We match what we need to match and leave out what we don’t. If the target fund doesn’t have a statistically significant “value” coefficient, then there’s no need to bother matching the HmL factor. Balance is the key. The clone will always work best when we have the statistically significant betas as closely matched as possible. Exactly matching one variable is generally a good solution, but not the best; try to align them all simultaneously. Even with an optimizer, it’s not possible to literally match each and every one of them exactly. With optimization, you can exactly match one thing (the objective function) and then subject that objective function to several constraints.
Capturing the market-beating advantage of “small” and “value” stocks is easy; the key is to measure it exactly if you want to match DFA exactly.3 There are three time periods of interest—the in-sample period, the out-of-sample period and the entire period—so we measure the factor loads and returns over each. Data is split into estimation and evaluation periods: The in-sample period is the portion of the data that is backtested. The out-of-sample period evaluates a model’s forward performance and provides confirmation of a model’s effectiveness. Out-of-sample tests help guard against data mining, so many researchers regard out-of-sample performance as the “ultimate test of a forecasting model.”
DFA’s Small Cap Value fund is a bit more “value-y” and more “small” than Vanguard, so we salt in a little microcap to increase the “small-cap” weighting (SmB) (Figure 5). Our DFA clone was built from Vanguard’s plain-Jane version of a small-cap value index (85 percent) and iShares’ Russell Microcap Index (15 percent).