Out-of-Sample Test of What Works on Wall Street (O’Shaughnessy’s Cornerstone Strategies)
Posted in Fundamental Valuation, Momentum Investing
August 3, 2009
In the mid-1990s, James O’Shaughnessy identified the”cornerstone value strategy” and the “cornerstone growth strategy” as best-of-breed equity investment strategies. The former emphasizes dividends, and the latter momentum/earnings growth. Based on Standard and Poor’s Compustat data, he found that the value (growth) strategy returned an average 15% (18%) per year over a backtesting period of 1952-1994, compared to 8.3% for the S&P 500 Index. He implemented these two strategies in late 1996 via mutual funds, and then publicized them in his 1998 book What Works on Wall Street: A Guide to the Best-Performing Investment Strategies of All Time. He subsequently sold the corresponding mutual funds (which apply slightly different portfolio formation rules from those specified in the original research) to Hennessy Funds in 2000, where they survive as the Hennessy Cornerstone Value Fund (HFCVX) and the Hennessy Cornerstone Growth Fund (HFCGX). Has a dozen years of out-of-sample performance of these two mutual funds confirmed the backtesting outputs? Using self-reported annual performance data for HFCVX and HFCGX, and annual returns for selected benchmark indexes for 1997-2008, we find that:
According to Hennessy Funds, they specify the holdings of HFCVX and HFCGX annually by “strictly adhering to the following time-tested, quantitative formula[s]” applied to the Compustat database, as follows:
HFCVX managers screen for large capitalization value firms (excluding utility companies) based on: (1) market capitalization above the average of the database; (2) number of shares outstanding above the average of the database; (3) 12-month sales 50% greater than the average of the database; and, (4) cash flow above the average of the database. They then select the 50 stocks with the highest dividend yield.
HFCGX managers pursue a strategy that “marries value with momentum” via screening based on: (1) market capitalization above $175 million; (2) price-to-sales ratio below 1.5; (3) annual earnings higher than the previous year; and, (4) positive returns over the past three and six months. They then select the 50 stocks with the highest 12-month past return. [Item (1) may have grown over the years.]
The following chart compares the average (arithmetic mean) annual returns of both mutual funds and various benchmark indexes over the entire 1997-2008 sample period.
HFCVX has materially underperformed both its benchmark Russell 1000 Value Index and the S&P 500 Index over the past decade. The fund underperformed the S&P 500 Index by about 2% per year, compared to the backtested average annual outperformance of about 7%. Its slightly lower standard deviation of annual returns (18.4% versus 19.9% for the Russell 1000 Value Index) seems not to justify the undesirability of the lower return. In any case, backtested past outperformance has not persisted over a 12-year out-of-sample implementation.
HFCGX has materially outperformed both its benchmark Russell 2000 Index and the S&P 500 Index over the past decade. The fund outperformed the S&P 500 Index by about 5% per year, compared to the backtested average annual outperformance of about 10%. Its slightly higher standard deviation of annual returns (23.1% versus 21.4% for the Russell 2000 Index) seems not to offset the benefit of the higher return. In any case, backtested past outperformance has persisted over a 12-year out-of-sample implementation.
For a different perspective, we look at cumulative returns.
The next chart tracks the year-end values of $10,000 initial investments in each of HFCVX, HFCGX and the selected benchmark indexes at the end of 1996 through the end of 2008. Results confirm that HFCVX has consistently underperformed its benchmark index, while HFCGX has consistently outperformed its benchmark index.
Possible reasons that the out-of-sample performances of these funds fall well short of their backtested past performance are:
- The assumptions made about trading frictions in the backtests are materially optimistic.
- The market changed. For example, the expansion of stock buybacks in lieu of dividends is largely absent from the backtest period and disrupts the dividend-based strategy of HFCVX.
- The market adapted, with more and more investors competing for the abnormal returns from the backtests justifying the cornerstone strategies.
- Data mining bias added material helpings of luck to the backtested outperformance of the fund strategies.
- Wildness (non-normality) of the distributions of stock returns makes the average past return unreliable as a measure of expected return.
- Deviations by the fund managers from the original backtest specifications materially diminished strategy performance.
In summary, as found for the “cornerstone” strategies, what worked on Wall Street in past decades may or may not work in future decades. Changing/adaptive investment environments, implementation compromises and frictions, bias derived from mining noisy data and non-normality of stock return distributions complicate strategy selection.
Occasionally, new books tackle investing in U.S. stocks along similar lines, optimizing based on characteristic-screened past performance (often without correction for data mining bias).
