With the 50-day moving average of the S&P 500 threatening to cross above the 200-day moving average, there are again mentions in the blogosphere of the predictive power of this “golden cross”. It has been a good buying signal over the past 20 years. I do not like this indicator and have just posted this in the comments elsewhere:

This only works because of the particular market pattern we have had in the past 20 years — specifically, two bear markets of approximately similar size and duration. I tested this years ago and found that it didn’t work in Japan — which has had a notably different pattern in its stock-index charts, with lots of short, quick rallies.

Because we have had those two big bear markets, any “golden cross” system will look good because it would have kept you out of them. Today I have tested a long-only system that buys when the 50-day average crosses above a longer average, and sells when it crosses below. 150-, 200-, 250-, 300- and 500-day moving averages all do the trick, with the system outperforming buy-and-hold over the past 20 years. There is not much to choose between 200-, 250-, 300- and 500-day moving averages. (A 750-day moving average does not work at all well.)

The problem with moving averages is that you don’t know in advance what length of average is going to be slow enough not to trade false rallies and temporary pullbacks but fast enough to get you out at the start of bear markets and in at the start of bull markets — because you don’t know how long and how sharp the rises and falls in the market are going to be. Thus, for example, a 200-day golden cross system underperforms buy-and-hold in the 20 years to October 2000 and the 20 years to October 1990.

Note that I have been generous to the system by not using it to short. Shorting using a golden cross system has pretty much broken even over the past 20 years. Analysis is based on price data only (underperformance of the system would be larger if dividends were taken into account) for the S&P 500. Data from Tradestation.