Momentum has one of the most consistent performance records across financial markets. Buying what has worked well over the last 3-15 months has been found to produce superior returns for individual stocks, industries, sectors, countries, currencies, and commodities, in short, just about anything that can be traded.
In most studies on the momentum effect, momentum is measured either by trailing performance (e.g. the performance over the last 12 months) or by trailing performance excluding the most recent data to avoid a short-term mean reversion effect (e.g. the performance over the 12 months ending one month ago).
In my opinion, these methods for measuring momentum are less than optimal. For example, take a look at the chart below with the 250 day rate of change of the S&P 500 (roughly equal to the performance over the last year).
When I first saw the periods when this momentum metric rose while the S&P was flat, and how low the correlation was in general, I was a bit shocked. This low correlation is because performance over a set time period depends on two price points, with just as much weight on the price at the beginning of the period as the price at the end.
So does it really make sense to use a simple trailing performance metric? I think the answer is obviously no. If you’re investing in a momentum based tactical asset allocation system, would you want to be using a metric that causes an asset class to appear significantly more attractive at the end than at the beginning of a period when the asset class itself has been flat?
So if we don’t want to use a simple trailing performance metric, what can we use instead. One solution is to average the performance over several time periods such as 12, 6 and 3 months. This is the method I have used for our market rotation and sector rotation strategies. However, while averaging the performance of several time periods largely solves the problem with using only one period, it doesn’t completely eliminate it.
I think the best solution is to use the distance from a moving average such as the 250 day moving average. The result has a much better correlation to the price than trailing performance does. In addition, if we want to avoid the short-term mean reversion effect and avoid whipsaws, than we can use the distance between a shorter moving average and a longer one.
In the end, what matters is whether performance is better or not. While I don’t have the numbers, I’m sure that using the distance between two moving averages will produce performance at least as high as using a trailing performance metric because the stocks with the highest momentum as measured by the two methods are mostly the same.
In the strategies on VectorGrader.com, we’ve been moving towards using the distance between two moving averages. We think the results are more robust and better at measuring momentum.