The Effects of the Organizational Structure on Asset Management Massimo Massa * Lei Zhang ** Abstract: We study how the strategies and performance of an asset management company are affected by its internal organizational structure. Relying on Stein’s (2002) theory of organizations, we argue that a more hierarchical structure reduces the incentives to collect “soft” information and to engage in proximity investment. This should lower portfolio concentration, increase managerial herding and reduce performance. We use information on the organizational structure of all the US mutual funds and insurance-managed funds investing in US corporate bonds. We show that more hierarchical structures invest less in firms located close to them and deliver lower performance. An additional layer in the hierarchical structure reduces the average performance by 24 basis points per month. At the same time, more hierarchical structures tend to herd more and to hold less concentrated portfolios. We also find that changes in fund structure quickly find their way into the behavior of fund managers. Overall, the organizational structure affects performance slightly more for mutual funds than insurance-managed funds, while it impacts proximity investment, herding and portfolio concentration more for insurance-managed funds than mutual funds. JEL classification: G23, G30, G32 Keywords: mutual funds, organization structure, performance, herding, proximity investment. ∗ Finance Department, INSEAD. Please address all correspondence to Massimo Massa, INSEAD, Boulevard de Constance, 77300 Fontainebleau France, Tel: +33160724481, Fax: +33160724045 Email: [email protected]. We thank an INQUIRE EUROPE Grant to make this research possible.
44
Embed
The Effects of the Organizational Structure on Asset ...faculty.insead.edu/massa/Research/FundStructure3new.pdf · The Effects of the Organizational Structure on ... The Effects of
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Effects of the Organizational Structure
on Asset Management
Massimo Massa * Lei Zhang **
Abstract:
We study how the strategies and performance of an asset management company are affected by its internal
organizational structure. Relying on Stein’s (2002) theory of organizations, we argue that a more hierarchical
structure reduces the incentives to collect “soft” information and to engage in proximity investment. This
should lower portfolio concentration, increase managerial herding and reduce performance. We use
information on the organizational structure of all the US mutual funds and insurance-managed funds
investing in US corporate bonds. We show that more hierarchical structures invest less in firms located close
to them and deliver lower performance. An additional layer in the hierarchical structure reduces the average
performance by 24 basis points per month. At the same time, more hierarchical structures tend to herd more
and to hold less concentrated portfolios. We also find that changes in fund structure quickly find their way
into the behavior of fund managers. Overall, the organizational structure affects performance slightly more
for mutual funds than insurance-managed funds, while it impacts proximity investment, herding and
portfolio concentration more for insurance-managed funds than mutual funds.
∗ Finance Department, INSEAD. Please address all correspondence to Massimo Massa, INSEAD, Boulevard de Constance, 77300 Fontainebleau France, Tel: +33160724481, Fax: +33160724045 Email: [email protected]. We thank an INQUIRE EUROPE Grant to make this research possible.
1
The Effects of the Organizational Structure
on Asset Management
Abstract:
We study how the strategies and performance of an asset management company are affected by its internal
organizational structure. Relying on Stein’s (2002) theory of organizations, we argue that a more hierarchical
structure reduces the incentives to collect “soft” information and to engage in proximity investment. This
should lower portfolio concentration, increase managerial herding and reduce performance. We use
information on the organizational structure of all the US mutual funds and insurance-managed funds
investing in US corporate bonds. We show that more hierarchical structures invest less in firms located close
to them and deliver lower performance. An additional layer in the hierarchical structure reduces the average
performance by 24 basis points per month. At the same time, more hierarchical structures tend to herd more
and to hold less concentrated portfolios. We also find that changes in fund structure quickly find their way
into the behavior of fund managers. Overall, the organizational structure affects performance slightly more
for mutual funds than insurance-managed funds, while it impacts proximity investment, herding and
portfolio concentration more for insurance-managed funds than mutual funds.
where ( ilat , ilon ), ( jlat , jlon ) are the (latitude, longitude) for fund i and bond issuer j in radian
degrees.2
We start with some univariate analysis. We report the results in Table II, Panel A. We break
down the sample into 4 different levels of hierarchy: from the lowest (1 layer) to the highest (4
layers). We then report the sample mean of fund portfolio distance at different levels of fund
hierarchy. The number of observations appears in parenthesis. We report the results for the
original sample and the two matching samples described earlier. We also provide univariate tests of
fund portfolio distance regarding to single vs. multi- fund hierarchy. Multi-hierarchy means the
number of fund hierarchies is greater than 1. The results show a monotonic increase in fund
portfolio distance as the number of layers increases. This holds regardless of the sample. A four-
layer fund tends to invest in bonds of firms on average 240 (175 and 210) km further away than a
one-layer fund in the case of the overall sample (sample based on matching within family and
sample based on matching across families).
We now move on to the multivariate analysis. We estimate:
titititi XHierarchyDis ,1,,, εδβα +×+×+= − , (1)
2 Information on bond issuer locations is from Compustat and SDC global new issue database. Since Lipper only provides county information of the managing firm, we use the location of the managing firm as the fund location. The county level coordinates (latitude, longitude) are from the Gazetteer Files of Census 2000.
12
where tiDis , represents the fund-bond distance of fund i at quarter t, t,iHierarchy is fund
hierarchy and 1, −tiX is the vector made of the other control variables defined above. We add the
fund type dummies across all specifications.
We report the results in Table II, Panel B for the entire sample. In Panels C and D, the sample
is based on the matching sample within and across fund families respectively. We include funds
owned by life insurance companies, mutual funds, property insurance companies and other
institutions (annuities and pension funds). Column (1) reports the results from an OLS regression
with standard errors clustered at fund level.
To address the possible endogeneity of fund structures, in Column (2), we implement an IV
regression, where family level structures are chosen as instruments. We instrument fund structure
variables using the following variables: family hierarchy (median of fund hierarchy within a
family), family employee specialty (median of employee specialty within a family), family team
(median of team dummy within a family), the interaction of family hierarchy with a financial
center dummy and the interaction of family employee specialty with a financial center dummy.
This latter variable resembles the instrument used by Chen et al. (2005) for the degree of
outsourcing. The intuition is the following. We know that the location in a financial center will
have a direct impact on proximity investment and performance, while the interaction with the
structure of the family needs not be so. At the same time, the interaction with the structure of the
family helps to explain the structure of the fund. Indeed, the incentive of the management family
to include many layers or many different areas of specialty depends on the availability of people.
Availability is higher in financial centers than in rural areas. Therefore, the desire of the family to
set up a specific structure is constrained by the location of the fund.
(Unreported) results show that the instruments help explain the organizational structure. Also,
they do not affect the dependent variable in the second stage through a channel different from the
impact on the instrumented variable. At the bottom of each IV specification we report the
Hansen’s J statistic (p-value). It always fails to reject the null, providing evidence for the quality of
our instruments.
As additional robustness check, In Column (3), we provide Fama-Macbeth (1973) estimates at
the fund level, while in Column (4), we provide the results of Fama-Macbeth estimates at the
family level. That is, we have first calculated family averages of all the variables. Column (5) and
(6) are estimated in the same way as in Column (3), but the sample is based on the funds owned
by life insurance companies and mutual fund families.
The results indicate that there is a strong positive relation between the average distance of the
firms in which the fund invests and fund hierarchy. This holds across the different specifications as
13
well as for different sub-samples. It also appears that the impact of the organizational structure on
insurance firms is higher than mutual funds.
The results are not only statistically significant but also economically relevant. An increase of
one layer in hierarchy raises the average distance (holding weighted distance) of the firms in which
the fund invests by 6% (Column (2) of Panel B). This supports our first hypothesis (H1), showing
that proximity investment is directly affected by the type of structure of the fund. If we consider
the other variables, we see that being managed by a team or by a sole manager does not affect the
decision to invest in closer firms. This suggests that our structure variables do not just proxy for
the mere fact that a fund is team-managed.
The other control variables are consistent with intuition. Being located in a financial center
increases the investment in closer firms. The same is true in the case the fund is more risk-
conscious and restricts itself to high-grade bonds. In the latter case, high risk prudence causes
funds to shorten their investment distances.
It is also interesting to note that funds that rotate their portfolio a lot (i.e., “high-turnover”
funds) are more likely to invest further away. This can be explained with the higher liquidity need
of these funds, not easy to meet in a more limited local area. There is scarce evidence in favor of an
impact of the degree of employee specialty.
As a further robustness check, for each fund, no matter whether it is single-hierarchy or multi-
hierarchy, we match it with some other fund similar in type, geographical location and size, but
different in terms of fund family and hierarchy. Then, we run regressions based on the differences
between the original fund and the matched fund. The idea is that funds located closely are more
likely to face a homogenous information set, and using differences, we can effectively cancel the
unobservable factors away. The matching procedure is as follows: for each fund-quarter we first
choose all the other funds of the same fund type but from different fund families and having
different fund hierarchy. Then we pick 20 funds located most closely and narrow them down to 10
according to similarity in fund size. From those 10 funds we select the final one with the smallest
geographical distances to the original fund. If there is more than one matched fund left meaning
that they are located at the same place, we choose the most similar one in terms of fund size. All
the variables except the financial center dummy, including both the dependent and independent
variables, are the differences between the original fund and its matched peer. The results are
reported in Panel E. We still find a strong positive relation between the difference in portfolio
distances and the difference in fund hierarchy.
Finally, it is worth mentioning that one possible criticism of our measure of distance is that it
measures the total distance that the fund in question resides from the bond issuers represented in
the portfolio. However, it may be that that it is not total distance that matters but how many
14
issuers are within a certain radius from this investor. We therefore also use an alternative approach
in which we define a radius (300 km) and close bonds are the ones of the firms located within the
radius and distant bonds the ones of the firms located outside the radius. We then assign a value of
0 to the close bonds and 1 to the distant ones. The (unreported) results based on this methodology
are consistent with the reported ones.
Overall our findings hold across different specifications as well as for different sub-samples.
They also hold in the specification based on “differenced” variables with another similar fund
located close-by. These findings show that proximity investment is directly affected by the type of
structure of the fund. We now move on to see if fund hierarchy affects other portfolio strategies
such as portfolio concentration and herding.
B. Organizational Structure and Portfolio Concentration
We now consider the impact of the organizational structure on portfolio concentration (H2). We
know that the availability of more information induces the manager to tilt the portfolio towards
the set of few assets over which he has more information (e.g., Kacperczyk et al., 2005). We
therefore expect that a more vertical hierarchy, by reducing the incentive to collect soft
information, reduces the degree of portfolio concentration. Restriction H2 posits that funds
characterized by a more hierarchical structure should display a lower degree of portfolio
concentration.
As in the previous analysis, we start with some univariate statistics. First, we define our
measure of portfolio concentration: t,iHerfin . It captures the degree of portfolio concentration in
bonds of fund i at quarter t. If we denote the set of bond issues held by fund i by Q and tjiw ,, be
the fraction invested in bond issue j, fund herfindahl is ∑∈
=Qj
2t,j,it,i wHerfin . We then break down
the sample into 4 different levels of hierarchy: from the lowest (1 layer) to the highest (4 layers).
In Table III, Panel A, we report the sample mean of fund portfolio concentration. We report
the results for the original sample and the two matching samples described earlier as well as tests
of fund portfolio differences in portfolio concentration between single and multi- fund hierarchy.
The results show a monotonic decrease in portfolio concentration as the number of layers increases.
This holds regardless of the sample. A four-layer fund tends to have a degree of concentration
equal to just 26% (41% and 37%) of the concentration of a one-layer fund in the case of the overall
sample (matching within family and matching across families).
We then employ a multivariate specification and estimate:
where t,iHerfin is the degree of portfolio concentration in bonds of fund i at quarter t, tiHierachy ,
is fund hierarchy and 1t,iX − is the vector of control variables as defined above.
We report the results in Table III, Panel B for the entire sample. In Panels C and D, the
sample is based on the matching sample within and across fund families respectively and with the
same specifications as in Panel B. Column (1) reports the results from an OLS regression with
standard errors clustered at the fund level. In Column (2), we report the results of an IV
regression, where family level structures are chosen as instruments. 3 The standard errors are
clustered at the fund level. In Column (3), we provide Fama-Macbeth (1973) estimates at the fund
level, while in Column (4), we provide the results of Fama-Macbeth estimates at the family level.
Column (5) and (6) are estimated in the same way as in Column (3), but they are restricted to the
funds managed by life insurance companies and mutual fund families. As in the previous
specifications, in Panel E, the regressions are based on the difference in fund concentration and the
difference in fund hierarchy between the original fund and another similar fund located close-by.
The matching procedure is the same as described before.
The results show a strong and consistent negative correlation between hierarchy and portfolio
concentration. This holds across the different specifications (OLS, IV and Fama-MacBeth) as well
as for the different sub-samples. An increase of one layer in the hierarchy reduces concentration by
48% (Column (2) of Panel B). The analysis based on the matching sample delivers consistent
results. These findings support H2 and are consistent with the previous ones on proximity
investment. They confirm an overall picture in which a more hierarchical structure reduces soft
information collection. It is also interesting to note a strong negative correlation between the
degree of employee specialty and portfolio concentration. This also holds across the different
specifications (OLS, IV and Fama-MacBeth). An increase of one layer in the degree of specialty
reduces concentration by 6% (Column (2) of Panel B).
Overall our findings show that the structure of the fund affects the degree of portfolio
concentration. We now move on to herding.
C. Organizational Structure and Herding
We argued that a higher hierarchy would stifle fund managers’ incentive to collect soft information
and would induce them to invest more in line with their peers (H3). To address this issue, we
study the relation between managerial herding and fund structure.
3 We instrument the proxies for fund structure using the following variables: family hierarchy (median of fund hierarchy within a family), family employee specialty (median of employee specialty within a family), family team (median of team dummy within a family), the interaction of family hierarchy with financial center dummy and the interaction of family employee specialty with financial center dummy.
16
We define a variable (Herdingi,t) which proxies for the tendency of a fund to follow the trading
behavior of its peers or to go against it. We employ the same methodology used by Lakonishok,
Shleifer and Vishny (1992) and Grinblatt, Titman and Wermers (1995). A detailed description of
the construction of this variable is reported in the section Variables Definitions at the end of the
paper. In Panel B of Table I, we report some descriptive statistics of the measure of herding. The
mean in our sample is 1.3%. It is higher than the findings of 0.84% by Grinblatt, Titman and
Wermers (1995). This can be seen as evidence that bond funds herd more with each other than
equity funds.
We start with some univariate analysis. We break down the sample into 4 different levels of
hierarchy: from the lowest (1 layer) to the highest (4 layers). We then report the sample mean of
fund herding. We also provide univariate tests of fund herding regarding to single vs. multi- fund
hierarchy. The results are reported in Table IV, Panel A. The results show a monotonic increase in
fund herding propensity as the number of layers increases. This holds regardless of the sample. A
four-layer fund tends to herd on average 51% (33% and 53%) more than a one-layer fund in the
case of the overall sample (matching within family and matching across families).
We then move on to the multivariate analysis and estimate:
where Herdingi,t represents the herding measure as defined above of fund i at quarter t,
t,iHierarchy is fund hierarchy and 1, −tiX are other control variables. The other variables are
defined as before. We include fund type dummies in all the specifications.
We report the results in Table IV, Panel B for the entire sample. In Panels C and D, the
sample is based on the matching sample within and across fund families respectively and with the
same specifications as in Panel B. In Panel E, we report the results of the specification based on
differences. We include funds owned by life insurance companies, mutual funds, property insurance
companies and other institutions (annuities and pension funds). The layout of the columns is the
same as in the previous analysis.
The results show a positive relation between hierarchy and herding. This holds across the
different specifications as well as for different sub-samples. An increase of one layer in the
hierarchy raises herding by 16% (Column (2) of Panel B). This is in line with H3. Hierarchy, by
reducing the incentives to collect soft information, translates in more herding.
It is also interesting to note a negative relation between herding and the degree of employee
specialty. An increase of one layer in the degree of employee specialty reduces herding by 12%
(Column (2) of Panel B). In the case of employee specialty, the effect of reduction of risk taking is
17
more than offset by the availability of more areas of specialty that increases the ability to beat the
peers and therefore discourages herding. This would be consistent with the fund relying more on its
private information because it has more areas of specialty available.
D. Organizational Structure and Performance
We now focus on performance. Restriction H4 posits that funds characterized by a less hierarchical
structure deliver higher performance. To address this issue, we study the relation between
performance and fund structure.
We start by defining the measure of performance ( t,iAlpha ) of fund i in month t. It is
constructed in the following way. First, for each fund-month (i,t), we estimate the monthly factor
loadings by running the following regression:
,UMDuHMLhSMBs)rr(mCVc
RSrTSt)rr(barr
s,is1t,is1t,is1t,is,rfs,m1t,is1t,i
s1t,is1t,is,rfs,dj1t,i1t,is,rfs,i
ε++++−+
+++−+=−
−−−−−
−−−−(4)
where 1ts30t −≤<− and we require a minimum of 25 observations for each regression. The
dependent variable is the monthly return of fund i in month s less the risk-free rate sfr , . The
independent variables include 8 factors: the excess return of Dow Jones Corporate Bond Index over
the risk-free rate ( s,fs,dj rr − ), the yield difference between the twenty-year constant maturity
treasury bonds and the two-year constant maturity treasury bonds (gs20-gs2, term spread, TS)
(Colin-Dufresne et. al., 2001), the yield difference between Moody’s BAA corporate bond index and
the thirty year constant maturity treasury bonds (Baa-gs30, risk spread, RS)4, the yield difference
between the five year constant maturity treasury bonds and the average yield of two year and ten
year constant maturity treasury bonds (gs5-(gs2+gs10)/2, curvature spread, CV), the excess return
of market return over the risk-free rate ( sfsm rr ,, − ), the return difference between small and large
capitalization stocks (SMB), the return difference between high and low book-to-market stocks
(HML), the return difference between stocks with high and low past returns (UMD). The return
data on Dow Jones Corporate Bond Index is from Dow Jones’ website. The data on treasury bond
yields and Moody’s Baa corporate bond yields are from the FRED database at the Federal Reserve
Bank of Saint Louis. The data on risk-free rate, market return, SMB, HML and UMD are from
Kenneth French’s website.
Then, using the estimated loadings, we calculate fund alpha in month t by:
4 Moody's includes bonds with remaining maturities as close as possible to 30 years. Moody’s drops bonds if the remaining life falls below 20 years, if the bond is susceptible to redemption, or if the rating changes.
18
.UMDuHMLhSMBs)rr(mCVc
RSrTSt)rr(brr
t1t,it1t,it1t,it,rft,m1t,it1t,i
t1t,it1t,it,rft,dj1t,it,rft,it,i
−−−−−
−−−
−−−−−
−−−−−−=α (5)
We start with some univariate analysis. We break down the sample into 4 different levels of
hierarchy: from the lowest (1 layer) to the highest (4 layers). We then report the sample mean of
fund alpha. We also provide univariate tests of difference in performance between single and multi-
fund hierarchy. The results are reported in Table V, Panel A. They show a monotonic decrease in
fund performance as the number of layers increases. This holds regardless of the sample. A four-
layer fund has a performance 41 bp (39 bp and 38 bp) lower than a one-layer fund in the case of
the overall sample (matching within family and matching across families).
We them move on to the multivariate analysis and estimate:
where t,imentFundManage∆ represents the change in fund behavior (portfolio distance, herding,
and portfolio concentration) as well as the change of fund performance. We consider two measures
of performance: the raw return (cumulative, quarterly) of the fund and the change of fund alpha
(cumulative, quarterly) respectively from quarter t -1 to quarter t. t,iHierarchy∆ is the change of
fund hierarchy and tiX ,∆ are the changes of other control variables from quarter t-1 to quarter t. A
more detailed definition of these variables is reported in the Appendix. 1,imentFundManage − is the
lagged dependent variable at quarter t-1. The standard errors are clustered at fund level and we
always include time dummies and fund type (e.g., life insurance, property insurance, ..) dummies.
The results are reported in Table VI. They are consistent with the previous ones. An increase
in the degree of hierarchy reduces proximity investment and portfolio concentration, while it raises
fund herding. Overall, this implies a lower performance. These findings are not only statistically
significant, but also economically relevant. An additional layer lowers the average distance from
firms whose bonds it invests in by 25 km, reduces portfolio concentration by 5% and raises fund
herding by 7%. It reduces fund performance by 14 bp in the case of raw returns and 23 bp in the
case of alpha (quarterly). These results provide an additional robustness check. They also show
that changes in the fund structure quickly find their way into the behavior of the fund managers.
V. Conclusion
We study how the internal organizational structure affects fund’s strategies and performance. We
focus mostly on mutual funds and insurance-managed funds. We argue that a more hierarchical
structure reduces the incentives to collect “soft” information and proximity investment. This
reduces the incentive to concentrate the investment in few bonds and makes the manager more
likely to herd. The net result is lower performance.
21
We show that that funds with more hierarchical structures tend to invest less in firms located
close to the funds. This has a direct negative effect on fund performance: more vertical structures
are characterized by worse performance. Funds with a more vertical structure tend to herd more
with the other funds and to hold less concentrated portfolios. We also find that changes in the fund
structure quickly find their way into the behavior of the fund managers.
These findings are consistent with Stein’s (2002) theory of organizations. They also indicate
that the organizational structure is an important determinant of fund strategies and performance.
22
References
Bär, M., A. Kempf, and S.Ruenzi, 2005, Team Management and Mutual Funds, Working Paper.
Berger, A.N. N.H., Miller, M.A. Petersen, R.G. Rajan, and J.C. Stein, 2005, Does Function Follow Organizational Form? Evidence from the Lending Practices of Large and Small Banks Journal of Financial Economics, 76(2), 237-69.
Brown,S.J., and W.N.Goetzmann, 1995, Performance Persistence, Journal of Finance, 50, 679-698.
Carhart, M., 1997, On Persistence in Mutual Fund Performance, Journal of Finance, 52, 57-82.
Chen, J., H. Hong, M. Huang, and J. Kubik, 2004, Does Fund Size Erode Mutual Fund Performance? The Role of Liquidity and Organization, American Economic Review, 94, 1276-1302.
Chen, J., H. Hong, M. Huang, and J. Kubik, 2005, Outsourcing Mutual Fund Management: Firm Boundaries, Incentives and Performance, Working Paper.
Chevalier, J., and G. Ellison, 1997, Risk Taking by Mutual Funds as a Response to Incentives, Journal of Political Economy, 105, 1167-1200.
Colin-Dufresne,P.,R.Goldstein, andJ.S. Martin, 2001. The Determinants of Credit Spread Changes. Journal of Finance 56, 2177—2207.
Coval, J.D. and T. Moskowitz, 1999, Home bias at home: Local equity preference in domestic portfolios, Journal of Finance 54, 2045-2073.
Coval, J.D. and T. Moskowitz, 2001, The geography of investment: Informed trading and asset prices, Journal of Political Economy, 109, 811-841.
Del Guercio, D., J. Reuter, and P.A. Tkac, 2007, Why Do Some Mutual Fund Families Pick Stocks for the Competition? Working Paper.
Elton, E.J., M.J. Gruber, and C.R. Blake, 1996, The Persistence of Risk-Adjusted Mutual Fund Performance, Journal of Business, 69, 133-157.
Gaspar, J.M., M. Massa, and P. Matos, 2006, Favoritism in Mutual Fund Families? Evidence on Strategic Cross-Fund Subsidization, Journal of Finance, 61, 73-104.
Graham, J.R., 1999, Herding Among Investment Newsletters: Theory and Evidence. Journal of Finance, 54, 237-268.
Grinblatt, M., Titman, S., and R. Wermers, 1995, Momentum Investment Strategies, Portfolio Performance, and Herding: A Study of Mutual Fund Behavior, American Economic Review, 85, 1088-1105.
Guedj, I., and J. Papastaikoudi, 2004, Can Mutual Fund Families Affect the Performance of Their Funds?, Working Paper.
Hong, H., Kubik, J.D. and J.C. Stein, 2004, Thy neighbor’s portfolio: word-of-mouth effects in the holdings and trades of money managers, Journal of Finance, 60, 2801-24.
Hong, H., Kubik, J.D. and J.C. Stein, 2005, The only game in town. Stock-price consequences of local bias. Working Paper.
Kacperczyk, M., C. Sialm, and L. Zheng, 2006, Unobserved Actions of Mutual Funds, Review of Financial Studies, forthcoming.
Kempf, A. and S.Ruenzi, 2004, Tournaments in Mutual Fund Families, Working Paper.
Khorana, A and H. Servaes, 1999, The Determinants of Mutual Fund Starts, Review of Financial Studies 12, 1043—1074.
Lakonishok, J, A. Shleifer, and R.W.Vishny, 1992, The Impact of Institutional Trading on Stock Prices. Journal of Financial Economics. Vol. 32. 23-43.
Mamaysky, H., M. Spiegel, 2001, A Theory of Mutual Funds: Optimal Fund Objectives and Industry Organization, Working Paper.
Massa M., J. Reuter and E. Zitzewitz, 2005, The Rise of Anonymous Teams in Asset Management, Working Paper.
Massa M., A. Yasuda and L. Zhang, 2007, Institutional Investors, Credit Supply Uncertainty, and the Leverage of the Firm, Working Paper.
23
Nanda, V., Wang, J.Z., and L. Zheng, 2004, Family Values and the Star Phenomenon: Strategies of Mutual Fund Families, Review of Financial Studies, 17, 667-698.
Stein, J.C., 2002, Information Production and Capital Allocation: Decentralized versus Hierarchical Firms, Journal of Finance, 57, 1891-1921.
Wermers, R., 1999, Mutual Fund Herding and the Impact on Stock Prices, Journal of Finance, 54, 581-622.
24
Variable Definitions Fund Hierarchy For each fund we first define fund vertical layer according to the employee’s job title in the following way:
Then fund hierarchy is counted as the distinct number of vertical layers. It is possible that a person can be entitled with multiple job functions, so in the cases where the distinct number of vertical layers exceeds the number of employees, we use the number of employees as the fund hierarchy. Employee Specialty For each fund we first determine the number of specialties according to the employee’s area of focus. We mainly look at the following sectors:
Asset Backed Any Corporate Sector Any Country Chairman Corporate Any Municipal Sector Canada President Government Any Specialty Emerging Markets Chief Executive Officer Mortgage backed Canadian Dollar Euroland Chief Financial Officer Local/Regional Euro Any State/Terr. Chief Investment Officer US firms investing non-domestically
U.S. Dollar United States Bond Department Head
Combination of Above Combination of Above Non-domestic Fixed-Income Head Combination of Above Portfolio Manger-General Portfolio Manger-Balanced Portfolio Manger-Convertibles Portfolio Manger-Eq/Pref Stock Portfolio Manger-High Yield Portfolio Manger-Inv Grade Corp Bd Portfolio Manger-Pvt Placements Portfolio Manger-Short-Term/MM
The number of specialties is counted as the number of total combinations of the above four sectors. Then we define employee specialty as the number of specialties divided by the number of employees. Team Dummy We define a team dummy equals 1 if the fund has more than 1 portfolio managers and 0 otherwise.
25
Fund Portfolio Distance Fund portfolio distance measures the distance between the fund and its bond portfolios. If we denote the set of bond
issues held by fund i by Q and tjiw ,, be the fraction of fund i invested in bond issue j, the fund portfolio distance
where ( ilat , ilon ), ( jlat , jlon ) are the (latitude, longitude) for fund i and bond issue j in radian degrees.
Information on locations of bond issuers is obtained from Compustat and SDC global new issue database. Since
Lipper only provides county information of the managing firm, we utilize location of the managing firm as the
location of the fund. The county level coordinates (latitude, longitude) are obtained from the Gazetteer Files of
Census 2000.
Fund Portfolio Concentration Fund portfolio concentration represents the fund’s concentration ratio (herfindal) of its bond portfolio. If we denote
the set of bond issues held by fund i by Q and tjiw ,, be the fraction invested in bond issue j, fund portfolio
concentration is defined as:
∑∈
=Qj
tjiti wHerfin 2,,, .
Fund Herding Fund Herding represents the tendency of a fund to “follow the crowd or to go against it”. We follow the same
methodology used by Lakonishok, Shleifer and Vishny (1992) and Grinblatt, Titman and Wermers (1995). The first
step is to define a measure of investor herding at the bond level. Let tkB , ( tkS , ) be the number of funds buying
(selling) bond issue k at quarter t, then the herding measure (UHM) is expressed as:
|][||][| ,,,,, tktktktktk pEpEpEpUHM −−−= ,
where )/( ,,,, tktktktk SBBp += is the proportion of funds trading issue-quarter k, t which are buyers. We use the
proportion of all trades by funds that are purchases during quarter t to proxy for ][ ,tkpE . The first part represents
the “extra” number of funds trading a bond issue during a given quarter as the proportion of the total number of
funds buying that issue-quarter minus the expected proportion of buyers. The second term is an adjustment factor
allowing for random variation around the expected proportion of buyers under the null hypothesis of cross-sectional
independence among fund trades. The expectation in the second term is calculated by assuming that tkB , follows
a binomial distribution with parameter ( tktk SB ,, + ) and ][ ,tkpE .
The second step is to define the signed herding measure (SHM) which indicates the tendency of whether fund i is
following the crowd or going against it in trading bond k. This is calculated as:
][ ,,,,,,,, tktkitktkitki UHMIEUHMISHM ×−×= ,
where tkiI ,, is an indicator variable:
26
=0 if |][||][| ,,,, tktktktk pEpEpEp −<− ; tkiI ,, =1 if |][|][ ,,,, tktktktk pEpEpEp −>− and fund i is a buyer
of bond k, or if |][|])[( ,,,, tktktktk pEpEpEp −>−− and fund i is a seller of bond k; tkiI ,, =-1 if
|][|][ ,,,, tktktktk pEpEpEp −>− and fund i is a seller of bond k, or if |][|])[( ,,,, tktktktk pEpEpEp −>−−
and fund i is a buyer of bond k. Additionally, we impose the restriction that 0,, =tkiSHM is there are fewer than
5 funds traded bond k during quarter t. Under the assumption that the number of buyers of bond k is binomially
distributed, the expectation term ][ UHMIE × can be calculated by the following formula:
∑∑−>−−−>−
−−−=×|][|)(:
,|][|:
,,,,,
)Pr()()12()Pr()()12(][tktktktk ppEppp
tkppEppp
tk ppUHMpppUHMpUHMIE ,
where Pr(p) is the probability of ( tktk SB ,, + )p occurrences assuming a binomial distribution with parameter
( tktk SB ,, + ) and ][ ,tkpE .
Finally, if we denote the set of bond issues held by fund i by Q and the fraction invested in bond k by tkiw ,, , the
herding measure of fund i at quarter t is:
∑∈
−−=Qk
tkitkitkiti SHMwwHerding ,,1,,,,, )( .
Fund Return: Fund monthly return refers to the raw investment return from its bond portfolio. The data on bond returns are obtained from Bloomberg. Fund Quarterly return refers to cumulative monthly returns in each quarter. Fund Performance For each fund-month (i,t), we first estimate the monthly factor loadings by running the following regression:
,)()( ,1,1,1,,,1,1,1,1,,,1,1,
,,
sistististisrfsmtistististisrfsdjtiti
srfsi
UMDuHMLhSMBsrrmCVcRSrTStrrbarr
ε++++−++++−+
=−
−−−−−−−−−
where 130 −≤<− tst . We require a minimum of 25 observations for each regression. The dependent variable is
the monthly return of fund i in month s less the risk-free rate sfr , . The independent variables include 8 factors: the
excess return of Dow Jones Corporate Bond Index over the risk-free rate ( sfsdj rr ,, − ), the yield difference between
the twenty-year constant maturity treasury bonds and the two-year constant maturity treasury bonds (gs20-gs2,
term spread, TS), the yield difference between Moody’s BAA corporate bond index and the thirty year constant
maturity treasury bonds (Baa-gs30, risk spread, RS), the yield difference between the five year constant maturity
treasury bonds and the average yield of two year and ten year constant maturity treasury bonds (gs5-(gs2+gs10)/2,
curvature spread, CV), the excess return of market return over the risk-free rate ( sfsm rr ,, − ), the return difference
between small and large capitalization stocks (SMB), the return difference between high and low book-to-market
stocks (HML), the return difference between stocks with high and low past returns (UMD). The return data on Dow
Jones Corporate Bond Index are from the Dow Jones’ website. The data on Treasury Bond yields and Moody’s Baa
corporate bond yields are from the FRED database at the Federal Reserve Bank of Saint Louis. The data on
risk-free rate, market return, SMB, HML and UMD are obtained from Kenneth French’s website. Then fund alpha
at month t is estimated by the following equation:
27
.ˆˆˆ)(ˆˆˆˆ)(ˆ1,1,1,,,1,1,1,1,,,1,
,,,
ttittittitrftmtittittittitrftdjti
trftiti
UMDuHMLhSMBsrrmCVcRSrTStrrb
rr
−−−−−−−− −−−−−−−−−−
−−=α
Fund Return Volatility: For each fund at month t, we calculate return volatility as the standard deviation of monthly returns during the past 12 months. Fund Portfolio Maturity: logarithm of the value-weighted average maturity of all the bonds held by each fund. Fund Size: logarithm of the par-amount of bond holdings held by each fund Family Size: logarithm of the par-amount of bond holdings held by each fund family (managing firm) Number of funds: the number of funds in each fund family Fraction in Investment-grade bonds: the fraction of bond portfolio invested in investment-grade bonds with S&P’s bond rating not lower than BBB. Financial Center Dummy: dummy variable taking a value of 1 if the fund is located at either of the following cities: New York, Chicago, Los Angeles, Boston and San Fransciso.
28
Table I Summary Statistics
This table presents summary statistics and univariate tests of the main variables used in the subsequent analysis.
Our primary database is Lipper’s eMAXX fixed income database. It contains information on quarterly bond
holdings of major U.S. insurance companies (life and property), mutual funds, annuities and pension funds from the
first quarter of 1998 to the second quarter of 2005. It also provides information on the fund employee’s job title,
market sector, credit sector and geographical focus which enables us to characterize the organizational structures.
Panel A: Summary Statistics of Fund Structure
Panel A reports the sample mean of fund hierarchy and employee specialty characterizing fund structures. The
detailed definitions can be found in the appendix. We separately report the results for mutual funds as well as funds
managed by life insurance companies, property insurance companies and other institutions (annuities and pension
funds). We construct two matching samples, one within fund family and one across fund families. The matching
procedure is performed as follows. The “matching within fund family” sample is constructed as follows. For each
multi-hierarchy fund, we first select another single hierarchy fund from the same fund family and most similar in
terms of fund size, then combine the matched single hierarchy funds with the original multi-hierarchy funds. The
“matching across fund families” sample is constructed similarly except that the matched single-hierarchy fund is
chosen from different fund families but belonging to the same fund type (mutual funds, insurance companies,
pension funds etc.). Panel A1 is based on the full sample, while Panel A2 and Panel A3 are for the matching sample
within and across fund families respectively. The number of observations (fund-quarter) is given in the parentheses.
Panel B: Summary Statistics of Fund Characteristics In Panel B we report the sample mean of fund characteristics with the number of observations (fund-quarter) given
in parenthesis. We separately report our results for mutual funds as well as funds managed by life insurance
companies, property insurance companies and other institutions (annuities and pension funds). The detailed
definition of each variable can be found in the Appendix.
(206) (189) (206) T-test: Multiple vs. Single Hierarchy 12.71*** 4.29*** 7.49*** Wilconxon Test: Multiple vs. Single Hierachy 11.55*** 3.00*** 6.34***
Panel B: Full Sample All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy 0.0384*** 0.0547*** 0.0381*** 0.0339*** 0.0610*** 0.0295*** (3.79) (3.97) (6.69) (5.06) (6.38) (3.73) Control Variables Employee Specialty 0.0067 0.0097 0.0065*** 0.0073*** -0.0056*** 0.0038 (1.09) (1.05) (3.90) (2.79) (-2.69) (1.16) Team Dummy -0.0132 -0.0025 -0.0134* 0.0265** -0.0465*** 0.0164 (-0.89) (-0.09) (-1.76) (2.03) (-3.08) (1.41) Fund Size 0.0272*** 0.0273*** 0.0261*** 0.0366*** 0.0207*** 0.0072*** (7.68) (7.70) (19.26) (10.86) (13.29) (4.11) Family Size -0.0076 -0.0072 -0.0066** -0.0069* -0.0064** 0.0041* (-1.60) (-1.51) (-2.31) (-1.89) (-2.07) (1.77) Portfolio Maturity 0.0067 0.0064 -0.0033 -0.0041 -0.0331*** -0.0342** (0.82) (0.78) (-0.52) (-0.30) (-3.12) (-2.13) Log(Number of Funds) -0.0083 -0.0082 -0.0102** 0.0080 -0.0333*** 0.0066 (-1.03) (-0.98) (-2.08) (1.39) (-5.79) (1.21) Fund Turnover 0.0277*** 0.0274*** 0.0209 0.0509** 0.0401*** 0.0045 (2.95) (2.91) (1.44) (2.63) (3.20) (0.32) Fund Return Volatility 0.5668 0.5650 0.1902 -1.0858 3.1276* -3.6036** (0.48) (0.48) (0.19) (-0.78) (1.85) (-2.40) Fund Return -0.1019 -0.0980 0.2944 0.3971 -0.2654 0.1410 (-0.92) (-0.89) (1.22) (1.25) (-0.48) (0.44) Fraction in Investment-grade Bonds -0.3158*** -0.3139*** -0.3153*** -0.3631*** -0.3720*** -0.3487*** (-13.65) (-13.49) (-20.64) (-16.75) (-10.50) (-21.06) Financial Center Dummy -0.0915*** -0.0905*** -0.0897*** -0.1200*** -0.0693*** -0.0981*** (-6.91) (-6.76) (-21.93) (-23.66) (-12.54) (-20.35) Const 6.8434*** 6.8128**** 6.8479*** 6.6748*** 6.8473*** 6.9682*** (107.03) (98.62) (277.11) (273.76) (134.52) (214.12) Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.30 - - - - (Average) R-squared 0.0791 0.0788 0.0862 0.0681 0.0674 0.0786 Number of observations 81342 81342 81342 24061 21245 18876
32
Panel C: Matching Within Family All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy 0.0267** 0.0601*** 0.0243*** 0.0298*** 0.0021 0.0419*** (2.07) (3.27) (3.70) (2.66) (0.19) (3.79) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.64 - - - - (Average) R-squared 0.1055 0.1023 0.1382 0.1795 0.1864 0.1479 Number of observations 11911 11911 11911 4893 3507 4338
Panel D: Matching Across Family All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy 0.0417*** 0.0624*** 0.0413*** 0.0498*** 0.0724*** 0.0395*** (3.89) (4.28) (5.77) (6.69) (6.73) (3.05) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.86 - - - - (Average) R-squared 0.0892 0.0880 0.1210 0.1255 0.1340 0.1300 Number of observations 16645 16645 16645 9147 4971 5962
Panel E: Regression based on Differenced Variable All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Difference in Fund Hierarchy 0.0394*** 0.0248*** 0.0366*** 0.0305*** 0.0334*** 0.0307*** (10.29) (4.32) (4.90) (3.54) (4.18) (3.27) Difference in Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.83 - - - - (Average) R-squared 0.0441 0.0425 0.0891 0.1275 0.0890 0.1569 Number of observations 61252 61252 61252 20886 17463 15721
33
Table III Fund Hierarchy and Portfolio Concentration
This table relates fund hierarchy to portfolio concentration. Panel A summarizes the sample mean of fund portfolio
concentration at different levels of fund hierarchy. We also provide univariate tests of fund portfolio distance
regarding to single vs. multi- fund hierarchy. Multi-hierarchy means the number of fund hierachies to be greater
than 1. We consider the overall sample as well as the “matching within fund family” and the “matching across fund
families”. We report the results for the full sample and the two matching samples separately. Both two tailed T-test
and Wilconxon rank-sum test are performed to test the differences. The number of observations (fund-quarter) is
given in the parentheses. In Panels B-E, we report the results of the multivariate analysis. We estimate:
titititi XHierarchyHerfin ,1,,, εδβα +×+×+= − ,
where tiHerfin , represents the portfolio herfindal of fund i at quarter t, tiHierarchy , is fund hierarchy and
1, −tiX are other control variables. The definitions are detailed in the appendix.
The analysis in Panel B is based on the full sample. We add the fund type dummies across all specifications.
Column (1) is based on OLS regressions with standard errors clustering at fund level. In Column (2), we report the
results of an IV regression, where family level structures are chosen as instruments. Specifically, we instrument fund
structure variables using the following variables: family hierarchy (median of fund hierarchy within a family),
family emplolyee specialty (median of employee specialty within a family), family team (median of team dummy
within a family), the interaction of family hierarchy with financial center dummy and the interaction of family
employee specialty with financial center dummy. Hansen’s J statistic (p-value) is reported to examine the quality of
instruments. The standard errors are clustered at fund level. Column (3) provides Fama-Macbeth (1973) estimates
at the fund level, while Column (4) provides the results of a Fama-Macbeth estimates at the family level. Column
(5) and (6) are estimated in the same way as in Column (3) but only based on funds owned by life insurance
companies and mutual fund families. Panel C and Panel D are based on the matching sample within and across fund
families respectively with the same specifications as in Panel B.
Panel E uses a “differenced” variable approach, where for each fund we match it with some other fund similar
in fund type, geographical location and size, but different in terms of fund family and fund hierarchy. The matching
procedure is as follows: for each fund-quarter we first choose all the other funds of the same fund type but from
different fund families and having different fund hierarchy. Then we pick 20 funds located most closely and narrow
them down to 10 according to similarity in fund size. From those 10 funds we select the final one with the smallest
geographical distances to the original fund. If there is more than one matched fund left meaning that they are
located at the same place, we choose the most similar one in terms of fund size. All the variables except the financial
center dummy, including both the dependent and independent variables, are the differences between the original
fund and its matched peer.
For the sake of brevity we only report the coefficients of fund hierarchy from Panel C to Panel E. ***, ** and
* represent significance levels at 1%, 5% and 10% respectively with t-statistics given in parentheses.
34
Table III (Cont’d)
Panel A: Univariate Results
Fund Portfolio Concentration by Fund Hierarchy Full Sample Matching Within
(213) (164) (178) T-test: Multiple vs. Single Hierarchy -27.87*** -14.44*** -17.38*** Wilconxon Test: Multiple vs. Single Hierachy -48.13*** -24.64*** -24.73***
Panel B: Full Sample All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy -0.0221*** -0.0329*** -0.0221*** -0.0272*** -0.0319*** -0.0063*** (-13.20) (-14.40) (-23.40) (-35.18) (-18.60) (-5.98) Control Variable Employee Specialty -0.0022*** -0.0042*** -0.0021*** -0.0045*** -0.0043*** 0.0007 (-3.24) (-4.17) (-11.19) (-16.90) (-12.25) (1.37) Team Dummy -0.0040* -0.0026 -0.0033*** -0.0042** -0.0022 -0.0079*** (-1.73) (-0.62) (-3.58) (-2.29) (-0.71) (-5.79) Fund Size -0.0236*** -0.0237*** -0.0234*** -0.0213*** -0.0219*** -0.0163*** (-38.44) (-38.60) (-69.54) (-23.41) (-58.32) (-22.08) Family Size 0.0034*** 0.0031*** 0.0033*** -0.0010 0.0042*** -0.0001 (4.60) (4.24) (15.48) (-1.34) (13.59) (-0.20) Portfolio Maturity -0.0093*** -0.0091*** -0.0140*** -0.0130*** -0.0118*** -0.0267*** (-7.45) (-7.32) (-8.88) (-6.97) (-4.83) (-10.91) Log(Number of Funds) -0.0024** -0.0028** -0.0021*** 0.0011 -0.0033*** 0.0029*** (-2.08) (-2.34) (-6.45) (1.38) (-5.44) (3.86) Fund Turnover 0.0062*** 0.0063*** 0.0073*** 0.0065** 0.0096** -0.0010 (4.33) (4.43) (3.60) (2.18) (2.40) (-0.58) Fund Return Volatility 0.2200 0.2238 1.0221*** 1.2464*** 0.3674 0.6780** (1.17) (1.19) (3.78) (3.68) (0.88) (2.49) Fund Return 0.0349* 0.0343* 0.0214 -0.0386 0.0245 0.0083 (1.84) (1.81) (0.28) (-0.39) (0.37) (0.07) Fraction in Investment-grade Bonds 0.0017 0.0011 0.0063* -0.0045 -0.0304*** 0.0238*** (0.47) (0.31) (1.67) (-1.40) (-6.28) (4.86) Financial Center Dummy -0.0061*** -0.0062*** -0.0056*** 0.0021** -0.0165*** 0.0064*** (-3.26) (-3.31) (-6.25) (2.60) (-12.01) (6.79) Const 0.2898*** 0.3093*** 0.3100*** 0.3592*** 0.3352*** 0.2782*** (30.50) (30.38) (46.38) (89.84) (58.26) (32.70) Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.35 - - - - (Average) R-squared 0.3021 0.2998 0.3114 0.3296 0.3757 0.2424 Number of observations 83998 83998 83998 24606 21105 19934
35
Table III (Cont’d)
Panel C: Matching Sample (Within Family) All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy -0.0109*** -0.0179*** -0.0109*** -0.0132*** -0.0125*** -0.0048*** (-11.05) (-12.06) (-16.19) (-12.39) (-11.53) (-5.93) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.75 - - - - (Average) R-squared 0.2375 0.2159 0.2715 0.3439 0.4204 0.2527 Number of observations 11576 11576 11576 4734 3367 4218
Panel D: Matching Sample (Across Family) All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy -0.0129*** -0.0190*** -0.0129*** -0.0150*** -0.0151*** -0.0076*** (-15.34) (-15.20) (-25.78) (-24.76) (-13.72) (-8.66) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.46 - - - - (Average) R-squared 0.2679 0.2664 0.2976 0.3273 0.4026 0.2785 Number of observations 16146 16146 16146 8904 4809
Panel E: Regression based on Differenced Variable All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Difference in Fund Hierarchy -0.0175*** -0.0196*** -0.0180*** -0.0209*** -0.0220*** -0.0060*** (-19.53) (-11.56) (-15.33) (-11.92) (-10.62) (-4.52) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.41 - - - - (Average) R-squared 0.0850 0.0848 0.1107 0.1527 0.1479 0.1277 Number of observations 62498 62498 62498 21246 17677 16226
36
Table IV Fund Hierarchy and Herding
This table related fund hierarchy to herding. Panel A summarizes the sample mean of fund herding at different
levels of fund hierarchy. We also provide univariate tests of fund herding regarding to single vs. multi- fund
hierarchy. Multi-hierarchy means the number of fund hierachies to be greater than 1. We consider the overall
sample as well as the “matching within fund family” and the “matching across fund families”. We report the results
for the full sample and the two matching samples separately. Both two tailed T-test and Wilconxon rank-sum test
are performed to test the differences. The number of observations (fund-quarter) is given in the parentheses. In
Panels B-E, we report the results of the multivariate analysis. We estimate:
titititi XHierarchyHerding ,1,,, εδβα +×+×+= − ,
where tiHerding , represents fund herding tendency of fund i at quarter t, tiHierachy , is fund hierarchy and
1, −tiX are other control variables. The definitions are detailed in the appendix.
The analysis in Panel B is based on the full sample. We add the fund type dummies across all specifications.
Column (1) is based on OLS regressions with standard errors clustering at fund level. In Column (2), we report the
results of an IV regression, where family level structures are chosen as instruments. Specifically, we instrument fund
structure variables using the following variables: family hierarchy (median of fund hierarchy within a family),
family employee specialty (median of employee specialty within a family), family team (median of team dummy
within a family), the interaction of family hierarchy with financial center dummy and the interaction of family
employee specialty with financial center dummy. Hansen’s J statistic (p-value) is reported to examine the quality of
instruments. The standard errors are clustered at fund level. Column (3) provides Fama-Macbeth (1973) estimates
at the fund level, while Column (4) provides the results of a Fama-Macbeth estimates at the family level. Column
(5) and (6) are estimated in the same way as in Column (3) but only based on funds owned by life insurance
companies and mutual fund families. Panel C and Panel D are based on the matching sample within and across fund
families respectively with the same specifications as in Panel B.
Panel E uses a “differenced” variable approach, where for each fund we match it with some other fund similar
in fund type, geographical location and size, but different in terms of fund family and fund hierarchy. The matching
procedure is as follows: for each fund-quarter we first choose all the other funds of the same fund type but from
different fund families and having different fund hierarchy. Then we pick 20 funds located most closely and narrow
them down to 10 according to similarity in fund size. From those 10 funds we select the final one with the smallest
geographical distances to the original fund. If there is more than one matched fund left meaning that they are
located at the same place, we choose the most similar one in terms of fund size. All the variables except the financial
center dummy, including both the dependent and independent variables, are the differences between the original
fund and its matched peer.
For the sake of brevity we only report the coefficients of fund hierarchy from Panel C to Panel E. ***, ** and
* represent significance levels at 1%, 5% and 10% respectively with t-statistics given in parentheses.
37
Table IV (Cont’d) Panel A: Univariate Results
Fund Herding by Fund Hierarchy Full Sample Matching Within
(194) (172) (192) T-test: Multiple vs. Single Hierarchy 3.49*** 0.38 3.44*** Wilconxon Test: Multiple vs. Single Hierachy 5.76*** 1.09 4.64***
Panel B: Full Sample All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy 0.0014*** 0.0021*** 0.0013*** 0.0016*** 0.0015** 0.0009** (4.31) (4.61) (4.47) (4.90) (2.45) (2.30) Control Variables Employee Specialty -0.0007*** -0.0010*** -0.0006*** -0.0007*** -0.0009*** -0.0007*** (-5.79) (-6.79) (-7.43) (-5.91) (-6.00) (-4.52) Team Dummy -0.0003 -0.0011* -0.0002 0.0006 0.0010 -0.0003 (-0.73) (-1.84) (-0.43) (1.33) (1.49) (-0.53) Fund Size -0.0002* -0.0001 -0.0002* -0.0003 -0.0002* 0.0004** (-1.80) (-1.43) (-1.77) (-1.46) (-1.74) (2.00) Family Size 0.0003** 0.0002 0.0003** 0.0005*** 0.0003* 0.0003 (2.51) (1.63) (2.32) (3.28) (1.90) (1.08) Portfolio Maturity 0.0010*** 0.0010*** 0.0014*** 0.0000 0.0000 0.0036*** (4.93) (4.87) (3.38) (0.10) (0.03) (4.82) Log(Number of Funds) 0.0007*** 0.0006*** 0.0006*** 0.0009*** 0.0004 0.0003 (3.60) (3.24) (3.53) (3.81) (1.30) (0.64) Fund Turnover 0.0087*** 0.0086*** 0.0098*** 0.0116*** 0.0177*** 0.0054*** (19.22) (19.18) (10.12) (9.45) (7.08) (5.18) Fund Return Volatility 0.0488 0.0493 0.0300 0.0103 -0.0879 0.0981* (1.48) (1.50) (0.60) (0.14) (-1.33) (1.68) Fund Return 0.0134** 0.0134** 0.0061 0.0061 0.0126 -0.0101 (2.52) (2.53) (0.41) (0.38) (0.71) (-0.51) Fraction in Investment-grade Bonds 0.0023*** 0.0023*** 0.0011 0.0011 -0.0007 0.0008 (3.77) (3.70) (0.95) (0.74) (-0.32) (0.65) Financial Center Dummy 0.0025*** 0.0024*** 0.0024*** 0.0017*** 0.0014*** 0.0016*** (7.20) (7.11) (7.35) (5.56) (2.80) (3.16) Const -0.0014 0.0052*** 0.0017 0.0036** 0.0050 -0.0127*** (-0.91) (3.16) (0.82) (2.20) (1.52) (-3.28) Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.26 - - - - (Average) R-squared 0.0717 0.0712 0.0757 0.1047 0.1092 0.0695 Number of observations 68606 68606 68606 20957 18575 18086
38
Panel C: Matching Within Family
All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy 0.0010** 0.0022*** 0.0010*** 0.0016*** 0.0012 0.0013** (2.45) (3.17) (2.78) (3.54) (1.45) (2.31) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.31 - - - - (Average) R-squared 0.0826 0.0790 0.1231 0.1833 0.2093 0.1651 Number of observations 10478 10478 10478 4555 3075 3932
Panel D: Matching Across Family All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy 0.0013*** 0.0025*** 0.0011*** 0.0013*** 0.0025*** 0.0004 (3.24) (4.28) (2.88) (3.56) (3.52) (0.50) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.16 - - - - (Average) R-squared 0.0810 0.0783 0.1147 0.1347 0.1896 0.1345 Number of observations 14717 14717 14717 8304 4410 5504
Panel E: Regression based on Differenced Variable All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Difference in Fund Hierarchy 0.0008*** 0.0010*** 0.0009*** 0.0008*** 0.0012*** 0.0015*** (4.17) (2.91) (3.24) (2.77) (2.77) (3.04) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.41 - - - - (Average) R-squared 0.0285 0.0283 0.0756 0.0847 0.1498 0.0974 Number of observations 47731 47731 47731 16965 13980 13922
39
Table V: Fund Hierarchy and Performance
This table relates fund hierarchy to performance. Panel A summarizes the sample mean of fund performance for different levels of fund hierarchy. We also provide univariate tests of
fund performance regarding to single vs. multi- fund hierarchy. Multi-hierarchy means the number of fund hierachies to be greater than 1. We consider the overall sample as well as
the “matching within fund family” and the “matching across fund families”. We report the results for the full sample and the two matching samples separately. Both two tailed T-test
and Wilcoxon rank-sum test are performed to test the differences. The number of observations (fund-quarter) is given in the parentheses. In Panels B-E, we report the results of the
multivariate analysis. We estimate: titititi XHierarchyAlpha ,1,,, εδβα +×+×+= − , where tiAlpha , represents the alpha of fund i at month t, tiHierachy , is fund
hierarchy and 1, −tiX are other control variables. The definitions are detailed in the appendix. The analysis in Panel B is based on the entire sample including life insurance companies,
mutual funds, property insurance companies and other institutions. We put fund type dummies in all specifications. Column (1) is OLS regression with standard errors clustering at
fund level. In Column (2), we have IV regressions, where family level structures are chosen as instruments. Specifically, we instrument fund structure variables using the following
variables: family hierarchy (median of fund hierarchy within a family), family employee specialty (median of employee specialty within a family), family team (median of team dummy
within a family), the interaction of family hierarchy with financial center dummy and the interaction of family employee specialty with financial center dummy. Hansen’s J statistic
(p-value) is reported to examine the quality of instruments. The standard errors are clustered at fund level. Column (3) provides Fama-Macbeth (1973) estimates at the fund level,
while Column (4) provides the results of a Fama-Macbeth estimates at the family level. From Column (7) to Column (10) we add the interaction term of fund hierarchy and a “close
investment” dummy. It equals 1 if the fund portfolio distance is below the sample median of the quarter and 0 otherwise. We also add the interaction term of employee specialty and
close investment dummy as additional controls. Panel C and Panel D are based on the matching sample within and across fund families respectively. Panel E is based on the
“differenced” variable as defined in the previous tables. For the sake of brevity we only report the coefficients of the fund hierarchy from Panel C to Panel E. In Panel F we compare
the impact of fund hierarchy on the fund performance of investing in high rated bonds with that of investing in low rated bonds. High rated bonds refer to bonds with Moody’s credit
rating above A3. Low rated bonds are bonds with Moody’s credit rating from B3 to BBB1. For each fund we estimate two portfolio alphas separately. One is based on the
value-weighted return of investing in high rated bonds while the other is based on the return of investing in low rated bonds. The estimation procedure is the same as described in the
appendix. From Column (1) to Column (4) we stack the high and low rated alphas together and create a rating category dummy which equals 1 if it is a low rated alpha and 0 otherwise.
Our focus is the interaction term of fund hierarchy and the rating category dummy. We also add the interaction of employee specialty and the rating category dummy as additional
controls. Standard erros are clustered at the fund level (Column (1)) as well as at the family level (Column (2)). Column (3) and (4) are estimated in the same way as in Column (1)
but only based on funds owned by life insurance companies and mutual fund families. In Column (5) and (6) we run Fama-Mecbeth regressions separately for the low rated alpha and
the high rated alpha.. ***, ** and * represent significance levels at 1%, 5% and 10% respectively with t-statistics given in parentheses.
Panel A: Univariate Results Fund Performance by Fund Hierarchy Full Sample Matching Within Family Matching Across Family
(250) (216) (228) T-test: Multiple vs. Single Hierarchy -9.59*** -1.76* -5.84*** Wilconxon Test: Multiple vs. Single Hierachy -9.71*** -1.34 -5.78***
40
Panel B: Full Sample
All Institutions Life Mutual All Institutions Life Mutual OLS IV FM FM Family FM FM OLS FM FM FM (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Fund Hierarchy -0.0015*** -0.0024*** -0.0012*** -0.0011*** -0.0008*** -0.0012*** -0.0008** -0.0008*** -0.0005** -0.0007*** (-4.61) (-4.77) (-8.47) (-4.15) (-4.03) (-4.86) (-2.28) (-4.69) (-2.05) (-2.68) Fund Hierarchy * Close Investment Dummy
Panel C: Matching Within Family All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy -0.0013*** -0.0029*** -0.0013*** -0.0013*** -0.0012*** -00011*** (-4.21) (-3.43) (-7.91) (-3.90) (-4.50) (-3.15) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.49 - - - - (Average) R-squared 0.3268 0.3238 0.3354 0.4208 0.3609 0.3884 Number of observations 19712 19712 19712 7133 6528 6118
Panel D: Matching Across Family All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Fund Hierarchy -0.0014*** -0.0021*** -0.0012*** -0.0012*** -0.0013*** -0.0011*** (-4.76) (-3.64) (-8.04) (-5.75) (-4.98) (-3.59) Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.34 - - - - (Average) R-squared 0.3550 0.3544 0.2911 0.3237 0.3222 0.3305 Number of observations 25560 25560 25560 14250 8794 8281
Panel E: Regression based on Differenced Variable All Institutions Life Mutual OLS IV FM FM Family FM FM (1) (2) (3) (4) (5) (6) Difference in Fund Hierarchy -0.0011*** -0.0013*** -0.0010*** -0.0011*** -0.0009*** -0.0015*** (-6.96) (-4.71) (-4.62) (-4.39) (-3.11) (-4.76) Difference in Control Variables Y Y Y Y Y Y Time Dummies Y Y - - - - Fund Type Dummies Y Y Y - - - Hansen J (p-value) - 0.93 - - - - (Average) R-squared 0.0315 0.0314 0.1689 0.1859 0.1993 0.2711 Number of observations 90659 90659 90659 35317 31630 14234