Do fund flows moderate persistence? Evidence from a global study

Abstract

We investigate whether fund flows eliminate future abnormal performance and persistence as in Berk and Green (2004. “Mutual Fund Flows and Performance in Rational Markets.” Journal of Political Economy 112: 1269–1295) using a sample of open-end domestic equity mutual funds from 32 countries. We show that flows have only a small moderating effect on persistence even in the United States, where fund industry conditions most closely resemble the Berk and Green assumptions. In fact, we find that most countries do not have decreasing returns to scale in fund management and, as a result, flows have limited impact on mutual fund performance persistence.

KEYWORDS:

• Mutual fund persistence
• fund flows
• returns to scale
• Berk and Green model

JEL Classifications:

• G15
• G23

Acknowledgement

I thank Miguel Ferreira, Helena Isidro, Aneel Keswani, Pedro Pires, Sofia Ramos, and David Stolin for comments. I also thank Chris Adcock (Editor), an Associate Editor, and two anonymous referees for their constructive comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 In Section 2, we provide a detailed discussion of the predictions of the Berk and Green (2004) model.

2 Ferreira et al. (2019) also find fund performance persistence in the majority of countries in their study. This is where the overlap between our work and their study ends. Ferreira et al. (2019) build on the US literature that shows that fund managers that face less competition are able to generate a more persistent alpha (e.g., Wahal and Wang 2011; and Hoberg, Kumar, and Prabhala 2018), to show that the level of competitiveness in the mutual fund industry is an important determinant of fund performance persistence. In our study, we run additional specifications of our tests where we control for the level of competition in the mutual fund industry and find that our main results are preserved.

3 Ferreira et al. (2013) examine the relation between fund size and performance internationally, which has also been examined for the US (e.g., Chen et al. 2004; Elton, Gruber, and Blake 2012; Reuter and Zitzewitz 2015; and Pastor and Stambaugh 2012). However, these studies do not analyze the impact of fund flows on mutual fund performance, which in the Berk and Green (2004) model is the key driver of performance persistence.

4 This dataset is used in Demirci et al. (2019).

5 See Ferreira et al. (2013), and Cremers et al. (2016) for a detailed description of Lipper's worldwide data coverage.

6 Factors are from AQR (https://www.aqr.com/Insights/Datasets).

7 This study analyzes persistence over a one-year frequency. Although performance persistence can be examined at different frequencies, most authors study persistence at a yearly frequency (e.g., Carhart 1997; Elton, Gruber, and Blake 2012; and Ferreira et al. 2019). This is because investors, fund managers, and investors tend to evaluate performance of mutual funds over annual time periods.

8 Barras, Scaillet, and Wermers (2010) propose a new methodology - the False Discoveries Rate (FDR) - to measure fund performance. In their influential study, they show that FDR allows to separate skill (alpha) from luck, which helps to identify the funds that generate the true alpha. The authors also show that controlling for false discoveries improves the ability of finding persistent performance. Andrikogiannopoulou and Papakonstantinou (2019) raise substantive concerns regarding the results in Barras, Scaillet, and Wermers (2010) and the FDR's applicability in fund performance evaluation. In a reply to Andrikogiannopoulou and Papakonstantinou (2019), Barras, Scaillet, and Wermers (2020) incorporate these concerns and propose revised parameter values. They acknowledge that the use of FDR in finance needs to be carefully implemented, particularly when the sample size is small. This discussion highlights the importance of a sufficient large sample to the accuracy of the FDR estimator. Therefore, using the FDR methodology in our study would potentially lead to incorrect estimations as our sample include many countries with a relative small number of observations.

9 In robustness tests we also use benchmark-adjusted returns as our performance measure.

10 In unreported results, we obtain similar results when clustering our standard errors by fund-year or country-year, when we run the regressions by country or pooled across countries, respectively.

11 We compute Total Shareholder Costs (TSC) by adding one-fifth of the front-end load to annual total expense ratio.

12 Ferreira et al. (2019) use data from 27 countries in the 2001–2010 period.

13 In unreported regressions we also determine the returns to scale of countries on the basis of the size of the coefficient on lagged size rather than its statistical significance. This is to address the possibility that countries may be classified as having constant returns to scale due to having few fund observations when in fact they have decreasing or increasing returns to scale in practice. When we use this alternative classification approach this does not alter the tenor of our results.

14 It is possible to test whether younger fund families in increasing industry returns countries perform better than older fund families from these countries. In unreported tests, we find that younger funds outperform older funds in these seven countries.

15 We exclude Judicial from US regressions in column (10) as this variable is time-invariant.

16 Previous studies, including Hunter et al. (2014), and Kosowski, Naik, and Teo (2007), evaluate funds using the alpha t-statistic rather than the estimated alpha. Kosowski et al. (2006) suggest that ranking funds by their t-statistics controls for heterogeneity in risk-taking across funds, which eliminates unusual nonnormalities in the cross-section of alphas. To check that our results are not affected by our methodology, in unreported tests we repeat the analysis in Table 5 with funds ranked on the t-statistic for their four-factor alpha. The results are similar to those presented in Table 5.

17 Pastor, Stambaugh, and Taylor (2015) find evidence of constant returns to scale in the US when they use data from 1993 to 2011, but find evidence of decreasing returns to scale using data from 1979 to 2011. We find, using the Pastor, Stambaugh, and Taylor (2015) methodology that there is decreasing returns to scale in the US, which is also consistent with the findings in Chen et al. (2004), Edelen, Evans, and Kadlec (2007), Yan (2008), and Ferreira et al. (2013). After their corrections, Reuter and Zitzewitz (2015) find that fund performance may improve with fund size. As we use an entirely different methodology and control variables to Reuter and Zitzewitz (2015), it is difficult to compare our returns to scale results for the US with those of their paper.

18 By doing so, the average fund age across countries in our sample increases from 13.4 years (see Table 2) to 22.7 years. In South Korea and China, the countries with lowest average fund age in our sample, the average fund age increases from 7.4 years to 10.4 years and from 7.8 years to 11.3 years, respectively.

19 Table 1 also shows substantial differences in TNA across countries, in unreported results we also use weighted least squares weighting by the inverse of the average TNA in each country-year, and the results remain similar. Because it can be argued that small industries with fewer funds are driving our results, as the results based on smaller fund industries could be less reliable, in unreported tests we also run the results in Table 4 excluding fund industries with less than 200 observations. The results show that this has no impact in our main findings.

20 For each fund, Lipper reports a self-declared ‘Fund Manager Benchmark’ and a ‘Technical Indicator Benchmark’, which are independently assigned by Lipper based on its assessment of the fund investment strategy. To avoid concerns that the fund may strategically choose its benchmark to rank higher in performance rankings, we follow Cremers et al. (2016) and use the Lipper ‘Technical Indicator Benchmark’ rather than the self-declared ‘Fund Manager Benchmark’.

Additional information

Funding

This work was supported by Fundação para a Ciência e a Tecnologia (PTDC/IIM–FIN/1500/2014 and grant UIDB/00315/2020).