Printer Friendly

Geography, informal information flows and mutual fund portfolios.

This paper explores how informal information channels impact mutual fund performance. We measure the strengths of two location-based information channels: 1) information transfers among fund managers (fund-fund links) and 2) transfers between managers and the companies in which they invest (fund-company links). We find that each channel increases investment performance in the absence of the other, but decreases it when acting in combination. Stock selection associated with the presence of one channel, but the absence of the other, earns positive future returns. Our results indicate that the economic benefits of informal information channels depend critically on the nature of their interactions.

**********

In financial markets where significant frictions exist in the dissemination of information, investors may acquire differential information through unobserved informal channels. Though these channels are typically opaque to outside observers, researchers have begun to theoretically and empirically characterize channels that might impact investors' belief formation and decision making. In particular, recent studies on mutual funds provide growing evidence on the presence of informal communication among market participants who are "neighbors," (i.e., located in social and geographical proximity). (1) For example, local fund managers have easier access to information about local companies as they can readily visit the operations, talk to employees, or collect information through their friends, acquaintances, and associates in the local community, giving them information advantages over managers in distant locations (Coval and Moskowitz, 2001). (2) Similarly, Hong, Kubik, and Stein (2005) posit that fund managers could obtain information about asset prices by "word-of-mouth" from their professional peers located in geographical proximity, resulting in correlated strategies among managers in the same city. (3)

Building on the existing literature, we focus our attention on two informal information channels known to be associated with mutual fund managers' portfolio decisions: 1) fund-fund (FF) links, which transfer information between fund managers about potential investment opportunities and 2) fund-company (FC) links, where the fund's geographical proximity to the company's headquarters facilitates the acquisition of information about the company. Similar to several prior studies, we use measures based on geographical distance to proxy for the existence of informal information channels. The main departure of our study from existing studies is that we use an empirical setting that accounts for more than one information channel simultaneously, thereby allowing us to disentangle the performance implications of each information channel in isolation and in combination with other channels. We argue that it is important to analyze the effects of these information channels in an integrated setting because, as we discuss further, various strands of research suggest that the economic impact of informal information channels varies depending on the volume and nature of communication. In addition, we seek to infer the performance implications of informal information channels in financial markets, complementing several related studies (Hong et al., 2005) that have focused on documenting the existence of such informal communication.

The economic value of informal communication remains an ambiguous issue in the literature. Empirical studies, such as Coval and Moskowitz (2001), suggest that certain types of informal communication enhance investment performance. Stein's (2008) model demonstrates that communication among competitors (e.g., mutual fund managers) can increase the value of investment ideas. Bala and Goyal (1998) develop models of information transmission and, under certain conditions, predict that agents arrive at optimal decisions in the presence of communication. Overall, these studies predict that investment performance increases with informal communication.

In contrast, Stein (2009) posits that "crowded trades" may reduce performance when too many investors act on similar information. Along similar lines, Colla and Mele (2009) and DeMarzo, Vayanos, and Zweibel (2003) predict that information exchanges have a detrimental impact on the informed traders' profitability when the information signals transmitted to the traders have positive correlations. Ellison and Fudenberg's (1995) model of word-of-mouth communication concludes that boundedly rational agents who communicate via word-of-mouth achieve efficient outcomes only if they each receive a little information, suggesting a negative relationship between the quantity of communication and the efficiency of outcomes. (4) These papers predict that more informal communication received by portfolio managers can diminish investment performance.

Given the ambiguity in the literature, the economic value of informal communication is an empirical issue that this paper addresses by focusing on mutual funds' investment performance. Using a large sample of holdings of actively managed equity mutual funds in the United States, we find that stronger informal information links have a significant impact on the fund manager's stock selection ability. (5) However, whether an informal information channel enhances or diminishes the profitability of investments depends critically on the nature (i.e., presence or absence) of other informal channels. We find that investments associated with stronger FF (FC) communication channels generate an outperformance of about 2.91% (3.02%) in annualized characteristic-adjusted returns in the absence of the other information channels. These findings are consistent with the value-added effects of informal channels on investment decisions documented by studies like Coval and Moskowitz (2001). In striking contrast, FF (FC) information links are associated with a substantial underperformance of about -3.74% (-3.63%) in annualized characteristic-adjusted returns when they act in combination with the other informal communication channels.

To further investigate the mechanism behind the finding that investment performance decreases with the volume of information, we calculate two measures, the absolute dollar ratio trade imbalance and the herding measure (HM), to capture the propensity of correlated trades using the methodology in Lakonishok, Shleifer, and Vishny (1992) and Wermers (1999). Our results, based on these two measures, provide evidence that "crowded trades" could explain the underperformance when strong FF and FC channels aggregate, since Stein (2009) suggests that profitability dissipates when crowded trades occur. Consistent with less reputed fund managers having less access to valuable informal communication, the marginal benefits of both FC and FF information links are smaller for funds from small families than for those from large families. Results are similar across subsamples based on fund size. The effect of the concentration of funds in geographical clusters that facilitate FF information channels is not subsumed by the "big city effect" (i.e., effect of population concentration) documented by Christofferson and Sarkissian (2009). Our results are robust to multivariate analyses that include a variety of model specifications and control variables.

Overall, we find that the holdings with ex post superior performance are those associated with: 1) strong FC information links in the absence of FF channels, and 2) strong FF information links in the absence of FC channels. Based on these stylized results, we conduct an initial examination of possible information that can be inferred about future asset prices from the investment choices of fund managers. Using fund holdings reported at the end of a quarter (which is stale information for all subsequent quarters), we construct a hypothetical portfolio of stocks associated with the two information settings that generate superior characteristic-adjusted returns. We call this the Best Information Portfolio (or BIP). The BIP replicates the equal-weighted holdings of the two superior information portfolios from the previous quarter, and holds it in subsequent quarters. For a wide range of holding periods, we find significant average monthly returns ranging from 1.35% to 1.81% for the BIP, and the returns remain significant even after adjusting for risk attributes. These results suggest that the stock selections made by fund managers among stocks associated with certain informal communication channels predict the future performance of those stocks.

This study broadly contributes to the literature on informed trading and informal communication in financial markets. We offer fresh empirical evidence regarding the effects of informal information channels on investment performance, an area where empirical evidence remains scarce. To the best of our knowledge, this study is the first to examine multiple information channels linked to the geographical location of mutual funds in a unified empirical setting, and to provide empirical evidence concerning the effects of the aggregation of informal communication. The results point to the important distinction between the volume versus the quality of informal communication, an aspect of information channels that is little understood in the context of financial markets. Overall, our findings characterize the conditions under which information channels related to geographical location add value for fund managers, but also reveal more complexity in information settings than acknowledged in previous empirical papers.

The paper is organized as follows. Section I describes the data and methodology. Our results are presented in Section II. Section III provides our concluding remarks.

I. Data and Methodology

A. Sample Description

The primary data source used in this study is the Centre for Research in Security Prices (CRSP) Survivor-Bias Free US Mutual Fund Database (MFDB), supplemented by data collected from other publicly available sources. CRSP MFDB added data regarding the location of management companies starting in 2000, with the availability of these data increasing in more recent years. For each fund listed in CRSP MFDB, we identify the fund's location as the city and state provided for the fund's management company. Our initial sample consists of US mutual funds investing in domestic stocks for which we were able to find location information in our sample period from the last quarter of 2003 to that of 2010. As is standard in the literature, we exclude index, sector, international, and bond funds, and funds that are not run by the same management company as the majority of funds in the same fund family.

In addition, we apply several other selection criteria to form our final sample. First, we only choose equity funds primarily investing in domestic equity having "growth," "aggressive growth," "growth and income," or "balanced" as stated objective categories. (6) We exclude funds with less than $5 million in total net assets (TNA). Our final sample consists of 2,931 unique equity funds spanning 571 fund families for which we have the headquarter location.

Next, we obtain the holdings data for the mutual funds in our final sample from the CRSP Holdings database. The CRSP Holdings database, introduced in 2005, gathers details of quarterly holdings dating back to July 2003 and uniquely maps to other CRSP fund data via Investment Company Data, Inc. (ICDI) codes. Portfolio holdings in a quarter are disclosed some time during the three months in the next quarter. For the stocks in the holdings database, we obtain the headquarter location of the company from the Compustat database.

The CRSP Holdings database assigns a portfolio code to each unique portfolio. We consolidate multiple ICDI codes representing different share classes of the same fund to the unique portfolio code if the underlying portfolios are identical. We then merge portfolio data with fund characteristics data like monthly TNA and management company. Finally, we obtain stock price and returns data for the portfolio companies from CRSP monthly stock files. The US Census Bureau's Gazetteer geographical data provide the latitude and longitude coordinates of the cities where the funds and companies are headquartered. These coordinates are then used to calculate the geographical distances between each pair of cities.

Table I presents summary statistics on the sample of funds and the characteristics of the stocks they hold. The median fund size (i.e., TNA) in the sample is $226.1 million and the fund age (based on the first offer date) is 10 years. Since our metrics of informal information links are based on geography, we report a summary geographical distribution of the portfolio companies and funds in Table I as well. In the median fund portfolio, 5.41% of the total amount invested in domestic equity holdings is concentrated in stocks of companies less than 100 km away from the fund location. Over half (66.72%) of the total amount invested is in companies more than 1,000 km from the fund. For the median fund in the sample, 10.23% of the funds are located within 100 km of each other, while 60.03% are located more than 1,000 km apart. While the mutual fund industry may be concentrated in a few cities, there is substantial dispersion in the data to allow for an examination of differences in location characteristics.

Table I also reports summary statistics on the stocks held in mutual fund portfolios during the sample period. The median fund portfolio holds 74 stocks, with a median company age of 37 years. We report the characteristics of the stocks along the three style dimensions in Daniel et al. (henceforth DGTW) (1997), namely, size, book-to-market (B/M) ratio, and momentum. The median stock holding in a value-weighted DGTW size quintile is 4.21, indicating that mutual funds, on average, concentrate their portfolios in larger companies. The median value-weighted DGTW B/M quintile of stocks held is 2.50, suggesting that mutual funds in the sample have dispersed investments across growth and value stocks. The median fund indicates a preference for stocks having higher past returns by investing in stocks slightly above the third DGTW momentum quintile.

B. Methodology

This section describes the empirical methodology used in this study. First, we outline a method for measuring information channels and then the framework to test their performance implications.

1. Measuring Information Channels

Ideally, data on real information channels through which each fund manager receives communication about investment opportunities would be employed to study the impact of informal communication on performance. However, it is extremely hard to pinpoint specific information sources with the available data. Researchers have used various proxies for unobservable communication channels, like shared educational backgrounds (Cohen, Frazzini, and Malloy, 2008), neighborhood effects, and geographical proximity. In this paper, we use measures of the likelihood of strong information channels based on geographical proximity.

a. FF Links. First, we consider information channels between a fund manager and other managers investing in similar asset classes (in this case, domestic stocks). Hong et al. (2005) document that proximately located mutual fund managers are more likely to engage in informal communication. We focus on information channels between fund managers from different fund families and explore the issues of intrafamily information exchanges separately in later sections.

Based on this premise, we measure the likelihood of fund-fund (FF) information links between fund manager j and fund manager i as

[L.sup.FF] = 1/(1 + [Distance.sub.j,t]). (1)

Here, [Distance.sub.j,t] is the natural logarithm of the geographical distance between the city locations of fund j and fund i, plus one. The likelihood of FF information channels between a fund manager and all other fund managers who do not belong to the same fund family are then aggregated for each fund in the sample. This provides an estimate of the quantity (or density) of communication fund manager j has with other fund managers as

[(Links).sup.FF.sub.j] = [N.summation over (i=1)][L.sup.FF][(Links).sub.j,t] = [N.summation over (i=1)][1/(1 + [Distance.sub.j,t])]. (2)

Here, i = 1, ..., N are all funds in the sample that do not belong to the same fund family as fund j.

Finally, we sort all funds in each quarter by the measure of FF information links ([Links.sup.FF]) and rank them into quartiles. The funds in the lowest quartile in a quarter are considered to be the funds with weak FF information links ([Q.sub.FF,weak]) while funds in the highest quartile are considered to be the funds with strong FF information links ([Q.sup.FF,strong]). Note that the notion of "strength" is based on the density of information links, and we do not make any assumptions about the precision or quality of the communication.

A few caveats are in order regarding the measure of FF information links used in this study. This measure includes only the funds in our sample, which is restricted based on various criteria like active management and predominantly domestic equity investments, among others. It does not account for all information links involving managers in the universe of mutual funds. For example, while a sector fund manager is excluded from the sample, she may be an information source for a manager included in the sample. However, we believe that the above measure of fund interactions is highly correlated with a similar measure derived from the universe of money managers, since the geographical cluster patterns are unlikely to change significantly. Likewise, the relative comparison of location-based interactions is also unlikely to change substantially. (7)

A second confounding effect that may impact the precision of measuring the information linkages is the issue of relocation and manager turnover. If information links arise out of a manager's networks and interactions, it is possible that they remain with the manager after she relocates. In that case, a manager "carries" the links wherever she goes. For instance, if a fund manager relocates to Iowa after many years as a Boston-based manager, it is possible that she continues to have access to information links in the Boston area that are not accounted for by the measures used. In fact, Parwada (2008) finds evidence indicating that fund managers who relocate continue to exhibit a preference for the formerly local stocks. Two crucial factors mitigate this problem in our empirical setting. First, our relatively shorter sample period makes the issue of relocation and turnover unlikely to be a substantial oversight. More importantly, while relocation or turnover may create noisy measures of information linkages, these events increase the similarity across information settings of managers in different locations, which may only cause a bias against finding systematic differences in the effect of geography-based information channels.

b. FC Links. The second type of informal communication considered in this paper is that between fund managers and companies in which they invest. We measure these fund-company (FC) information links for each stock in a fund's portfolio in a given quarter. The likelihood of an FC information channel existing between fund manager j and the stock n held in/s portfolio is

[L.sup.FC][(Links).sub.j,n] = 1/(1 + [Distance.sub.j,n]). (3)

Here, [Distance.sub.j,n] is the natural logarithm of the geographical distance between the city locations of fund j and company n, plus one.

Next, we sort the stocks in each fund's portfolio in each quarter by measures of FC information links ([L.sub.FC](Links)) and rank them into quartiles. Within each fund's portfolio, the lowest FC quartile is considered the information portfolio with the weakest FC links ([Q.sup.FC,weak.sub.j]). Alternatively, the holdings in the highest FC quartile are considered to be the portfolio of stocks with the strongest FC information links ([Q.sup.FC,strong.sub.j]). We rescale to sum to one the portfolio weights of the holdings within each quartile, thus creating four portfolios for each fund manager in each quarter. We also consider alternative rankings, like deciles and quintiles, which generate qualitatively similar results.

Our main metric of the strength of FC links has a continuous range as opposed to Coval and Moskowitz's (2001) dichotomous local versus nonlocal status of holdings. (8) The main distinction of our setting is that a part of each fund's portfolio is guaranteed to be regarded as relatively proximate by definition, while Coval and Moskowitz (2001) allow managers to not maintain any local stock holdings. For example, a fund manager located in Alaska may have the nearest holding in a company headquartered 500 km away. This holding may be categorized in the nearest FC quartile in our setting. In contrast, a New York manager's nearest FC quartile may only be comprised of companies located within 100 km. While we consider the dichotomous measure for robustness checks, we retain the continuous measure of [L.sup.FC] (Links) as our main metric for two reasons. First, the dichotomous measure of localness, by definition, is likely to be a restrictive measure when used to proxy for the intensity of information channels in geographical proximity. In contrast, our continuous variable can better measure the potential information links attributable to geographical proximity beyond the threshold of "localness." Specifically, regional externalities like public transportation and regional educational and business networks, among other factors, may facilitate an elevated level of information exchanges and interaction beyond the customary 100-km threshold for the dichotomous variable. For example, Amtrak's Northeast Corridor (NEC) is a well-developed high-speed train system connecting Boston, New York City, Washington DC, and other areas in the region. While most of these cities are more than 100 km apart in geographical distance, they are well connected by public transportation and facilitate frequent travel for business meetings and conferences. The pairwise likelihood of economic interaction between two fund managers located in the NEC coverage area may be substantially higher than between a manager from these areas and one from the West Coast, either due to ease of transportation, commonalities in economic networking events (like asset management conferences), or shared educational backgrounds. In addition, our continuous measure allows for the fact that managers who do not have enough investment opportunities within 100 km may increase their effort in acquiring information on the relatively proximate holdings.

2. Measuring Marginal Impacts of Information Channels

Based on the empirical metrics of information channels discussed previously, we develop a framework for studying the marginal impact of informal information channels (FF and FC) on fund performance. Our goal is to empirically disentangle the impact and investigate the interactions between these two informal information channels in the context of investment performance.

Figure 1 presents the outline of our information metric in a univariate setting. We form four groups of investments by assigning them to portfolios P1, P2, P3, and PA. The funds having strong (weak) FF links form the quartile group [Q.sup.FF,strong] ([Q.sup.FF,weak]). Here, [Q.sup.FF,strong] is the highest quartile, [Q4.sup.FF], ranked by measures of FF links and [Q.sup.FF,weak] is the lowest quartile [Q1.sup.FF] ranked by measures of FF links. Within each fund j's portfolio, the stocks with strong (weak) FC links form the quartile group [Q.sup.FC,strong.sub.j] ([Q.sup.FC,weak.sub.j]). Here, [Q.sup.FC,strong.sub.j] is the highest quartile [Q4.sup.FC] ranked by measures of FC links and [Q.sup.FC,weak.sub.j] is the lowest quartile [Q1.sup.FC] ranked by measures of FC links.

In Figure 1, the main portfolios of interest ranked based on FF and FC information links are:

* Portfolio P1: Portfolio of funds with weak FF links ([Q.sup.FF,weak]) investing in stocks with strong FC links ([Q.sup.FC,strong.sub.j]).

* Portfolio P2: Portfolio of funds with weak FF links ([Q.sup.FF,weak]) investing in stocks with weak FC links ([Q.sup.FC,weak.sub.j])

* Portfolio P3: Portfolio of funds with strong FF links ([Q.sup.FF,strong]) investing in stocks with strong FC links ([Q.sup.FC,strong.sub.j]).

* Portfolio P4: Portfolio of funds with strong FF links ([Q.sup.FF,strong]) investing in stocks with weak FC links ([Q.sup.FC,weak.sub.j])

Developing these categories of information channel combinations lays the foundation for disentangling the performance impact of different information channels using empirical data. The manner of construction ensures that each fund in each FF quartile has at least one stock holding that can be assigned to the four FC quartiles based on FC information channels. For example, in Figure 1, the holdings in Portfolio P1 are located relatively close to the fund, and have a higher likelihood of FC communication as compared to stocks in Portfolio P2, which are geographically distant and likely to have fewer information transfers. The information processes related to Portfolio P1 may be different from P2 by the additional FC information links that are present in PI, but absent in P2. Note that the method of disentangling PC information channels makes it possible to avoid confounding effects due to heterogeneities in other fund-, manager-, family- and time-specific factors. Similarly, Portfolio P1 is different from P3 in that P3 is associated with FF information channels generating an additional informal link, while P1 is not. While it is necessary to make comparisons across funds to observe the marginal impact of FF information links, doing so introduces other confounding fund-specific factors, like managerial skills, to the analysis. Therefore, we later perform further bivariate and multivariate analyses to compare funds that have similar attributes along dimensions other than information channels. It is also interesting to note that in this framework, Portfolio P2 can be viewed as a control group with the lowest density of both forms of informal communication.

Following Daniel et al. (1997), we compute the characteristic-adjusted return (i.e., DGTW return) that measures a fund manager's stock selection ability. We use the DGTW return as the main measure of return in this paper. Our main objective is to uncover the impact of informal communication on fund managers' before-cost investment ability in specific investments. As such, we do not examine after-cost net returns that mingle the effects of manager skills, fees, and transaction costs. For each fund in Quartiles [Q1.sup.FF] through [Q4.sup.FF] we compute the mean monthly raw returns for quartile subportfolios [Q1.sup.FC] through [Q4.sup.FC] in each quarter as

[R.sup.raw] = [1/3][S.summation over (i=1)][3.summation over (t=1)] [w.sub.i,t] [R.sub.i,t] (4)

where [w.sub.i,t] is the portfolio weight of stock i in month t in that information portfolio, and S is the number of stocks in the portfolio. (9) Similarly, we compute characteristic-adjusted information portfolio returns for each fund in each quarter as

[R.sup.adj] = [1/3][S.summation over (i=1)][3.summation over (t=1)] [w.sub.i,t] ([R.sub.i,t] - [R.sup.bench.sub.i], (4)

where [R.sup.bench.sub.i] is the value-weighted monthly return of stock i's DGTW benchmark portfolio (following Daniel et al., 1997). The differences in the portfolio returns form the basis of initial examination of the marginal impact of these information channels in a reduced-form setting. The baseline analyses identify the marginal impact of informal information links as follows:

* For holdings with strong FC channels, the difference in characteristic-adjusted returns [R.sup.adj] (P1) between Portfolios P3 and P1 reflects the marginal impact of FF information links.

* For holdings with weak FC channels, the difference [R.sup.adj] (P4) - [R.sup.adj] (P2) between Portfolios P4 and P2 gives the marginal impact of FF information links.

* For the funds with weak FF channels, difference [R.sup.adj] (P1) - [R.sup.adj] (P2) measures the marginal performance impact of PC information links for the fund.

* For the funds with strong FF channels, [R.sup.adj] (P3) - [R.sup.adj] (P4) measures the marginal impact of FC information links for the fund.

Finally, we report the value-weighted average returns (weighted by TNA) across funds in order to provide conservative estimates that avoid results driven by small funds.

II. Results

In this section, we present the empirical results of the relationship between informal information channels and the stock selection ability exhibited by mutual fund managers in specific information settings.

A. Summary Statistics: Portfolio Characteristics

Table II reports the summary statistics of the funds and stock holdings for a 4 x 4 matrix of portfolios formed from the two information channels. In Panel A, we present various fund characteristics for the group of funds forming the four FF quartiles. Funds with strong FF measures (Quartile [Q4.sup.FF]) tend to be larger and somewhat younger than those with weak FF measures (Quartile [Q1.sup.FF]). However, expenses and turnover are similar across the different FF quartiles. The number of funds managed by the fund family is substantially more for funds in [Q4.sup.FF] (median of 59 funds) as compared to those in [Q1.sup.FF] (median of 28 funds). Managers of funds in [Q4.sup.FF] tend to have shorter tenures than fund managers in [Q1.sup.FF].

Panel B presents the portfolio characteristics of the stocks that form the FC quartiles for the funds in different FF quartiles. The overall DGTW size, B/M, and momentum factors across the portfolios appear comparable. Based on the value-weighted averages reported, most of the portfolios hold stocks around the fourth DGTW size quartile, between the second and third DGTW B/M quartile and third DGTW momentum quartile. These summary statistics indicate that fund managers, on average, hold stocks of similar characteristics (on a dollar-weighted basis) across different FF and FC information portfolios. The percentage of portfolio dollars invested across the different information portfolios are also of similar magnitude for the overall sample. For instance, funds in the weakest FF quartile ([Q1.sup.FF]), on average, invest 24.67% of the total portfolio amount in companies forming the strong FC portfolio ([Q4.sup.FC]), while investing 25.79% in the weak FC portfolio ([Q1.sup.FC]). Funds in the strongest FF quartile ([Q4.sup.FF]) invest about 26.0% in companies forming the strong FC portfolio ([Q4.sup.FC]), while investing 24.29% in the weak FC quartile ([Q1.sup.FC]).

B. Information Channels and Portfolio Performance

1. Univariate Results

Table III reports the baseline univariate tests on the correlation between informal information channels and portfolio returns. We present average annualized returns on quarterly holdings. We compute value-weighted returns for each fund's four FC information portfolios in each quarter. The average values reported in the table for [Q1.sup.FC] - [Q4.sup.FC] are portfolio returns weighted by fund TNA. Panel A of Table III presents raw returns on the portfolios forming the 4 x 4 matrix of the FF and FC quartiles. We make comparisons between the paired ([Q1.sup.FF], [Q4.sup.FC]), ([Q1.sup.FF], [Q1.sup.FC]), ([Q4.sup.FF], [Q4.sup.FC]), and ([Q4.sup.FF], [Q1.sup.FC]) subportfolios outlined in Figure 1 as our main results in these reduced-form analyses. For funds in [Q1.sup.FF], the raw returns increase when the FC links grow stronger, and are the highest for [Q4.sup.FC]. However, the returns across FC links indicate different patterns for funds in the stronger FF link quartiles (i.e., [Q2.sup.FF], [Q3.sup.F], and [Q4.sup.FF]). In the presence of strong FC links ([Q4.sup.FC]), the returns decrease from columns [Q1.sup.FF] to [Q4.sup.FF]. In contrast, in the absence of the FC links ([Q1.sup.FC]), the returns increase from columns [Q1.sup.FF] to [Q4.sup.FF]. The results of raw returns demonstrate the significant marginal impact of the conditional FF and FC links. Conditional upon the weak FF links (i.e., within the [Q1.sup.FF] column), the marginal impact of the FC links is 2.63% and is significant at the 5% level. Conditional upon the weak FC links (i.e., within the [Q1.sup.FC] row), the marginal impact of the FF links is 3.53% and is significant at the 1% level. Conditional upon the strong FF links (i.e., within the [Q4.sup.FF] column), the marginal impact of the FC links is a negative and economically significant -5.12%. Conditional upon the strong FC links (i.e., within the [Q4.sup.FC] row), the marginal impact of the FF links is a negative and economically large -4.22%. Nonetheless, the raw returns may simply be a reflection of different levels of risk borne by the managers in these portfolios.

In Panel B of Table III, we report characteristic-adjusted returns for the information portfolios as the more appropriate measure for judging stock selection ability due to the presence of style-related fixed effects in the returns. A comparison of Panels A and B finds that style-adjusted returns have a pattern qualitatively similar to that of raw returns. The FC links have a significantly positive marginal impact on returns in the absence of the FF links (i.e., funds in [Q1.sup.FF]). Funds with weak FF links ([Q1.sup.FF]) generate positive characteristic-adjusted returns on their holdings with strong FC links ([Q4.sup.FC]) that exceed the returns on holdings with weak FC links ([Q1.sup.FC]) by 3.02% per year at the 1 % significance level. In contrast, in the presence of strong FF links (i.e., funds in [Q4.sup.FF]), the FC links have a significantly negative marginal impact on portfolio returns. For funds in [Q4.sup.FF], the portfolios with strong FC links ([Q4.sup.FC]) generate significantly lower characteristicadjusted returns when compared to the portfolios with weak FC links ([Q1.sup.FC]). For these funds, the marginal impact of FC channels on characteristic-adjusted returns is -3.63% and is significant at the 1% level. In other words, when the strong FF links act in combination with the strong FC links, the result is a reversal in informational benefits and subsequent underperformance.

Using across-fund comparisons (presented in rows), the FF links have a significantly positive marginal impact on returns in the absence of the FC links (i.e., stocks in [Q1.sup.FC]). Note that unlike the analyses of the marginal impact of FC links where within-fund portfolio decompositions are used, examining the marginal impact of FF links necessitates across-fund comparisons. This gives rise to the possibility that other heterogeneities between the funds drive performance differences. We explore these issues in later analyses. For the baseline results, we use a means comparison test between the information portfolio returns for each quarter to determine the marginal impact of FF information channels, holding the nature of FC channels constant. Funds with strong FF links ([Q4.sup.FF]) underperform funds with weak FF links ([Q1.sup.FF]) by 3.74% in characteristic-adjusted returns (significant at 1% level), for holdings where strong FC links exist ([Q4.sup.FC]). Interestingly, in the absence of FC links ([Q1.sup.FC]), the strong FF funds outperform the weak FF funds by 2.91% annually (at the 1 % level of significance).

To summarize, the salient feature of the results is that the two information channels, while beneficial when acting in isolation, have a negative impact on the fund manager's stock selection ability when they act in combination. As such, an aggregation of information channels has a detrimental effect on the portfolio outcomes of fund managers. These findings may arise due to the value-reducing impact of "crowded trades" discussed by Stein (2009), which reduce returns for fund managers located in areas with high fund manager density investing in local stocks (which are also local to many other investors). In contrast, remotely located firms that are far from areas with high fund manager density are less likely to attract a high number of correlated trades. The results are also consistent with the theoretical models predicting that people have a poor ability to aggregate information efficiently due to limitations in cognitive capacity (Tversky and Kahneman, 1974; Ellison and Fudenberg, 1995).

2. Crowded Trades Effect

The puzzling results regarding the underperformance of the portfolios associated with strong information channels on both FF and FC dimensions could be attributed to the effect of crowded trades. To explore this issue, we calculate two variables, the absolute dollar ratio trade imbalance ([absolute value of DRatio]) and the HM, to measure the likelihood that fund holdings with two strong information channels are associated with elevated correlated trading (e.g., correlated purchases or sales) by mutual fund managers. These two measures, first introduced in Lakonishok et al. (1992) to capture feedback trading or trading in herds among institutional investors, are later used extensively in Wermers' (1999) study of mutual fund herding. The essence of these two measures is to capture the imbalance between the number of buyers and sellers, which increases with crowded trades.

Following Wermers (1999), we define the absolute dollar ratio trade imbalance measure ([absolute value of DRatio]) for each stock i held by a fund in quarter t as

[absolute value of ([DRatio.sub.i,t])] = [absolute value of ([$buy.sub.i,t] - [$sell.sub.i,t]/ [$buy.sub.i,t] + [$sell.sub.i,t])], (6)

where [$buy.sub.i,t] ([$sell.sub.i,t]) equals the total purchases (sales) by all mutual funds, in dollars, of stock i during quarter t, applying the average of the beginning- and end-of-quarter prices to aggregate increases (decreases) in share holdings for that stock quarter. We compute the HM for each stock i held by a fund in quarter t as

[HM.sub.i,t] = [absolute value of ([p.sub.i,t] - E[[p.sub.i,t])]] - E[absolute value of ([p.sub.i,t] - E[[p.sub.i,t]])] (7)

where [p.sub.i,t] is the proportion of funds trading stock i during quarter t that are buyers. The proxy used for E[[p.sub.i,t]] is the proportion of all stock trades by mutual funds during quarter t that are buys. E[absolute value of ([p.sub.i,t] - E[[p.sub.i,t]])] is calculated under the null hypothesis of herding only by random chance (see Lakonishok et al., 1992 and Wermers, 1999).

In Table IV, we report our empirical tests of the conjecture that crowded trades may be a driver of the underperformance observed when strong information channels act in combination. In general, higher values of [absolute value of DRatio] and HM indicate a greater propensity for correlated trades. As demonstrated in Table IV, Panels A and B present a picture consistent with the possible crowded trades effect on portfolio performance. For funds with relatively weak FF links ([Q1.sup.FF] to [Q2.sup.FF]), the difference in [absolute value of DRatio] and HM between the strongest ([Q4.sup.FC]) and the weakest ([Q2.sup.FC]) FC quartile portfolios is not statistically significant suggesting that their performance differentials are not driven by the variance in correlated trades. We observe similar patterns for performance differentials between the funds with the strongest ([Q4.sup.FF]) and weakest ([Q1.sup.FF]) FF links for holdings with relatively weak FC channels ([Q1.sup.FC] to [Q2.sup.FC]). In sharp contrast, the differences of [absolute value of DRatio] and HM measures between Portfolios [Q4.sup.FF] and [Q1.sup.FF], including stocks with strong FC channels, are positive and statistically significant at the 1% level. These results confirm that the stocks held in the Q4FC portfolio of the funds with the strongest FF links ([Q4.sup.FF]) are associated with more crowded trades than the [Q4.sup.FC] portfolio of the funds with the weakest FF links ([Q1.sup.FF]). Similarly, the differences in these two measures between Portfolios [Q4.sup.FC] and [Q1.sup.FC] with the strongest FF links are also positive and statistically significant at the 1% level. Taken together with the earlier results on returns, these findings are consistent with Stein's (2009) conjecture that crowded trades reduce the performance of investments substantially. They strongly suggest that crowded trades could explain the negative marginal impact of FF and FC information channels when they act in combination.

3. Fund and Family Size

The baseline results support the significant marginal impact of FC and FF information channels on managers' stock selection ability. However, the strengths of information channels may be proxying for other fund attributes that affect performance. For example, funds with stronger FF links may be part of larger family complexes, while those with weaker FF links belong to small families. Moreover, family size is likely to be a proxy for the level of intrafamily information channels that are available to a fund manager. It may be the case that external FF links are only important for managers lacking intrafamily information channels. We address the issue of family size by scrutinizing small and large fund families separately. We sort the sample of funds and rank them into terciles by the number of funds within the family for each quarter, independent of prior sorts on FF links. We consider other rankings like quartiles, but the ranking method does not have any bearing on our results.

Panel A of Table V presents the results for the smallest fund families, while Panel B reports corresponding statistics for the largest families. The fund managers from small families are less likely to gain from the intrafamily interactions that facilitate information acquisition. Fund managers from large fund families are likely to have more intrafamily information channels, in addition to more internal resources (e.g., research units). Alternatively, fund managers from larger families may also be more reputed among peers with stronger networks with other managers that lead to more information links. Thus, while fund managers from larger families may have the least need for information from informal external channels, they are likely to have the most informal information channels available to them.

In Panel A of Table V for small families, the marginal impact of FC links, while still positive, ceases to be statistically significant for funds with weak FF links in contrast with the baseline results. The funds in the strong FF quartile ([Q4.sup.FF]) underperform in their strong FC holdings ([Q4.sup.FC]) when compared to weak FC holdings ([Q1.sup.FC]) by 3.29% (at the 1% level of significance), which is consistent with the baseline results. The empirical evidence suggests that unlike typical funds from the entire sample, fund managers from small families are unable to draw economic benefits from strong FC information sources. Interestingly, these findings lend more credence to the word-of-mouth mechanism of information transfer than localized information acquisition. There is no clear reason to expect managers from small fund families to have less access to localized information via, for instance, regional media when compared to managers from large fund families. The qualitative results for the marginal impact of FF channels are similar to those in the baseline results. FF links help in the absence of FC channels, but not in the presence of strong FC links.

In Panel B of Table V, we present the results for the funds belonging to the largest fund families. They broadly echo the baseline results in Table III. For weak FF funds from large families, the marginal impact of strong FC links on characteristic-adjusted returns is positive and significant (at 5%) with a magnitude of 1.83% annually. In contrast, the strong FF funds from large families underperform in their strong FC holdings when compared to weak FC holdings by 2.43%. In addition, the marginal impact of FF channels remains significant even for comparisons between funds from large FF families. Therefore, the evidence indicates that intrafamily communication channels do not subsume the effects of external information links. For holdings with strong FC measures, large family funds with weak FF measures outperform large family funds with strong FF measures by 2.53%. For these funds, the marginal impact (1.73% and 0.88%) of FF on returns from holdings where there are weak FC ([Q1.sup.FC] and [Q2.sup.FC]) is larger and more significant than those from funds belonging to small families.

Overall, the results suggest that the marginal impact of information links is larger for funds belonging to large families, and intrafamily channels do not subsume the effects of external channels. These results suggest that funds from large families are able to leverage their reputation and visibility among their peers to form more advantageous communication channels than those from small families.

Fund size is another potentially critical factor influencing the role of external information links, since it is likely to proxy for various unobservable fund- and manager-specific factors and the impact on informal information links (largely "soft" information that cannot be immediately verified) may differ across funds of different sizes. Theoretical research seems to suggest that smaller funds may be able to generate better marginal benefits from informal information channels than larger funds (Stein, 2002; Berk and Green, 2004). (10) Alternatively, managers of large funds are likely to have more reputational and social capital. (11) In the empirical tests that follow, we investigate the relationship between fund size and marginal benefits from informal information channels. Comparisons within fund size categories also act as a robustness check to verify whether the baseline results hold across the spectrum of fund sizes.

Table VI presents the analyses of characteristic-adjusted returns from portfolios across fund size terciles. We rank funds into size terciles for each quarter based on TNA in the last month of the previous quarter. Panel A reports the analyses for the subsample of funds forming the smallest fund size tercile. Consistent with the findings thus far, the marginal impact of FC channels on performance continues to be positive (but insignificant) for weak FF funds and significantly negative for strong FF funds (-2.16%), supporting the prediction that information advantages are higher for more exclusive FC channels. In addition, the marginal impact of FF channels remains significantly negative (-1.99%) for holdings with strong FC information channels, and significantly positive (1.59%) in the absence of FC channels.

Panel B of Table VI presents the subsample of funds forming the largest fund size tercile. Again, weak FF funds show significant stock selection ability in holdings where they have strong FC channels. Strong FF funds underperform weak FF funds by 2.47% annually when investing in companies associated with strong FC measures. In contrast, the strong FF funds outperform the weak FF funds by 1.85% annually in investments lacking FC channels. Due to space considerations, we omit the results for medium-sized funds, which are qualitatively similar to those of large funds. Overall, the evidence regarding the marginal impact of information channels holds across different fund size categories.

4. Multivariate Results

Thus far, we have presented our univariate results for the total sample and various subsamples. These analyses allow a more focused examination of the nature of informal information links for various types of funds and provide a clear demonstration of the salient features of our results. However, the univariate analyses are subject to the concern that information channels proxy for other factors that are underlying drivers of investment performance. Therefore, we proceed to multivariate regressions with additional controls related to the portfolio manager's ability and decision-making process.

In the following regressions, we use the value-weighted portfolio return of holdings in each FC quartile as the dependent variable. As a result, one fund in a quarter accounts for four observations. The choice of this dependent variable achieves several purposes. It reduces the problem of dependence across observations in a large panel data sample without removing the option of studying performance differences within a fund's portfolio. It makes the multivariate regression results more comparable to the findings in the previous sections, while retaining the validity of a measure that aims to capture the performance generated from a portfolio in a particular information environment.

Table VII presents the results of fixed-effect regressions for various model specifications. Specifically, the value-weighted characteristic-adjusted portfolio returns for quarter t are regressed on dummy variables that capture the strength of the FF and FC links (and their interactions) in addition to a variety of control variables. Among the main explanatory variables, Strong (Weak) FF Dummy assumes a value of one if the fund is in the strongest (weakest) FF quartiles of the sample of funds in the quarter, and zero otherwise. The strongest (weakest) FF quartiles are the third and fourth (first and second) quartiles of funds formed by the sorted and ranked FF measure. Strong (Weak) FC Dummy assumes a value of one if the portfolio is the strongest (weakest) FC portfolio for the fund quarter. The strongest (weakest) FC portfolios are the third and fourth (first and second) quartiles of holdings formed by the sorted and ranked FC measure. Strong FF x Strong FC is the interaction term for Strong FF Dummy and Strong FC Dummy and represents the portfolio where the funds with strong FF links invest in stocks with relatively strong FC links. Weak FF x Strong FC is the interaction term for Weak FF Dummy and Strong FC Dummy and represents the portfolio where the funds with the weak FF links invest in stocks with relatively strong FC links.

Model (1) in Table VII is the baseline specification that includes only the information links-related variables as the explanatory variables. Model (1) demonstrates that the coefficient of Strong FF Dummy is positive and significant at the 5% level, with an economically significant magnitude of 0.013. This suggests that the strong FF links improve portfolio performance by 1.3% per year, ceteris paribus. The coefficient on Strong FF x Strong FC is significantly negative, with an economically significant magnitude of -0.021. Ceteris paribus, strong FF and strong FC channels, when combined, reduce style-adjusted returns by a net 0.8% per year. The results imply that the information links among fund managers have a positive marginal impact on stock selection ability independently, but the combination of the two channels has a detrimental impact on portfolio performance. The sign and the significance of the coefficient on the Weak FF x Strong FC variable point to a similar interpretation for the marginal impact of FC channels; that is, the FC information links have a positive impact on performance in the absence of FF channels. Overall, the results from the multivariate regressions are consistent with those from the univariate analyses presented in previous sections.

Model (2) adds a set of fund and family characteristics as control variables. The fund-level control variables in Model (2) include fund size, expenses, and turnover, among others. In Model (2), the magnitude and significance of the FF and FC measures are qualitatively similar to the baseline regression. Model (3) includes a set of variables that capture the effect of city size based on general population. As noted in Christofferson and Sarkissian (2009), the urban location of funds significantly affects their performance. It is possible that the FF and FC measures capture heterogeneities related to the city of location instead of the strength of informal communication channels. Big City Dummy is a dummy taking a value of one if the fund is headquartered in one of the 20 largest populated cities as defined by the US Census Bureau in the report year, and zero otherwise. Model (3) also includes interaction terms of the city size, FF, and FC dummies. Notably, the variables Strong FF Dummy, Strong FF x Strong FC, and Weak FF x Strong FC remain significant and are not subsumed by the city size dummies. In tact, none of the city size variables are statistically significant. In unreported regressions, we use alternative city size definitions and obtain similar results. (12) This result is interesting as it suggests that portfolio performance is related to the density of the mutual fund population, but not the density of the general population, supporting the theory that the more relevant information channels influence performance. Model (4) of Table VII includes another set of control variables measuring the local bias exhibited by fund managers, the interaction of local bias with the FF and FC measures. The results for the two information channels are consistent with the baseline regression and all of the local bias variables are insignificant. (13) Model (5) includes all of the control variables employed in Models (2) to (4), and generates results consistent with the sparser specifications. (14) Thus, after controlling for a variety of factors and various fixed effects in multivariate analyses, the FF and FC measures continue to be significantly related to stock selection ability. In sum, the multivariate analysis confirms the correlation between informal information channels, the nature of their interactions, and stock selection ability revealed in the univariate tests.

C. Copycat Portfolios and Future Asset Prices

Our analyses indicate that the portfolios with ex post superior performance are associated with: 1) strong FC information links in the absence of FF links, and 2) strong FF information links in the absence of FC links. It is possible that fund investments made based on the information acquired via these channels are indicative of future stock prices, beyond the horizon of the three-month quarterly portfolios of funds. If so, fund holdings that become publicly available information in a report quarter could be used to infer the future value of assets by observing mutual funds' portfolio strategies.

Table VIII presents an initial examination of superior information channels of mutual fund managers and their relationship with future stock returns. At the beginning of each quarter, using fund holdings reported in the previous quarter, we construct a hypothetical "copycat" portfolio, called the Best Information Portfolio (or BIP), including the stock holdings that were part of the two information environments that were (ex post) identified as beneficial in mutual fund investments. BIP contains the holdings of the two information portfolios with superior performance from reports in the most recently disclosed holdings using equal weights across stocks, and holds it starting the first month of the next quarter for up to one year.

Panel A of Table VIII presents returns for 3- to 12-month holding strategies for equal-weighted BIP portfolios, with three-month increments. We also present the returns for the subsamples of funds that vary in family size and fund size since these fund attributes have an impact on the relation between returns and information channels. The three-month holding period for the BIP generates statistically significant average monthly returns of 1.63%. The corresponding returns for the 6-, 9- and 12-month holding periods for BIP are positive and have magnitudes of 1.81%, 1.35%, and 0.85%, respectively, and remain significant until the nine-month holding period strategy. Moreover, the three- and six-month holding period returns of the BIP strategy are positive and significant for subsamples of funds with varying family and fund sizes. In summary, the overall positive returns from BIP are representative of patterns within fund subsamples, and do not appear to be driven by outliers. The evidence suggests that the information channels that are associated with better quarterly holdings performance for mutual funds seemingly select stocks that continue to outperform in the future.

While Panel A of Table VIII reports the positive returns from BIP strategies, they may be explained by the underlying characteristics of the stocks in the portfolio. Panel B presents the average monthly characteristic-adjusted returns for the BIP. The monthly returns of the BIP are significantly positive for two holding periods and are similar across fund subsamples, providing similar conclusions as the results based on raw returns. Overall, the evidence suggests potential asset pricing implications of identifying investor information channels that outperform others, where stocks picked by certain information channels are more likely to have superior future returns. Note that we have not factored in the transaction costs or established the feasibility of the strategy. A more comprehensive investigation of these asset pricing implications is beyond the scope of this study and is left for future research.

III. Concluding Remarks

This study broadly contributes to the literature on informed trading that links the geographical location of investors to informal communication channels that influence their investment decisions. Our focus is on two forms of information linkages associated with mutual fund managers that have been documented in previous studies: 1) FF information links, which transfer information about potential investment opportunities between fund managers across fund families, and 2) FC information links, which facilitate a manager's acquisition of differential information about a company via links with the company. While some previous studies have explored the performance impact of the informal information channels associated with mutual fund managers by focusing on one type of informal communication (e.g., communication between local companies and fund managers as in Coval and Moskowitz, 2001), we make the first attempt to disentangle the performance implications of more than one type of informal communication in a setting that accounts for their interactions. Overall, our results provide several novel empirical insights.

First, strong FF information links have a significantly positive marginal impact on returns in the absence of FC information links, and vice versa. As such, informal communication via either channel is beneficial for the managers' stock selection ability in isolation. However, when the two types of information channels are present together, they generate substantial underperformance from portfolio holdings and appear distinctly counterproductive in generating economic value. There is some evidence that the crowded trades effect can explain this apparently puzzling finding. Multivariate analyses confirm the results after controlling for a multitude of factors like fund size, family size, fund age, manager tenure, degree of local bias, and the size of cities in which the fund is located. We also conduct an initial examination of the link between the information environments that generate superior investment performance and future asset returns. Our findings suggest that investment decisions in stocks associated with certain beneficial information environments may predict asset returns that persist into the future.

The empirical evidence presented in our study raises intriguing questions related to the study of informed trading and suggests some interesting avenues for future research. While our findings provide some insight as to how information links related to geographical location can be valuable for fund managers, they also reveal substantial complexity in the informational benefits. As in Coval and Moskowitz (2001), our results seem to suggest that, in equilibrium, fund managers should maximize performance by concentrating their portfolios in holdings located in informationally advantageous environments, perhaps without having to substantially increase risk. Additional research that accounts for alternative informal information channels (e.g., educational networks) and identifies the complexities in the effect of informal communication in financial markets could provide further insight regarding the economic value of communication in financial markets.

References

Abdel-Khalik, A.R. and K.M. El-Sheshai, 1980, "Information Choice and Utilization in an Experiment on Default Prediction," Journal of Accounting Research 18, 325-342.

Bala, V. and S. Goyal, 1998, "Learning from Neighbors," Review of Economic Studies 65, 595-621.

Berk, J. and R.C. Green, 2004, "Mutual Fund Flows and Performance in Rational Markets," Journal of Political Economy 112, 1269-1295.

Bikchandani, S., D. Hirshleifer, and I. Welch, 1992, "A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades," Journal of Political Economy 100, 992-1026.

Cao, H., B. Han, and D. Hirshleifer, 2011, "Taking the Road Less Traveled By: Does Conversation Eradicate Pernicious Cascades T' Journal of Economic Theory 146, 1418-1436.

Christofferson, S.E.K. and S. Sarkissian, 2009, "City Size and Fund Performance," Journal of Financial Economics 92, 252-275.

Cohen, L., A. Frazzini, and C. Malloy, 2008, "The Small World of Investing: Board Connections and Mutual Fund Returns," Journal of Political Economy 116, 951-979.

Colla, P. and A. Mele, 2009, "Information Linkages and Correlated Trading," Review of Financial Studies 23, 203-246.

Coval, J.D. and T.J. Moskowitz, 2001, "The Geography of Investment: Informed Trading and Asset Prices," Journal of Political Economy 109, 811-841.

Daniel, K., M. Grinblatt, S. Titman, and R. Wermers, 1997, "Measuring Mutual Fund Performance with Characteristic-Based Benchmarks," Journal of Finance 52, 1035-1058.

Demarzo, P.M., D. Vayanos, and J. Zwiebel, 2003, "Persuasion Bias, Social Influence and Uni-Dimensional Opinions," Quarterly Journal of Economics, 118, 909-968.

Duflo, E. and E. Saez, 2002, "Participation and Investment Decisions in a Retirement Plan: The Influence of Colleagues' Choices," Journal of Public Economics 85, 121-148.

Duflo, E. and E. Saez, 2003, "The Role of Information and Social Interactions in Retirement Plan Decisions: Evidence from a Randomized Experiment," Quarterly Journal of Public Economics 118,815-842.

Ellison, G.D. and D. Fudenberg, 1995, "Word of Mouth Communication and Social Learning," Quarterly Journal of Economics 110, 93-125.

Fama, E.F. and J.D. MacBeth, 1973, "Risk, Return, and Equilibrium: Empirical Tests," Journal of Political Economy 81,607-636.

Feng, L. and M.S. Seasholes, 2004, "Correlated Trading and Location," Journal of Finance 59, 2117-2144.

Hogarth, R.M., 1980, Judgment and Choice: The Psychology of Decision, New York, NY, John Wiley and Sons.

Hong, H., J.D. Kubik, and J.C. Stein, 2004, "Social Interaction and Stock-Market Participation," Journal of Finance 59, 137-163.

Hong, H., J.D. Kubik, and J.C. Stein, 2005, "Thy Neighbor's Portfolio: Word-of-Mouth Effects in the Holdings and Trades of Money Managers," Journal of Finance 60, 2801-2824.

Ivkovic, Z., C. Sialm, and S. Weisbenner, 2008, "Portfolio Concentration and the Mutual Fund Performance," Journal of Financial and Quantitative Analysis 43, 613-656.

Ivkovic, Z. and S. Weisbenner, 2005, "Local Does as Local Is: Information Content of the Geography of Individual Investors' Common Stock Investments," Journal of Finance 60, 267-306.

Ivkovic, Z. and S. Weisbenner, 2007, "Information Diffusion Effects in Individual Investors' Common Stock Purchases: Covet Thy Neighbors' Investment Choices," Review of Financial Studies 20, 1327-1357.

Lakonishok, J., A. Shleifer, and R.W. Vishny, 1992, "The Impact of Institutional Trading on Stock Prices," Journal of Financial Economics 32, 23-44.

Massa, M. and A. Simonov, 2006, "Hedging, Familiarity and Portfolio Choice," Review of Financial Studies 19, 633-685.

Ng, L. and F. Wu, 2010, "Peer Effects in Investor Trading Decisions: Evidence from a Natural Experiment," Financial Management 39, 807-831.

Ozsoylev, H.N., 2005, "Asset Pricing Implications of Social Networks," University of Oxford, Said Business School, Working Paper.

Parwada, J.T., 2008, "The Genesis of Home Bias? The Location and Portfolio Choices of Investment Company Start-Ups," Journal of Financial and Quantitative Analysis 43, 245-266.

Shiller, R.J., 2000, Irrational Exuberance, Princeton, NJ, Princeton University Press.

Stein, J.C., 2002, "Information Production and Capital Allocation: Decentralized versus Hierarchical Firms," Journal of Finance 57, 1891-1921.

Stein, J.C., 2008, "Conversations Among Competitors," American Economic Review 98, 2150-2162.

Stein, J.C., 2009, "Presidential Address: Sophisticated Investors and Market Efficiency," Journal of Finance 64, 1517-1548.

Tversky, A. and D. Kahneman, 1974, "Judgment under Uncertainty: Heuristics and Biases," Science 185, 1124-1131.

Wermers, R., 1999, "Mutual Fund Herding and the Impact on Stock Prices," Journal of Finance 54, 581-622.

We thank Professor Raghavendra Rau (Editor) and an anonymous referee for their help in significantly improving the paper. All errors remain ours.

Richard Fu and Swasti Gupta-Mukherjee *

* Richard Fu is an Associate Professor of Finance in the Collat School of Business, University of Alabama at Birmingham in Birmingham, AL. Swasti Gupta-Mukherjee is an Assistant Professor of Finance in the Quinlan School of Business Administration, Loyola University Chicago in Chicago IL

(1) For empirical evidence concerning informal communication in financial markets, see Shiller (2000), Hong, Kubik, and Stein (2004), and Duflo and Saez (2002, 2003). For theoretical studies drawing attention to the impact of social communication on asset prices, see Ozsoylev (2005) and Colla and Mele (2009).

(2) Ivkovic and Weisbenner (2005), Massa and Simonov (2006), and Ivkovic, Sialm, and Weisbenner (2008) also find similar evidence. In addition, Parwada (2008) uses a sample of entrepreneurial fund managers to confirm that managers tend to locate near their former employers, perhaps to access professional, social, and family networks.

(3) Some other studies also uncover the effects of informal communication in investment decisions. Ivkovic and Weisbenner (2007) find word-of-mouth communication among individual investors as reflected in the trades of neighbors. Ng and Wu (2010) note similar evidence for investors in the same trading room. Duflo and Saez (2002,2003) confirm the impact of social interaction on individuals' retirement plan decisions. Feng and Seasholes (2004) document correlated trading among investors in the same region.

(4) This notion is also consistent with cognitive studies indicating that human beings find it difficult to combine a lot of information simultaneously (Tversky and Kahneman, 1974; Abdel-Khalik and El-Sheshai, 1980; Hogarth, 1980). Some other related studies regarding informal communication include Bikchandani, Hirshleifer, and Welch (1992) and Cao, Han, and Hirshleifer (2011), which posit that it is possible for economic outcomes to be suboptimal when communication occurs.

(5) Throughout the paper, we use the notion of "strength" of information flows in the context of volume (or quantity) of information flows, while remaining agnostic about the quality implications.

(6) The main results remain materially unchanged if we exclude balanced funds from the sample.

(7) We calculated an alternative fund-fund (FF) link measure including sector funds. The alternative measure has a correlation close to one with the FF measure used in the study.

(8) In unreported robustness checks, we use Coval and Moskovitz's (2001) binary measure and the results are qualitatively similar.

(9) Using equal weight yields qualitatively similar results.

(10) Berk and Green (2004) argue that larger funds have managers with more managerial skills. In their model, one of the explanations for a skilled manager's failure to consistently outperform passive benchmarks is that the manager spreads her information acquisition activities too thin across various assets while managing sufficiently large funds. Stein (2002) relates organizational form to firm size and posits that "soft" or unverifiable information generates better performance in smaller, single manager, decentralized firms as compared to large hierarchical firms.

(11) Managers of larger funds may be the originators of the information that is being transmitted via fund-fund channels. In this case, these managers are more likely to have taken long positions in stocks before the information diffuses to other managers resulting in increased stock prices. This would be reflected as higher returns and, consequently, have a greater marginal impact of fund-fund flows for these managers. However, this process cannot clearly explain the relationship between the marginal impact of fund-company channels and fund size since managers of smaller funds can also develop direct ties with companies.

(12) Alternative measures of city size that were also considered were remote city dummies (as defined by Coval and Moskowitz, 2001) and big city as defined by the top 10 cities by population.

(13) In unreported results, the cross-sectional regression methodology outlined in Fama and MacBeth (1973) was also used as an alternative. The results were qualitatively similar to the reported pooled panel data regressions using robust standard errors. Since the magnitude of the economic impact of FC and FF information flows are difficult to interpret from the Fama-MacBeth (1973) methodology, only the pooled panel data regressions are reported and interpreted in this section for brevity.

(14) In an unreported robustness check, we replicate the multivariate regressions reported above for a subsample of large cap stocks (the two largest DGTW size quintiles) to address the possibility that informal communication may not play a significant role in the stock holdings of companies that have the most publicly available information, visibility, and analyst following. However, the qualitative results are indistinguishable from the representative regressions suggesting that public information availability does not preclude the impact of informal information flows.
Table I. Summary Statistics on US Mutual Fund

Fund age in the holding quarter is computed as the difference between
the year of the first month in the quarter and the year of the fund's
first offer date. Proximity distribution of a fund's holdings is
reported as the percentage of the total amount invested in domestic
equity in a quarter, in stocks of companies that are 0 to 100, 100 to
500, 500 to 1,000 km, and more than 1,000 km from the location of the
fund. Proximity distribution of mutual funds is computed for each fund
as the percentage of funds in the universe of funds located 0 to
185 100,100 to 500, 500 to 1,000 km, and more than 1,000 km from each
specific fund, and reported as an average across all funds. Median
number of companies in a fund's portfolio is the average across all
quarters. Daniel et al. (DGTW) (1997) is followed to compute the size,
book-to-market, and momentum quintiles for the stocks held. The
value-weighted characteristics are reported for the median fund.
Company age is the difference between the year of the first month of
the quarter and the year in which a company was established, reported
in years.

Total number of funds in sample                   2,931
Total number of mutual fund families in sample    516
Median total net assets ($ million)               226.1
Median fund age (in years)                        10.0
Proximity distribution of fund investments:
  Avg. % invested in stocks: 0 to 100 km          5.41
  Avg. % invested in stocks: 100 to 500 km        12.56
  Avg. % invested in stocks: 500 to 1,000 km      15.31
  Avg. % invested in stocks: > 1,000 km           66.72
Proximity distribution of fund managers:
  Avg. % of funds located 0 to 100 km             10.23
  Avg. % of funds located 100 to 500 km           20.12
  Avg. % of funds located 500 to 1,000 km         9.62
  Avg. % of funds located > 1,000 km              60.03
Portfolio characteristics:
  Median number of companies in portfolio         74
  Median value-weighted DGTW size quintile        4.21
  Median value-weighted DGTW B/M quintile         2.50
  Median value-weighted DGTW MOM quintile         3.13
  Median company age (in years)                   37.0

Table II. Summary Statistics on Portfolio Characteristics

The likelihood of a fund-company (FC) information channel existing
between fund manager j and each company issuing stock n that is held
in j's portfolio is [L.sup.FC] [(Links).sup.j,n] = 1/(1+
[Distance.sub.j,n]). Here, [Distance.sub.j,n] is the natural
logarithm of the geographical distance (in km) between the city where
fund j is located and the city in which company n is headquartered.
Every quarter, each fund's holdings are ranked into quartiles
[Q1.sup.FC] through [Q4.sup.FC] based on [L.sup.FC] [(Links).sup.j,n],
with [Q4.sup.FC] being the Strong Fund-Company Links portfolio and
[Q1.sup.FC] being the Weak Fund-Company Links portfolio. The
likelihood of an information channel existing between fund manager j
and fund manager i is [L.sup.FF] [(Links).sup.j,i] = 1/(l +
[Distance.sub.j,i]) where [Distance.sub.j,i] is the natural logarithm
of the geographical distance between the city where fund j is located
and the city in which fund i is located. The strength of fund-fund
(FF) informal information links a fund manager j has with other fund
managers is computed as

[(Links).sup.FF.sub.j] = [N.summation over (i=1)] [L.sup.FF]
[(Links).sup.j,t] = [N.summation over (i=1)] [1/(1 +
[Distance.sub.j,t])]

Here, i = 1, ..., N are all funds in the sample that do not belong to
the same fund family as fund j. Every quarter, the funds are ranked
into quartiles [Q1.sup.FF] through [Q4.sup.FF] based on
[(Links).sup.FF.sub.j], with [Q4.sup.FF] being the Strong Fund-Fund
Links portfolio and [Q1.sup.FF] being the Weak Fund-Fund Links
portfolio. Fund Age is computed from the first offer date. Mgr. tenure
is the length of manager tenure at the fund. # of funds in family is
the number of funds in the family. Size Q, BE/ME Q, Mom Q are the
value-weighted Daniel et al. (DGTW) (1997) size, book-to-market and
momentum quintiles for the stocks in the portfolio averaged across all
funds. % invested is calculated as the portfolio weight of stocks in
an FC quartile for a fund by market value of holdings, averaged across
all funds.

                                                  Weak Fund-
                                                Fund (FF) Links

                                                  [Q1.sup.FF]

                                                Mean     [Median]

A. Fund Characteristics:
TNA (in $mill)                                 558.9     [206.32]
Fund age (in years)                             13.2      [10.0]
Expenses                                         0.011     [0.012]
Turnover                                         0.97      [0.88]
# of funds in family                            52       [28]
Manager tenure (in years)                        5.9       [5.0]
B. Portfolio Characteristics:
Strong Fund-Company (FC) Links
[Q4.sup.FC]                       Size Q         3.95
                                  BE/ME Q        2.38
                                  Mom Q          3.17
                                  % invested    24.67
[Q3.sup.FC]                       Size Q         4.02
                                  BE/ME Q        2.63
                                  Mom Q          3.20
                                  % invested    25.04
[Q2.sup.FC]                       Size Q         3.92
                                  BE/ME Q        2.80
                                  Mom Q          3.20
                                  % invested    25.57
[Q1.sup.FC]                       Size Q         3.86
Weak Fund-Company (FC) Links      BE/ME Q        2.39
                                  Mom Q          3.24
                                  % invested    25.79

                                                   [Q2.sup.FF]

                                                 Mean      [Median]

A. Fund Characteristics:
TNA (in $mill)                                 1,011.2     [274.67]
Fund age (in years)                               12.7      [10.0]
Expenses                                           0.012     [0.012]
Turnover                                           1.41      [0.93]
# of funds in family                               40        [38]
Manager tenure (in years)                          6.3       [5.0]
B. Portfolio Characteristics:
Strong Fund-Company (FC) Links
[Q4.sup.FC]                       Size Q           4.13
                                  BE/ME Q          2.83
                                  Mom Q            3.09
                                  % invested      24.40
[Q3.sup.FC]                       Size Q           3.82
                                  BE/ME Q          3.01
                                  Mom Q            3.26
                                  % invested      25.78
[Q2.sup.FC]                       Size Q           4.13
                                  BE/ME Q          2.71
                                  Mom Q            3.37
                                  % invested      26.02
[Q1.sup.FC]                       Size Q           3.98
Weak Fund-Company (FC) Links      BE/ME Q          2.12
                                  Mom Q            3.01
                                  % invested      25.01

                                               [Q3.sup.FF]

                                                 Mean      [Median]

A. Fund Characteristics:
TNA (in $mill)                                 1,084.1     [201.71]
Fund age (in years)                               12.8      [10.0]
Expenses                                           0.011     [0.012]
Turnover                                           1.22      [0.94]
# of funds in family                               60        [31]
Manager tenure (in years)                          5.8       [5.0]
B. Portfolio Characteristics:
Strong Fund-Company (FC) Links
[Q4.sup.FC]                       Size Q           4.02
                                  BE/ME Q          2.84
                                  Mom Q            3.12
                                  % invested      24.93
[Q3.sup.FC]                       Size Q           4.11
                                  BE/ME Q          3.02
                                  Mom Q            3.29
                                  % invested      24.39
[Q2.sup.FC]                       Size Q           4.01
                                  BE/ME Q          3.02
                                  Mom Q            3.41
                                  % invested      24.14
[Q1.sup.FC]                       Size Q           4.02
Weak Fund-Company (FC) Links      BE/ME Q          2.71
                                  Mom Q            3.33
                                  % invested      24.91

                                               Strong Fund-
                                               Fund (FF) Links

                                               [Q4.sup.FF]

                                                Mean     [Median]

A. Fund Characteristics:
TNA (in $mill)                                 602.2     [285.56]
Fund age (in years)                             12.1      [10.0]
Expenses                                         0.012     [0.012]
Turnover                                         0.98      [0.86]
# of funds in family                             71        [59]
Manager tenure (in years)                        4.8       [4.0]
B. Portfolio Characteristics:
Strong Fund-Company (FC) Links
[Q4.sup.FC]                       Size Q         4.18
                                  BE/ME Q        2.60
                                  Mom Q          3.16
                                  % invested    26.00
[Q3.sup.FC]                       Size Q         4.13
                                  BE/ME Q        2.60
                                  Mom Q          3.09
                                  % invested    24.79
[Q2.sup.FC]                       Size Q         3.98
                                  BE/ME Q        2.40
                                  Mom Q          3.02
                                  % invested    24.27
[Q1.sup.FC]                       Size Q         4.08
Weak Fund-Company (FC) Links      BE/ME Q        2.67
                                  Mom Q          3.33
                                  % invested    24.29

Table III. Performance of Mutual Funds and Informal Information
Channels (Percentage Annualized Returns)

Every quarter, the likelihood of a fund-company (FC) information
channel existing between fund manager j and each company issuing stock
n that is held in j's portfolio is calculated as
[L.sup.FC][(Links).sub.j,n] = 1/(1 + [Distance.sub.j,n]). Here,
[Distance.sub.j,n] is the natural logarithm of the geographical
distance (in km) between the city where fund j is located and the city
in which company n is headquartered. Each fund's holdings are ranked
into quartiles [Q1.sup.FC] through [Q4.sup.FC] based on
[L.sup.FC][(Links).sub.j,n], with [Q4.sup.FC] as the Strong
Fund-Company Links portfolio and [Q1.sup.FC] as the Weak Fund-Company
Links portfolio. The likelihood of an information channel existing
between fund manager j and fund manager i is
[L.sup.FF][(Links).sub.j,i] = = 1/(1 + [Distance.sub.j,i]), where
[Distance.sub.j,i], is the natural logarithm of the geographical
distance between the city where fund j is located and the city in
which fund i is located. The strength of fund-fund (FF) informal
information links fund manager j has with other fund managers is
computed as

[(Links).sup.FF.sub.j] = [N.summation over (i=l)] [L.sup.FF]
[(Links).sub.j,t] = [N.summation over (i=l)] [1/(1 +
[Distance.sub.j,t])]

Here, i = 1, ..., N are all of the funds in the sample that do not
belong to the same fund family as fund j. Every quarter, the funds are
ranked into quartiles [Q1.sup.FF] through [Q4.sup.FF] based on
[(Links).sup.FF.sub.j], with [Q4.sup.FF] as the Strong Fund-Fund Links
portfolio and [Q1.sup.FF] as the Weak Fund-Fund Links portfolio. For
each fund, returns are computed in each quarter for an FC links
portfolio as

[R.sup.raw] = [1/3] [S.summation over (i=l)] [3.summation over (t=l)]
[w.sub.i,t] [R.sub.i,t],

where [w.sub.i,t] is the portfolio weight of stock i in month /, and S
is the number of stocks in the portfolio. Characteristic-adjusted FC
links portfolio returns are computed in each quarter as

[R.sup.adj] = [1/3] [S.summation over (i=l)] [3.summation over (t=l)]
[w.sub.i,t] ([R.sub.i,t] - [R.sup.bench.sub.i]),

where [R.sup.bench.sub.i] is the value-weighted monthly return of
stock i's DGTW quintile benchmark portfolio (following Daniel et al.,
1997). Reported returns (expressed in annual percentage rates) are
value-weighted by fund total net assets (TNA) within each FF quartile
and averaged across quarters

                                       Weak Fund-      [Q2.sup.FF]
                                     Fund (FF) Links
                                       [Q1.sup.FF]

                                     Panel A. Raw Returns

Strong Fund-Company (FC) Links
[Q4.sup.FC]                          18.43              19.01
[Q3.sup.FC]                          18.17              17.45
[Q2.sup.FC]                          17.02              17.10
[Q1.sup.FC]                          15.80              18.24
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]-[Q1.sup.FC])    2.63 **            0.77
[t-statistic]                        [2.32]             [1.07]

                          Panel B. Characteristic-Adjusted Return

Strong Fund-Company (FC) Links
[Q4.sup.FC]                           2.34               0.92
[Q3.sup.FC]                           1.33               0.03
[Q2.sup.FC]                           0.11              -0.34
[Q1.sup.FC]                          -0.68               1.83
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]-[Q1.sup.FC])    3.02 ***          -0.91
[t-statistic]                        [2.98]            [-1.03]

                                     [Q3.sup.FF]    Strong Fund-
                                                   Fund (FF) Links
                                                     [Q4.sup.FF]

                                      Panel A. Raw Returns

Strong Fund-Company (FC) Links
[Q4.sup.FC]                           15.71             14.21
[Q3.sup.FC]                           17.20             17.50
[Q2.sup.FC]                           19.01             18.91
[Q1.sup.FC]                           18.83             19.33
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]-[Q1.sup.FC])    -3.12 ***         -5.12 ***
[t-statistic]                        [-3.10]           [-4.39]

                          Panel B. Characteristic-Adjusted Return

Strong Fund-Company (FC) Links
[Q4.sup.FC]                           -0.40             -1.40
[Q3.sup.FC]                           -0.26             -0.35
[Q2.sup.FC]                           -0.30              0.17
[Q1.sup.FC]                            1.40              2.23
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]-[Q1.sup.FC])    -1.80 **          -3.63 ***
[t-statistic]                        [-2.42]           [-4.12]

                                          Marginal Impact:
                                              FF Links
                                     ([Q4.sup.FF]-[Q1.sup.FF])
                                           [t-statistic]

Panel A. Raw Returns

Strong Fund-Company (FC) Links
[Q4.sup.FC]                            -4.22 ***        (-3.57]
[Q3.sup.FC]                            -0.67            [-0.83]
[Q2.sup.FC]                             1.89 **          [2.10]
[Q1.sup.FC]                             3.53 ***         [3.40]
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]-[Q1.sup.FC])
[t-statistic]

Panel B. Characteristic-Adjusted Return

Strong Fund-Company (FC) Links
[Q4.sup.FC]                            -3.74 ***        [-3.94]
[Q3.sup.FC]                            -1.68 **         [-2.31]
[Q2.sup.FC]                             0.06             [0.23]
[Q1.sup.FC]                             2.91 ***         [2.62]
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]-[Q1.sup.FC])
[t-statistic]

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table IV. Crowded Trades and Informal Information Channels (Dollar
Ratio Trade Imbalance and Herding)

Every quarter, the likelihood of a fund-company (FC) information
channel existing between fund manager j and each company issuing stock
n that is held in j's portfolio is calculated as
[L.sup.FC][(Links).sub.j,n] = 1-(1+ [Distance.sub.j,n]). Here,
[Distance.sub.j,n] is the natural logarithm of the geographical
distance (in km) between the city where fund j is located and the city
in which company n is headquartered. Each fund's holdings are ranked
into quartiles [Q1.sup.FC] through [Q4.sup.FC] based on
[L.sup.FC][(Links).sub.j,n] with [Q4.sup.FC] as the Strong
Fund-Company Link's, portfolio and [Q1.sup.FC] as the Weak
Fund-Company Links portfolio. The likelihood of an information channel
existing between fund manager j and fund manager i is
[L.sup.FF][(Links).sub.j,i] = 1-(1 + [Distance.sub.j,i]) where
[Distance.sub.j,i] is the natural logarithm of the geographical
distance between the city where fund j is located and the city in
which fund i is located. The strength of fund-fund (FF) informal
information links a fund manager j has with other fund managers is
computed as

[(Links).sup.FF.sub.j] = [N.summation over (i=1)]
[L.sup.FF][(Links).sub.j,t] = [N.summation over (i=1) 1/(1 +
[Distance.sub.j,t]).

Here, i = 1, ..., N are all funds in the sample that do not belong to
the same fund family as fund j. Every quarter, the funds are ranked
into quartiles [Q1.sup.FF] through [Q4.sup.FF] based
on[(Links).sup.FF.sub.j], with [Q4.sup.FF] as the Strong Fund-Fund
Links portfolio and [Q1.sup.FF] as the Weak Fund-Fund Links portfolio.
For each stock i held by a fund in quarter t, the absolute dollar
ratio trade imbalance measure ([absolute value of DRatio]) for the
stock quarter is computed as

[absolute value of D [Ratio].sub.i,t]] = [absolute value of
$[buy.sub.i,t] -$[sell.sub.i,t]]-[absolute value of $[buy.sub.i,t] -
$[sell.sub.i,t]]

where $[buy.sub.i,t] ($[sell.sub.i,t]) equals the total purchases
(sales) by all mutual funds, in dollars, of stock i during quarter t
(applying the average of the beginning-and end-of-quarter prices to
aggregate increases (decreases) in shareholdings for that stock
quarter following Wermers, 1999). Panel A reports the time-series
equal-weighted average of the absolute dollar ratio trade imbalance
measure across all stocks in the fund's portfolio, averaged across all
funds in the sample in a quarter. For each stock i held by a fund in
quarter t, the herding measure (HM) for the stock quarter is computed
following Wermers (1999) as

[HM.sub.i,t] = [absolute value of [p.sub.i,t] -E[[p.sub.i,t]] -
E[absolute value of [p.sub.i,t] -E[[p.sub.i,t]]

where [p.sub.i,t] is the proportion of funds trading stock i during
quarter t that are buyers. The proxy used for E[[p.sub.i,t]] is the
proportion of all stock trades by mutual funds during quarter t that
are buys. E[absolute value of [p.sub.i,t] -E[[p.sub.i,t]]] is
calculated under the null hypothesis of herding only by random chance.
Panel B reports the time-series equal-weighted average of the HM
across all stocks in the fund's portfolio, averaged across all funds
in the sample in a quarter.

                                        Weak Fund-      [Q2.sup.FF]
                                      Fund (FF) Links
                                        [Q1.sup.FF]

Panel A. Absolute Dollar Ratio ([absolute value of DRatio]) Trade
Imbalance Measure

Strong Fund-Company (FC) Links
[Q4.sup.FC]                             0.113            0.092
[Q3.sup.FC]                             0.089            0.085
[Q2.sup.FC]                             0.090            0.077
[Q1.sup.FC]                             0.108            0.084
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])    -0.005            0.008
[t-statistic]                         [-0.22]           [0.63]

Panel B. Herding Measure (HM)

Strong Fund-Company (FC) Links
[Q4.sup.FC]                             1.845            3.102
[Q3.sup.FC]                             1.566            2.140
[Q2.sup.FC]                             2.093            3.092
[Q1.sup.FC]                             2.271            2.834
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])    -0.426            0.268
[t-statistic]                         [-0.37]           [0.42]

                                      [Q3.sup.FF]       Strong Fund-
                                                       Fund (FF) Links
                                                         [Q4.sup.FF]

Panel A. Absolute Dollar Ratio ([absolute value of DRatio]) Trade
Imbalance Measure

Strong Fund-Company (FC) Links
[Q4.sup.FC]                             0.080           0.142
[Q3.sup.FC]                             0.121           0.105
[Q2.sup.FC]                             0.103           0.101
[Q1.sup.FC]                             0.091           0.099
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])    -0.011           0.043 ***
[t-statistic]                         [-1.29]          [3.43]

Panel B. Herding Measure (HM)

Strong Fund-Company (FC) Links
[Q4.sup.FC]                             2.830           2.952
[Q3.sup.FC]                             2.261           2.551
[Q2.sup.FC]                             1.733           2.020
[Q1.sup.FC]                             1.568           1.520
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])     1.262 *         1.432 ***
[t-statistic]                          [1.90]          [2.57]

                                      Marginal Impact:
                                      FF Links ([Q4.sup.FF]
                                      -[Q1.sup.FF])
                                      [t-statistic]

Panel A. Absolute Dollar Ratio ([absolute value of DRatio]) Trade
Imbalance Measure

Strong Fund-Company (FC) Links
[Q4.sup.FC]                            0.029 **   [2.41]
[Q3.sup.FC]                            0.016 *    [1.99]
[Q2.sup.FC]                            0.011      [1.43]
[Q1.sup.FC]                           -0.009      [-1.21]
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])
[t-statistic]

Panel B. Herding Measure (HM)

Strong Fund-Company (FC) Links
[Q4.sup.FC]                            1.107 **   [1.98]
[Q3.sup.FC]                            0.985 *    [1.74]
[Q2.sup.FC]                           -0.073      [-0.22]
[Q1.sup.FC]                           -0.751      [-1.12]
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])
[t-statistic]

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table V. Performance of Mutual Funds and Informal Information Channels
(By Family Size)

Every quarter, the likelihood of a fund-company (FC) information
channel existing between fund manager j and each company issuing stock
n that is held in j's portfolio is calculated as
[L.sup.FC][(Links).sub.j,n] = 1-(1 + [Distance.sub.j,n]). Here,
[Distance.sub.j,n] is the natural logarithm of the geographical
distance (in km) between the city where fund j is located and the city
in which company n is headquartered. Each fund's holdings are ranked
into quartiles [Q1.sup.FC] through [Q4.sup.FC] based on [L.sup.FC]
[(Links).sub.j,n], with [Q4.sup.FC] as the Strong Fund-Company Links
portfolio and [Q1.sup.FC] as the Weak Fund-Company Links portfolio.
The likelihood of an information channel existing between fund manager
j and fund manager j is [L.sup.FF][(Links).sub.j,i] = 1-(1 +
[Distance.sub.j,i]) where [Distance.sub.j,i] is the natural logarithm
of the geographical distance between the city where fund j is located
and the city in which fund i is located. The strength of fund-fund
(FF) informal information links a fund manager j has with other fund
managers is computed as

[(Links).sup.FF.sub.j] = [N.summation over (i=1)]
[L.sup.FF][(Links).sub.j,t] = [N.summation over (i=1)] 1/
(l+[Distance.sub.j,t]).

Here, i = 1, ..., N are all funds in the sample that do not belong to
the same fund family as fund j. Every quarter, the funds are ranked
into quartiles [Q1.sup.FF] through [Q4.sup.FF] based
on[(Links).sup.FF.sub.j], with [Q4.sup.FF] as the Strong Fund-Fund
Links portfolio and [Q1.sup.FF] as the Weak Fund-Fund Links portfolio.
For each fund, returns are computed in each quarter for an FC links
portfolio as

[R.sup.raw] = 1/3 [S.summation over (i=1)] [S.summation over
(i=1)][w.sub.i,t] [R.sub.i,t],

where [w.sub.i,t] is the portfolio weight of stock i in month t, and S
is the number of stocks in the portfolio. Characteristic-adjusted FC
links portfolio returns are computed in each quarter as

[R.sup.adj] = 1-3 [S.summation over (i=1)] [S.summation over (t=1)]
[w.sub.i,t]([R.sup.i,t] -[R.sup.bench.sub.i]),

where [R.sup.bench.sub.i] is the value-weighted monthly return of
stock i's DGTW quintile benchmark portfolio (following Daniel et ah,
1997). Every quarter, the funds are sorted into terciles by the number
of funds in the family, with the lowest and highest terciles as the
small and large fund families, respectively. Panel A (Panel B) reports
returns for funds in small (large) families. Returns (expressed in
annual percentage rates) are value-weighted by fund total net assets
(TNA) within each FF quartile and averaged across quarters

                                        Weak Fund-       [Q2.sup.FF]
                                       Fund (FF) Links
                                         [Q1.sup.FF]
Panel A. Small Fund Families

Strong Fund-Company (FC) Links              1.13            -0.37
[Q4.sup.FC]
Weak Fund-Company (FC) Links
[Q3.sup.FC]                                 1.02            -0.12
[Q2.sup.FC]                                 0.54            0.45
[Q1.sup.FC]                                 0.13            1.13
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])         1.00           -1.50 *
[t-statistic]                              [1.49]          [-1.90]

Panel B. Large Fund Families

Strong Fund-Company (FC) Links              1.86            1.40
[Q4.sup.FC]
[Q3.sup.FC]                                 0.82            0.53
[Q2.sup.FC]                                 0.14            -0.12
[Q1.sup.FC]                                 0.03            0.23
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])        1.83 **          1.17
[t-statistic]                              [2.18]          [1.23]

                                       [Q3.sup.FF]     Strong Fund-
                                                     Fund (FF) Links
                                                       [Q4.sup.FF]
Panel A. Small Fund Families

Strong Fund-Company (FC) Links            -0.53           -1.84
[Q4.sup.FC]
Weak Fund-Company (FC) Links
[Q3.sup.FC]                               -0.72           -0.28
[Q2.sup.FC]                                1.03            0.05
[Q1.sup.FC]                                0.91            1.45
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])       -1.44 *         -3.29 **
[t-statistic]                            [-1.84]         [-3.52]

Panel B. Large Fund Families

Strong Fund-Company (FC) Links            -0.93           -0.67
[Q4.sup.FC]
[Q3.sup.FC]                               -0.76            0.30
[Q2.sup.FC]                                0.41            1.02
[Q1.sup.FC]                                1.29            1.76
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])       -2.22 ***       -2.43 ***
[t-statistic]                            [-2.58]          [2.75]

                                             Marginal Impact:
                                                 FF Links
                                       ([Q4.sup.FF]--[Q1.sup.FF])
Panel A. Small Fund Families                  [t-statistic]

Strong Fund-Company (FC) Links           -2.97 ***       [-2.60]
[Q4.sup.FC]
Weak Fund-Company (FC) Links
[Q3.sup.FC]                              -1.30           [-1.43]
[Q2.sup.FC]                              -0.49           [-1.03]
[Q1.sup.FC]                               1.32            [1.50]
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])
[t-statistic]

Panel B. Large Fund Families

Strong Fund-Company (FC) Links           -2.53 **        [-2.51]
[Q4.sup.FC]
[Q3.sup.FC]                              -0.52            [0.95]
[Q2.sup.FC]                               0.88            [1.62]
[Q1.sup.FC]                               1.73 **         [2.34]
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])
[t-statistic]

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table VI. Performance of Mutual Funds and Informal Information
Channels (by Fund Size)

Every quarter, the likelihood of a fund-company (FC) information
channel existing between fund manager j and each company issuing stock
n that is held in j's portfolio is calculated as [L.sup.FC]
[(Links).sub.j,n] = 1-(1 + [Distance.sub.j,n]). Here,
[Distance.sub.j,n] is the natural logarithm of the geographical
distance (in km) between the city where fund j is located and the city
in which company n is headquartered. Each fund's holdings are ranked
into quartiles [Q1.sup.FC] through [Q4.sup.FC] based on
[L.sup.FC][(Links).sub.j,n] with [Q4.sup.FC] as the Strong
Fund-Company Links portfolio and [Q1.sup.FC] as the Weak Fund-Company
Links portfolio. The likelihood of an information channel existing
between fund manager j and fund manager i is [L.sup.FF]
[(Links).sub.j,i] = 1-(1 + [Distance.sub.j,i]), where
[Distance.sub.j,i] is the natural logarithm of the geographical
distance between the city where fund j is located and the city in
which fund i is located. The strength of fund-fund (FF) informal
information links a fund manager j has with other fund managers
is computed as

[(Links).sup.FF.sub.j] = [N.summation over (i=1)]
[L.sup.FF][(Links).sub.j,t] = [N.summation over (i=1)] 1/
(1+[Distance.sub.j,t]).

Here, i = 1, ..., N are all funds in the sample that do not belong to
the same fund family as fund j. Every quarter, the funds are ranked
into quartiles [Q1.sup.FF] through [Q4.sup.FF] based
on[(Links).sup.FF.sub.j], with [Q4.sup.FF] as the Strong Fund-Fund
Links portfolio and [Q1.sup.FF] as the Weak Fund-Fund Links portfolio.
For each fund, returns are computed in each quarter for an FC links
portfolio as

[R.sup.raw] = 1/2 [S.summation over (i=1)][3.summation over (t=1)]
[w.sub.i,t] [R.sub.i,t],

where [w.sub.i,t], is the portfolio weight of stock i in month t, and
S is the number of stocks in the portfolio. Characteristic-adjusted FC
links portfolio returns are computed in each quarter as

[R.sup.adj] = 1/3 [S.summation over (i=1)][3.summation over (t=1)]
[w.sub.i,t] [R.sub.i,t] - [R.sup.bench.sub.i],

where [R.sup.bench.sub.i] is the value-weighted monthly return of
stock i's DGTW quintile benchmark portfolio (following Daniel et al.,
1997). Every quarter, the funds are sorted into terciles by total net
assets (TNA) in the month prior to the beginning of the quarter. The
lowest and highest terciles are the small and large funds,
respectively. Panel A (Panel B) reports returns for small (large)
funds. Returns (expressed in annual percentage rates) are
value-weighted by fund total net assets (TNA) within each FF quartile
and averaged across quarters

                                         Weak Fund-      [Q2.sup.FF]
                                      Fund (FF) Links
                                        [Q1.sup.FF]

Panel A. Small Funds

Strong Fund-Company (FC) Links
[Q4.sup.FC]                                 1.76             2.03
[Q3.sup.FC]                                 1.23             1.20
[Q2.sup.FC]                                 0.94            -0.07
[Q1.sup.FC]                                 0.34             1.57
Weak Fund-Fund (FF) Links
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])         1.42             0.46
[t-statistic]                              [1.23]           [0.37]

Panel B. Large Funds

Strong Fund-Company (FC) Links
[Q4.sup.FC]                                 1.34             1.65
[Q3.sup.FC]                                 0.71             0.66
[Q2.sup.FC]                                -0.40            -0.09
[Q1.sup.FC]                                -0.57            -0.13
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])         1.91 **         1.78 **
[t-statistic]                              [2.23]          [2.01]

                                      [Q3.sup.FF]     Strong Fund-
                                                    Fund (FF) Links
                                                      [Q4.sup.FF]

Panel A. Small Funds

Strong Fund-Company (FC) Links
[Q4.sup.FC]                               1.83           -0.23
[Q3.sup.FC]                               0.75            0.14
[Q2.sup.FC]                              -0.30           -0.18
[Q1.sup.FC]                               1.28            1.93
Weak Fund-Fund (FF) Links
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])       0.55           -2.16 **
[t-statistic]                            [0.68]         [-2.31]

Panel B. Large Funds

Strong Fund-Company (FC) Links
[Q4.sup.FC]                              -1.02           -1.13
[Q3.sup.FC]                              -0.50           -0.82
[Q2.sup.FC]                               0.32           -0.07
[Q1.sup.FC]                               1.24            1.28
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])      -2.26 ***       -2.41 ***
[t-statistic]                           [-2.72]         [-2.77]

                                      Marginal Impact:
                                      FF Links
                                      ([Q4.sup.FF]--[Q1.sup.FF])
                                      [t-statistic]

Panel A. Small Funds

Strong Fund-Company (FC) Links
[Q4.sup.FC]                           -1.99 **       [-2.21]
[Q3.sup.FC]                           -1.09          [-1.38]
[Q2.sup.FC]                           -1.12          [-1.42]
[Q1.sup.FC]                            1.59 *         [1.99]
Weak Fund-Fund (FF) Links
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])
[t-statistic]

Panel B. Large Funds

Strong Fund-Company (FC) Links
[Q4.sup.FC]                           -2.47 ***      [-2.93]
[Q3.sup.FC]                           -1.53 *        [-1.72]
[Q2.sup.FC]                            0.33           [0.62]
[Q1.sup.FC]                            1.85 **        [2.15]
Weak Fund-Company (FC) Links
Marginal Impact:
FC Links ([Q4.sup.FC]--[Q1.sup.FC])
[t-statistic]

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table VII. Multivariate Regressions

Every quarter, the likelihood of a fund-company (FC) information
channel existing between fund manager j and each company issuing stock
n that is held in j's portfolio is calculated as
[L.sub.FC][(Links).sub.j,n] = 1-(1 + [Distance.sub.j,n]). Here,
[Distance.sub.j,n] is the natural logarithm of the geographical
distance (in km) between the city where fund j is located and the city
in which company n is headquartered. Each fund's holdings are ranked
into quartiles [Q1.sub.FC] through [Q4.sub.FC] based on [L.sup.FC]
[(Link).sub.j,n], with [Q4.sup.FC] as the Strong Fund-Company Links
portfolio and [Q1.sup.FC] as the Weak Fund-Company Links portfolio.
The likelihood of an information channel existing between fund manager
j and fund manager i is [L.sub.FF][(Links).sub.j,i] = 1-(1 +
[Distance.sub.j,i]), where [Distance.sub.j,i] is the natural logarithm
of the geographical distance between the city where fund j is located
and the city in which fund i is located. The strength of fund-fund
(FF) informal information links a fund manager j has with other fund
managers is computed as

[(Links).sup.FF.sub.j] = [N.summation over (i=1)][L.sup.FF]
[(Links).sub.j,t] = [N.summation over (i=1)] [1/(1 +
[Distance.sub.j,t])].

Here, i = 1, ..., N are all funds in the sample that do not belong to
the same fund family as fund j. Every quarter, the funds are ranked
into quartiles [Q1.sup.FF] through [Q4.sup.FF] based on
[(Links).sup.FF.sub.j], with [Q4.sup.FF] as the Strong Fund-Fund Links
portfolio and [Q1.sup.FF] as the Weak Fund-Fund Links portfolio. For
each fund, returns are computed in each quarter for an FC links
portfolio as

[R.sup.raw] = [1/3] [S.summation over (i=1)] [3.summation over (t=1)]
[w.sub.i,t] [R.sub.i,t]

where [w.sub.i,t] is the portfolio weight of stock i in month t, and S
is the number of stocks in the portfolio. Characteristic-adjusted FC
links portfolio returns are computed in each quarter as

[R.sup.adj] = [1/3] [S.summation over (i=1)] [3.summation over (t=1)]
[w.sub.i,t] ([R.sub.i,t] - [R.sup.bench.sub.i]),

where [R.sup.bench.sub.i] is the value-weighted monthly return of
stock i's DGTW quintile benchmark portfolio (following Daniel et al.,
1997). The dependent variable in the regressions is the annualized
value-weighted [R.sup.adj] for an FC quartile in a fund's portfolio in
a given quarter. Strong (Weak) FF Dummy assumes a value of one if the
fund is in the third or fourth (first and second) quartiles of funds
formed by the sorted and ranked FF measure, and zero otherwise. Strong
(Weak) FC Dummy assumes a value of one if the portfolio in the
dependent variable is the third or fourth (first and second) quartiles
of holdings formed by the sorted and ranked FC measure. Strong FF x
Strong FC is the interaction term for Strong FF Dummy and Strong FC
Dummy. Weak FF x Strong FC is the interaction term for Weak FF Dummy
and Strong FC Dummy. Big City Dummy assumes a value of one if the fund
is located in one of the top 20 cities by general population, and zero
otherwise. Local Bias Dummy assumes a value of one if the fund is in
the highest quartile based on Coval and Moskowitz's (2001) local bias
measure, and zero otherwise. Log (TNA), Log (Family Size), Log
(Tenure), and Log (Fund Age) are the natural logarithms of fund total
net assets in the month prior to the quarter, the number of funds in
the family, the length of manager tenure, and the fund age since the
first offer, respectively. Expenses and Turnover are the fund's
expense and turnover ratios, respectively. Fund, Time, Objective, and
State Fixed Effect regressions include fund dummies, quarter dummies,
1CDI objective dummies, and state of location dummies, respectively.
The coefficients and p-values based on robust standard errors are
reported.

                                              (1)

                                        Coeff.     p-Value

Strong FF Dummy                        0.013 **     0.02
Strong FF x Strong FC                 -0.021 ***    0.00
Weak FF x Strong FC                    0.019 ***    0.00
Big City Dummy
Big City x Strong FF x Strong FC
Big City x Weak FF x Strong FC
Local Bias Dummy
Local Bias x Strong FF x Strong FC
Local Bias x Weak FF x Strong FC
Log (TNA)
Log (Family Size)
Log (Tenure)
Log (Fund Age)
Expenses
Turnover
Fund fixed effects                       YES
Time fixed effects                       YES
Objective fixed effects                   NO
State fixed effects                       NO
N                                      169,896
Adjusted [R.sup.2]                      0.078

                                              (2)

                                        Coeff.     p-Value

Strong FF Dummy                        0.015 ***    0.01
Strong FF x Strong FC                 -0.018 **     0.03
Weak FF x Strong FC                    0.020 ***    0.00
Big City Dummy
Big City x Strong FF x Strong FC
Big City x Weak FF x Strong FC
Local Bias Dummy
Local Bias x Strong FF x Strong FC
Local Bias x Weak FF x Strong FC
Log (TNA)                             -0.008 **     0.04
Log (Family Size)                      0.003        0.34
Log (Tenure)                          -0.010        0.49
Log (Fund Age)                        -0.001        0.82
Expenses                              -0.109        0.28
Turnover                               0.023        0.50
Fund fixed effects                        NO
Time fixed effects                       YES
Objective fixed effects                  YES
State fixed effects                      YES
N                                      169,663
Adjusted [R.sup.2]                      0.046

                                              (3)

                                        Coeff.     p-Value

Strong FF Dummy                        0.018 ***    0.00
Strong FF x Strong FC                 -0.016 ***    0.01
Weak FF x Strong FC                    0.021 ***    0.00
Big City Dummy                         0.003        0.54
Big City x Strong FF x Strong FC       0.009        0.48
Big City x Weak FF x Strong FC         0.004        0.66
Local Bias Dummy
Local Bias x Strong FF x Strong FC
Local Bias x Weak FF x Strong FC
Log (TNA)                             -0.008 **     0.09
Log (Family Size)                      0.002        0.23
Log (Tenure)                          -0.009        0.60
Log (Fund Age)                         0.000        0.93
Expenses                              -0.112        0.22
Turnover                               0.019        0.71
Fund fixed effects                        NO
Time fixed effects                       YES
Objective fixed effects                  YES
State fixed effects                      YES
N                                      169,663
Adjusted [R.sup.2]                      0.049

                                              (4)

                                        Coeff.     p-Value

Strong FF Dummy                        0.018 ***    0.01
Strong FF x Strong FC                 -0.027 ***    0.00
Weak FF x Strong FC                    0.013 ***    0.03
Big City Dummy
Big City x Strong FF x Strong FC
Big City x Weak FF x Strong FC
Local Bias Dummy                       0.003        0.53
Local Bias x Strong FF x Strong FC    -0.002        0.79
Local Bias x Weak FF x Strong FC       0.004        0.61
Log (TNA)
Log (Family Size)
Log (Tenure)
Log (Fund Age)
Expenses
Turnover
Fund fixed effects                       YES
Time fixed effects                       YES
Objective fixed effects                   NO
State fixed effects                       NO
N                                      169,663
Adjusted [R.sup.2]                      0.083

                                              (5)

                                        Coeff.     p-Value

Strong FF Dummy                        0.021 ***    0.00
Strong FF x Strong FC                 -0.025 ***    0.01
Weak FF x Strong FC                    0.014 ***    0.00
Big City Dummy                         0.001        0.82
Big City x Strong FF x Strong FC      -0.012        0.21
Big City x Weak FF x Strong FC         0.013        0.27
Local Bias Dummy                       0.001        0.91
Local Bias x Strong FF x Strong FC    -0.002        0.80
Local Bias x Weak FF x Strong FC       0.004        0.57
Log (TNA)                             -0.008 *      0.08
Log (Family Size)                      0.002        0.38
Log (Tenure)                          -0.009        0.63
Log (Fund Age)                        -0.001        0.88
Expenses                              -0.111        0.27
Turnover                               0.020        0.62
Fund fixed effects                        NO
Time fixed effects                       YES
Objective fixed effects                  YES
State fixed effects                      YES
N                                      169,663
Adjusted [R.sup.2]                      0.062

*** Significant at the 0.01 level.
** Significant at the 0.05 level.
* Significant at the 0.10 level.

Table VIII. Copycat Portfolio Strategies

Every quarter, the likelihood of a fund-company (FC) information
channel existing between fund manager j and each company issuing stock
n that is held in j's portfolio is calculated as [L.sup.FC]
[(Link).sub.j,n] = 1-(1 + [Distance.sub.j,n]). Here,
[Distance.sub.j,n] is the natural logarithm of the geographical
distance (in km) between the city where fund j is located and the city
in which company n is headquartered. Each fund's holdings are ranked
into quartiles [Q1.sup.FC] through [Q4.sup.FC] based on [L.sup.FC]
[(Link).sub.j,n], with [Q4.sub.FC] as the Strong Fund-Company Links
portfolio and [Q1.sub.FC] as the Weak Fund-Company Links portfolio.
The likelihood of an information channel existing between fund manager
j and fund manager i is [L.sup.FF] [(Link).sub.j,i] = 1-(1 +
[Distance.sub.j,i]),  where [Distance.sub.j,i] is the natural
logarithm of the geographical distance between the city where fund j
is located and the city in which fund i is located. The strength of
fund-fund (FF) informal information links a fund

[(Links).sup.FF.sub.j] = [N.summation over (i=1)][L.sup.FF]
[(Links).sub.j,t] = [N.summation over (i=1)] [1/(1 +
[Distance.sub.j,t])].

Here, i = 1, ..., N are all funds in the sample that do not belong to
the same fund family as fund j. Every quarter, the funds are ranked
into quartiles [Q1.sup.FF] through [Q4.sup.FF] based on
[(Links).sup.FF.sub.j], with [Q4.sup.FF] as the Strong Fund-Fund Links
portfolio and [Q1.sup.FF] as the Weak Fund-Fund Links portfolio. A
hypothetical equally weighted portfolio, called the Best Information
Portfolio (BIP), comprised of stocks that are classified as: 1) Strong
Fund-Company Links stocks for Weak Fund-Fund Links funds and 2) Weak
Fund-Company Links stocks for Strong Fund-Fund Links funds is created.
BIP is formed at the beginning of month t based on the publicly
released information on mutual fund holdings for a quarter comprised
of months t -3, t -2, and t -1. BIP (Qtr t) is the BIP constructed in
quarter t in the sample period, where t = 1, ..., 10. Panel A (Panel
B) presents the average monthly raw returns (characteristic-adjusted
returns) of an equally weighted portfolio of BIP stocks for 3-, 6-, 9-
, and 12-month holding periods. Characteristic-adjusted returns are
computed following Daniel et al. (1997). p-Values based on robust
standard errors are reported in parentheses.

Hypothetical          3-Mth        6-Mth        9-Mth        12-Mth
Portfolio           Return (%)   Return (%)   Return (%)   Return (%)

                       Panel A. Average Monthly Raw Returns

BIP (Overall)        1.63 ***     1.81 ***     1.35 **      0.85
                    (0.01)       (0.00)       (0.03)       (0.17)
BIP (Small Fund      1.49 **      1.52 ***     0.88 *       0.79
  Families)         (0.05)       (0.01)       (0.10)       (0.25)
BIP (Large Fund      1.82 ***     2.13 ***     1.59 **      0.97
  Families)         (0.00)       (0.00)       (0.02)       (0.19)
BIP (Small Funds)    1.32 **      1.45 **      0.69         0.62
                    (0.04)       (0.03)       (0.32)       (0.43)
BIP (Large Funds)    2.02 ***     2.28 ***     1 73         1.03 *
                    (0.00)       (0.00)       (0.00)       (0.09)

            Panel B. Average Monthly Characteristic-Adjusted Returns

BIP (Overall)        0.82 ***     1.02 ***     0.53        -0.05
                    (0.00)       (0.00)       (0.13)       (0.95)
BIP (Small Fund      0.72 **      Q 'J'J       0.24        -0.48
  Families)         (0.03)       (0.00)       (0.49)       (0.35)
BIP (Large Fund      1.13 ***     1.25 ***     0.69 *       0.28
  Families)         (0.00)       (0.00)       (0.10)       (0.62)
BIP (Small Funds)    0.80 **      0.91 **      0.42        -0.59
                    (0.02)       (0.02)       (0.39)       (0.32)
BIP (Large Funds)    0.95 ***     1.27 ***     0.70 *       0.45
                    (0.00)       (0.00)       (0.09)       (0.40)

*** Significant at the 0.01 level.
** Significant at the 0.05 level.
* Significant at the 0.10 level.

Figure 1. Marginal Impacts of Informal Information Channels

This figure reports a metric for forming portfolios of stocks held by
mutual funds according to information links associated with the
holding. The likelihood of FC information links existing between fund
manager j and the company issuing stock n that is held in j's
portfolio is [L.sup.FC] [(Link).sub.j,n] = 1/(1 + [Distance.sub.j,n])
where, [Distance.sub.j,n] is the natural logarithm of the geographical
distance between the city where fund j is located and the city in
which company n is headquartered. For each fund j, each quarter the
holdings are sorted by [L.sup.FC] [(Link).sub.j,n] and ranked into
quartiles with [Q.sup.FC,strong.sub.j] ([Q.sup.FC,weak.sub.j]) being
the highest (lowest) [i.e. [Q4.sup.FC] ([Q1.sup.FC])] quartile and
form the portfolios Strong (Weak) Fund-Company Links in each quarter.
The likelihood of an information channel existing between fund manager
j and fund manager i is [L.sup.FF][(Links).sub.j,i] = 1/(1 +
[Distance.sub.j,i]) where [Distance.sub.j,i] is the natural logarithm
of the geographical distance between the city where Hind j is located
and the city in which fund / is located. The strength of the informal
information channels fund manager j has with other fund managers is
computed as:

[(Links).sup.FF.sub.j] = [N.summation over(i=1)][L.sup.FF]
[(Links).sub.j,t] = [N.summation over(i=1)] [1/(1 +
[Distance.sub.j,t])].

Here, i = 1, ..., N are all of the funds in the sample that do not
belong to the same fund family as fund j. Each quarter, the funds are
sorted by [(Links).sup.FF.sub.j] and ranked into quartiles with
[Q.sup.FC,strong.sub.j] ([Q.sup.FC,weak.sub.j]) being the highest
(lowest) [i.e. [Q4.sup.FF] ([Q1.sup.FF])] quartile and form the
categories Strong (Weak) Fund-Fund Links. [R.sup.adj] is the
value-weighted characteristic-adjusted return generated from a
portfolio of holdings.

FF Links                         Weak Fund-Fund (FF) Links
FC Links                             ([Q.sup.FF,weak])

Strong Fund-Company (FC)                Portfolio P1
Links                                ([Q.sup.FF,weak],
([Q.sup.FC,strong.sub.j])         [Q.sup.FC,strong.sub.j])

                                     Information Links:
                                        Fund-Company

Weak Fund-Company (FC)                  Portfolio P2
Links                                ([Q.sup.FF,weak],
([Q.sup.FC,weak.sub.j])            [Q.sup.FC,weak.sub.j])

                                     Information Links:
                                            None

Marginal Impact:                        Difference:
FC Links                    [R.sup.adj] (P1) - [R.sup.adj] (P2)
([Q.sup.FC,strong.sub.j])
([Q.sup.FC,weak.sub.j])

FF Links                        Strong Fund-Fund (FF) Links
FC Links                            ([Q.sup.FF,strong])

Strong Fund-Company (FC)                Portfolio P3
Links                               ([Q.sup.FF,strong],
([Q.sup.FC,strong.sub.j])         [Q.sup.FC,strong.sub.j])

                                     Information Links:
                                         Fund-Fund
                                        Fund-Company

Weak Fund-Company (FC)                  Portfolio P4
Links                               ([Q.sup.FF,strong],
([Q.sup.FC,weak.sub.j])            [Q.sup.FC,weak.sub.j])

                                     Information Links:
                                         Fund-Fund

Marginal Impact:                        Difference:
FC Links                    [R.sup.adj] (P3) - [R.sup.adj] (P4)
([Q.sup.FC,strong.sub.j])
([Q.sup.FC,weak.sub.j])

FF Links                              Marginal Impact:
FC Links                                  FF Links
                            ([Q.sup.FF,strong] - [Q.sup.FF,weak])

Strong Fund-Company (FC)                 Difference:
Links                        [R.sup.adj] (P3) - [R.sup.adj] (P1)
([Q.sup.FC,strong.sub.j])

Weak Fund-Company (FC)                   Difference:
Links                        [R.sup.adj] (P4) - [R.sup.adj] (P2)
([Q.sup.FC,weak.sub.j])

Marginal Impact:
FC Links
([Q.sup.FC,strong.sub.j])
([Q.sup.FC,weak.sub.j])
COPYRIGHT 2014 Financial Management Association
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2014 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Fu, Richard; Gupta-Mukherjee, Swasti
Publication:Financial Management
Geographic Code:1USA
Date:Mar 22, 2014
Words:18561
Previous Article:Emerging market exposures and the predictability of hedge fund returns.
Next Article:Competition, efficiency, and stability in banking.
Topics:

Terms of use | Copyright © 2018 Farlex, Inc. | Feedback | For webmasters