An important question for investors in actively managed mutual funds is how to gauge the proportion of returns that is attributable purely to the skill of the fund manager compared with more easily replicable factor exposures, such as value, small size, or momentum. But one challenge to evaluating fund manager skill using factors is that factor loadings, or their weights in both the fund manager’s discretionary portfolio and the total market portfolio, vary over time. One new approach explores whether these troublesome issues of attribution analysis might be solved by introducing analysis of cross-sectional risk characteristics.

A straightforward example of the market exposure factor is beta, which calibrates the relationship between the market and its component parts. Beta naturally varies with the economic environment, changing individual company characteristics, and the fluctuating risk aversion of market participants or agents. So, the standard quantitative models investors may use—such as the capital asset pricing model (CAPM) or econometric and statistical models—cannot always rely on constant beta or constancy in other factor loadings.

If analysts consider time-varying factors, traditionally they use rolling time-series regressions—an approach that is attractive because of its simplicity, given that it requires information only about returns rather than risk. But the time-series estimates have a significant drawback: They update only slowly and may result in excessively smoothed coefficients. Consequently, regression-based attribution estimates can be misleading when managers adjust their exposures dynamically.

To better evaluate fund managers, the authors generate factor benchmarks that are both dynamic and investable. Instead of using time-series returns, they set out to show how to estimate time-varying factor loadings using cross-sectional risk characteristics. Their model uses 12 style characteristics—including momentum, volatility, and size—as well as 55 individual sector exposures. Their research is one of the first large-scale studies of fund performance to use holdings rather than returns-based performance attribution only.

The authors’ approach involves calculating factor loadings for a fund at a given time by finding the combination of factors whose characteristics most closely match the fund’s holdings.

The authors apply this methodology to the quarterly holdings of 1,267 US-based active equity mutual funds from September 2010 to June 2015, which together represent more than US$3 trillion in assets. They note that they have chosen such a short dataset because their methodology has advantages in cases where there are short time periods or where managers rotate their stock and factor positions aggressively. They break down the active returns—defined as the fund’s return minus its benchmark return—into three components, attributable to

• constant factor exposures (such as a tilt to value or momentum stocks),
• time-varying factor exposures, or
• security selection.

The authors find that in many cases, the static factor, time-varying factor, and security selection components are each distinct. The analysis also shows a striking diversity in factor concentration across managers and styles. Large-cap growth funds tend to be concentrated in two factors—momentum and quality—while large-cap blend funds exhibit the most factor diversity. The allocation to the value factor is greatest for large-cap value funds, and value-type funds across the size categories load heavily on the value factor. Across all funds, the most important investable factor tilts used by active managers are, in order, momentum, value, and quality.

In contrast to the findings of some previous studies, no evidence is found—after controlling for factor loadings and other fund characteristics—that funds with higher active shares produce larger active returns. In fact, across the whole sample, the reverse is true. The authors find that some of the worst-performing managers have the highest expense ratios. The authors note that high active share corresponds to higher risk taking, which is unlikely by itself to result in outperformance of benchmarks.

The authors show that because factor loadings are dynamic, time-series returns-based regression estimates of past factor loadings are inadequate. Their new methodology—among the first to use holdings data and cross-sectional risk characteristics—could be useful to investment professionals analyzing funds.

This study offers views into how managers generate excess returns. It suggests that across managers and styles, there is considerable diversity in factor concentration. Additionally, using active share to predict outperforming active managers may lead to misleading results. In other words, high-conviction managers don’t always outperform. The authors point out that there are clearly skilled managers who beat their benchmark but that active share is probably not a way to identify them preemptively.