Active bond selection performance is typically assessed by comparing a fund’s excess returns with the returns of a broadly diversified bond or treasury market index, using a linear time-series regression. The index represents the interest rate risk premium.

However, this approach assumes a linear relationship between interest rate risk exposure (the linear regression’s beta) and expected return, a relationship that is known to be nonlinear. As a result, active bond selection performance (alpha) is mismeasured.

The authors first demonstrate the nature and extent of the mismeasurement of alpha by looking at the performance of passive US Treasury bond total return indexes with different maturity ranges. They propose a new maturity-matched performance measure (MM alpha) that matches each fund to an individual benchmark treasury index based on the fund’s reported durations.

The authors test their new measure on samples of government and corporate bond funds and compare it with the usual linear regression methods. They use a sample of 127 active US domestic government bond funds and 291 corporate bond funds from 1990 to 2014. They run a linear multifactor regression model on 10 different treasury indexes to demonstrate the alpha deviation problem. For the MM alpha, the authors individually match funds and treasury indexes by the minimum difference between the average duration of the fund and the respective index.

The MM alpha measure cannot be applied to funds that do not report their duration, so the authors consider alternative matching criteria between funds and benchmarks. They evaluate alternatives, including the funds’ self-reported primary benchmark indexes from Morningstar, their beta, and their R² values.

Finally, they demonstrate the practical implications of the alpha deviation in active bond selection performance. They test whether popular performance metrics (represented by the Morningstar ratings) and investor decisions (represented by investor flows) are more closely related to the flawed linear regression measures or to their new MM alpha approach.

The authors show that using the broad index for all funds systematically overestimates the average active bond selection performance. The corporate bond fund sample shows similar results, with a statistically significant alpha deviation of 0.22% per annum.

The study’s key finding is that by virtue of their duration, some funds are overestimated and others are underestimated. The average absolute alpha deviation for the sample is more than twice the average alpha deviation for the sample of government bond funds and almost four times the average alpha deviation for corporate bond funds. This difference indicates that on the individual fund level, very pronounced positive and negative alpha deviations exist and depend systematically on the size and direction of their maturity mismatch.

As an alternative to duration matching, beta matching works well for government bond funds but is less effective for corporate bond funds. Using R² works reasonably well for government bond funds and produces the best results for corporate bond funds.

The authors find that both performance ratings (Morningstar ratings) and investor decisions are more closely related to the erroneous measure of alpha than to the MM alpha.

The common linear regression estimate of alpha overestimates the average bond fund’s active performance. This is a problem not only in government bond funds—where interest rate risk is the critical determinant of expected return—but also in corporate fund bonds, where other systematic risks, such as default risk, are significant.

Both the commonly used Morningstar ratings and investor flows are influenced by this alpha deviation because they are more sensitive to maturity-mismatched performance measures than to maturity-matched performance measures.

The authors advocate matching funds and benchmarks using MM alpha to more reliably assess active bond selection performance. This approach has the advantage of combining the widely used regression approach with the matching of bonds and benchmarks on their durations and also requires much less data.