In the Part 1 of this article, we introduced the research design of our study. Here, let’s look at the results.
Results
Exhibit 1 shows the annualized performance of the stock- and industry-level versions of each factor. While the stock-level value factor earned a respectable 2.36% annualized return over the 50-year sample period, the industry-level value factor posted a small negative return. This suggests that industry tilts did not contribute significantly to the performance of the broad value strategy. Despite the gap between the two value factor returns, they were highly correlated. (The correlation coefficient was 0.78.)
.png)
Industry tilts did not help the value factor's performance, likely because they were fairly static, as illustrated in Exhibit 2. This table shows the percentage of time each industry appeared in the long side of each factor portfolio. To illustrate, the energy industry ranked in the cheaper half of the 12 industries 86% of the time, while the healthcare industry only fell in the cheaper half 2% of the time. Industry performance can diverge, particularly over short time frames. But it appears that returns across industries are not strongly related to their (price/book) valuations. Differences in business models, asset intensity, and accounting practices can make such valuations tough to compare across industries. Therefore, industry-relative valuations should contain more information about expected returns than valuations across industries. Our results are consistent with this pattern.
.png)
The industry-level momentum factor performed much better than its value counterpart. It actually generated a slightly higher return than the stock-level momentum factor with marginally lower volatility. Momentum's industry tilts are more dynamic (as shown in Exhibit 2) than the value or low-volatility factors and seem to contain greater information about future expected returns. For example, differences in growth rates and capital structures will almost always cause the financial services and utilities industries to trade at lower valuations than the technology and healthcare industries. However, industry performance leadership is often a short-term phenomenon. The high-turnover momentum factor captures this effect, while the value factor is slow-moving. These results suggest that ignoring industry momentum would reduce the efficacy of a momentum signal.
Neither the stock- nor the industry-level low-volatility factors in this study generated positive returns during the full sample period. However, absolute performance is not the best way to measure their efficacy. By definition, less volatile stocks and industries are not as risky as their more volatile counterparts, so it is better to assess the low-volatility factor's success by its risk-adjusted performance. Exhibit 3 shows the performance data for the low- (long) and high-volatility (short) portfolios used to create the stock- and industry-level factors.
.png)
The 50-year compounded return spread between the low- (long) and high- (short) volatility portfolios differs slightly from the factor returns. This is because the factors were calculated from the monthly spreads between the long and short portfolios, which allows them to compound at a slightly different rate than full period spread between the two portfolios. The two industry-level portfolio returns were comparable, but the low-volatility portfolio exhibited a lower standard deviation, resulting in better risk-adjusted returns (as measured by its Sharpe ratio). However, the stock-level low-volatility portfolio had a much stronger absolute performance relative to its more volatile counterpart. Not only did it enjoy a substantially lower standard deviation, but it also posted stronger returns.
These results suggest that while a low-volatility factor can be successfully implemented at both the individual stock and industry level, it has historically been more effective at the stock level. This is likely because there is greater variability in volatility across stocks than industries. While stock-level low-volatility strategies still tend to skew toward defensive industries, this dispersion allows them to realize a greater reduction in volatility than the industry-level strategy.
Consistent with the value and momentum factor results, the industry- and stock-level low-volatility factors were highly correlated during the sample period. (The correlation coefficient was 0.66.) However, the industry-level low-volatility factor underperformed (generated a negative alpha) after controlling for its exposure to the market risk, size, value, and stock-level low-volatility factors. This further indicates that the low-volatility factor has been more effective at the stock level.
Not surprisingly, the industry-level low-volatility factor consistently favored the consumer nondurable (consumer defensive) and utilities industries, and generally avoided the business equipment (which includes technology stocks) and consumer durable (consumer cyclical) industries. However, its industry tilts were not as consistent as the value factor's.
Conclusion
The results of this study suggest that it is sensible to make industry adjustments for factor strategies with persistent industry tilts, like value and low volatility, because these tilts don't contribute significantly to these strategies' long-term success. However, it is best to leave industry tilts unconstrained for momentum strategies, because they can capture short-term industry performance persistence. Even without a performance benefit, industry constraints allow investors to mitigate unintended bets. This comes at the cost of lower factor style purity, higher turnover, and greater complexity. In many cases, it's worth the trade-off.
The full report is available in the Research Library on Morningstar's corporate website.
