Some stock selection schemes seem silly. This weekend’s Wall Street Journal reports the results of two hypothetical portfolios which are clearly intended to be nonsensical. One example, the so-called “Graham and Buffett Portfolio” comprises stocks whose ticker symbols consist only of the letters found in the names “Benjamin Graham” and “Warren Buffett.” Silly it may sound, but a 20-year backtest of the Graham-Buffett portfolio shows that it handily outperformed the S&P 500.
The Graham-Buffett portfolio is a more complicated variation on a theme sounded earlier by Vanguard Group. The Vanguard “AlphaBet” portfolios depend only on the first letter of each stock’s ticker. Like the Graham-Buffett portfolio, the backtested Vanguard AlphaBets also outperformed the S&P 500. (This is especially impressive given that active large-cap U.S. managers typically underperform.)
The Journal argued that these results depend on data mining — that is, the ability of backtesters to keep trying various rules until they find “something, anything, that would produce groups of stocks with high returns.” “Data mining” thus connotes a degree of intellectual dishonesty. We agree that investors should regard any backtest results with a fair degree of skepticism, but the explanation for the Journal and Vanguard results may be much simpler — and reveal an important truth.
Building a portfolio, in either a backtest or real time, is a two-stage process:
- The first step is security selection— the process by which we determine which stocks should be part of the portfolio. Selection mechanisms can be fundamental or technical or quantitative — or some combination. They can also be nonsensical, as with the Journal and Vanguard examples. Importantly, when we say that a stock selection mechanism is nonsensical, we’re also saying that you’d do just as well by selecting stocks at random.
- The second step is portfolio construction — the process of weighting and constraining that combines the stocks we’ve selected into a final portfolio.
The most important thing about the Journal and Vanguard portfolios is not their (somewhat similar) stock selection processes. The most important thing is that after they determine what stocks they want to own, the portfolio construction process weights each stock equally.
Equally-weighted indices tell us the performance of the average stock in the selection universe. What return should we expect if we pick stocks randomly (or with a selection rule that is no better than random)? With enough trials and enough time, random selection will produce the return of the average stock in the index. That means that the best estimate of the return of a randomly-selected portfolio is the return of an equal-weight index. Over time, and certainly during the interval covered by the Journal and Vanguard portfolios, equal-weight indices outperformed their cap-weighted counterparts.
The Journal/Vanguard results are not only, and arguably not most importantly, about the perils of relying on backtests. The most important point they illustrate is that equal weighting creates a powerful tailwind for a strategy’s performance. Astute portfolio construction can mask flaws in security selection.