A Common Error, Revisited

I just ruined a couple hours of work by committing one of the common errors that always seem to crop up when doing analysis of market data. As I discussed in this post, it's easy to generate trading signals that look too good to be true by not correlating your dates properly. To quote myself, "I'd been using a signal derived from one day's market activity to predict that same day's activity, rather than activity the following day."

Well, I did it again. In an effort to branch out from forex trading into other markets, I've been doing research into how the behavior of certain stock market indices (or indexes, if you prefer) affects the movement of their component stocks. So I was looking at how the Philadelphia Semiconductor Index affects the activity of Advanced Micro Devices stock (ticker symbol: AMD). And I came up with the most amazing signal, an almost perfectly straight line shooting upwards and promising to make me filthy rich within a year or so. Or so it appeared. (I won't post a picture of the chart in question because it's just too painful to look at.)

After congratulating myself on finding such a perfect correlation, I began to look more closely at this signal that my wishful thinking had assured me was the real thing. Well, it turned out that early on in my table of historical prices, one day was out of place, and my signal had shifted from reflecting the previous day's prices, to the current day's price. So if the index went up 1% on a given day, surprise, usually AMD did too. The lesson being, if your signal looks too good to be true, there's a good chance it is...so take a closer look at the underlying data and check your work to make sure it's the real thing.

For the record, there are significant correlations between the semiconductor index and AMD stock - just not as significant as I'd first thought. Oh well. At least I caught the error early, before urging all my friends to sign over their bank accounts.

Related topic:
Common Errors When Designing a Trading Strategy