It's always nice to test one's hypotheses "out of sample" to guard against data mining: maybe results were just an accident of the time period studied. So I decided to look at DL prospectively, starting with a survivorship-bias-free sample of monthly data beginning January 2000, for the following asset classes and their respective indexes:

As in the previous study, I ranked from 1 (best) to 10 (worst) the performance of each index, and plotted it against the percentile performance for the index in each style box versus the active funds. Purists will chafe at the use of an index instead of an actual fund, but since funds were not available for all of the indexes, I wanted to be internally consistent. Further, of the six funds available from Vanguard, four have managed to equal or surpass their benchmarks by small amounts. This analysis produced 120 data points, plotted below:

This is a bit of a scattergram, but the "northwest to southeast" bias of the results is still fairly discernible. (Note that I've inverted the percentile scale on the y-axis, placing the "good results" at the top of the graph. This is the opposite of the convention used in Mr. Rekenthaler's graph.) The statistical power of the data, even after only 12 months, is staggering: a t-stat of 6.15 and a p value of 10-8. The individual monthly plots are fascinating. In some months, like January, March, August, and December, the points line up like the Rockettes, with highly statistically significant results. (In these four months, the t-stats were 4.00, 3.53, 4.44, and 6.53, respectively, with p values of .004, .008, .002, and .0002.) In other months, the relationship is nonexistent. I've posted the individual plots on a separate page. Those of you who would like the study data may contact me.

<<Previous                       Next >>