Wednesday, October 04, 2006

Moving Correlations: A Tool for Examining Sector Relationships

Traders are familiar with moving averages, in which we track the price of a stock or index against a moving window of previous prices. Suppose, however, we want to track the price of a stock or index vis a vis another stock or index? One way to accomplish this is to create a moving correlation of prices. For instance, in this exercise, we'll look at a 10-day moving correlation between the price of the Dow Jones Industrial Average (DIA) and the Russell 2000 Index (IWM).

Recently, this correlation has dipped into negative territory, as the Dow is up over the time period, but the Russell is down. When the two averages are moving in sync (high positive correlation) during a rally are expectations for future prices different than when they're moving in opposite directions (low positive or negative correlation)?

Overall, since 2004 (N = 684 trading days), the average 10-day correlation between the Dow and Russell has been .73. That means that about half of all the movement in those indices can be attributed to a general "buy the market, sell the market" effect. The other half of variance in the movement can be attributed to factors independent of the relationship between the two averages.

We have only had 31 occasions in which the 10-day correlation has been outright negative--and only four in which the negative correlation occurred as a result of the Dow outperforming the Russell. Although this sample size is too small for reliable generalization, it is interesting that all four occasions led to gains in IWM over the following five days--perhaps a kind of catch-up effect, as the average gains in IWM exceeded those in DIA.

When we look at occasions in which (as at present) the Dow has been up more than 1% over the past 10 days (N = 249), we find that the ten-day correlation with the Russell has impacted returns for the Dow only modestly over the next five days. When the Russell has been highly correlated with the rising Dow (N = 124), the next five days in the Dow average a loss of -.05% (61 up, 63 down), which is weaker than the average five-day gain of .08% (362 up, 322 down) for the Dow sample overall. When the Russell has been weakly correlated with the rising Dow (N = 125), the next five days in the Dow have averaged a gain of .07% ( 64 up, 61 down), in line with overall performance.

When we examine the impact of the correlation on the Russell, however, we see more of a pattern. On those occasions in which the Dow has been up over 1% during the past ten sessions, when the Russell has been highly correlated (N = 124), the next five days in the Russell have averaged a gain of .09% (67 up, 57 down), which is weaker than the average five-day gain of .20% (376 up, 308 down). When the Russell has been relatively weak in its correlation with the Dow (N = 125), the next five days in the Russell have averaged a gain of .39% (80 up, 45 down)--a much more bullish edge.

If we look just at times in which the Dow has been up over 1% in the past 10 sessions and the correlation between the Dow and Russell has been under .50 (N = 25), the next five days in the Dow and Russell have been quite bullish. The Dow has averaged a gain of .47% (18 up, 7 down); the Russell has averaged a gain of 1.15% (20 up, 5 down).

What this suggests is that it isn't just how an index moves that is important; it's how it moves relative to other sectors. Moving correlation may be a worthy tool for market analysis. More research to follow here and in the Trading Psychology Weblog.

Note: I want to thank Instant Bull for its kind recognition of TraderFeed as a 2006 "Best of Breed" blog.