anomaly?

The market has been on a roller coaster ride lately. It’s been six sessions since the initial bounce off the July lows topped out. In the last six days the S&P 500 has closed either up at least 1.25% or down at least 1.25% five times. Three times it’s been to the downside. All three times the big drops came on lower volume. As you might suspect, seeing the S&P drop 1.25% on lower volume 3 times in 6 days is a fairly rare event. Below is a summary of how the market has traded following such events:

The number of instances is too low to make much out of but the size of the returns is certainly eye-popping. An average return of over 5% in the next 7 sessions. That’s a substantial pop to say the least. For those who would like to review the dates, they are all listed below. Notice how 2002 dominated the study.

While not statistically significant, I find the results noteworthy and interesting. So to the header question: Are we looking at a massive edge or a statistical anomaly?

I'll let you decide.


Johan Lindén said...

Most of us agree that one should make studies based on relativity more often than absolute values.

With all the studies you have that you put in % gain or lost as a criterion, which means most of your studies, wouldn't it be more accurate to normalize the lost or gain % depending on the type of volatility there was at that point.

For instance, if the 14D Average True Range is 0,75% and the SPX makes a 2% move it is much more significant than if the SPX make a 2% move and the 14D ATR is at 1,5%.

I can imagine that this takes a lot more work (or maybe not), but it would be interesting to try this hypothesis at least in 1 or 2 of your future (or past) studies to see if this makes the system even better.

Damian said...

Interesting Rob - I went back and looked at all those instances - I found it interesting that the setup was profitable during the 2002-2003 period (highly) but not so much outside of that period. What this suggests to me is that it was a particular characteristic of that sell-off period, while the period we're in now is rather different. So I would say "no" - and be ready to eat my words.

katallaxia said...

Sell 7 days later = 8 occurrences
Sell 1 day later = 23 occurrences

Why do you always [sic!] ignore "overlapping signals"? All [sic!] your "studies" catch the first occurrence/signal/trade only. If your EasyLanguage Code (I assume that's the prob) can't do the trick ==> Excel.