The Geography of Information Acquisition

Using detailed data about company visits by Chinese mutual funds, we provide direct evidence of mutual fund information acquisition activities and the consequent informational advantages mutual funds establish in local firms. Mutual funds are more likely to visit local and nearby firms, though ease of travel between fund and firm locations alleviates geographic distance constraints. Company visits by mutual funds are related to both fund trading activities and fund trading performance. The relation between company visits and fund trading performance is particularly strong in newly initiated stock positions. Consistent with this result, we find evidence of intensive information acquisition by mutual funds in stocks not in their current portfolios. Our results suggest that mutual fund local preference in portfolio choice is at least partially driven by geographic constraints of information acquisition and the ensuing information asymmetry.

1 Investors exhibit a strong preference for locally headquartered firms in their portfolio choices. This phenomenon, often referred to as "local bias," is prevailing not only for individual investors (Ivkovic and Weisbenner 2005), but also for professional investors such as mutual funds (Coval and Moskowitz 1999). Despite the prevalent evidence of local bias, there is no consensus on the causes of such bias. The local preference could arise from investors' familiarity with local firms or from the informational advantages enjoyed by investors in these firms. Pool, Stoffman, and Yonker (2012) find evidence of "familiarity effects"-mutual funds overweigh stocks from fund managers' home states even though the managers do not possess a comparative information advantage in these stocks. Consistent with an information-based explanation, a strand of the literature finds that local portfolios tend to deliver better performance than non-local portfolios for mutual funds (Coval and Moskowitz 2001), hedge funds (Teo 2009), and other types of institutional investors (Baik, Kang, and Kim 2010).
The superior local portfolio performance documented in the literature suggests that investors exploit their informational advantage in local investments. However, important questions about why and how investors achieve and maintain such information advantages in local firms remain largely unanswered. For example, although both Coval and Moskowitz (2001) and Baik, Kang, and Kim (2010) conjecture that access to private information of geographically proximate firms and improved monitoring of these firms could be associated with the local information advantage and better investment performance, it is unclear how geographic location affects information access and information acquisition by investors. Because few of the information acquisition activities associated with investment decisions are readily observable, answering these questions remains an empirical challenge.
In this paper, we study the information acquisition activities of mutual funds in local and non-3 kilometers [KM] of the fund). The frequency of mutual fund visits doubles for companies that are located in the same city or close to the city of the mutual fund as compared to their non-local counterparts. Mutual funds are more likely to visit local firms both in and outside of their existing portfolios. In multivariate regression analyses that control for a large set of fixed effects and firm characteristics, we continue to find robust evidence that mutual funds conduct more visits to local than non-local companies.
It is possible that geographic proximity could be related to fund location choice and existing economic and culture ties across the regions. To address these concerns, we exploit the rapid change in travel mode in China during the sample period to identify the causal effect of geographic distance on information acquisition. We collect information on the establishment of direct highspeed (HS) train connections and use a difference-in-difference approach to examine whether the introduction of a HS train connection between the fund city and the firm city affects mutual fund visits. Because the treatment is determined by pairs of fund-firm cities, we can control for all timeseries variations at the city level and identify a clean treatment effect. We find that fund managers significantly increase their site visits to a city after the establishment of a HS train connection.
Further dynamic tests support the causal effects of HS train connection on fund site visits. The difference-in-difference tests help establish a causal relation between distance and information acquisition activities.
We next examine whether mutual fund site visits affect investment decisions. If site visits represent an important form of information gathering by mutual funds, these site visits should lead to changes in mutual fund portfolio positions in these stocks. To investigate such a link, we study how fund site visits affect subsequent fund trading activities in the visited stocks. The results show that site visits have a positive and highly significant impact on mutual fund trading activity. This Electronic copy available at: https://ssrn.com/abstract=3371978 positive relation exits for both buy and sell decisions for stocks the mutual funds currently own and is particularly important for initiating new positions in stocks the mutual funds did not own.
Company visit decisions and investment decisions are likely determined jointly. For example, a fund manager may visit a company for which she is already considering a change in portfolio position. Furthermore, a significant relation between company visits and position changes does not by itself suggest the company visits yield valuable investment information. To answer these questions, we examine post-visit trading performance in the 1-, 3-, and 6-month periods after each company visit. We find significant performance differences between post-visit buy and sell trades in stocks that mutual funds already own and between post-visit buy and non-traded stocks for stocks that mutual funds do not previously own. The results hold for all three evaluation periods and are stronger for initial than for incremental purchase decisions. Although the performance differences persist for at least 6 months after portfolio formation, the differences noticeably level off during the second half of the 6-month period. The performance results confirm that mutual fund company visits affect investment decisions because such visits provide valuable information to mutual funds.
Our paper contributes to the growing literature of geography of investment decisions. The investment performance evidence in Coval and Moskowitz (2001), Teo (2009), and Baik, Kang, and Kim (2010) suggests that fund managers enjoy an informational advantage in local holdings.
Schumacher (2017) shows that mutual fund managers exhibit home-industry bias in foreign investments because these managers are better informed about industries in the domestic market.
Compared with these studies that infer a local information advantage for fund managers through investment performance, we study the effects of geographic location on information acquisition activity directly and explore the implications for investment decisions and performance. Our Electronic copy available at: https://ssrn.com/abstract=3371978 evidence reveals that the local information advantage enjoyed by mutual fund managers is at least partly derived from their more intensive information-gathering efforts for local firms. Such information acquisition efforts have a direct impact on mutual fund trading decisions, portfolio choice, and investment performance. Because differences in the costs of information acquisition likely remain for local and non-local stocks, the information-induced "local bias" in investment decisions could stay a robust phenomenon even in well-developed financial markets.
Our examination of mutual fund information acquisition activity also yields three notable findings that shed further light on the possible causes and consequences of investor local preference. First, mutual funds are more likely to visit local companies regardless of whether these companies are in their portfolios. In a theoretical model of "home bias" based on endogenous information acquisition, Van Nieuwerburgh and Veldkamp (2009) argue that investors who enjoy some initial information advantage in a market may further develop this advantage through greater information acquisition effort. Our results provide direct support for this argument. To the extent that geographic proximity provides some initial information advantage or differential information endowment to local fund managers, these managers acquire more information in local stocks they do not own and continue to exert greater effort in acquiring information after they own the stocks.
Second, we find that mutual funds conduct extensive site visits to companies they do not own; in fact, they devote on average 80% of their visits to these firms. These information acquisition activities may not be immediately reflected in investment decisions and thus are overlooked in studies that examine information advantage through fund returns and portfolio holdings. For example, extensive information gathering efforts could help explain the superior investment performance of stocks that are newly acquired by mutual funds (Alexander, Cici, and Gibson 2007) and the poor performance of local stocks that are not owned by mutual funds (Coval and Electronic copy available at: https://ssrn.com/abstract=3371978 6 Moskowitz 2001). Third, the literature suggests that better local investment performance could also be driven by enhanced monitoring by local investors (Baik, Kang, and Kim 2010). Company visits may be part of the monitoring activity. However, the intensive visit activity in stocks not held by mutual funds is unlikely to be motivated by monitoring. This evidence, combined with the evidence on post-visit trading performance, suggests that information rather than monitoring reasons largely explain local investment decisions and investment performance.
Our paper is related to the literature on the role of direct communication in financial markets.
Soltes (2014) and Brown et al. (2015) find that analysts' private communication with management plays an important role in analyst research in the United States. Direct communication is found to be informative and can take the form of conference presentations by management (Bushee, Jung, and Miller 2011), broker-hosted investor conferences (Green et al. 2014a(Green et al. , 2014b, investor/analyst day hosted by companies (Kirk and Markov 2016), and private in-house meetings (Bowen et al. 2018). Using SZSE company visit data, Cheng et al. (2016) and Han, Kong, and Liu (2018) study company visits by Chinese financial analysts and find that these visits facilitate analysts' information acquisition and improve earnings forecast accuracy. Without considering the role of geographic location, Liu, Dai, and Kong (2017) find that site visits of Chinese mutual funds are informative and predict fund trading and firm earnings surprise. We show that site visits by mutual funds play an important role in their investment decisions and highlight how geographic location can affect this form of information acquisition.
Finally, our paper adds to the fast-growing literature on the effect of transportation on financial markets. Bradley, Gustafson, and Williams (2018)  show that the introduction of new airline routes affects venture capitalists' involvement with their portfolio companies (e.g., Mao, Tian, and Yu 2014;Bernstein, Giroud, and Townsend 2016), broadens the investor base of firms, lowers their cost of equity (e.g., Da et al. 2018), affects mutual fund holdings (Ellis, Madureira, and Underwood 2019), and leads to an increase in plant-level investment from the headquarters and to total factor productivity (Giroud 2013). These studies all suggest geographic distance affects communication and information gathering. Our paper provides direct evidence that distance affects a well-defined, important form of information-gathering activity and that transportation infrastructure development can improve information efficiency in financial markets.
The remainder of the paper is organized as follows. Section 1 provides the institutional background of the mutual fund industry in China and the regulation and disclosure of company visits. Section 2 introduces the data. Section 3 examines local preference in mutual fund portfolio choice and the relation between location and site visits. Section 4 studies how ease of travel affects site visits to establish a causal effect of location on information acquisition. Section 5 investigates how site visits influence mutual fund investment decisions and performance. Section 6 concludes.

Mutual fund industry in China
The modern era of the Chinese mutual fund industry started in 1997 after the passage of the The growth of actively managed open-end equity funds and balanced equity funds, or equity funds as a whole, largely parallels the overall growth of the investment fund industry but is also substantially influenced by the ups and downs of the Chinese stock market. For our sample, the number of equity funds increases from 3 in 2001 to 591 by the end of 2016. The total net assets of these funds varies over time. They start at 11.8 billion RMB in 2001, peak at 2.33 trillion RMB during the stock market boom in 2007, and plummet to 1.03 trillion RMB during the market crash in 2008. The ratio of total net assets of equity funds to total market capitalization ranges between 0.3% and 8.5% from 2001 to 2016, and its variation largely mirrors that of the total net assets. As we discuss later in Section 3.1, an important reason for the relatively low ownership by equity funds is the high state ownership of publicly listed state-owned companies. These shares are transferable and counted as part of the public float, but are practically not available to the public and are seldom traded in the open market.

Mutual fund site visits
Investors and financial intermediaries in China are allowed to visit the listed companies. Article 41 of the "Guidelines of Investor Relations Management,'' issued by SZSE in 2003, states that "the listed companies should accommodate the requests from investors, financial analysts and fund managers to visit the companies or the project site.'' In August 2006, SZSE issued the "Fair Information Disclosure Guidelines for Listed Companies in the Shenzhen Stock Exchange'' ("Disclosure Guidelines''), which establishes detailed regulations on site visits by investors and financial analysts. In particular, the listed companies should timely record information about the site visits such as the time and location of the visits, the names of the visitors, and the content of the discussion. Furthermore, the listed companies should disclose these visit records in their periodic financial reports. The companies cannot provide and the visitors cannot request material non-public information during the visits.
To standardize the implementation of the Disclosure Guidelines, SZSE issued the "No. 1 Memo on Information Disclosure'' in February 2007. This memo specifies the disclosure format of the site visits in the annual financial reports. Appendix Table A.1 provides an example of a typical visit record from an annual financial report. After the establishment of these guidelines in 2007, visit information became available.
To schedule site visits, investors typically initiate the requests and firms usually accommodate these requests. The visit dates are negotiated by both parties. The company can arrange visits by different groups of investors on the same day, and visitors can coordinate their trips. During the visits, investors meet with mid-and high-level corporate executives, but do not usually meet with top executives such as chief executive officers (CEOs). In addition to meetings, investors can take a field tour of the production facilities and observe firms' operations.
As shown in Appendix Table A.1, fund managers ask questions about the firm's core business model, development strategy, and competitiveness in the industry. Face-to-face communications and interactions with firm management during site visits can help fund managers obtain information that is different from what is reported in financial statements. Equally important, observations through direct communications, or "seeing is believing," can help fund managers update their prior beliefs about the firm and improve their understanding of the operations and performance of the firm. Cheng et al. (2016), for example, conclude that site visits help analysts improve their forecast accuracy. Fund managers, particularly those who are skilled at incorporating "soft information" in their investment decision, can take advantage of site visits and acquire valuable investment-relevant information.

Data and sample characteristics
We obtain the data set of investors' site visits from SZSE for January 2007 to June 2017. The visitors include not only different types of financial institutions such as brokerage firms, mutual funds, hedge funds, insurance companies, and other types of asset management companies, but also individual investors, news media, and others. 2 About 35% of the visits are by brokerage firms, followed by mutual funds at 25% of the visits. Our analyses focus on visits by mutual funds at the fund-family level because company visits are typically carried out by both portfolio management and investment research teams of the mutual fund companies.
The mutual fund data we employ come from several sources. Data on mutual fund stockholdings and fund location are from the China Stock Market and Accounting Research (CSMAR) database. Chinese mutual funds are required to disclose their equity holding information, including number of shares and market value of holdings, at the end of each quarter with different reporting requirements. More specifically, mutual funds must report only the top 10 stockholdings in their portfolios in the first and third calendar quarters, but must disclose their full list of stockholdings in the second and fourth quarters. Therefore, we focus on the two semiannual reports with complete stockholding information. The sample period for fund equity holdings is from December 2006 to December 2016.
Information on mutual fund net asset value is from the RESSET Database, one of the leading financial data vendors in China. It provides the daily fund unit price and number of units for each 2 Companies may accommodate the visit requests by individual investors if they are large shareholders. fund, as well as the classification of mutual fund type. For our study, we retain only actively managed equity funds and balanced equity funds with at least 60% asset allocation in stocks, and exclude other types of funds such as index funds, balanced funds with low equity allocation, and bond funds. We merge the mutual fund data from CSMAR with the data from RESSET using fund code numbers, which are uniquely assigned by the Chinese Securities and Regulatory Commission.
We aggregate fund holdings information at the fund-family level every half year. The site visit activities of each fund family are also aggregated semiannually. We further obtain accounting information, financial analyst coverage, and stock return information of public companies in the sample from CSMAR.  In terms of mutual fund investment, the holding number (portfolio weight) of SZSE stocks 12 increases from 37.7% (32.8%) to 56.3% (60.4%) over the sample period, reflecting the growing importance of SZSE stocks in Chinese mutual fund portfolios. Because of the large state ownership of publicly listed SOEs, mutual fund holdings of SZSE-listed stocks represent a greater share of mutual fund investment portfolio than the aggregate market weight of SESE-listed stocks. Overall, SZSE-listed stocks represent about half of the mutual fund portfolio in both the number of stocks and the value of holdings in the portfolio (Columns 8 and 9 of Table 1).
We observe a steady increase in site visits by mutual funds over the sample period. In the first half of 2007, each fund family makes 11.9 site visits on average to companies listed on SZSE. 3 By the second half of 2016, each fund family conducts more than 79.5 visits on average. The increase in number of visits can be partly explained by the increase in SZSE-listed stocks and the increase in these stocks in mutual fund portfolios. However, the ratio of total visits per fund family to number of SZSE stocks varies from 1.9% to 5.6% and increases significantly over the same period. The trend is indicative of the growing importance of site visits in mutual funds' investment decisions.
Because our analysis is at fund-family level, it would be helpful to examine the statistics for individual funds within a family. The results are reported in the last two columns of Table 1 The number of funds per family increases from 3.1 in 2007 to 7.8 in 2016. Finally, the product of 3 If a fund family visits the same company more than once on different dates during the half year, we count them as separate visits. As we show in the analysis on unique and repeat visits in the next section, repeat visits of the same firm by the same fund within 1 year are uncommon.
Electronic copy available at: https://ssrn.com/abstract=3371978 the number of stocks per fund and the number of funds per family is much larger than the total number of holdings stocks in Column 2, suggesting that there is an overlap in holdings across funds within the same family.

Mutual fund local preference and information advantage
In this subsection, we examine whether mutual funds in China exhibit a preference for local stocks and whether such local preference is related to the information advantage these mutual fund possess in local firms. We limit the sample to SZSE-listed stocks because site visit information is available only for these companies. For each fund-company pair, we compute the distance between the city of the mutual fund family and the city of the corporate headquarters of the company. 4 We use three definitions for local firms. We first define a company as a local firm for a fund if the company and the fund are in the same city; that is, the distance between their cities is zero. We then change the definition for a local firm by extending the distance to less than 100 KM, and then less than 200 KM. In untabulated results, all results are qualitatively similar if a threshold of 300 KM is used.
Using these definitions, we follow Coval and Moskowitz (2001) and aggregate fund holdings into local and non-local holdings for each fund family. For comparison, we construct two benchmarks for local holdings and non-local holdings for each fund family. The first benchmark follows most mutual fund studies that use the ratio of total market value of all local stocks to the 4 We use the headquarter locations of the mutual fund families. Although mutual funds in China may have subsidiaries in other cities, the main function of those subsidiaries is fund sales. Investment managers of funds typically work at fund headquarters. total market value of all stocks. The second benchmark is based on the holdings of all mutual funds.
To construct this second benchmark, we calculate the total fund assets invested in the stocks located in a local area across all fund families, and divide by the holding value of all stocks across all fund families. This approach takes into consideration mutual fund preference for various stock characteristics and can further address one shortcoming for using the market weight of a stock as the benchmark weight in the Chinese stock market. Publicly listed SOEs, for which the state is the controlling shareholder and holds a significant proportion of shares, are an important component of the Chinese stock market. Although state-owned shares can be traded in the secondary market, for political and control considerations, these shares are rarely sold to the public and for practical purposes are not available to mutual funds. As a result, the market weight of these SOEs is much higher than the actual weight available for mutual funds and other investors. The second approach therefore provides a better approximation of the proportion of firms in the invested assets for all funds.
We present the results on mutual fund local preference in Table 2. Column 1 reports the average portfolio weight of local stocks across fund families over our sample period for the three definitions of local firms. Columns 2 and 3 present the benchmark weights based on market value and aggregate fund holdings, respectively. For all definitions of local firms, we find robust and consistent patterns of local preference. The mutual funds in China hold disproportionally more local stocks than non-local stocks. On average, mutual funds allocate 9.3% of their assets to firms located in the same city, 2.2 (1.2) percentage points greater than that of 7.1% (8.1%) for the benchmarks based on the total market value (fund holdings universe). The differences between fund local holdings and the benchmarks are highly significant. As we relax the definition of locality and expand the radius, the weight of local stocks in mutual fund portfolio increases, as does the weight of local stocks in the benchmark portfolios. However, we continue to find that mutual funds allocate more investments to local stocks than the benchmarks.
The pattern and magnitude of local bias in Chinese mutual funds documented here are similar to those based on U.S. data. Using a sample of mutual funds from 1975 to 1994, Coval and Moskowitz (2001) find that an average fund manager invests about 7% of assets locally, whereas local stocks constitute only 6.2% of market capitalization. Baik, Kang, and Kim (2010) study a more recent sample from 1995 to 2007 and find that the actual fraction of local holdings by institutional investors is 8.2%, whereas the fraction of market weight in the same area is 6.6%. Our results suggest that local bias in mutual fund investments in China is robust and largely comparable in magnitude with the well-documented evidence in the United States.
Next, we examine whether mutual funds exhibit an informational advantage in their local holdings. As suggested by Coval and Moskowitz (2001), the existence of such an informational advantage can be reflected in superior returns generated by the local holdings in a fund portfolio.
Our analysis proceeds as follows. At the end of each half-year period t, we separate a fund's holdings into a local portfolio and a non-local portfolio based the three definitions of local firms.
For each portfolio, we calculate the average monthly raw return in period t+1 (the next 6 months) using the stockholding weight at the end of period t.
In addition to raw returns, we calculate abnormal holding-period returns based on benchmark portfolios based on size and book-to-market. To construct the benchmark, we first sort all firms into five size portfolios at the beginning of July in year t using the market value of firms at the end of June in the same year. Within each size portfolio, we sequentially sort all firms into five bookto-market portfolios at the beginning of July in year t based on the book value of equity measured at the end of fiscal year t−1 and the market value of equity measured at the end of December in year t−1. We rebalance the benchmark portfolios every year and calculate the equally weighted benchmark portfolio returns. An equally weighted benchmark is preferred because of the strong size effect in the Chinese equity market that is due partly to the poor performance of large SOEs (see, e.g., Lee, Qu, and Shen 2017;Hu et al. 2018). Results, however, are robust regardless of whether we use value-weighted benchmark portfolio returns or the alternative size and earnings-toprice ratio benchmark in Liu, Stambaugh, and Yuan (2019). 5 We compare the performance of local and non-local holdings of the same fund family and report the results in Table 3. Columns 1 and 2 show the time-series average monthly raw returns (in percentage) of local and non-local portfolios, and the difference between the two is reported in Column 3. For all three definitions of local firms, local holdings deliver a higher return than nonlocal holdings. The annualized difference in raw returns ranges from 4.9% to 5.7% and is statistically significant. Columns 4 and 5 show benchmark-adjusted abnormal returns for the local and non-local portfolios, respectively. After controlling for firm size and book-to-market ratio, local portfolios continue to deliver significantly higher returns than non-local portfolios. The annualized abnormal returns of local portfolios are between 4.2% to 5.2% and exceed those of non-local portfolios by 5.5 to 6.3 percentage points. These results are striking given that we update the holdings information at a low frequency (every half year) and examine portfolio performance based on fund holdings rather than trading activities. The superior performance of local portfolios in our sample of mutual funds suggests that these funds do enjoy substantial informational 5 We do not further partition the portfolios into momentum portfolios for two reasons. First, there is no evidence of a momentum effect in the Chinese stock markets (see, e.g., Lee, Qu, and Shen 2017). Second, further partition leads to less reliable benchmark portfolio returns in the early sample period because of the small number of stocks. Nonetheless, our results are robust if we follow the method in Daniel et al. (1997) that includes the momentum factor in benchmark construction. advantage in their investments of the local stocks.
Our results are consistent with the evidence documented in Coval and Moskowitz (2001) and Baik, Kang, and Kim (2010) based on U.S. data, which finds that the local holdings of fund managers (or institutional investors) deliver better performance than their non-local holdings. After documenting both the local preference and local informational advantage exhibited by these mutual funds, we next explore one channel through which mutual fund managers may obtain and maintain their informational advantage.

Company location and mutual fund visits
We hypothesize that local information advantage arises from the more intensive mutual fund information acquisition activity in local firms. Location could affect investor information acquisition activities because the cost of acquiring local information is lower due to geographic proximity to local firms. More important, as Van Nieuwerburgh and Veldkamp (2009) argue, investors who enjoy some initial information advantage due to geographic proximity can further maintain and expand their advantage through greater information acquisition effort in these local firms. The unique data on mutual fund site visits allow us to directly examine and compare one such type of information acquisition activity by mutual funds in local and non-local firms.
To study the relation between fund-firm distance and company visits by mutual funds, we first conduct two univariate tests and present the results in Table 4. We independently sort stocks into 2×2 groups based on whether they are local stocks to the funds and whether they are held by the funds. In panel A, we report the average ratio of the number of visits to each group of stocks to the total number of SZSE stocks in that group, which takes into consideration the number of potential firms in each group for each fund family to visit. Table 4 show the average percentages of local and non-local visits by funds for the full sample of stocks. We find that mutual funds do indeed visit a greater fraction of local versus non-local companies. On average, mutual funds visit about 7.5% of local companies each half year, more than double their visits to non-local companies (3.7%). The differences are highly significant for all three definitions of local firms. As expected, as the definition of being local includes more distant companies, the ratio of local visits decreases monotonically.

Columns 1-3 in panel A of
We next split the full sample into two subsamples based on whether stocks are held by the funds at the end of period t−1. Columns 4-6 in panel A of Table 4 present the results for companies that are held by funds, and Columns 7-9 presents results for companies that are not held by funds a time t−1. Consistent with the full sample results, in both subsamples, mutual funds visit a greater fraction of local versus non-local companies. The differences in fund visits are highly significant and the results hold for all definitions of local companies. For both local and non-local companies, mutual funds are more likely to visit companies in their portfolios than those that are not in their portfolios. As shown in Columns 4 and 7, the frequency of mutual fund visits to local companies in their portfolios is more than double the frequency of visits to local companies not in their portfolios (16.0% vs. 6.5% for companies located in the same city). Columns 5 and 8 show that mutual funds are also much more likely to visit non-local companies in their portfolios than nonlocal companies in their portfolios (10.1% vs. 3.0% for companies located in the same city).
Interestingly, mutual funds are more likely to visit non-local companies in their portfolios (Column 5) than local companies not in their portfolios (Column 7).
These results are consistent with the endogenous information acquisition model in Van Nieuwerburgh and Veldkamp (2009). In their model, because of budget or resource constraints, investors who enjoy some initial information advantage in stocks may further develop their information advantage through greater information acquisition effort in these stocks. In our setting, managers may have some initial information advantage in local stocks, and they are indeed more likely to visit those stocks to acquire further information. Similarly, managers may have established some information advantage in the stocks they hold, and they are more likely to visit these stocks to bolster their information advantage. Our results indicate that both information endowment and continuous information acquisition can be greatly influenced by geographic location.
The results in panel A of Table 4 provide evidence on the likelihood of a company receiving visits from a mutual fund given its geographic proximity to the fund. However, because there are typically far more non-local stocks than local stocks, the stock-level results do not provide sufficient information about how a mutual fund allocates its site visits between local and non-local firms. We conduct a comparison of such efforts in panel B by reporting the average percentage of visits to each group of stocks by each fund family within its total number of visits. The ratios in panel B describe the allocation of information acquisition efforts to each category of stocks by mutual funds. Columns 1-3 show that on average, local (non-local) companies account for 13.3% (86.7%) to 26.6% (73.4%) of all mutual funds visits, depending on which definition local firm is used. Mutual funds visit far more non-local than local firms. This split in visit efforts is not totally surprising because there are more non-local than local firms for mutual funds. The result does reveal that even with the more intensive information acquisition efforts in local firms (panel A), mutual funds conduct a far greater portion of their visits to non-local firms.
Columns 4-6 in panel B of Table 4 report results for companies held by mutual funds, and Columns 7-9 report results for companies not held by mutual funds. In aggregate, mutual funds allocate slightly more than 20% of their visits to companies they hold (sum of Columns 4 and 5) and close to 80% of their visits to companies they do not hold (sum of Columns 7 and 8).
Depending on the definition of local company, between 57.1% (20.5%) and 67.4% (10.2%) of total visits are made to non-local (local) companies that are not held by mutual funds at the end of the previous periods, whereas these numbers are between 16.4% (6.1%) and 19.3% (3.1%) for nonlocal (local) companies that are held by mutual funds. Based on the number of visits, the splits between local and non-local stocks in both groups heavily favor non-local stocks. Most important, though, these results demonstrate that mutual funds devote more effort to visiting stocks not in their portfolios than those in their portfolios.
The results in the panels A and B of Table 4 provide an interesting contrast. At the stock level, mutual funds are more likely to visit stocks they currently hold and stocks that are located nearby.
However, based on their overall visits, mutual funds devote a larger percentage of their efforts to non-local stocks and to stocks they do not own. Because information acquisition activities are typically not directly observable, studies on information acquisition or information advantage based on mutual fund portfolio holdings are unlikely to capture the information acquisition activities in stocks that mutual funds do not hold. Consequently, analyses based on portfolio holdings likely substantially underestimate mutual fund information acquisition activities. The results also suggest that mutual funds engage in considerable information gathering and analysis before initiating a new position in a stock. Such information acquisition activity could help explain the highly positive performance in newly initiated stocks (Alexander, Cici, and Gibson 2007).
We now study the relation between fund-firm distance and site visits of mutual funds in a multivariate framework. To control for the fund and firm characteristics and other factors that may affect a fund's site visit decision, we estimate the following model: where i denotes a fund family, j denotes a company, and t denotes the period. In each period t, a company is included in the regression if it is visited by any fund.
The dependent variable is the logarithm of one plus the number of visits by fund family i for company j in period t. The variable of interest is Local, which is an indicator variable that equals 1 if the distance between the city of a mutual fund and the city of a company is within a specified range, and 0 otherwise. As before, we consider three definitions for local firms. The variable Holding is the portfolio weight of the company's stock in the fund overall holdings at the end of period t−1. FirmChar represents an array of firm characteristics variables that include firm size (market value of equity), return on assets, a dummy for SOE, firm age, analyst coverage, assets growth, abnormal returns of the company's stock, standard deviation of stock returns, and share turnover. All these variables are measured at the end of period t−1. 6 The company characteristics may affect mutual fund investment decisions as well as visit decisions. We use the log value for the dependent variable and an ordinary least squares (OLS) regression because a large set of fixed effects can be included in the model. The time fixed effects capture all macroeconomic factors that could influence fund visits, including the time trends in the number of funds and in the listed companies. The fund×time fixed effects control for all cross-sectional and time-series variations in mutual fund characteristics pertaining to company visits, such as fund size, fund performance, style, and so on. Funds may prefer to visit a city that has greater economic importance and public company agglomeration, or historical and cultural ties between the cities, so we include firm-city×time fixed effects which control for the effects of companies' location that may influence site visits. The standard errors are clustered at the fund×firm-city level.
We observe in Table 4 that mutual fund site visit decisions can differ depending on whether they hold stocks in a company. Therefore, we study the determinants of fund visits for the two subsamples of stocks separately. This approach also allows us to assess whether the size of fund holdings affects company visit decisions. 7 Regression results are presented in Table 5. Columns 1-3 present the results for stocks that are held at the end of period t−1. In Column 1, a company is considered local for a fund family if both are in the same city. We find that the coefficient on the local dummy is positive and highly significant, indicating mutual funds conduct more site visits to local firms after controlling for an array of firm characteristics. Given the coefficient of 0.081, mutual funds conduct 8.4% more visits to local versus non-local companies during every period. 8 The economic magnitude is significant given that we control for the effects of fund family, company, and location. The coefficient on Holding is positive and significant, suggesting mutual funds make more visits to firms in which they have larger stakes. The result is consistent with the univariate evidence in panel A of Table 4.
In Columns 2 and 3 of Table 5, we vary the definition of Local and consider a company local if it is located within 100 KM and 200 KM of the fund city, respectively. The results are quantitatively similar to those in Column 1. As expected, the coefficient on Local decreases in magnitude as the local definition expands to cover more distant areas. 7 We also obtain results based on the full sample of observations. The results are largely consistent with the subsample results and are especially similar to the results for the subsample of not-held stocks, as this subsample constitutes a large portion of the full sample. Electronic copy available at: https://ssrn.com/abstract=3371978 23 Columns 4-6 of Table 5 present results for stocks that are not held by the mutual funds. The sample size is much larger than in Columns 1-3. Again, we find that mutual funds conduct more visits to local versus non-local firms for stocks they do not own. Holding is omitted in these models because it is zero for all firms in this sample by design. The findings are robust across the two subsamples.
A few comparisons stand out between the results of these two subsamples. Mutual funds are more likely to visit large companies when the stocks of these firms are not held by the funds. The preference for large firms may reflect general mutual fund preference for liquidity, but it may also reflect concentrated information acquisition efforts in a small number of stocks that allow mutual funds to establish greater information advantage (Van Nieuwerburgh and Veldkamp 2010). In comparison, when mutual funds own the stocks, portfolio weight rather than firm size affects visit decisions. Mutual funds are also more likely to visit companies with greater analyst coverage, and this holds for stocks that are held and not held by mutual funds. Financial analyst coverage and fund information acquisition may be driven by similar factors. Specifically, greater analyst coverage may indicate stronger demand for information by institutional clients (see, e.g., O'Brien and Bhushan 1990). For both subsamples, funds are more likely to visit firms with good stock price performance, younger firms, and firms with greater return volatility. The latter two firm characteristics are proxies for information uncertainty and hence value of information acquisition.
Overall, the key finding in Table 5 is that mutual funds are more likely to visit local versus nonlocal companies. The results remain robust after controlling for firm characteristics and various fixed effects, and they hold for stocks both held and not held by funds.
In the Internet Appendix, we conduct two robustness checks. First, our sample focuses on the firms listed on SZSE. Therefore, mutual funds in Shenzhen may acquire information differently from funds located in other cities because of the location of SZSE. To address this concern, we perform the same tests as in Table 5 and exclude funds located in Shenzhen, which represent about one-fourth of the sample. As shown in Internet Appendix Table A2, excluding the Shenzhen-based mutual funds does not affect the main findings, and the results are similar to those in Table 5.
Second, not all visits are the same or provide the same amount of information. We consider one aspect of the heterogeneity in company visits that is particularly relevant in our setting: unique versus repeat visits by the mutual funds. A unique (repeat) visit in month t is defined as a fund visit to a firm in month t without (with) prior visits by the same fund in the same firm in the past 12 months (from month t−1 to t−12). The unique visits, which include first-ever visits to a firm, may provide more information to fund managers than repeat visits, so the former may be more important for subsequent investment decisions. However, repeat visits may indicate a greater need for more information acquisition and hence greater value for the acquired information. To differentiate the impact of firm location on the two types of visit activities, we split the sample into unique and non-unique visits and repeat the analyses in Table 5 for the two subsamples.
As shown in the Internet Appendix, unique visits (Table A3) are far more common than nonunique visits (Table A4), and the sample size of the latter is about one-tenth that of the former. The results may indicate the resource or effort constraints fund managers face in conducting company visits: They rarely engage in repeat visits. Nonetheless, in both subsamples, mutual funds are more likely to visit local companies than non-local companies, and the results hold regardless of whether the stocks are in the fund portfolios.

Distance, Travel, and Site Visits
If geographic distance affects the information acquisition activities of mutual funds because long distance increases information acquisition costs, we expect that ease of travel between cities will affect site visits. As the ease of travel alleviates the geographic distance constraint by reducing the overall cost (time, effort, etc.) of non-local visits, there should be more site visits to firms located in cities that are easier to reach by fund managers. Investigating how travel between fund and firm cities affects fund visits allows us to determine the causality of the results documented in the previous section, and it offers further insights into the effects of geographic constraints on information acquisition.
In this section, we formally test this idea by exploiting the rapidly changing mode of transportation that significantly altered ease of travel between Chinse cities: the introduction of the HS train. We consider the introduction of HS train connections between fund and firm cities a shock to ease of travel, and use a difference-in-difference approach to examine the causal impact of distance on information acquisition. The change in HS train connection allows us to identify the channel through which a firm's geographic location affects information acquisition efforts by mutual funds. In the research design, the treatment is determined for a pair of cities that established a direct HS train connection, allowing us to control for other local trends and shocks and identify According to a survey in the study, about 40% of HS train passengers travel for business purposes.
HS railways mainly compete with air travel for long-distance trips and with traditional railways for short lines, and they quickly became the preferred choice for short-to-median distance business travel. Lin (2017)  We estimate the following difference-in-difference model for each pair of fund-firm cities: log(1 + , , ) = 0 + 1 , , −1 + The dependent variable is the number of visits by fund i to city k in period t. The variable of interest is HS, which is an indicator variable that equals 1 if a direct HS train connection between the fund city and the firm city is established before period t, and 0 otherwise. As in Equation (1) Panel A of Table 7 reports the results for samples of non-local visits in Columns 1-3. We exclude cities within a specified distance range from the sample, as visits to these cities are considered local visits. In all columns, the coefficients of HS are positive and significant. Therefore, after direct HS train connections were introduced between the fund and firm cities, fund managers increased visits to firms. As we control for fund×firm-city fixed effects and firm-city×time fixed effects, the result is not driven by any established economic or cultural ties between the cities or other shocks to the cities of firms. As in previous regressions, we include fund×time fixed effects to ensure that the result is not driven by changing company visit policies by mutual funds. The evidence shows that ease of travel has a causal effect on mutual fund site visit decisions. The finding supports our hypothesis that distance and its associated cost affect the information acquisition decisions of mutual funds.
The empirical identification in Equation (2) comes from comparing the change in visits between a pair of cities that experience a shock to their transportation mode with the change in visits between a pair of cities that do not experience a shock. One concern with the difference-indifference approach is that the estimated treatment effect could be due to pretreatment differences in the characteristics of treated and control groups. To address this concern, we examine the dynamics of mutual fund site visits around the introduction of new HS train connections. In particular, we add the leads (before treatment) and lags (after treatment) of HS in Equation (2).
Including the leads, HS(−1) and HS(−2), can control for pre-treatment effects, and controlling for the lags, HS(+1) and HS(+2), can trace the treatment effects in the periods after the initial shock.
The estimates based on this new specification are reported in panel B of Table 7. The dynamics of site visits around HS train connections strongly support our hypothesis. First, we do not find an anticipatory effect. The lead HS variables are not significant, indicating that funds do not pay more visits before an HS train connection is established. Second, funds pay more visits to a city right after an HS train connection and continue to do so in the following periods. The significant effect of the lag HS variables suggests that the impact of the shock to ease of travel is long-lasting and that the effect of HS train connections strengthens over time.
Overall, the evidence in Table 7 shows that the establishment HS train connections has a strong impact on the information acquisition activities of mutual funds. Giroud (2013)  Next, we examine whether the new mode of transportation alleviates the geographic distance constraint and how distance and the mode of travel jointly affect mutual fund visits. We study mutual fund visits to individual firms by estimating the following model: log(1 + , , ) = 0 + 1 , + 2 , , −1 + 3 , × , , −1 + 4 , , −1 + ℎ , −1 + where the number of visits is at the fund-firm level. Near is −log(1 + distance), which is log of the inverse of one plus the distance between the fund city and the firm city. The greater the value of this variable, the shorter the distance. All other variables are defined as earlier. The main variable of interest is the interaction term between HS and Near. The ease of travel should attenuate the costs associated with geographic distance, so given the same distance, funds could pay more visits to companies in cities with HS train connections. The effects of fund holdings and firm characteristics are also similar to those in Table 5. The results reveal that geographic distance has a significant impact on fund visit activities. The ease of travel attenuates this effect but does not eliminate geographic distance constraints.
Columns 4-6 of Table 8 present the results for stocks that are not held by mutual funds. We find the same evidence as in Columns 1-3. The coefficients on Near, HS, and Near×HS are positive and significant for all definitions of non-local areas. Efforts by mutual funds to obtain information about stocks that are not in their portfolios are not directly observable. The earlier results confirm that geographic proximity significantly affects mutual fund visit decisions in these firms as well.
The results here, in combination with the results in panel B of Table 4, show that mutual funds allocate considerable resources to stocks they do not own, highlighting the effects of geographic distance on overall information acquisition by mutual funds.

Site Visits, Investment Decisions, and Performance
In Section 2, we present evidence that Chinese mutual funds exhibit a strong local preference in their portfolio holdings and that these mutual funds seem to benefit from such local preference, as their local holdings outperform non-local holdings, on average. In Sections 3 and 4, we show that geographic proximity affects mutual fund site visits. In this section, we study how site visits affect mutual fund investment decisions and performance. The analysis offers direct support for the assumption that site visits are a form of information acquisition activity by mutual funds and can provide the missing link between local preference and local holding performance.

Site visits and mutual fund trading
We first examine whether company visits are associated with mutual fund investment decisions in the visited stocks. As shown in the previous sections, mutual funds devote substantial effort visiting firms inside and outside of their portfolios. If such site visits represent important information gathering by mutual funds, through which the funds can reduce the noise of their information signals or acquire new information, these site visits should lead to changes in mutual fund portfolio positions in the visited stocks. To formally test this conjecture, we first estimate the following model: | ℎ | , , = 1 , + 2 × log (1 + , , ) + ℎ , −1 + + where the notations follow the previous models. The dependent variable is the absolute value of the holding change of a mutual fund i on the stock of company j in period t. In particular, where ℎ is the number of shares of firm j held by fund i at the beginning of the period, ℎ is the number of shares of firm j held by fund i at the end of the period, is the share price of stock i at the beginning of the period, and is the total net asset value of the mutual fund at the beginning of the period. A company is included in the estimation if it is held by the mutual fund at either the beginning or the end of the period.
In the regression Equation (4), we use site visits to firms to explain fund holding changes in the same period, which allows us to study the impact of mutual fund visits on mutual fund investment decisions. Ideally, we should use site visits from the previous period to explain subsequent changes in fund holdings. However, given the low frequency of the holding data (every half year), we can only observe fund holding changes over a half-year period. It is reasonable to assume that if there is a relation between site visits and holding changes, mutual funds are likely to make investment decisions after rather than before site visits. Consequently, for the full sample, we examine the relation between site visits and investment decisions in the same period. In unreported results, we also examine the relation between early site visits (first 2 months in the half- year period) and holding changes in the same period, and the relation between late site visits (last two months in the half-year period) and holding changes in the later period. These subsample results are consistent with the full sample results.
Columns 1-3 of Table 9 present results for stocks that are held by mutual funds at the end of the previous period, so the holding change during the period can represent either buying or selling.
Columns 4-6 present the results for stocks that are not held by mutual funds, so the holding change can only be buying. In all columns, the number of visits has a positive and significant impact on the holding changes of mutual funds. In Internet Appendix Table A5, we further divide the previously held stocks into groups in which the holding change is positive (buy) and negative (sell) and re-estimate Equation (4). Again, we find that visits are positively and significantly associated with subsequent buying and selling activities.
As in earlier results, we use different specifications for local areas. The coefficients on Local are close to zero and generally insignificant in all specifications, suggesting that mutual funds do not trade the stock of a company more frequently simply because the company is local. Note that mutual funds are more likely to visit local firms, for both stocks they own and stocks they do not own. The highly significant effect of site visits on fund holding changes could suggest that mutual funds are likely to trade local stocks more frequently because they are also more likely to update and acquire information about these stocks. 9 This result provides an important insight into mutual fund portfolio holding decisions. The local bias documented in Table 2 is not driven by distance per se. Rather, because geographic proximity lowers the cost of information acquisition and facilitates the information acquisition process, it is information acquisition that at least partly drives the mutual fund investment decisions in local stocks.

Site visits and mutual fund trading performance
We next study whether and how site visits affect the trading performance of mutual funds.
Again, because of the low frequency of observed fund holding changes, we calculate post-visit cumulative abnormal returns based on the trading activities of visiting mutual funds during the concurrent period. This approach reflects the returns that mutual funds could earn through their trades shortly after the visits. 10 The computation is as follows. At the end of each month j within reporting period t, we group all stocks visited by funds in month j into five portfolios based on whether the stock is held by the fund at the beginning of period t and how the fund trades the stock in period t. We divide the visited stocks that are not held into portfolios of visit-buy and visit-noact, and stocks that are held into portfolios of visit-buy, visit-sell, and visit-no-act. We examine both equally weighted and trade-value-weighted portfolio returns in month j+1, from months j+1 to j+3, and from months j+1 to j+6. The trade value in the value-weighted return result is the dollar value of fund holding changes in period t, as defined in Equation (5). The returns are adjusted by size and book-to-market benchmark returns. The benchmark portfolios are equally weighted, and the results are robust when value-weighted benchmark portfolios are used. For brevity, we report results based on one definition of locality, where a firm is considered local for a fund if the distance between their cities is less than or equal to 100 KM. Results are similar when alternative definitions of locality are used.
Panel A of Table 10 presents post-visit 1-month performance. For stocks that are not held at the beginning of period t, the equally (trade-value) weighted abnormal returns for visit-buy stocks are 1.13% (1.15%) and statistically significant, and both local and non-local stocks have positive and significant abnormal returns. Column 2 presents abnormal returns for visit-no-act stocks that are not owned before visits, and the returns do not differ significantly from their benchmark portfolio returns. Column 3 reports the differences between Columns 1 and 2. The visit-buy stocks outperform the visit-no-trade stocks in the full sample, and the outperformance is concentrated in the non-local stocks where the difference in the 1-month return is 1.37%. For the large number of 10 An alternative approach is to examine stock returns over the next period after observing visit and trading activities in the previous (half-year) period. Because this incurs a substantial lag in computed trading returns, we find that the tests lack statistical power in untabulated results.
Electronic copy available at: https://ssrn.com/abstract=3371978 stocks that mutual funds do not hold, site visit decisions are unlikely to be exogenous and the information acquisition decisions and investment decisions could be determined jointly. By comparing stocks that mutual funds visit but take different actions after the visits, we can evaluate the effect of these visits on investment performance through different investment decisions.
The results provide strong evidence that site visits provide important information to mutual funds. For firms that funds visit but do not buy, their stocks underperform the visit-buy stocks during the 1-month period after the visiting month. This effect is more prominent for non-local stocks for two possible reasons. First, site visits are costly, especially for non-local stocks. The visits and buy decisions of non-local, not previously held stocks could be triggered by much stronger information signals both before and during the visits. Second, mutual funds may be able to acquire substantial information about local stocks through channels other than site visits (e.g., Gurun and Butler 2012). Interestingly, local stocks that mutual funds visit but do not buy substantially outperform their non-local counterparts, suggesting possible differences in pre-visit information of the two groups of stocks.
Columns 4-6 of Table 10 present post-visit abnormal returns for stocks owned by the funds.
The results are similar to those in Columns 1-3 but the statistical significance is weaker. Here, we can directly compare the trading performance between post-visit mutual fund purchases and mutual fund sales, and report both equally weighted and value-weighted results. Visit-buy stocks tend to have positive abnormal future returns, and Column 7 shows that visit-buy stocks outperform visit-sell stocks. The results hold for both equally weighted and trade-value-weighted portfolios. There is little difference in the returns of visit-sell and visit-no-act stocks. Column 8 provides a comparison of post-visit purchases of stocks owned and not owned by the mutual funds.
The results show that initial purchase of local and non-local stocks outperforms accumulation purchase of those stocks, though the differences are not statistically significant.
In panels B and C of Table 10, we conduct the same tests as in panel A and examine the postvisit 3-and 6-month cumulative abnormal returns. All findings in panel A continue to hold in panel B with a similar economic magnitude and a stronger statistical significance. For example, Columns 4-6 present abnormal returns for stocks that are initially owned by mutual funds. The visit-buy stocks have positive and significant abnormal future returns, and outperform the visit-sell stocks significantly. Panel C further shows that the overall return patterns for the 1-and 3-month periods hold for the 6-month period, but the economic magnitude is nevertheless weaker for the second 3month period. Overall, the superior performance of visit-buy stocks does not disappear quickly and persists for at least 6 months after portfolio formation. However, the outperformance levels off during the second half of the 6-month period.
In Internet Appendix Table A6, we split the sample into unique and repeat visits, and compute 1-month post-visit abnormal returns as in panel A of Table 10 using the two subsamples separately.
We continue to find that the visit-buy stocks deliver positive and significant abnormal returns if they are not initially owned in both subsamples. The visit-buy stocks still outperform both the visit-sell and visit-no-act stocks. Therefore, the main findings in Table 10 hold for both unique and repeat visits.
Overall, the results provide support for the argument that mutual funds conduct costly site visits because they can acquire valuable investment-relevant information during these company visits. The results in this section show that mutual fund site visits affect mutual fund investment decisions and that these information acquisition activities contribute significantly to mutual fund investment performance.

Conclusion
Using a unique data set of Chinese mutual fund site visits to companies, we provide the first direct evidence of mutual fund information acquisition activities and the relation between information acquisition and investment decisions. Mutual funds are more likely to visit geographically proximate firms. These visits provide valuable information to fund managers and affect fund trading decisions. We establish the causal effects of geographical proximity on information acquisition by exploring shocks to ease of travel between cities and adopting a difference-indifference research design. Ease of travel mitigates but does not eliminate the geographic constraints in information acquisition.
Our findings suggest that the local preference in mutual fund portfolio decisions and the superior performance of local holdings are at least partially driven by more intensive information acquisition about local stocks by mutual funds. Geographic proximity could provide mutual fund managers some initial information advantage. The initial advantage, combined with the lower costs of information acquisition about local firms, leads to greater information acquisition efforts in these firms. Our results provide direct support for the endogenous information acquisition explanation of local bias (Van Nieuwerburgh and Veldkamp 2009).
We also uncover evidence of substantial information acquisition activities by mutual funds in stocks they do not own. Based on company visits, mutual funds on average devote 80% of their efforts to stocks outside of their portfolios. These actions are typically hidden from researchers and cannot be inferred directly from analyses of mutual fund portfolio holdings. Such information-gathering efforts can have important implications for the joint information acquisition and investment decision process. For example, the evidence helps explain the superior performance of newly initiated stock positions in mutual fund portfolios (Alexander, Cici, and Gibson 2007) and the poor performance of local stocks that local mutual funds do not own (Coval and Moskowitz 2001   Electronic copy available at: https://ssrn.com/abstract=3371978

Variable Definition City level
Near −log(1 + distance), where distance is measured between the city of a fund and the city of a firm HS Equal to 1 if the direct high-speed train connection between the city of fund and the city of firm is established, and 0 otherwise Fund-family-firm level Visits Number of visits of the fund family to a firm in each period

Local
Equal to 1 if the distance between the city of a mutual fund and the city of a firm is within a range, and 0 otherwise Holding Portfolio weight of a firm's stock in the mutual fund holdings |HoldingChg| Absolute value of holding change of a fund family on a firm's , where ℎ is the number of shares held at the beginning of period, ℎ is the number of shares held at the end of period, is the share price of the stock at the beginning of the period, and is the total net asset value of the mutual fund at the beginning of the period Firm level Size log(market value of firm equity)

ROA
Earnings before interest and taxes (EBIT) scaled by total assets SOE Equal to 1 if the firm is state owned, and 0 otherwise Age log[(Fiscal year-ending date − initial listed date)/365] Coverage log(1 + Report Num), where Report Num is the total number of financial analyst reports for the firm in a period AssetGrowth (Total assets − lagged total assets)/lagged total assets   This table presents  Electronic copy available at: https://ssrn.com/abstract=3371978 (1)   This table presents results on the local preference of mutual funds. The sample includes all actively managed equity funds and equity balanced funds from January 2007 to June 2017. The fund-level information is aggregated at the fund-family level every half year. Only stocks listed on the Shenzhen Stock Exchange are included in the analysis. Three definitions of local firms are used. A firm is considered local for a fund if the distance between their cities is 0 kilometers (KM), less than 100 KM, or less than 200 KM. Column 1 reports the average portfolio weight of local stocks across fund families. Column 2 presents the benchmark weight based on the market capitalization of firms. To calculate this benchmark, we divide the total market value of local stocks by the total market value all stocks. Column 3 presents the benchmark weight based on the holdings of all fund families. To calculate this benchmark, we divide the holding value of local stocks across all fund families by the holding value of all stocks across all fund families. Column 4 shows the differences between Columns 1 and 2. Column 5 shows the differences between Columns 1 and 3. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.
(1)  Table 3: Local preference and information advantage This table presents the relation between local preference and future fund portfolio returns. The sample includes all actively managed equity funds and equity balanced funds from January 2007 to June 2017. The fund-level information is aggregated at the fund family level every half year. At the end of period t, we split a fund's holdings into the local portfolio and the non-local portfolio. Three definitions of local firms are used. A firm is considered local for a fund if the distance between their cities is 0 kilometers (KM), less than 100 KM, or less than 200 KM. For each portfolio, we calculate raw returns as the average monthly returns in period t+1 (the next 6 months) using the stockholding weight at the end of period t and present the results in columns 1-3. We construct the benchmark portfolios based on size and book-to-market (BM) and present the abnormal returns in Columns 4-6. Column 3 shows the differences between Columns 1 and 2. Column 6 shows the differences between Columns 4 and 5. Returns are in percentage points. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

4: Local and non-local site visits
This table presents the relation between fund-firm distance and site visits of mutual funds. The sample includes all actively managed equity funds and equity balanced funds from January 2007 to June 2017. The fund-level information is aggregated at the fund-family level every half year. We independently sort stocks into 2×2 groups by whether they are local stocks and whether they are initially held by funds. Three definitions of local areas are used. A firm is considered local for a fund if the distance between their cities is 0 kilometers (KM), less than 100 KM, or less than 200 KM. In panel A, we report the average ratio of the number of visits to each group of stocks to the total number of Shenzhen Stock Exchange (SZSE) stocks in that group. In panel B, we report the average ratio of the number of visits to each group of stocks to the total number of visits by each fund family. Columns 1-3 show the results for local and non-local visits in full sample. Columns 4-6 show the results for local and non-local visits for stocks that are initially held. Columns 7-9 show the results for local and non-local visits for stocks that are not initially held. Column 3 shows the differences between Columns 1 and 2. Column 6 shows the differences between Columns 4 and 5. Column 9 shows the differences between Columns 7 and 8. ***, **, and * correspond to significance at the 1%, 5%, and 10% levels, respectively.
Full sample With initial position Without initial position (1) (2)    The sample includes all actively managed equity funds and equity balanced funds from January 2007 to June 2017. The fund-level information is aggregated at the fund-family level every half year. The dependent variable is the logarithm of one plus the number of visits to a city by a fund family in each period. In panel A, HS is a dummy variable that equals 1 if the fund city and firm city are directly connected by the highspeed train network, and 0 otherwise. In panel B, lags and leads of HS are used. Three definitions of non-local areas are used in Columns 1-3. A firm is considered non-local for a fund if the distance between their cities is more than 0 kilometers (KM), more than 100 KM, or more than 200 KM. The time, fund×firm-city, fund×time, and firm-city×time fixed effects are included. The t-values are reported in parentheses. The standard errors are clustered at the fund×firm-city level. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.
(1)    At the end of each month j within period t, we sort all stocks visited by funds in month j into 5 portfolios based on whether the stock is held by the fund at the beginning of period t, and how the fund trades the firm's stock in period t. We examine both the equally weighted (EW) and trade value-weighted (VW) portfolio performance in month j+1 in panel A, from months j+1 to j+3 in panel B, and from months j+1 to j+6 in panel C, using the size and book-to-market adjusted cumulative abnormal returns. Returns are in percentage points. In each panel, we present the results based on the full sample, the sample of local and non-local stocks, and the sample of initially held and not held stocks. A firm is considered local for a fund if the distance between their cities is less than 100 kilometers. The t-values are reported in parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Online Appendix Table A4: Location and Non-Unique Visits
This table presents the estimates from a series of OLS regressions. The dependent variable is the logarithm of the number of visits plus one. The sample include all actively managed equity funds and equity balanced funds from January 2007 to June 2017. The fund level information is aggregated at the fund family level every half year. Different definitions of local areas are used. A firm is considered as local for a fund if the distance between their cities is 0 KM, is less than 100 KM, and is less than 200 KM. A non-unique visit in month t is defined if a fund visits the firm in month t, and has visited the same firm from month t-1 to t-12. Columns (1) to (3) present the results for stocks that are initially held by funds. Columns (4) to (6) present the results for stocks that are not initially held by funds. The time fixed effects, fund×time fixed effects, and firm-city×time fixed effects are included. The variable definitions are provided in Appendix Table A2. The t-values are reported in parentheses. The standard errors are clustered at the fund×firm-city level. ***, **, and * correspond to statistical significance at the 1%, 5%, and 10% levels, respectively. With Initial Position Without Initial Position (1)  (6) present the results if the holding change is negative. Different definitions of local areas are used. A firm is considered as local for a fund if the distance between their cities is 0 KM, is less than 100 KM, and is less than 200 KM. The time fixed effects, fund×time fixed effects, and firm-city×time fixed effects are included. The variable definitions are provided in Appendix  Table A2. The t-values are reported in parentheses. The standard errors are clustered at fund×firm-city level. ***, **, and * correspond to statistical significance at the 1%, 5%, and 10% levels, respectively. This table presents the performance of stock portfolios based on mutual fund trading activities and site visits. At the end of each month j within a period t, we sort all stocks visited by funds in month j into 5 portfolios based on whether the stock is held by the fund at the beginning of period t, and how the fund trades the firm's stock in period t. We examine both the equally weighted and value weighted portfolio performance in month j+1, using the size and book-to-market adjusted abnormal returns. We present the results based on the full sample, the sample of local and non-local stocks, and the sample of initially held and not held stocks. Returns are in percentage points. A firm is considered as local for a fund if the distance between their cities is less than 100 KM. A unique (non-unique) visit in month t is defined if a fund visits the firm in month t, and does not visit (also visits) the same firm from month t-1 to t-12. The t-values are reported in parentheses. ***, **, and * correspond to statistical significance at the 1%, 5%, and 10% levels, respectively.