Fund flow
To test whether external product markets play an active role in responding to changing in managerial structure, one way is to examine whether shareholders redirect their money away from fund family, reflecting through fund flow measurement. We analyze two measures of net flow. The first measure is the net percentage flow, scales net flows by the total net assets in year t-1 and can be interpreted as an asset growth rate net of appreciation. While most previous papers in the mutual fund flow-performance literature have analyzed only percentage flows, we also focus the dollar measure. We focus on dollar flows for three reasons. First, it is the response of investors to performance (or lack thereof), not the percentage increase in fund size, in which we are interested. Fund manager incentives should be driven by dollar flows and not percentage increases in fund size. Second, the use of dollar flows is more amenable to calculation of elasticities of fund flows with respect to performance. Third, the autocorrelation of fund flows is more directly investigated without the confounding effects of autocorrelation in assets that is present using percentages. The second measure annual net dollar flow in or out of a fund, defined as the annual changes in total net assets minus appreciation. These two net manager flow measures can be viewed as the aggregation of allocation decisions of all the manager’s clients.
Fund flow percentage measure
Regarding to the first fund flow measurement utilising existing data available on the Morningstar direct platform, most flow data are reported as total assets of the fund at the end of the year, these figures contain nuisance in calculating our targeted key variable net inflow/outflow when the returns generated by the mutual fund manager(s) during the year are also part of the component of empirical analysis variables. Therefore I follow most of the previous literatures examining the relationship between fund flows and performance in mitigating nuisance variables, including Jain and Wu (2000); Patel et al. (1994); Sirri and Tufano (1998), where net flow is defined as the net growth in fund assets beyond reinvested dividends. Formally, it is calculated as,
- NetFlow,t = [TNAi,t – TNAi,t-1* (1+Ri,t)]/ TNAi,t-1, (1)
Where TNA is the year end t’s total net asset of fund I, Rit is the fund i’s return during previous year t. The assumption is that all investor earnings will be reinvested and there is no new fund inflow. Based on this formula, the target variable NETFLOWit is derived which reflects the percentage growth of a fund due to new investment.
Fund flow dollar value measure
We follow dollar value fund flow measure of Fant and O’Neal (2000)’ paper which takes into account all new money invested and any reinvested distributions and is measured in dollars.
- 〖FLOW〗_(i,t)=〖EA〗_(i,t)- 〖BA〗_(i,t) (〖ENAV〗_(i,t)/〖BNAV〗_(i,t) )
Where
- EA = ending net assets;
- BA = beginning net assets;
- ENAV = ending net asset value; and
- BNAV = beginning net asset value.
Fund flow-performance relationship
The baseline regression model for testing the fund flow performance relationship is to determine whether fund flows are simultaneously a function of their own returns. In this paper, we examine this question mainly in the context of lagged linkages. Prior studies find both fund age and size have a significant impact on flow-performance sensitivity (Chevalier & Ellison, 1997; Jain & Wu, 2000). Size, for example, is important in the flow relationship because larger funds are more likely to attract larger funds. Thus, fund size will correlate for any scale effects induced by the fund flow measure. Similarly, the older, more established funds have a greater potential to attract money flow than do newer funds. Therefore, the age of fund is included in our model. We follow Wang et al. (2015)’s paper and ensure that the results are not biased by age and size factors. Expense has also been incorporated in the literature (Barber et al. 2005; Gil-Bazo and Ruiz-Verdú 2009; Huang et al. 2007); A fund with a higher expense ratio, ceteris paribus, is less attractive to investors compared to that of lower fees. Huang et al. (2007) also consider the effect of fund volatility on money flows. The derivation of fund size, fund age, and expense ratio is very straightforward, however, we need to consider the effect from a market perspective and incorporate the return on the U.S. equity market as a measure for volatility.
Volatility measure
To measure volatility, we construct the standard deviation of monthly excess returns or raw returns in the 12-month prior to each quarter t (based on past 12 months):
- σ_(i.t)= √(∑_(i=1)^T▒〖(r_(i,t)-μ_(i,t))〗^2 ) (2)
Where ri,t denotes standard deviation for fund i for each quarter t. ri,t represents monthly raw return or excess return for fund I for the previous 12 months, and〖 μ〗_(i,t) denotes the average raw return/excess return of fund i during the past 12 months.
Three performance measure
There are many issues that surface when deciding on a set of performance measures to study. The performance evaluation literature is large, and there is considerable debate on as to which measures are most appropriate. Therefore, to provide a more comprehensive approach for fund performance, we use three different measures, namely raw return, the objective-adjusted returns using the capital asset pricing model (CAPM) as the underlying model, and Carhart (1997)’s four-factor model.
Raw return is the most commonly adopted measure of performance. (Benson et al. 2010; Chevalier and Ellison 1997; O’Neal 2004; Sirri and Tufano 1998). We likewise consider this measure because most investors view it as the primary performance measure. The monthly raw return is adjusted for dividend distribution and retrieved from Morningstar Direct database. For each fund in month t, we calculate the fund’s moving average in the past 12 months and use it as a proxy for fund i’s performance in the past year.
The objective-adjusted measure the abnormal performance of a fund relative to the mean performance of other funds within the corresponding investment objective. We adopt Jun (2014)’s method and control for risk differentials across the respective funds by using the capital asset pricing model (CAPM). Therefore, we perform market model regression using fund’s return for the past 24 months to estimate the model parameter α and β. Specifically:
- R_it – R_ft= α + β_i (R_mt – R_ft) + ε_it, (3)
Where Rit is the ith fund’s raw return in month t, Rft is the risk-free rate (i.e. the one-year interest rate for certified deposits in the U.S.) in month t. and Rmt is the monthly return of the U.S. stock market composite index after dividend distribution adjustment. This regression can be further interpreted that Rit – Rft is he excess return on the security and Rmt – Rft is the market premium. After deriving the target variable α, which is the excess risk-adjusted return (commonly referred to as Jensen’s α), the next step is to compute the monthly risk-adjusted return. Formally, it is calculated as:
- α_it^CAPM=α_i+e_it (4)
The risk-adjusted abnormal return of the CAPM model can be calculated as the moving average of α_it^CAPM for the past 12 months.
The third measure is the risk-adjusted abnormal return using Carhart’ (1997)’ four-factor model. A number of factors have been shown to influence a fund’s cross-sectional variation in performance, including the fund portfolio s exposure to a market (β) factor, momentum factor, size factor, and market-to-book ratio. Carhart’s methodology is standard for mutual fund studies and incorporates these factors into the performance analysis in order to compute a fund’s risk-adjusted alpha:
- Rit = αit + bit RMRFt + sit SMBt + hitHMLt + pitPR1YRt+ eit (5)
Where Rit is the fund return in excess of the one-month T-bill return; RMRF is the value-weighted return on aggregated market index of all NYSE, AMX, and NASDAQUE firms over risk-free rate benchmark; SMB (small minus big) is the excess returns for small stock portfolios over that for big stock portfolios, ceteris paribus (e.g. weighted average book-to-market equity); HML (high minus low) is the return differences between high and low book-to-market equity portfolios. PRYR is the momentum factor calculated in Carhart (1997), defining as the difference between the equal-weighted average of firms with the lowest 30% 11-month return lagged one month and the one month lagged equal-weighted average of firms with the highest 30% 11-month return. Specifically, in each month t, we estimate αit, bit, sit, hit, and pit for each fund using the returns over the past 24 months. We then compute the monthly risk-adjusted return of the Carhart’ (1997)’ four-factor model as:
- α_it^Ch=c_i+ μ_it (6)
The final step is to compute the average of α_it^Ch for the past 12 months.
Previous literature documents that mutual fund flow reacts to past performance in an asymmetric way, customers invest disproportionately in funds with higher performance in the previous period. To provide an overview of the flow-performance relationship, we follow Ippolito (1992)’ paper and rank each fund at the beginning of quarter t based on the average three performance measures in the past 12 months. Each fund is assigned a ranking score from zero for the funds in the worst-performing group to 1 for the funds in the best-performing group. The three segments divide the fund universe in each year by ranked return into the top quintile (HIGH), the middle three quintiles (MID), and the bottom quintile (LOW):
- LOW=min(〖Rank〗_(t-1),0.2)
- MID=min(〖Rank〗_(t-1)- LOW,0.6)
- HIGH=max〖(0,〖Rank〗_(t-1),-0.8)〗
Flows are regressed on performance ranking in the low, medium, and high performance ranged using control variables in regression models.
- 〖Netflow〗_it= α+ β_1 〖TEAM〗_(i,t-1)+β_2 σ_(i.t-1)+β_3 〖TE〗_(i,t-1)+β_4 〖FL〗_(i,t-1)+β_5 〖BL〗_(i,t-1)+β_6 B_(i,t-1)+β_7 LOW+β_8 MID+β_9 HIGH+β_10 〖Log(TNA〗_(i,t-1))+β_11 〖Log(Age〗_(i,t-1))+β_12 〖Log(Familysize〗_(i,t-1))+〖β_13 V_(i,t-1) P_(i,t-1) 〖VM〗_(i,t-1)+β_14 V_(i,t-1) P_(i,t-1) 〖VH〗_(i,t-1)+β_15 〖FEE〗_(i,t-1) P_(i,t-1)+ε〗_(i,t) (7)
The explanatory variables are defined as: Net flowi,t: Net flows into fund i at quarter t, calculated from Equation (1), Vi,t-1: Riskiness or volatility of fund i at quarter t-1, obtained from Equation (2). TEi,t-1: Total expense ratio incurred for fund I at quarter t-1, which is ratio of total investment that investors pay for the fund operating expenses, FLi,t-1:Fundi’s front-end load at quarter t-1, BLi,t-1:Fundi’s back-end load at quarter t-1, Bi,t-1: Fundi’s 12b_1 expense at quarter t-1, LOW, MID, HIGH: Performance ranks, each quintile captures funds with the corresponding to weighted average of three performance measures as defined earlier this section, Log (TNAi,t-1): Control variable, logarithm of fundi’s size at quarter t-1, Log(Agei,t-1): Control variable, logarithm of fundi’s age at quarter t-1, Log(Familysizei,t − 1) is the log transformation of total net assets under management in the fund family to which the ith fund belongs at the end of quarter t − 1, excluding the net assets of the ith fund; α and ε_(i,t): Constant and residual terms, respectively. Pi,t-1: Fund i’s quarterly excess return or raw return at quarter t-1. All fund and manager controls are lagged by 1 period to exclude their potentially current effect on fund flows.
Sample funds are ranked into three volatility ranks: low (VL), volatility (VM) and high (VH), based on their riskiness calculated using Equation (2). VL captures the bottom third, and VM and VH the middle and top thirds. The volatility rank takes the value of 1 if the fund falls in that rank, 0 otherwise. Two interaction terms between volatility, past quarterly performance and volatility rank dummies: (V_(i,t-1) P_(i,t-1) 〖VM〗_(i,t-1)and V_(i,t-1) P_(i,t-1) 〖VH〗_(i,t-1)) are included. FEEi,t-1:Fundi’s fee charges, which includes of front-end load (FLi,t-1), back-end load (BLi,t-1), 12b_1 expense (Bi,t-1) and pure operating expense (Oi,t-1).FEEi,t-1 represents quantitative variables.
To fulfill a more comprehensive understanding of managerial structure’s effect on fund flow, we hypothesize that team size are also important to fund flow. This conjecture is inspired by Patel and Sarkissian (2017)’s paper, which find that that largest gains in risk-adjusted performance are observed among funds with 3 managers. Therefore, we run the following regression model with a more, substituting team dummy variable with more specific team size:
- 〖Netflow〗_it= α+ β_1 〖2FM〗_(i,t-1)+β_2 〖3FM〗_(i,t-1)+β_3 〖4FM〗_(i,t-1)+β_4 〖5FM〗_(i,t-1)+β_5 V_(i,t-1)+β_6 〖TE〗_(i,t-1)+β_7 〖FL〗_(i,t-1)+β_8 〖BL〗_(i,t-1)+β_9 B_(i,t-1)+β_10 LOW+β_11 MID+β_12 HIGH+β_13 〖Log(TNA〗_(i,t-1))+β_14 〖Log(Age〗_(i,t-1))+β_15 〖Log(Familysize〗_(i,t-1))+〖β_16 V_(i,t-1) P_(i,t-1) 〖VM〗_(i,t-1)+β_17 V_(i,t-1) P_(i,t-1) 〖VH〗_(i,t-1)+β_n 〖FEE〗_(i,t-1) P_(i,t-1)+ε〗_(i,t) (8)
where 2FMi,t-1 , 3FMi,t-1 , 4FMi,t-1 , and 5FMi,t-1 are dummies that equal 1 if the fund has 2 managers, 3 managers, 4 managers, and 5 or more managers, respectively, at the end of the previous calendar year and 0 otherwise. Other variables are defined as before.
Institutional investor and retail investor (optional)
Previous studies also show that the money flows experienced by retail and institutional funds are likely to vary because of the different markets these sectors serve. Karceski and James (2002) find that while a difference in the performance of retail and institutional funds exists, institutional fund flows are less sensitive to performance than are retail fund flows. Further, they find that the lack of a flow-performance linkage in the institutional funds can be explained by the more sophisticated performance measures that these groups of investors implement. Given that institutional investors are thought to be more professional, experienced, and sophisticated than retail investors, we expect the former to be less likely fooled by performance manipulation. The following equation shows the basic model with manipulation dummy and institutional holdings proportion:
- 〖Netflow〗_it= α+ β_1 〖TEAM〗_(i,t-1)+β_2 〖TEAM〗_(i,t-1)*〖Lnshld〗_(i,t-1)+β_3 V_(i,t-1)+β_4 〖TE〗_(i,t-1)+β_5 〖FL〗_(i,t-1)+β_6 〖BL〗_(i,t-1)+β_7 B_(i,t-1)+β_8 LOW+β_9 MID+β_10 HIGH+β_11 〖Log(TNA〗_(i,t-1))+β_12 〖Log(Age〗_(i,t-1))+β_13 〖Log(Familysize〗_(i,t-1))+〖β_14 V_(i,t-1) P_(i,t-1) 〖VM〗_(i,t-1)+β_15 V_(i,t-1) P_(i,t-1) 〖VH〗_(i,t-1)+β_16 〖FEE〗_(i,t-1) P_(i,t-1)+ε〗_(i,t)
where 〖Lnshld〗_(i,t-1) is the proportion of institutional holdings of fund i at time t-1. We further incorporate 〖2FM〗_(i,t-1),〖3FM〗_(i,t-1),〖4FM〗_(i,t-1),5〖FM〗_(i,t-1) to replace 〖TEAM〗_(i,t-1).
Managerial structure
Fund complex structure
This figure depicts a standardized mutual fund complex system. Within this governance structure, two relationships of decision delegation are prominently displayed. First, portfolio managers are assigned by the management company under the delegation of shareholders. Second, portfolio managers make decision on the composition of investment portfolios under the delegation of management company. Under these two channels, management companies or fund families may choose single or multiple managers in responsible for the portfolio choice decisions.
Risk-taking behaviour: Single manager versus multiple managers
Besides the possible reason for the employment of multiple managers to provide a stable management (e.g. if a manager A leaves the job, manager B can still run the fund), the primary motive for employing more than one manager is to make the portfolio choice decision diversified over the style and judgment of the managers. The diversification of style are most commonly exhibited in team management – where there are multiple individuals who manage the fund together and this style refers that a fund is divided into several sub-accounts that are allocated to different managers who manage their sub-accounts independently. The final investment decision is the result of the aggregation of analyses of the management team rather than of a single manager. Often, it is hard to distinguish between the diversification of style and judgment. For instance, in team management, each team member might specialize in specific sectors so their decisions may be relatively independent. Furthermore, in the case where multiple managers manage different sub-accounts of a fund, they might analyze different, but not completely diverse, subsets of securities, and they would communicate with each other when making an investment decision. Sharpe (1980) first proposed some theoretical justification for the decentralization of investment management. He argued that the employment of multiple managers could reduce the danger of overall fund performance being damaged by the serious decision errors of a single manager. In a follow-up paper, Barry and Starks (1984) proposed an alternative motivation for the employment of multi-managers. They argued that due to risk-sharing arrangements between multiple managers, investors might benefit from the higher risk taken by multiple managers. This section focuses on the analysis of how multiple-manager arrangements affect the risk-taking behaviour of a mutual fund. There are at least two reasons to believe that the number of managers will affect a manager’s ability and willingness to alter the risk of her portfolios. First, a manager who is solely responsible for an investment decision usually is well recognized in the industry. It is unlikely that a person who has been in the industry for a few years only or who has had a poor record could become the sole manager of a fund. The termination risk for these individuals might not be a serious concern for them. Therefore, mid-year loser funds managed by single managers may have a greater incentive to take on higher risk. Second, while a multiple-manager fund, which is a mid-year winner fund, would try to become the top fund by adjusting its portfolio risk, it may be unable to make the adjustment in a timely manner because its managers have to coordinate their decisions. Thus, we would expect that single managers have greater incentives and ability to alter the risk of their portfolios to a larger degree relative to the multiple managers. Loser single managers will be even more aggressive than loser multiple managers. To test this hypothesis, we first estimate the following parametric model,
- σ_(i,t)= 〖RPM〗_(i,t)+ β_1 σ_(i,t) + β_2 〖AGE〗_(i,t) +β_3 〖Log(SIZE)〗_(i,t) +β_4 〖LG〗_i+ β_5 〖GI〗_i+β_6 〖MULTI〗_(i,t)+β_7 〖MULTI〗_(i,t)*〖RPM〗_(i,t) ∑_(t=1992)^2017▒μ_1 YEARUM_T + ε_(i,t),
where RPMi,t is the relative performance of fund managers i at time t adopting Carhart (1997)’s four-factor model, which is defined in Equation (5). The σi,t is the standard deviation of the return of fund i in period j of the year t. AGEi,t is the age of fund i at time t. 〖Log(SIZE)〗_(i,t) is the logarithm of fund i’s total net assets in period 1 in year t. LGi is a dummy variable which is equal to 1 if fund i’s objective is long-term growth and 0 otherwise. GIi is also a dummy variable which is equal to 1 if fund i’s objective is growth and income and 0 otherwise. In order to examine the difference in risk-taking behaviour of single and multiple managers, we include a dummy variable MULTIi,t which takes the value of one if fund i is managed by multiple managers in year t and zero otherwise. To allow for relative performance sensitivity of adjustment of risk to be different between single managers and multiple managers, we also include an interaction term between RPMit and the dummy variable MULTIi,t. YEARDUMt is the dummy variable which is equal to 1 if the year is year t and 0 otherwise.
Manager turnover
Turnover and fund flow prediction
The preliminary condition for our second hypothesis is that whether turnover has any marginal explanatory power to predict future flow. Kostovetsky and Warner (2015)’ paper use predictive model and find that turnover is associated with improved flow for poor performing funds, suggesting that investors not only pay attention to past returns but also to management changes. Thus, turnover in response to prior poor performance benefits investors, even though the underlying mechanism is not improved return performance. The finding that investor flow responds to manager changes is consistent with evidence presented elsewhere. For example, Massa, Reuter, and Zitzewitz (2010) show that flow falls when the manager of a good performing fund departs. Although flow may largely reflect irrational return chasing, it would not be surprising to also find that such irrational investors pay attention to manager changes. Anecdotal evidence also supports the plausibility of the view that investors pay attention to mutual fund manager changes, which lead to flow fluctuation. Morningstar sometimes has articles about specific changes, and their analysts give facts and opinions about both departing managers and their replacements. Furthermore, changes in Morningstar fund ratings predict fund flow (Del Guercio and Tkac (2008)), so what Morningstar says appears to influence some investors.
Model of multiple managers in manager turnover and fund flow relationship
Similar to a traditional event study of stock returns, an event study on fund flow aims to parsimoniously purge raw fund flow of the influence of all performance and non-performance characteristics other than the manager turnover and thereby isolate the incremental flow due to the managerial structure change from sole to team. To compute flow we estimate a time-series benchmark regression for each individual fund i, in which we take flow and performance control variables into account in the previous period.
- F_(i,t)=γ+ β_1 〖SF〗_(i,t) +β_2 〖RET〗_(i,t) + β_3 R_(i,t-1)+β_4 〖turnover〗_(t-1)*team+ β_5 〖turnover〗_(t-1)+β_6 team+β_7 F_(t-1)+ε_t,
where Fi,t is the net dollar flow to fund i at month t, RETi,t-1 is fund i’s raw return in month t-1. SFi,t is the aggregate net flow to all funds in the same style category as fund i at month t, Ri,t−1 is fund i’s performance at month t calculated by using Carhart (1997)’s four factor model, and Ft−1 is the net flow to fund i at month t − 1. Team is a dummy variable equals to 1 if fund i has a multiple manager structure such as 2, 3, 4 or more managers take charge of portfolio investment decisions, 0 if fund i is managed by sole manager. The coefficient on turnover, β_5, capture the effect of manager turnover event to fund flow for fund i at month t-1. The coefficient β_4 captures the changes in the metrics that is uniquely associated with the effect of team managerial structure in the event of manager turnover to investors’ perception reflected in fund flow changes. In accordance with our hypothesis two, we expect that if managerial structure has an effect to fund flow when there is managerial turnover, the coefficient β_4 will have a significant negative value.
Team size’s effect on manager turnover event
Since previous study shows that there is a relationship between team size and fund performance. For instance, Mueller (2012) documents that smaller teams display superior performance than larger teams. They further provide evidence that any group composed of 4 or more individuals will see significant increase in coordination costs within the group and diminishing motivation across members of the group. Hamilton et al. (2003) also present a nonlinear benefits of team size, finding that the largest increases in productivity of workers when they join the teams at the early stages of team formation. A profound study of Laughlin et al. (2006) find that when dealing with highly intellective problems, 3-person groups performs better than the best individuals, but more members do not add extra performance gains. In a similar vein, it is reasonable to assume that team size has a nonlinear effect on fund flow when there is manager turnover. Now we examine this effect by running the following regression model:
- F_(i,t)=γ+ β_1 〖SF〗_(i,t) +β_2 〖RET〗_(i,t) + β_3 R_(i,t-1)+β_4 〖turnover〗_(t-1)*team+ β_5 〖turnover〗_(t-1)+β_6 2FM+〖β_6 3FM+β_6 4FM+β_6 5FM+β〗_7 F_(t-1)+ε_t,
Where 2FMit, 3FMit, 4FMit, and 5FMit are dummies that equal 1 if the fund has 2, 3, 4, and 5 or more managers, respectively, at the end of the calendar year and 0 otherwise. Other variables are defined as before.