Data and methodology
A model to represent multi-asset risk factors
Our study applies a return-generating (i.e., risk factor) model to a large sample of multi-asset funds for the 30-year period from 1986 through 2016, and uses monthly returns to estimate the model in Figure 1 below.
A four-factor return-generating model allows for control of specific risk factors that investors can gain exposure to relatively cheaply via passive investments. The three Fama-French factors represent characteristics of broad market exposure as well as market capitalization and style diversification. Since multi-asset funds can have varied exposure to fixed income, the model includes the Bloomberg Barclays U.S. Intermediate Government Bond Index.
The study uses multi-asset fund data from Morningstar's U.S. Category Group defined as "Allocation." This group was simplified by removing all target-date strategies as well as convertible funds. This resulted in seven remaining subcategories
Smaller and newer funds were also removed; the study focused on funds with net assets at the strategy level of at least $50M and with five years (60 months) of performance history. To minimize the impact of survivorship bias, funds that were defined as "no longer active" in the Morningstar database are part of the study.
These screening steps arrived at a total of 2,483 individual funds, when including multiple share classes, provided that they had at least 60 consecutive months of returns between 1986 and 2016.
Study parameters: Rolling-return periods over multiple market cycles
With the sample determined, the study considers monthly net-of-fee return data for every fund as well as for each of the four risk factors from the equation in Figure 1, for the period from January 1986 through December 2016. With the first 5-year/60-month period, which ended December 1990, represented as month t, the regression outlined in Figure 1 is calculated for each multi-asset fund. The parameters bjM, bjSMB, bjHML, and bjBOND in Figure 1 represent estimates of average sensitivity or exposure to each of the four factors. The intercept of the regression, ajt, represents an estimate of the average incremental return a specific fund has added over five years. This incremental return represents the portfolio alpha (positive or negative) that an investor would have received separately from the returns generated by the four risk factors.
For every fund, a new alpha term is calculated for every month from December 1990, or from the end of the fund's first full five-year performance period through December 2016. Funds that closed or merged remained in the study, represented by only those five-year periods that they completed. Funds that existed at the beginning of our study, and continued in operation as of December 2016, would have a maximum of 312 alpha values (e.g., 12 months x 26 years). Starting in December 1990, and for every subsequent month through December 2016, an average alpha return across the funds with the necessary 60 months of return history is then calculated. With those results, we can then calculate an average of this new series allowing us to assess not only the average alpha of multi-asset funds, but also how it has changed over time.
Treatment of outliers
To eliminate outlier managers, the average regression r-squared from equation in Figure 1 is another filter. To accomplish this, we rank all funds by r-square and symmetrically trim the top and bottom 5%. The goal is to remove managers that are essentially "closet-indexers" or are not truly representative of the sample group. Screening out the tails leaves the trimmed sample of active multi-asset funds, which is the middle 90% of our initial sample. The funds in the trimmed sample have an average r-squared coefficient in the range of 66%–98%. (Note: In a subsequent step of this analysis, we will return to this 10% cohort to compare the average alpha with the larger trimmed sample.)