Why Does Correlation Matter?
Source: Ken French, JQR Capital
Why Does Correlation Matter?
Two weeks ago we explored the inverse relationship between the size of a company (measured using market capitalization) and their annualized stock performance between the years 1964 and 2022. Last week we expanded our view to look at all seven individual factors with data sourced from the Ken French website. In this post, we will address several important questions.
Should we use more than one factor in our stock selection model?
Which factors tend to “play nice” without stepping on others toes?
Is there a limit to the number of factors we can add to our model?
Which Factors Performed Best?
The chart displayed above shows the performance of the seven individual factors posted on the Ken French Website. A visual review tells us that two of the seven factors had an inverse relationship with performance (the quintile bars slope downward), two of the factors had mildly upward slope, and three of them had strongly upward slope. A downward slope is actually not a bad thing - just an indication that the payoff from exposure to that factor was opposite to that for the one with a positive slope. Again, the rankings for these factors are shown in the table below.
Source: Ken French Data, JQR Capital calculations
The ranked absolute slopes place the book equity to market equity factor as the most powerful between 1964 and 2022 with the investment to total assets factor a close second. One might wonder: what if we use more than one factor in our stock selection model? Would that be more powerful than just one factor? One thing to note is that these factor payoff slopes change over time. Perhaps we should first examine the way they change and their correlation to each of the other factors.
What Is This Correlation?
Correlation is a measure for how two things (in our case, two factor payoff slopes) move in the same - or the opposite - direction over time. These factor payoff slopes have changed over the past 59 years. Here is the measure showing how these factor payoff slopes have correlated with each other over this period when looking at annual return patterns for the five quintile groupings.
Source: JQR Capital calculations
The first thing of note (which is a general characteristic of all correlation tables) is that there is a perfectly positive correlation between each of the seven factors with itself. This is why the diagonal to the table is displayed in red cells with a value of 1.00. The second thing you may notice is that the table is symmetrical about this diagonal axis. The correlation coefficients are bounded by -1.00 on the lower end and +1.00 on the upper end. We color coded the table cells to highlight “better” correlations between the various factor payoff slopes over time.
Why Does Correlation Matter?
This correlation “matrix” highlights opportunities for us to combine factors in our stock selection process without having the newest factor member effectively step on the “toes” of the existing members. In an ideal world, our stock selection factors would have (1) strong payoff from exposure to them (some do and some do not), (2) be stable in their payoff through time (they are not), and (3), they would be uncorrelated through time. This last feature is only helpful when the factor payoffs are not stable over time - since uncorrelated payoffs tend to smoothen out the “ride” over time. Again, investing trends are almost like fashion trends. It is basically impossible to predict when the latest “hot” fad will become stone cold and the long lost stone cold fashion will become the next runway star. Next time we will dig deeper into this idea of what is called a multi-factor stock selection model.
The wisest rule in investment is: when others are selling, buy. When others are buying, sell. Usually, of course, we do the opposite. When everyone else is buying, we assume they know something we don't, so we buy. Then people start selling, panic sets in, and we sell too. - Jonathan Sacks
References
https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html
https://www.investopedia.com/terms/c/correlation.asp
Comments
Post a Comment