What Triggers Corporate Site Visits, and Do Investors Care? A Comparison of Buy-Side and Sell-Side Analyst Site Visits in China

Abstract

Taking advantage of a recently established dataset that records financial analysts’ firm visits, this study examines the factors that determine the number of buy-side and sell-side analyst firm visits in a given year and how investors respond to these visits. Both buy-side and sell-side analysts seem motivated to visit firms when they are considering buying or recommending that investors buy shares rather than when they are considering selling or when they are recommending that investors sell shares. However, there are still significant differences in the firms that buy-side and sell-side analysts choose to visit. Using a binomial count model, this study shows that buy-side analyst firm visits are heavily focused on industry leaders—firms with a high share of the total revenue in their given industry—while sell-side analyst firm visits are not. By looking at the response of investors to buy-side or sell-side analyst firm visits, a regression analysis uncovers evidence that investors trust buy-side analysts, responding to firm visits by buy-side analysts by purchasing more shares in the firm. Investors do not have the same level of confidence in sell-side analysts. Investors respond to firm visits by sell-side analysts by selling more shares in the firm. This phenomenon is even more significant for firms that are outperforming the market. The results suggest that there is a significant cost to the conflict of interest inherent in sell-side analysts’ research. These costs increase when analysts recommend outperforming firms to the public.

1. Introduction

Investors rely on professional analyses when making financial investment decisions. Some investors, particularly individual investors, utilize managed mutual funds, but a substantial proportion of market participants prefer to manage their investments themselves. Reasons for the self-management of one’s investments include a lack of trust of mutual fund managers due to high agency costs (Ferris and Yan 2009), high costs to researching and understanding fund performance (Kothari and Warner 2001), and high fees, particularly for funds with a strong performance (Luo 2002). If individual investors prefer to self-manage their investments, they often rely on professional recommendations in the form of financial analyst reports or financial press coverage (Nagy and Obenberger 1994).

Financial analyses are conducted by both “buy-side” and “sell-side” analysts in the financial market. The fundamental objective of buy-side and sell-side analysis is the same: both conduct research on companies in order to make investment recommendations. However, there are some significant differences between the two. Buy-side analysts work for fund managers and usually do not disclose their work to the public. They keep a low profile when performing market research to avoid publicizing their purpose (Frey and Herbst 2014). Buy-side analysts usually focus more on fundamental financial report analyses (Brown et al. 2016). There is evidence that the firms they select for coverage outperform the firms selected by sell-side analysts (Rebello and Wei 2014; Cheng et al. 2006). Sell-side analysts usually work for investment banks. They make their investment recommendations—whether to buy, sell, or hold certain stocks—available to institutional and retail clients and often even to the general public. One of the roles of sell-side analysts is to retain clients at the firm by helping them make better investment returns. However, sell-side analysts in China are not compensated based on investment performance; they are expected to cover the costs of their research through commissions on stock trades. This compensation structure creates pressure on sell-side analysts to generate trading volume and demand for new issues that their employer may be underwriting or distributing. Clearly, this creates conflicts of interest for sell-side analysts: their compensation is based on the profitability of their employer, the investment bank, rather than the quality of their investment advice. This is because there are two main types of business in the investment banking industry. One is the intermediation of stock and derivative trading, and the other is advising on initial public offerings when privately held firms shift to public listings on a stock exchange. The conflicts of interest inherent in these two roles of investment banks leave sell-side analysts open to criticism (Arand and Kerl 2015). Additionally, since the sell-side analysts publish their personal ideas, they may have a self-interest bias (Gu et al. 2013). For example, they may give an optimistic evaluation to a potential future employer but a pessimistic evaluation to firms they do not expect will hire them (Lourie 2019). Sell-side analysts also keep good relationships with managers in the firms they cover (Soltes 2014), which could potentially influence their judgment. The compensation of sell-side analysts is also always questioned. Their wage is not related to the quality of their work. Unlike fund managers, whose historical performance is publicly tracked, sell-side analysts do not have a research performance history that can be checked.

It has been well-documented that a buy-and-hold strategy of market indices performs at least as well, if not better, than professional money management, yet investors still rely on professional advice (Reiter-Gavish et al. 2021a). When the uncertainty of stock performance seems high, investors read investment recommendations from sell-side analysts and observe buy-side analyst firm visits for signals on their investment recommendations (Kim et al. 2021). Individual investors believe professional analysts possess more advanced industry knowledge and experience than themselves and, through their access to on-site research, have better, firm-specific information (Kong et al. 2021). This is especially true for investors with low financial literacy or financial sophistication (Reiter-Gavish et al. 2021b, 2022) or under distressed market conditions. At the same time, regulators have questioned the quality and integrity of professional financial analyses, at least in developed economies. A better understanding of investors’ reactions to investment recommendations from different sources will contribute to our understanding of financial markets and shape policies that enhance financial ethics in emerging markets as well.

The development of China’s financial markets accelerated after the Global Financial Crisis of 2007–2009. As a result, the market has experienced a rapid increase in the number of equity investment funds and financial analysts. Currently, investors can access recommendations from sell-side analysts by simply “following” the stock on a financial data stream platform in China. Buy-side recommendations are not directly accessible to investors in the Chinese market. However, since 2017, both the Shanghai and Shenzhen exchanges have required firms to disclose any research firm visits in a timely manner. Investors can observe the number of company visits performed by buy-side analysts and deduce buy-side interest in those firms.

The development of China’s financial markets and the increasing number of investment funds and financial analysts presents some ethical challenges. Many institutional investors do not trust the investment recommendations from sell-side financial analysts in the Chinese market (Ding et al. 2014). Chinese culture emphasizes personal relationship networks, which cultivates in the dispersion of information through those networks. Analysts with close connections to firms work as a channel through which information is disseminated to other sell-side analysts (Li et al. 2020). If the information is not verified or is difficult to verify, the sell-side analysts should not use it to provide a recommendation. However, the number of recommendations sell-side analysts make is large, and there are few consequences to making incorrect recommendations. The compensation of sell-side analysts is linked to the commissions generated from their investment advice and the performance of the investment bank employing them. The compensation and job security of buy-side analysts are linked to the performance of the funds’ investment portfolio. Therefore, it can be argued that the compensation of buy-side analysts is closely linked to the quality of their research. Another potential conflict of interest stems from the fact that mutual funds are often important clients of investment banks’ stock brokerage services and sell-side analysis is one tool investment banks use to maintain good relationships with buy-side funds (Spence et al. 2019). There is evidence that sell-side analysts issue optimistic recommendations that favor large buy-side clients (Gu et al. 2013). Even though there are five categories of recommendations the sell-side analysts can make to the public, it is extremely rare to see a recommendation to sell in the Chinese financial market (Hirshleifer et al. 2022).

Against this institutional background, this study investigates the factors that trigger corporate site visits by buy-side and sell-side analysts in the Chinese financial market. Further, this research explores investors’ reactions to corporate site visits by buy-side and sell-side analysts. This research contributes to the existing literature in several ways. Firstly, although much existing research has covered sell-side analysis, there are few studies that include buy-side analysis since it is generally not publicly disclosed, and it usually does not include clear buy or sell recommendations. Secondly, the current literature does not include an appropriate comparison of the incentives for buy- and sell-side analysts’ site visit preferences. We propose a new technique for comparing the firm characteristics that trigger a buy-side vs. a sell-side analyst’s visit to a firm. The existing literature has concluded that the number of firm visits by analysts significantly increases their forecast accuracy (Cheng et al. 2016), and there is evidence that analysts’ firm visits act as an external monitor that can increase firms’ performance (Cheng et al. 2016), but there is little distinguishment between the effects of buy- and sell-side analysts. Our results show that who conducts such a visit is an important qualifier. Buy-side analysts are more likely to visit firms that are industry leaders, whereas sell-side analysts are less likely to conduct visits to firms that are industry leaders. Investors respond positively to firm visits conducted by buy-side analysts but negatively to firm visits conducted by sell-side analysts. Especially when firm shares outperform the market, firm visits by sell-side analysts negatively impact investor sentiment.

The rest of the paper is organized as follows. The next section introduces the existing literature and formulates several hypotheses to investigate. The data and research methodology are explained in Section 3. Section 4 reports, interprets, and discusses the regression outcomes. The final section concludes the paper, providing some policy suggestions and directions for future research.

2. Literature Review and Hypotheses

This section reviews the existing related literature on what factors influence analyst visits, which we use to develop Hypotheses 1 below, and how analyst visits and other factors influence interest from investors as measured by the change in the number of shareholders in a given firm, which we use to develop Hypotheses 2a and 2b below.

2.1. Analyst Corporate Site Visits

2.1.1. The Efficient Market Hypothesis

Theoretically, we begin with the fundamental efficient market hypothesis of Fama (1970), which asserts that current asset prices already reflect all publicly available market information. If the efficient markets hypothesis is correct, there should be no arbitrage opportunities available: investors cannot make abnormal returns by picking certain stocks or timing their trades. This implies that, to minimize costs, the best strategy for investors is to passively follow a chosen index. In fact, while mutual fund managers may target a certain declared index, they deviate from the index strategically, looking to add value to their fund through stock picks and market timing (Kacperczyk et al. 2014). There is also evidence that mutual funds also learn from their competitors and show some level of investment herding (Jiang and Verardo 2018).

Applying the efficient markets hypothesis to an analysis of analyst visits or reports, we might expect there to be no benefit from conducting company visits. However, there is empirical evidence that firm visits increase analysts’ forecast accuracy (Cheng et al. 2016). Even if we accept that there may be a role for analyst firm visits for information gathering, unless analysts have insider information, both buy-side and sell-side analysts should provide similar information to investors based on firm visits. So, theoretically, there should not be any difference in the factors that predict company visits by buy-side or sell-side analysts. Conventional wisdom in the industry, however, is that buy-side and sell-side analysts give quite different investment advice.

One explanation for these deviations from the efficient markets hypothesis is that managerial skill increases the reaction speed toward information updating, which increases the profitability of investments (Kacperczyk and Seru 2007). Another explanation may be differences in the incentives of sell-side analysts and buy-side analysts. For example, sell-side analyst compensation is usually tied to brokerage fees, which may incentivize sell-side analysts to encourage larger trading volumes (Irvine 2004). Sell-side analysts may also provide upward-biased recommendations, especially when they initiate a new recommendation (Ertimur et al. 2011), so that the investment banking division of their employer can attract more business (Lin and McNichols 1998).

2.1.2. Comparison between Buy-Side and Sell-Side Analysts

Buy-side and sell-side analysts differ somewhat in their investment strategies. Buy-side analysts work for large-scale mutual funds and are often looking for strategic deviations from the consensus based on stock selection or market timing. Previous research has found evidence that funds target stocks with high visibility and liquidity, with a low level of idiosyncratic risk (Falkenstein 1996). In China, funds gravitate toward large firms that meet those criteria (Yang et al. 2014).

Another difference between buy-side analysts and sell-side analysts is that while sell-side analyst recommendations are often public, buy-side analyst reports are generally kept private and therefore unobservable by researchers. For this reason, research investigating the differences between buy-side and sell-side analyst recommendations is rare. One exception is Hobbs and Singh (2015), which extrapolates unobservable buy-side recommendations by assuming that mutual fund performance outcomes fully reflect buy-side analysts’ recommendations using Equation (1) below.

$$Ins{t}_{i,t}=\frac{S_{i,t }{P}_{i,t }-S_{i,t-1 }{P}_{i,t-1 }\left(1+r\right)}{S_{i,t-1 }{P}_{i,t-1 }}-\frac{{{\displaystyle \sum }}_{i=1}^{N=m}{S}_{i,t }{P}_{i,t}-{{\displaystyle \sum }}_{i=1}^{N=m}{S}_{i,t }{P}_{i,t-1}\left(1+r\right)}{{{\displaystyle \sum }}_{i=1}^{N=m}{S}_{i,t-1 }{P}_{i,t-1}}$$

(1)

where $Ins{t}_{i,t}$ is the change in the position of stock i in a portfolio at time t, while $S_{i,t}$ and $P_{i,t}$ are the number of shares of company i held at time t and the price of the shares of company i at time t, respectively. When the position change of company i in the portfolio at time t, $Ins{t}_{i,t}$, is positive, it is interpreted as meaning that a buy-side analyst of that mutual fund made a buy recommendation for company i. When the position change of company i in a portfolio at time t, $Ins{t}_{i,t}$, is negative, it is interpreted as meaning that a buy-side analyst of that mutual fund made a sell recommendation for company i. The first term of Formula (1) therefore represents the net monetary position change in stock i. The second term represents the portfolio’s overall return. The difference represents how much the individual stock “beat” the overall portfolio.

Using U.S. data, Hobbs and Singh (2015) then compared portfolio performance based on sell-side analysts’ recommendations and the extrapolated buy-side analysts’ recommendations and concluded that sell-side analysts outperform buy-side analysts.

This approach is novel and interesting, but it has certain flaws. First, if both the share number and price increase, it is accurate to conclude that the analysts made a buy recommendation. However, when the increase in the first term of Equation (1) is from a price increase, without any simultaneous increase in the number of shares held, it may be incorrect to interpret the change as reflecting a buy recommendation by analysts. Secondly, when there is a small increase in price but a large increase in shares, there is a large increase in the first term. The large increase in the number of shares held probably does reflect a recommendation to buy. However, the second term in Equation (1) is also difficult to interpret. The second term is large and positive if all other stocks perform well. Following the methodology put forth by Hobbs and Singh (2015), even if there is an increase in the number of shares held, the interpretation would be that the analysts put out a recommendation to sell. This would again likely be an incorrect interpretation of events. Finally, the comparison between the recommendations by buy-side and sell-side analysts is not really accurate to begin with as, following the methodology, there are only two decision options for buy-side analysts (buy or sell), but there are more than two possible recommendations from sell-side analysts (buy, sell, or hold; in some cases, there are other qualifications on the recommendation, such as “strong buy” or “strong sell”, etc.).

The current study improves on techniques used in the existing literature, using a count model to identify and compare buy-side and sell-side analysts’ decisions to make on-site firm visits. We propose the following hypotheses based on the efficient markets hypothesis and the existing literature (Hobbs and Singh 2015).

Hypothesis 1 (H1). 

Buy-side analysts are more likely than sell-side analysts to visit firms that are industry leaders.

2.2. Investor Reaction

2.2.1. The Free Rider Problem

The theoretical framework for understanding how investors react to sell-side vs. buy-side analysts’ on-site corporate visits begins with the free rider problem. Unlike investment banks, which simply give buy or sell recommendations to third parties, mutual funds “put their money where their mouth is,” invest in the firms, and hold their investments. Mutual fund managers conduct their research before making an investment decision. Once they have invested, they become insiders, able to access the firm’s financial and operational data. Unless the invested fund wants to further understand the operational efficiency of the firm for possible additional investments, there is no need for an invested “insider” mutual fund to conduct an “external” visit to a firm. Therefore, individual investors likely interpret a mutual fund analyst’s visit to a firm as a buy signal.

There is significant evidence that mutual fund ownership increases firm performance (Yuan et al. 2008). After investing in them, if mutual fund managers hold significant shares in diversified firms, they can actively influence the management of the firms. In general, mutual funds have two options for improving management quality. One is to directly intervene, and the other is to threaten to exit. There is evidence that long-term institutional investors tend to intervene in corporate governance more often and are more concerned about firm operation strategy and quality than other short-term investment objectives (McCahery et al. 2016). Additionally, free-riding individual investors expect management quality improvements at firms in which mutual funds are invested (Smith 1996).

2.2.2. The Conflict-of-Interest Problem

In general, market players tend to believe that financial analysts have stock-picking and market-timing abilities (Womack 1996). Therefore, financial analysts’ coverage of a firm is sometimes regarded as third-party monitoring of the firm’s performance. There is evidence that when more analysts cover a firm, the firm usually enjoys lower capital costs (Derrien and Kecskés 2013). The number of financial analysts covering certain firms also affects the competition among analysts. Research quality and accuracy are negatively affected when the competition is weak. Likewise, more analyst coverage increases positive externalities (Merkley et al. 2017). When financial analysts perform market research and make stock recommendations, they learn from each other (De Bondt and Forbes 1999), resulting in herding behavior in both their recommendations and actual investment decisions (Hirshleifer and Hong Teoh 2003).

As mentioned in the introduction, however, sell-side analysts face an inherent conflict of interest in their research. In some cases, this intrinsic conflict of interest is addressed using investor feedback to evaluate the quality of sell-side analysts’ research. There is evidence that if investor evaluations are economically important to analysts, feedback reduces the bias and alleviates some of the inherent conflicts of interest in sell-side analyses (Barradale et al. 2022). However, individual investors, aware of this inherent conflict of interest, may suspect that investment recommendations from sell-side analysts are designed to increase the volume of trading and therefore brokerage commissions (Jackson 2005; Brown et al. 2015). Individual investors may not have much confidence in investment recommendations from sell-side analysts since the main incentive behind those recommendations may be higher commissions.

Considering the free-rider problem and sell-side analysts’ inherent conflict of interest, the following hypotheses are made.

Hypothesis 2a (H2a).

Investors believe mutual fund analysts’ interests are closely aligned with investor interests, so the number of investors increases with the number of buy-side analysts’ visits to a given firm.

Hypothesis 2b (H2b).

Investors suspect sell-side analysts face an inherent conflict of interest, so the number of investors decreases with the number of sell-side analysts’ visits to a given firm.

3. Data

The data used in this research include annual financial statements and disclosures of on-site firm visits by buy-side and sell-side analysts from the Choice database. Prior to 2018, the disclosure of analyst visits was not compulsory. In 2018, both the Shanghai and Shenzhen exchanges began requiring firms to disclose analysts’ corporate visits. The firms included in the analysis cover a range of 2556 firms categorized into 90 sub-industries listed in the Chinese stock market on the Shanghai and Shenzhen stock exchanges over the three-year period between 2018 and 2020, the most recent data available at the time of this research study. After excluding any firms suffering from financial distress, the total sample is 7668 observations. Summary statistics and detailed variable definitions are reported in

Table 1. Summary Statistics.

Variable	Unit	Observation	Mean	Standard Deviation	Min	Percentile (25%)	Percentile (75%)	Max
BUY	Number of Visits	7668	0.937	2.991	0	0	0	54
SELL	Number of Visits	7668	2.611	7.804	0	0	2	142
∆Shareholders	10,000 accounts, then take LN	7656	0.014	0.282	−1.473	−0.130	0.101	2.571
∆AvgShares	Number of shares, then take LN	7668	0.155	0.437	−2.571	−0.032	0.258	3.370
∆FundShares	One million shares, then take LN	7174	0.008	2.056	−12.381	−0.833	0.877	13.666
EXRETURN	Percentage	7668	−12.833	39.900	−110.49	−34.970	−0.775	374.239
MKTSHARE	Percentage	7668	3.130	8.691	0	0.185	2.232	100
LIAB	Percentage	7668	46.299	30.669	0.836	30.032	60.404	1879.036
EPS	Percentage	7668	−43.571	1247.936	−93100	−27.586	41.051	14100
BPS	CNY	7668	4.986	3.734	−6.14	2.811	6.195	108.271
TOP	Percentage	7668	34.384	15.007	2.87	23.038	44.560	89.09
Inst Hold	Number of Institutional Investors	7668	88.993	176.999	0	8	94	2364

and

Table 2. Variable Definitions.

Variable	Symbol	Variable Treatment
Fund analysts visiting	BUY	Number of fund analysts visiting a firm
Investment bank analysts visiting	SELL	Number of investment bank analysts visiting a firm
Shareholder change	∆Shareholders	Natural log value of 10,000 of change in number of shareholders in a firm
Average shares hold	∆AvgShares	Average shares held by each shareholder
Fund position change	∆FundShares	Natural log value of the mutual fund share position change in a firm
Firm share excess return	EXRETURN	Firm’s share return minus the HS300 index return
Market share	MKTSHARE	Revenue of firm/sum of the total revenue of all firms in the same industry
Liability-to-asset ratio	LIAB	Liability/total assets
Earning per share growth	EPS	Earning per share growth rate
Book value per share	BPS	Total equity book value/total shares outstanding
Top shareholder’s ownership	TOP	Ownership of largest shareholder/total shares
Number of institutional investors	Inst Hold	Number of institutional shareholders

.

To address potential heteroscedasticity, the natural log, ln, is taken of the number of shareholder changes and the fund position change. In calculating an individual stock’s excess return, the HUSHEN 300 (CSI 300) index is used as the benchmark. The top shareholder’s ownership is from the capital structure disclosure, which reflects the top shareholder’s potential voting concentration. The liability ratio, earnings per share growth, and book value per share are collected from the firm’s financial reports and notes to reflect the firm’s operational leverage, growth, and the book value of the firm.

The summary statistics reported in Table 1 show that on-site firm visits from investment bank analysts (sell-side analysts) are more frequent than visits from mutual fund analysts (buy-side analysts). Comparing each individual stock’s return to the benchmark index return, the excess return is negative on average, indicating that most firms performed below the market during the sample period. The market share is calculated as each firm’s total revenue divided by the sum of the total revenue for the respective firm’s sub-industry. The summary statistics demonstrate that the market share is right-skewed, indicating some firms experience larger than usual market power in each sub-industry. Summary statistics on the number of shareholders and the funds’ share position change show the features of a normal distribution. Most variables collected from annual financial statements and notes are normally distributed, except the earnings per share growth.

4. Empirical Methodology

The data described in the previous section are used to test the hypotheses put forth in Section 2. A binomial count model is applied to the first set of empirical models to investigate the factors that influence the number of buy-side or sell-side analysts’ visits to a firm in a given year—Hypothesis 1. A regression analysis is applied to the second set of empirical models to analyze the effect of analyst firm visits on the number of shareholders of the firm—Hypotheses 2a and 2b.

4.1. Analyst Corporate Site Visits

Hypothesis 1, which posits that the factors that prompt analyst corporate site visits are different for buy-side vs. sell-side analysts, is tested using a negative binomial count model. The baseline model, Equations (2) and (3) below, is based on the Fama and French (1993) three-factor model. The empirical analysis investigates to what extent the number of firm visits conducted by analysts annually is captured by the firm size as measured by firm revenue as a ratio to the total revenue in the sub-industry, firm performance as measured by earnings per share, and net worth as measured by the book value per share.

$$BU{Y}_{i,t}={\beta }_0+{\beta }_{1}MKTSHAR{E}_{i,t}+{\beta }_{2}EP{S}_{I,t-1}+{\beta }_{3}BP{S}_{i,t-1}+{\epsilon }_{i,t}$$

(2)

$$SEL{L}_{i,t}={\beta }_0+{\beta }_{1}MKTSHAR{E}_{i,t}+{\beta }_{2}EP{S}_{I,t-1}+{\beta }_{3}BP{S}_{i,t-1}+{\epsilon }_{i,t}$$

(3)

Taking into consideration some of the unique features of China’s financial markets, where a relatively large percentage of the listed firms are family-owned or majority-state-owned enterprises (SOEs) which may have poor internal governance mechanisms, we also control for ownership concentration as measured by the percentage of shares held by the largest shareholder. Our final and preferred specifications, Equations (4) and (5) below, also control for firm capital structure as measured by the total liabilities as a ratio to the total assets.

$$BU{Y}_{i,t}={\beta }_0+{\beta }_{1}MKTSHAR{E}_{i,t}+{\beta }_{2}EP{S}_{I,t-1}+{\beta }_{3}BP{S}_{i,t-1}+{\beta }_{4}TO{P}_{i,t}+{\beta }_{5}LIA{B}_{i,t}+{\epsilon }_{i,t}$$

(4)

$$SEL{L}_{i,t}={\beta }_0+{\beta }_{1}MKTSHAR{E}_{i,t}+{\beta }_{2}EP{S}_{I,t-1}+{\beta }_{3}BP{S}_{i,t-1}+{\beta }_{4}TO{P}_{i,t}+{\beta }_{5}LIA{B}_{i,t}+{\epsilon }_{i,t}$$

(5)

The firm’s earnings, book-to-price ratio, ownership concentration, and capital structure could all potentially affect analysts’ decision to visit a firm. Existing research shows that analysts target firms with high earnings and low earning volatilities for “buy” recommendations (Bricker et al. 1995). Buy recommendations are also more likely for stocks with higher price-to-book values (Lin et al. 2011). A firm’s capital structure also influences analyst coverage. Highly leveraged firms face high credit costs, which may explain why a negative relationship between default risk and financial analysts’ coverage has been documented (Cheng and Subramanyam 2008). Finally, the Chinese market in particular has many state-owned enterprises (SOE) listed on the stock exchange and SOE ownership has been shown to negatively influence analysts’ consensus earning forecasts in the Chinese financial market (Huang and Wright 2015). To control for the effect of state ownership on the number of firm visits performed by analysts, we include the ownership concentration as a control variable.

4.2. Investor Reaction

Hypotheses 2a and 2b, in which investors respond differently to firm visits by buy-side vs. sell-side analysts, are tested in a model regressing the change in the number of shareholders for a given firm on the number of annual visits to the firm from buy-side and sell-side analysts. Our baseline specification, given in Equation (6), is based on findings in the existing literature. Regulators and industry insiders are well-aware of the conflicts of interest that sell-side analysts face. However, despite the upward bias in sell-side analyst forecasts (Huang and Wright 2015), it is well documented that traders react to analyst reports and small investors in particular do not fully account for the effects of analysts’ incentives on the credibility of analyst reports (Mikhail et al. 2007). Investor reactions to analyst reports are especially strong in countries with weak investor protection, such as China (Defond and Hung 2007), even when analyst recommendations have been shown to be inconsistent (Barniv et al. 2009).

If, as explained above, investors expect to free ride on increased monitoring of firm management after a firm visit by a buy-side analyst, Hypothesis 2a posits that the coefficient estimate on the number of annual visits by buy-side analysts, ${\beta }_1$, will be positive and statistically significant. If the conflict of interest inherent in the investment banking business model, in which the same institution simultaneously engages in commission-based stock trading and investment-bank-based stock and debt-issuing lines of business, leads investors to mistrust sell-side analysts, then Hypothesis 2b posits that the coefficient estimate on the number of annual visits by sell-side analysts, ${\beta }_2$, will be statistically insignificantly different from zero or even negative.

We refine our initial baseline specification with controls for the firms’ excess returns relative to the benchmark, the top shareholder concentration, and mutual fund share position change in the firm. If the investments from the mutual funds are large, there is a significant price push from below. A final specification, Equation (8), also includes an interaction term between the excess returns and sell-side analysts’ visits to investigate how sell-side analyst visits affect the number of shareholders when the stock outperforms the market index.

$$\Delta Shareholder{s}_{i,t}={\beta }_0+{\beta }_{1}BU{Y}_{i,t}+{\beta }_{2}SEL{L}_{i,t}+{\epsilon }_{i,t}$$

(6)

$$\begin{array}{ll}\Delta Shareholder{s}_{i,t}& \\ & ={\beta }_0+{\beta }_{1}BU{Y}_{i,t}+{\beta }_{2}SEL{L}_{i,t}+{\beta }_{3}EXRETUR{N}_{i,t}+{\beta }_{4}TO{P}_{i,t}\\ & +{\beta }_5\Delta FundShare{s}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(7)

$$\begin{array}{ll}\Delta Shareholder{s}_{i,t}& \\ & ={\beta }_0+{\beta }_{1}BU{Y}_{i,t}+{\beta }_{2}SEL{L}_{i,t}+{\beta }_{3}EXRETUR{N}_{i,t}+{\beta }_{4}TO{P}_{i,t}\\ & +{\beta }_5\Delta FundShare{s}_{i,t}+{\beta }_6\left(EXRETUR{N}_{i,t}\ast SEL{L}_{i,t}\right)+{\epsilon }_{i,t}\end{array}$$

(8)

5. Results

5.1. Analyst Corporate Site Visits

Table 3. Negative Binomial Count Models.

	Dependent Variable
	BUY	SELL	BUY	SELL	BUY	SELL
	(1)	(2)	(3)	(4)	(5)	(6)
MKTSHARE	0.006	0.002	0.007 *	0.003	0.010 ***	0.005
	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)
EPSt−1	0.001 ***	0.001 ***	0.001 ***	0.001 ***	0.001 ***	0.001 ***
	(0.0001)	(0.0001)	(0.0001)	(0.0001)	(0.0001)	(0.0001)
BPSt−1	0.066 ***	0.055 ***	0.072 ***	0.059 ***	0.072 ***	0.060 ***
	(0.009)	(0.008)	(0.009)	(0.008)	(0.009)	(0.008)
TOP			−0.008 ***	−0.006 ***	−0.009 ***	−0.006 ***
			(0.002)	(0.002)	(0.002)	(0.002)
LIAB					−0.006 ***	−0.004 ***
					(0.002)	(0.001)
Constant	−0.447 ***	0.653 ***			0.076 ***	1.007 ***
	(0.058)	(0.052)			(0.121)	(0.108)
Observations	7668	7668	7668	7668	7668	7668

Note: ***, **, and * denote the statistical significance at 1%, 5%, and 10%; standard errors are shown in parentheses.

reports the results of a count model analysis of Equations (1)–(4), factors that explain the incidence of analyst corporate site visits. In the count model, a negative binomial model is preferred over a Poisson model to address the over-dispersion problem. Columns 1 and 2 report the results of the negative binomial count model analysis of Equations (1) and (2), which explain the number of analyst visits to a firm by a buy-side (column 1, Equation (1)) or a sell-side (column 2, Equation (2)) analyst based on factors similar to the Fama–French three-factor model: firm size, as measured by firm revenue as a share of the total industry revenue; performance, as measured by earnings per share; and net value, as measured by the book value per share. In columns 3 and 4, the ownership concentration as measured by the percentage of shares held by the largest shareholder is included as a control variable. Our final and preferred specification, based on Equations (3) and (4) above, is reported in columns 5 and 6. This final specification also accounts for the firm’s capital structure, as measured by the total liabilities as a ratio to the total assets.

Looking at the results in Table 3, we note that, at least in our baseline specification, the market share, as measured by the firm revenue as a ratio to the total industry revenue, does not statistically significantly affect the incidence of firm visits by analysts. Higher earnings per share, reflecting better idiosyncratic firm performance, statistically significantly increase the incidence of firm visits by both buy-side and sell-side analysts in all specifications. The book value per share, reflecting the underlying value of the stock, also statistically significantly increases the incidence of firm visits by both buy-side and sell-side analysts across all specifications. Both results are highly statistically significant at the 1% level and suggest that on-site firm visits are motivated by an interest in purchasing firm shares in the case of a firm visit by a buy-side analyst or in making a recommendation to buy in the case of a firm visit by a sell-side analyst.

Given that firm visits seem to be motivated by an interest in buying or recommending buying, it is perhaps not surprising that both buy-side and sell-side analysts conduct fewer firm visits to firms with highly concentrated ownership. This is demonstrated by the highly statistically significant and negative coefficient estimate on the top shareholders’ ownership concentration ratio in columns 3–6 of Table 3. We expect that the ownership concentration by the top shareholder will be the highest for China’s many family-owned firms and state-owned enterprises (SOEs). Analysts may be reluctant to buy or recommend buying those firms because of suspected corporate governance issues and therefore do not make many firm visits. Note that in column 4, once the top shareholder’s ownership concentration is controlled for, the market share emerges as a marginally statistically significant factor in determining the number of firm visits by buy-side analysts.

The final and preferred specification, capital structure, as measured by the total liabilities as a ratio to the total assets, is also included as a control variable. Highly leveraged firms have fewer firm visits by both buy-side and sell-side analysts, as indicated by the highly statistically significantly negative coefficient estimates on the capital structure variable. This is as expected given that our results thus far have suggested that analyst visits are motivated by an interest in buying or recommending buying the firms visited. In this final specification, the coefficient estimate on the market share is positive and highly statistically significant in column 5, indicating that firms with a larger market share receive more on-site firm visits from buy-side analysts. In contrast, in column 6, the market share does not have a statistically significant impact on the number of on-site firm visits from sell-side analysts. The results confirm Hypothesis H1, in which buy-side analysts focus their visits on industry leaders. We speculate that this may be because, as documented in previous research, fund investors place greater value on the liquidity and low idiosyncratic risk of their holdings (Falkenstein 1996). A higher share of market revenue statistically significantly increases the number of firm visits performed by buy-side analysts but does not statistically significantly affect the number of firm visits performed by sell-side analysts.

5.2. Investor Reaction

Next, we turn to an analysis of investor reactions to analyst on-site visits. With the change in the number of shareholders as the dependent variable,

Table 4. Shareholder Change.

	Dependent Variable
	∆Shareholders
	(1)	(2)	(3)	(4)	(5)
BUY	0.010 ***	0.008 ***	0.009 ***	0.007 **	0.009 ***
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
SELL	−0.003 ***	−0.002 ***	−0.003 ***	−0.002 ***	−0.002 ***
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
EXRETURN			0.0005 ***	0.0004 ***	0.001 ***
			(0.0001)	(0.0001)	(0.0001)
TOP			−0.002 ***	−0.002 ***	−0.002 ***
			(0.0002)	(0.0002)	(0.0002)
∆FundShares			−0.016 ***	−0.016 ***	−0.016 ***
			(0.002)	(0.002)	(0.002)
EXRETURN*SELL					−0.0001 ***
					(0.00001)
Constant	0.011 ***	−0.007	0.078 ***	0.023 *	0.055 ***
	(0.003)	(0.006)	(0.008)	(0.013)	(0.010)
Year Control	N	Y	N	Y	Y
Observations	7656	7656	7162	7162	7162
R2	0.006	0.011	0.028	0.035	0.040
Adjusted R2	0.005	0.010	0.027	0.034	0.039
Residual Std. Error	0.282 (df = 7653)	0.281 (df = 7651)	0.278 (df = 7156)	0.277 (df = 7154)	0.276 (df = 7153)
F Statistics	21.519 *** (df = 2; 7653)	20.418 *** (df = 4; 7651)	40.691 *** (df = 5; 7156)	36.906 *** (df = 7; 7154)	37.474 *** (df = 8; 7153)

Note: ***, **, and * denote the statistical significance at 1%, 5%, and 10%; standard errors are shown in parentheses.

reports the results of the analysis of the reaction of investors to buy-side and sell-side analyst firm visits: the regression analysis of Equations (6)–(8).

Across all specifications, we see that a higher number of firm visits by buy-side analysts has a statistically significantly positive impact on the number of shareholders of the firms’ visited, but a higher number of firm visits by sell-side analysts has a statistically significantly negative impact on the number of shareholders of the firms’ visited. Thus, the results support Hypotheses H2a and H2b. In columns 3–5, we note that the idiosyncratic firm performance as measured by the firm’s excess returns over the benchmark return statistically significantly positively contributes to the number of the shareholders of the firm. The concentration of ownership of the top shareholder statistically significantly negatively affects the number of shareholders of the firm, presumably because of concerns about the corporate governance of family-owned firms and state-owned enterprises as discussed above. The theory of free riding suggests that individual investors may expect more institutional investor ownership of a firm to increase the level of monitoring and therefore they themselves would also increase the number of shares held. The negative coefficient estimate on the change in shareholding by mutual funds contradicts this. We are somewhat puzzled by this result but suspect that it is driven by the opposite relationship. Since institutional investors often hold quite large positions in any given firm, a decrease in institutional investor ownership may put downward pressure on the firm’s share price, making individual investors reluctant to reduce their ownership shares at the same time. If individual investors believe institutional investors are preparing to sell in the near future, they may be motivated to reduce their own holdings in advance.

In the final specification, Equation (7), reported in column 5, the interaction term between a firm’s excess returns and the incidence of sell-side analysts’ firm visits is statistically significantly negative. This means that for stocks that are outperforming the market, as the number of sell-side analyst firm visits increases, the number of investors decreases. This last finding further supports the idea that investors are concerned about conflicts of interest inherent to the role of sell-side analysts making stock purchase recommendations.

5.3. Robustness Checks

Finally, we subject the results of our panel data analysis presented in Table 4 to several robustness checks to possible heterogeneity in our data sample, spuriousness in the relationship between the dependent variable and the independent variable of interest, measurement error in the dependent variable and potential endogeneity in our empirical specification.

5.3.1. Sample Heterogeneity

To explore the possibility of heterogeneity across listing exchanges, the regression models from Equations (6) to (8) are also run-on sub-samples of firms listed on the Shanghai and Shenzhen stock exchanges separately. The results are reported in

Table 5. Sub-sample, Shanghai exchange listed firms.

	Dependent Variable
	∆Shareholders
	(1)	(2)	(3)	(4)	(5)
BUY	0.011 ***	0.008 **	0.011 ***	0.007 **	0.004
	(0.003)	(0.003)	(0.003)	(0.003)	(0.004)
SELL	−0.004 ***	−0.002	−0.003 ***	−0.002 ***	−0.002 *
	(0.002)	(0.002)	(0.001)	(0.001)	(0.002)
EXRETURN			0.0005 ***	0.0001 ***	0.001 ***
			(0.0001)	(0.0001)	(0.0001)
TOP			−0.001 ***	−0.001 ***	−0.001 ***
			(0.0003)	(0.0003)	(0.0003)
∆FundShares			−0.019 ***	−0.018 ***	−0.018 ***
			(0.002)	(0.002)	(0.002)
EXRETURN*SELL					−0.0001 ***
					(0.00004)
Constant	−0.003	−0.039 ***	0.063 ***	0.023 *	0.024 *
	(0.004)	(0.007)	(0.011)	(0.013)	(0.013)
Year Control	N	Y	N	Y	Y
Observations	3873	3873	3643	3643	3643
R2	0.003	0.019	0.032	0.052	0.055
Adjusted R2	0.002	0.018	0.031	0.050	0.053
Residual Std. Error	0.259 (df = 3870)	0.257 (df = 3868)	0.254 (df = 3637)	0.252 (df = 3635)	0.251 (df = 3634)
F Statistics	5.028 *** (df = 2; 3870)	18.975 *** (df = 4; 3868)	24.205 *** (df = 5; 3637)	28.237 *** (df = 7; 3635)	26.590 *** (df = 8; 3634)

Note: ***, **, and * denote the statistical significance at 1%, 5%, and 10%; standard errors are shown in parentheses.

and

Table 6. Sub-sample, Shenzhen exchange listed firms.

	Dependent Variable
	∆Shareholders
	(1)	(2)	(3)	(4)	(5)
BUY	0.009 ***	0.009 **	0.09 ***	0.008 **	0.011 ***
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
SELL	−0.003 ***	−0.003 ***	−0.003 ***	−0.003 ***	−0.004 ***
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
EXRETURN			0.0003 *	0.0003 ***	0.001 ***
			(0.0001)	(0.0001)	(0.0001)
TOP			−0.001 ***	−0.001 ***	−0.001 ***
			(0.0004)	(0.0004)	(0.0004)
∆FundShares			−0.013 ***	−0.014 ***	−0.014 ***
			(0.002)	(0.002)	(0.002)
EXRETURN*SELL					−0.0001 ***
					(0.00001)
Constant	0.029 ***	0.035 ***	0.087 ***	0.085 ***	0.088 ***
	(0.006)	(0.010)	(0.013)	(0.015)	(0.015)
Year Control	N	Y	N	Y	Y
Observations	3783	3783	3519	3519	3519
R2	0.007	0.008	0.022	0.023	0.031
Adjusted R2	0.006	0.007	0.020	0.021	0.028
Residual Std. Error	0.303 (df = 3780)	0.302 (df = 3778)	0.299 (df = 3513)	0.299 (df = 3511)	0.298 (df = 3510)
F Statistics	13.315 *** (df = 2; 3780)	7.681 *** (df = 4; 3778)	15.466 *** (df = 5; 3513)	11.995 *** (df = 7; 3511)	13.866 *** (df = 8; 3510)

Note: ***, **, and * denote the statistical significance at 1%, 5%, and 10%; standard errors are shown in parentheses.

, respectively.

The results of the analysis on both sub-samples confirm our analysis of the entire sample. Both support the original finding: on-site firm visits performed by buy-side analysts increase the number of shareholders, but firm visits performed by sell-side analysts decrease the number of shareholders. The better an individual stock’s performance relative to the market benchmark, the larger the decrease in the number of shareholders when the number of sell-side analyst visits increases.

The analysis of heterogeneity across listing exchanges illustrates that the results are robust to differences in firms’ characteristics. The Shanghai stock exchange’s listed firms are larger-sized firms in traditional industries, while the Shenzhen exchange’s listed firms are generally mid-sized firms in high-tech industries. The relationship between shareholders’ sentiments and the buy-side or sell-side analysts’ firm visits holds across firm listings in both exchanges.

As another robustness check against heterogeneity in the data sample, we re-run our regressions of Equations (6)–(8) on a subsample excluding the many firms that were not visited at all by an analyst, neither a sell-side nor a buy-side analyst, in the year in question. The remaining sample includes only those firms that had an on-site corporate visit from either a sell-side or a buy-side analyst in the given year, for a total of 1752 observations. The results of the analysis on this sub-sample, reported in

Table 7. Sub-sample, firms experiencing a corporate visit.

	Dependent Variable
	∆Shareholders
	(1)	(2)	(3)	(4)	(5)
BUY	0.007 ***	0.003	0.008 ***	0.004 *	0.005 **
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
SELL	−0.004 ***	−0.002 **	−0.004 ***	−0.002 **	−0.002 **
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
EXRETURN			−0.00004	−0.00005	0.00004 *
			(0.0002)	(0.0002)	(0.0002)
TOP			−0.002 ***	−0.002 ***	−0.002 ***
			(0.001)	(0.0005)	(0.0005)
∆FundShares			−0.028 ***	−0.028 ***	−0.029 ***
			(0.004)	(0.004)	(0.004)
EXRETURN*SELL					−0.0001 ***
					(0.00002)
Constant	0.067 ***	0.010	0.144 ***	0.085 ***	0.088 ***
	(0.010)	(0.016)	(0.019)	(0.023)	(0.023)
Year Control	N	Y	N	Y	Y
Observations	1752	1752	1735	1735	1735
R2	0.016	0.037	0.051	0.072	0.079
Adjusted R2	0.015	0.035	0.049	0.068	0.075
Residual Std. Error	0.318 (df = 1749)	0.315 (df = 1747)	0.312 (df = 1729)	0.309 (df = 1727)	0.307 (df = 1726)
F Statistics	14.181 *** (df = 2; 1749)	16.675 *** (df = 4; 1747)	18.747 *** (df = 5; 1729)	19.136 *** (df = 7; 1727)	18.490 *** (df = 8; 1726)

Note: ***, **, and * denote the statistical significance at 1%, 5%, and 10%; standard errors are shown in parentheses.

, also confirm our original findings.

5.3.2. Spurious Correlation

Next, we turn to robustness checks on our original, full sample. Our first robustness check is for a spurious relationship between the explanatory variables of interest, the number of buy-side or sell-side analyst firm i visits in year t, $BU{Y}_{i,t}$, and $SEL{L}_{i,t}$, and the dependent variable, the change in the number of shareholders of firm i in year t. To check for spuriousness in the relationship between these main variables of interest, we re-run the regressions after including the lagged dependent variable, the change in the number of shareholders of a given firm in a given year, by one period, so that Equations (6)–(8) become Equations (9)–(11):

$$\Delta Shareholder{s}_{i,t}={\beta }_0+{\beta }_{1}BU{Y}_{i,t+1}+{\beta }_{2}SEL{L}_{i,t+1}+{\epsilon }_{i,t}$$

(9)

$$\begin{array}{ll}\Delta Shareholder{s}_{i,t }& \\ & ={\beta }_0+{\beta }_{1}BU{Y}_{i,t+1}+{\beta }_{2}SEL{L}_{i,t+1}+{\beta }_{3}EXRETUR{N}_{i,t}\\ & +{\beta }_{4}TO{P}_{i,t+1}+{\beta }_5\Delta FundShare{s}_{i,t+1}+{\epsilon }_{i,t}\end{array}$$

(10)

$$\begin{array}{ll}\Delta Shareholder{s}_{i,t}& \\ & ={\beta }_0+{\beta }_{1}BU{Y}_{i,t+1}+{\beta }_{2}SEL{L}_{i,t+1}+{\beta }_{3}EXRETUR{N}_{i,t+1}\\ & +{\beta }_{4}TO{P}_{i,t+1}+{\beta }_5\Delta FundShare{s}_{i,t+1}\\ & +{\beta }_6\left(EXRETUR{N}_{i,t+1}\ast SEL{L}_{i,t+1}\right)+{\epsilon }_{i,t}\end{array}$$

(11)

The results of a regression of Equations (9)–(11), reported in

Table 8. Robustness Check—Spurious Correlation.

	Dependent Variable
	∆Shareholders
	(1)	(2)	(3)	(4)	(5)
BUYt+1	0.001	0.001	0.002	0.002	0.002
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
SELLt+1	0.0004	0.0004	0.001	0.001	0.001
	(0.001)	(0.001)	(0.002)	(0.002)	(0.002)
EXRETURNt+1			−0.0003 ***	−0.0003 ***	−0.0002 ***
			(0.0001)	(0.0001)	(0.0001)
TOPt+1			−0.002 ***	−0.002 ***	−0.002 ***
			(0.0002)	(0.0002)	(0.0002)
∆FundSharest+1			0.002	0.002	0.002
			(0.002)	(0.002)	(0.002)
EXRETURN*SELLt+1					−0.00001
					(0.00001)
Constant	−0.022 ***	−0.022 ***	0.034 ***	0.034 ***	0.034 ***
	(0.003)	(0.006)	(0.009)	(0.010)	(0.013)
Year Control	N	Y	N	Y	Y
Observations	7656	7656	7612	7612	7612
R2	0.0004	0.001	0.009	0.010	0.010
Adjusted R2	0.0001	0.0002	0.009	0.009	0.009
Residual Std. Error	0.287 (df = 7653)	0.287 (df = 7651)	0.283 (df = 7156)	0.283 (df = 7154)	0.283 (df = 7153)
F Statistics	1.392 (df = 2; 7653)	1.360 (df = 4; 7651)	13.416 *** (df = 5; 7156)	10.315 *** (df = 7; 7154)	9.198 *** (df = 8; 7153)

Note: ***, **, and * denote the statistical significance at 1%, 5%, and 10%; standard errors are shown in parentheses.

, below, show that, as expected, there is no statistically significant relationship between the change in the number of shareholders at firm i in year t and the number of analyst firm visits in year t + 1. This finding should alleviate concerns that our original finding was based on a spurious relationship or due to the autocorrelation of the data.

5.3.3. Measurement Error

Next, to address concerns about a possible measurement error in our dependent variable, the change in the number of shareholders, we explore the same phenomenon using a different variable: the change in the average number of shares held by each shareholder, ∆AvgShares. Equations (6)–(8) thus become Equations (12)–(14) below:

$$\Delta AvgShare{s}_{i,t}={\beta }_0+{\beta }_{1}BU{Y}_{i,t}+{\beta }_{2}SEL{L}_{i,t}+{\epsilon }_{i,t}$$

(12)

$$\begin{array}{ll}\Delta AvgShare{s}_{i,t}& \\ & ={\beta }_0+{\beta }_{1}BU{Y}_{i,t}+{\beta }_{2}SEL{L}_{i,t}+{\beta }_{3}EXRETUR{N}_{i,t}+{\beta }_{4}TO{P}_{i,t}\\ & +{\beta }_5\Delta FundShare{s}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(13)

$$\begin{array}{ll}\Delta AvgShare{s}_{i,t}& \\ & ={\beta }_0+{\beta }_{1}BU{Y}_{i,t}+{\beta }_{2}SEL{L}_{i,t}+{\beta }_{3}EXRETUR{N}_{i,t}+{\beta }_{4}TO{P}_{i,t}\\ & +{\beta }_5\Delta FundShare{s}_{i,t}+{\beta }_6\left(EXRETUR{N}_{i,t}\ast SEL{L}_{i,t}\right)+{\epsilon }_{i,t}\end{array}$$

(14)

The average number of shares held by shareholders is an investigation into the concentration of the shareholder ownership of firms, rather than a measure of the dispersion of ownership as in our original measure, the change in the total number of shareholders of the firm. If our original hypothesis that investors free ride on increased monitoring of firm management after a firm visit by a buy-side analyst, Hypothesis 2a, is true, then with our new measure of concentrated ownership, we would expect the coefficient estimate on the number of annual visits by buy-side analysts, ${\beta }_1$, to be the opposite sign, negative. If the conflict of interest inherent in the investment banking business model leads investors to mistrust sell-side analysts, then Hypothesis 2b posits that the coefficient estimate on the number of annual visits by sell-side analysts, ${\beta }_2$, will be statistically insignificantly different from zero or even the opposite sign, positive.

The results of a regression of Equations (12)–(14), reported in

Table 9. Robustness Check—Measurement Error.

	Dependent Variable
	∆AvgShares
	(1)	(2)	(3)	(4)	(5)
BUY	−0.011 ***	-0.006 ***	−0.009 ***	−0.004	−0.005 ***
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
SELL	0.004 ***	0.001	0.003 ***	0.001	0.001
	(0.001)	(0.001)	(0.001)	(0.002)	(0.002)
EXRETURN			−0.001 ***	−0.001 ***	−0.0002 ***
			(0.0002)	(0.0002)	(0.0001)
TOP			0.004 ***	0.004 ***	0.004 ***
			(0.0003)	(0.0003)	(0.0003)
∆FundShares			0.033 ***	0.033 ***	0.033 ***
			(0.002)	(0.002)	(0.002)
EXRETURN*SELL					0.00005 ***
					(0.00001)
Constant	0.156 ***	0.219 ***	0.007	0.074 ***	0.073 ***
	(0.005)	(0.009)	(0.013)	(0.015)	(0.015)
Year Control	N	Y	N	Y	Y
Observations	7668	7668	7174	7174	7174
R2	0.003	0.012	0.053	0.063	0.064
Adjusted R2	0.003	0.011	0.052	0.062	0.063
Residual Std. Error	0.436 (df = 7665)	0.435 (df = 7663)	0.427 (df = 7168)	0.425 (df = 7166)	0.425 (df = 7165)
F Statistics	11.250 *** (df = 2; 7665)	22.836 *** (df = 4; 7663)	80.167 *** (df = 5; 7168)	68.733 *** (df = 7; 7166)	61.217 *** (df = 8; 7165)

Note: ***, **, and * denote the statistical significance at 1%, 5%, and 10%; standard errors are shown in parentheses.

, below, confirm Hypothesis 2a and 2b, even using this alternative measure. Across all specifications, we see that a higher number of firm visits by buy-side analysts has a statistically significantly negative impact on the average number of shares held by shareholders, while a higher number of firm visits by sell-side analysts has a statistically significantly positive impact on the average number of shares held by shareholders. Thus, the results using this new measure of shareholder concentration confirm our original findings.

5.3.4. Endogeneity

We are not overly concerned about issues with endogeneity in our main results. In the count model analysis, the dependent variable, firm visits by analysts, cannot influence the market power of the firms, the main independent variable of interest. It is possible that a firm visit by a financial analyst might trigger changes in the firm’s own accounting and therefore affect its reported earnings. To alleviate concerns about potential endogeneity in our analysis of the effect of the earnings per share on analyst visits to the firm, we use the one-year lagged earnings as the independent variable in Equations (2)–(5). In the panel data regression analysis of the effect of analyst firm visits on the change in the number of firm shareholders, Equations (6)–(8), the nature of the data disclosure naturally addresses concerns about possible endogeneity. Analyst firm visits are disclosed in a timely manner, but the change in shareholders is only reported at the end of the year. Therefore, it is possible for shareholders or potential shareholders to observe analyst firm visits, but at the time that they perform firm visits, there is no way the visiting analysts could know about any changes in shareholder statistics.

Although we are not very concerned that buy-side or sell-side analyst visits are endogenous to changes in the number of shareholders over the year, as a robustness check to alleviate concerns about potential correlation between the explanatory variable in our main finding, the number of buy-side analyst firm visits in year t, $BU{Y}_{i,t}$, with the error term, ${\epsilon }_{i,t}$, in Equations (6)–(8), we conduct an instrumental variable estimation.

The ideal instrumental variable will be correlated with the explanatory variable of interest, in this case, the number of on-site corporate visits made to firm i in year t, but will affect the outcome variable, in this case the change in the number of shareholders of firm i in year t, only through the explanatory variable, so that it is uncorrelated with the error term ${\epsilon }_{i,t}$.

We use number of institutional investors investing in firm i in the following year, year t + 1, as an instrument for the number of on-side firm visits from buy-side analysts in year t. The first column of

Table 10. Robustness Check—Endogeneity—2SLS Instrumental Variable Estimation.

	Dependent Variable
	Inst Hold	∆Shareholders
	OLS	Instrumental Variable
	(1)	(2)	(3)	(4)	(5)	(6)
BUY	12.041 ***	0.125 ***	0.136 ***	0.146 ***	0.172 ***	0.170 ***
	(0.661)	(0.013)	(0.016)	(0.018)	(0.025)	(0.024)
SELL		−0.034 ***	−0.038 **	−0.039 ***	−0.048 ***	−0.049 ***
		(0.004)	(0.004)	(0.005)	(0.007)	(0.007)
EXRETURN				−0.0005 ***	−0.001 ***	0.0002
				(0.0002)	(0.0002)	(0.0002)
TOP				−0.001 ***	−0.001 ***	−0.001 ***
				(0.0003)	(0.0004)	(0.0003)
∆FundShares				−0.019 ***	−0.021 ***	−0.021 ***
				(0.002)	(0.003)	(0.003)
EXRETURN *SELL						−0.0003 ***
						(0.00004)
Constant	67.124 ***	−0.015 ***	0.010	0.144 ***	0.085 ***	0.088 ***
	(3.476)	(0.005)	(0.016)	(0.019)	(0.023)	(0.023)
Year Control	Y	N	Y	N	Y	Y
Observations	7668	7656	7656	7162	7162	7162

Note: ***, **, and * denote the statistical significance at 1%, 5%, and 10%; standard errors are shown in parentheses.

reports our statistical check that the first condition is met. As reported in Column 1 of Table 10, first-stage coefficient estimates on the instrument are highly statistically significant, with a p-value less than 0.01. Given the findings reported above, which indicate that buy-side analysts generally conduct an on-site firm visit when they are considering buying stocks (and not when they are considering selling), we are not surprised by this statistical relationship. Since institutional investors only make up a small fraction of the total number of investors in China’s financial markets, we are assured that the instrumental variable, the number of institutional investors in firm i in year t + 1, is uncorrelated with the error term, ${\epsilon }_{i,t}$, in Equations (6)–(8), thereby meeting the second criterion as well.

The result of the two-stage least-squares estimation using the number of institutional investors in firm i in year t + 1 as an instrumental variable for the number of on-site firm visits to firm i by buy-side analysts in year t are reported below in Table 11. The two-stage least-squares instrumental variable estimation results again support our original findings. Across all specifications, we see that a higher number of firm visits by buy-side analysts has a statistically significantly positive impact on the number of shareholders of the firms visited, but a higher number of firm visits by sell-side analysts has a statistically significantly negative impact on the number of shareholders of the firms visited. Thus, the instrumental variable regression results again support Hypotheses H2a and H2b.

5.4. Summary of Findings

Table 11 summarizes the empirical results, whether they validate or refute the stated hypotheses, and discusses our interpretation of the results.

Table 11. Summary of Findings.

Hypotheses	Validation	Discussion
H1. Firm revenue as a share of industry revenue and buy-side analyst firm visits	Supported	Larger firm revenue as a share of industry revenue attracts more buy-side analyst firm visits.
H2a. Buy-side analyst firm visits and the change in the number of investors	Supported	Additional buy-side analyst firm visits increase the number of shareholders.
H2b. Sell-side analyst firm visits and the change in the number of investors	Supported	Additional sell-side analyst firm visits decrease the number of shareholders.

Table 11. Summary of Findings.

Hypotheses	Validation	Discussion
H1. Firm revenue as a share of industry revenue and buy-side analyst firm visits	Supported	Larger firm revenue as a share of industry revenue attracts more buy-side analyst firm visits.
H2a. Buy-side analyst firm visits and the change in the number of investors	Supported	Additional buy-side analyst firm visits increase the number of shareholders.
H2b. Sell-side analyst firm visits and the change in the number of investors	Supported	Additional sell-side analyst firm visits decrease the number of shareholders.

The current literature demonstrates that firm characteristics influence analysts’ decisions about whether to conduct on-site visits to those firms. Our binomial count analysis of the factors that determine the number of buy-side and sell-side analyst visits contributes to the literature by identifying significant heterogeneity in on-site firm visits across buy-side and sell-side analysts. Our regression analysis results on how analyst visits affect the number of investors in those firms are consistent with the theory that sell-side analysts face significant conflicts of interest and that these agency issues lead investors to mistrust sell-side analyst signals.

6. Conclusions and Directions for Future Research

This study is one of the few empirical investigations into the differences between buy-side and sell-side market analysis. Using a recently established dataset on firm visits by financial analysts, this research uses a negative binomial count model to examine the factors that lead to buy-side and sell-side analyst corporate site visits. The regression analysis is then used to analyze investor response to those corporate site visits.

Examining the factors that lead to buy-side and sell-side analyst firm visits in the Chinese financial market, we find evidence that buy-side and sell-side analysts have different incentives for firm visits. There is evidence that both buy-side and sell-side analyst firm visits are motivated by an interest in purchasing or recommending the purchase of stock rather than selling or recommending the sale of stock. However, buy-side analysts, whose sole purpose is to look for valuable investment opportunities, focus their firm visits on industry leaders: firms with a high share of revenue within their given industry. Sell-side analysts are not more likely to visit industry leaders.

Next, we explore investor reactions to firm visits by buy-side and sell-side analysts. We find evidence that investors have confidence in buy-side analysts but give little credence to sell-side analysts. Firm visits by buy-side analysts attract new investors, whereas the firm visits by sell-side analysts prompt shareholders to wind down their holdings. This phenomenon is exacerbated when firms outperform the market.

The results suggest that market players have serious reservations about the conflicts of interest inherent in the investment banking business model, which misaligns sell-side analysts’ incentives from the investors they advise. In China, as in other developed financial markets, there is a legal firewall separating sell-side analyses on financial products from investment banking business activities, but this system seems insufficient to assuage investors. Other systems are needed to ensure impartial financial analyses. For example, if the compensation of sell-side financial analyses was based on research quality, analyst incentives would be more closely aligned with that of investors. If financial analyses were published anonymously, similar to academic journals’ double-blind review system, sell-side financial analysts would not face negative repercussions from issuing a negative recommendation for firms with which they have established a good working relationship. Continuing education on financial ethics in the financial analysis industry would also help build investor confidence in such analyses.

Due to limitations in data availability, the research results presented here may include confounding effects from other events. If changes in shareholders are disclosed more frequently, future research could isolate and control for these confounding effects. Additionally, finding a qualified instrument variable would enable further testing of the robustness of the results reported here. As more data are added to the source used in this study, one extension of the research presented here would be to investigate whether the analyst recommendation quality increases with the analysts’ firm coverage period. The longer a given analyst has followed a given firm, the more firm visits we may expect to have been performed and therefore the more “soft information” about the firm the analyst has presumably gathered. Such “soft information” should result in higher-quality research and more accurate forecasts. Herding effects are another potentially interesting area for future research. Given the wide dissemination of sell-side analyses, there may be incentives to follow investment recommendations from industry analyst leaders, which would be consistent with theories of herd behavior (Scharfstein and Stein 1990; Healy and Palepu 2003).