We provide new evidence for dis-economies of scale at the mutual fund level. Building on Berk and Green (2004) and allowing for gradual adjustment to equilibrium, we show that (quarterly) changes in fund performance are strongly negatively related to lagged predicted fund flows. We find that alphas would be more cross-sectionally dispersed and more persistent without the damping effect of flows. Thus, flows are an important factor behind the lack of predictability in mutual fund performance. This flow mechanism is strongest for smaller and more active funds with higher expense ratios, suggesting that it is related to the stock illiquidity and costly search for investment opportunities.
When Fund Flows Take The Fun (Alpha) Away – Introduction
It is well established that money chases mutual fund performance,1 but puzzling that past performance, as measured by fund alpha, is not a good predictor of future performance. The neoclassical view, advanced by Berk and Green (2004), argues that these observations can be reconciled with the simple assumption of decreasing returns to scale. Investors crowd into high-performing funds, diminishing their future performance, and poorly performing funds improve due to out flows. When rational investor flows attenuate performance in this way, managers can be skilled and yet we may observe little performance persistence.
The validity of this view hinges upon the assumption of decreasing returns to scale. To date, however, the evidence is mixed about whether individual fund returns vary with fund size. Empirical tests relating fund performance to fund size face several statistical challenges. First, fund size is persistent, making inference dicult. Second, size is endogenous to the history of a fund’s performance, both directly through changes in portfolio value and indirectly through performance-induced flows (Pastor, Stambaugh, Taylor, 2004, McLemore, 2014).
Warren Buffett’s 2018 Activist Investment
We circumvent these problems by modeling changes in performance and size rather than levels. Our methodology enables us to test explicitly for the eect of flow-induced changes in fund size on performance, which we call the ow mechanism. In particular, in the Berk and Green (2004) model, prices are normalized to one and fund
ows are the sole performance-equilibrating mechanism. Empirically as well, changes in fund size are mostly due to flows.
Thus, the test for decreasing returns to scale can alternatively be formulated by analyzing the relation between changes in performance and changes in size, most cleanly measured by flows. Since flows are less persistent than size, our estimates are not prone to the small sample bias. Also, our focus on changes substantially diminishes the problem of endogeneity for two reasons. First, changes in size depend only on the most recent fund performance rather than the whole history of fund performance. Second, we only focus on the part of the change in size that is due to flows. Still, performance changes contain lagged performance, which may be related to lagged flows. To address this concern, we do not use lagged flows directly, but instrument lagged ows with further lags of flows and performance.
A second important benet of our methodology is that an analysis of changes in performance is better suited to capture partial adjustments in performance than an analysis of levels. If per-period adjustment is partial, funds with high past alphas and high past inflows may still have relatively high current alphas. Analyzing changes may therefore be necessary to capture this partial adjustment due to flows.
We also examine an empirical model where we test both directions of the flow-performance relation simultaneously. This provides more ecient estimates and enables to quantify the overall importance of the flow-performance mechanism. Last, separating inflows and outflows allows to test for any asymmetries in performance response to changes in fund size.
We test our empirical model on domestic active funds from the CRSP mutual fund database from January 1999 to June 2014. Quarterly market model and four-factor alphas are estimated using daily fund returns (net-of-fees). We choose a quarterly horizon as this is the highest frequency at which alphas can be reliably estimated with daily data. To circumvent endogeneity issues, we instrument flows using four lags of past flows and past alphas.
See full PDF below.