OK, here's a start. I mentioned I got curious about the predictive power of the so-called "death cross," which is used as a technical indicator and occurs when the 50-day moving average drops below the 200-day moving average. So far as I can tell, technical analysts use this as a sell signal, with the presumption that it indicates an expectation that stocks are about to drop. There is also its reverse, the so-called "golden cross," which happens when the 50-day average moves back above the 200 day average, and is a signal to buy because stocks are expected to rise. The following image illustrates the two concepts:
[IMG]
http://i68.tinypic.com/2nbhpxy.jpg[/IMG
]
I downloaded S&P 500 daily data from 1950 onward and did some basic analysis. There were 17167 trading days over that period, during which there were 53 death cross occurrences and 54 golden cross occurrences. If these signal something, we should be able to discern a difference between random expectation (i.e. expectation of forward index value for any randomly selected day) and that expected after one of these holy cross events. I looked at 1, 3 and 6 month, 1, 2, 3, 5 and 10 year changes following each cross occurrence, and did the same for every trading day, then compared the datasets. Here are some simple stats:
[IMG]
http://i66.tinypic.com/wrj31u.jpg[/IMG
]
If you look carefully, you should conclude that the cross patterns are perfectly useless for predicting the future. There is absolutely no meaningful statistical difference between expected forward-looking change following a holy cross versus that of any random day.
So what can these patterns tell us? They average past index values, so they ought to be able to tell us something about past market behaviour. To check that, I looked rearward to see the changes leading up to the cross events, and compared with average. Here is what I found:
[IMG]
http://i63.tinypic.com/2dlk8dt.jpg[/IMG
]
Here we can see that the death cross is usually preceded by a market decline (typically ~ 5% over 1 to 2 months), and a golden cross is usually preceded by a ~ 5% rally over the previous month or two.
So there you have it. These cross patterns have some "predictive" power if you're interested in hind-casting. They have absolutely zero predictive power for forecasting future index values.
---
I learned some cooler stuff during this exploration and will share that when I find time. Need to put in a full day at work first...