Do analysts cater to investor beliefs via target prices

International Review of Economics and Finance 44 (2016) 232–252

Contents lists available at ScienceDirect

International Review of Economics and Finance journal homepage: www.elsevier.com/locate/iref

Do analysts cater to investor beliefs via target prices An-Sing Chen a, Chong-Chuo Chang b, Lee-Young Cheng a, Hsing-Yu Tu a a b

Department of Finance, National Chung Cheng University, 168 University Rd., Min-Hsiung, Chia-Yi 62102, Taiwan Department of Finance, Asia University, 500, Lioufeng Rd, Wufeng, Taichung 41354, Taiwan, ROC

a r t i c l e

i n f o

Article history: Received 1 April 2015 Received in revised form 29 March 2016 Accepted 18 April 2016 Available online 21 April 2016 JEL classification: G11 G14 G17

a b s t r a c t This study examines analyst target price bias within the framework of catering theory. Given that analyst catering is more probable when the clients of the analyst forecasts are less sophisticated investors who are less likely detect it, we focus on a unique stock market where individual investors rather than institutions are the predominant group and account for 75% of the total security trading. Our results show that analysts do cater to investors via overshooting actual end-of-forecast-period prices even after controlling market sentiment index, analyst and company characteristics. Furthermore, results show that foreign analysts produced more biased target prices compared to domestic peers. © 2016 Elsevier Inc. All rights reserved.

Keywords: Target prices Investor sentiment Catering theory

1. Introduction Previous research shows that analysts' target prices are consistently biased. Asquith, Mikhail, and Au (2005) report that approximately 54% of analysts' target prices are achieved or exceeded during the year following the publication of an analyst report. Bradshaw, Brown, and Huang (2013) find that 38% of target price forecasts are met at the end of the 12-month period and 64% are met sometime during the period. They attribute their results to the absence of incentives for analysts to make more accurate target price forecasts than earnings forecasts. Analyst earnings forecast accuracy is monitored by the market and it affects analyst compensation. In contrast, target price accuracy is neither subjected to market scrutiny nor is it related to compensation. Bonini, Zanetti, Bianchini, and Salvi (2010) argue that less informed investors tend to incorporate analyst target prices in their investment strategies more frequently than informed investors. If this is so, analysts can transfer their own risk to less informed investors by delivering biased target price on purpose. Early studies have focused on analysts' earnings forecasts accuracy and documented that the sources of analyst earnings forecast biases are caused by two incentives. One of them is the distorted incentives of analysts. Analysts face conflicting tradegenerating incentives and must trade off the short term incentive to optimistically bias forecasts and recommendations with the long term incentive to build their own reputation through accurate forecasts and recommendations (Irvine, 2004 and Jackson, 2005). Hong and Kubik (2003) argue that, since optimistic forecasts relative to the consensus increase the chances of favorable job performance evaluation, analyst forecast optimism bias is likely due to incentives to promote stocks. Cowen, Groysberg, and Healy (2006) find that analyst optimism is, at least, partially driven by trading incentives. The other source is the distorted incentives of managers in companies covered by analysts. Mittendorf and Zhang (2005) demonstrate that optimal incentive contracts between analysts and companies they cover leads to biased earnings guidance and E-mail addresses: fi[email protected] (A.-S. Chen), [email protected] (C.-C. Chang), fi[email protected] (L.-Y. Cheng), [email protected] (H.-Y. Tu).

http://dx.doi.org/10.1016/j.iref.2016.04.005 1059-0560/© 2016 Elsevier Inc. All rights reserved.

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

233

biased analyst forecasts. Baik and Jiang (2006) find that the likelihood of management guidance increases in the optimism of analysts' consensus forecasts before the announcement. Assessed as a whole, the arguments in these papers can be described as pertaining to “supply side” sources of analyst earnings forecast bias. Understanding the use of earnings forecast slant as a measure of analyst performance is important because earnings forecast is often considered as a crucial factor when evaluating corporations. However, understanding analysts' target prices forecasting accuracy is also important for several reasons. First, given that target prices provide an up-front estimate of the potential value of a stock, they may have an influence on investors' investment strategies. Therefore, how accurate these analyst target prices are in predicting future stock prices is important to investors. Second, under the current structure of the investment industry, there is no explicit control on the quality of target prices being enforced by the consumers (investors) of these target prices. This means that analysts may have an incentive to provide biased target prices to less informed investors, since doing so can produce a riskshifting effect favorable to the provider of the target prices whose company may also have a position in the stock. If consumers of target prices have a better picture of the accuracy or potential biases in analyst target prices, then they can be in a better position to develop strategies to arbitrage away these inefficiencies. Third, understanding analysts' forecasting accuracy of target prices helps to shed light on the question of whether target prices are incrementally informative since the accuracy of target prices may affect the value investors assign to price targets. Fourth, compared to earnings forecasts which cover only quarter by quarter periods and stock recommendations which are discrete (e.g., buy, hold, sell), target price forecasts offer several advantages. They are continuous in nature, have well-defined forecast horizons, and exhibit higher frequency of revision than earnings forecasts. Understanding the forecast accuracy of analysts' target prices is, thus, important for investors and researchers as target prices are potentially useful inputs for firm valuation and indicators of future returns. Finally, by combining analysts' target prices with market prices and forming a ratio of target-to-market prices, a proxy for ex ante expected return can be constructed. Understanding the accuracy of analysts target prices is therefore also important for asset pricing research which often requires a measure of ex ante expected return, since the target price/market price ratio can serve as a check on more commonly used measures of ex ante expected returns generated from asset pricing models such as the CAPM and Fama–French type multi-factor asset price models. Target prices provide market participants with analysts' most clear and precise statements on the magnitude of the company's expected value (Brav & Lehavy, 2003). Target prices comprise a straightforward measure of the potential change in value of the underlying stock that may have an influence on investors' investment strategies. However, Bradshaw et al. (2013) point out that unlike earnings forecasts, target price forecasts are subject to no market examination and surveillance.1 Recently, catering theory argues that analysts' forecast bias might be a result of their catering to “demand side” influence, that is, their clients—the readers of analysts' reports—who are the purported victims of analyst bias (Lai, 2005). Lai (2005) shows that analysts cater to what investors believe when forecasting EPS. Hribar and McInnis (2012) find that when investor sentiment is high, analysts' forecasts of one-year-ahead earnings and long-term earnings growth are relatively more optimistic for certain types of companies. Therefore, analysts' forecast bias might be a result of their catering to “demand side” influence, that is, the readers of analysts' reports who are the purported victims of analyst bias. Bilinski, Cumming, Hass, Stathopoulos, and Walker (2015) discuss incentives which may cause analysts to cater to investor beliefs via target price. They argue that analysts working for brokers who are not affiliated with investment banks have incentives to cater to investor beliefs via target price rather than earnings estimates because the marginal cost of reputation loss from issuing biased target prices is lower for these analysts compared to issuing biased earnings estimates. These brokers, in turn, benefit from issuing biased target prices since retail investors on average fail to see the bias and short-term institutional investors can, thus, take advantage of temporary stock price increases caused by the biased target prices and sell their shares to the retail investors. These same short-term institutional investors can then reward the brokers with analysts engaging in catering via higher future trades channeled through the broker. Bilinski et al. (2015) find empirical evidence supporting this incentive mechanism. Their results show that for stocks with high short-term institutional ownership, analysts strategically bias their target prices, but not their earnings estimates. They also find that the bias in the target prices, when there are short-term institutional investors, concentrates mainly among analysts working for brokers that are not affiliated with investment banks. Also, they find evidence consistent with short-term investors rewarding brokers engaging in catering with higher future trades channeled through the broker. To explore the source of analyst target price bias, this study investigates whether or not analysts cater to investor beliefs via target prices. Target prices are forecast of future prices that analyst expects will gravitate to fundamentals in the future. But, if the analyst expects prices to deviate from fundamental value, for example, because of speculative fever sweeping the market, he may issue a target price that differs from his target value. This consideration underscores an important point in applying investor sentiment to target price bias. We use analysts' report data from the Taiwan stock market to analyze analyst behavior and identify whether or not biased target price is caused by investor beliefs. The Taiwan stock market (TWSE) provides a unique environment to examine analyst target price bias. Compared to more developed markets such as the United States where institutional investors are the main players, individual investors are the predominant group in the Taiwan stock market in that they account for 75% of the total security trading. Individual investors being less “sophisticated” investors as a whole, tend to rely more on target prices rather than EPS forecasts, since they may not have enough computational sophistication to be able to translate an EPS forecast into an appropriate target price suitable for their investment purposes. Moreover, the TWSE is known as one of the most active exchanges in the

1

See Stickel (1990, 1992, 1995); Cooper, Day, and Lewis (2001), and Bernhardt, Campello, and Kutsoati (2004).

234

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

Asia Pacific region, price target recommendations from foreign financial institutions and foreign securities analysts play a major part in driving stock market swings in Taiwan. Therefore, the use of this stock market for analyzing analyst target price bias and catering should yield several unique insights that may be difficult to extract using standard datasets. In this study, we adopt different proxies to quantify investor beliefs. Following Kumar and Lee (2006), we calculate buy–sell imbalance (BSI) for each company covered by analysts as measures of investor beliefs for future growth (decline) of underlying share prices. Kumar and Lee (2006) show that trades of individual investors are systematically correlated. This phenomenon predicts that individuals tend to buy (or sell) stocks in conjunction with each other. If a group of individuals want to buy shares of a specific corporation synchronously, the buy-sell trade volume (or shares) of individual investors would be imbalanced. If the imbalance led by these investors continues long enough, it would become a signal indicating that these investors believe share price of the company is going to rise. Analysts would notice and follow the signal of investor beliefs then issue outright optimistic price targets for that company. The targets would therefore systematically overshoot the maximum market prices. The more intense the signal, the more bias the target prices issued by analysts. On the other hand, if a group of investors begin offering sell orders for a period of time, analysts would notice that signal and issue more pessimistic price targets. The targets would naturally overshoot the minimum market prices. We argue that strong investor beliefs will lead to target price bias. That is, when investors believe the price of underlying share is going to increase (or decrease), analysts will follow investor beliefs and report an outright optimistic (or pessimistic) target price. Our results show that analysts do cater to investors via overshooting target prices when investor beliefs are proxied in the company level. This effect remains even after controlling market sentiment index, analyst and company characteristics. In addition, previous studies on international analysts mainly test for accuracy differences between local and foreign analysts with respect to their earnings forecasts or buy-sell recommendations. The empirical findings are mixed. Orpurt (2006), for example, finds that local analysts issue more accurate earnings forecasts than other analysts in Europe. Sonney (2009) reports that country-specialized analysts produce more accurate EPS forecasts. Conversely, Bacmann and Bolliger (2002) show that foreign analyst earnings forecasts outperform local analyst forecasts in Latin American countries. Chang (2006) reports that in Taiwan, foreign analysts with local presence issue upgrade or downgrade recommendations that outperform those issued by local and other foreign analysts. Lai and Teo (2008) find that in Asian emerging markets, domestic analyst upgrade or downgrade recommendations are consistently more optimistic than those of foreign analysts. In Taiwan, most retail investors transact at domestic brokerage firms and investment banks. If domestic analysts know their clients' demand better than foreign analysts do and are under greater pressure to generate commissions, it is reasonable to extrapolate that domestic analysts may cater to investors more severely than their foreign peers do. Specifically, domestic analysts will issue a more optimistic (pessimistic) target price than foreign analysts when investors believe the price of underlying shares is going to rise (decline). Our results show that this is not the case. For the Taiwan stock market, foreign analysts produced more biased target prices compared to domestic peers, which is not consistent with Lai and Teo's (2008) findings that domestic analyst recommendations are consistently more optimistic than those of foreign analysts. In contrast, the results provide support for Bradshaw et al.'s (2013) notion that unlike earnings forecasts, target price forecasts are subject to no market examination and surveillance. Our results also suggest the possibility that foreign analysts can cater more because they have clients outside Taiwan who may be less knowledgeable concerning the peculiarities of the Taiwan stock market and thus easier for these analysts to cater according to their client's demand. Investigating the determinants of analysts' target price bias has received limited attention in academic research. Our paper's main contribution to the literature is twofold. First, prior research finds analysts' target prices are consistently biased. However, the relationship between investor beliefs and target price bias has not yet been analyzed in the literature. Using the market sentiment index, prior studies show that managers cater to investors through several channels such as dividends (Baker & Wurgler, 2004a, 2004b), and stock price (Baker, Greenwood, & Wurgler, 2009). However, since sentiment for the whole market cannot tell us whether or not an analyst caters to investors of an underlying company covered by the analyst's report, we extend the theory by treating investor beliefs not only in the market level, but in the company level, as a proxy of investor sentiment. Second, this study also contributes with respect to target price accuracy of foreign versus domestic analysts. Stock recommendations have been found to be particularly valuable to investors in developed countries (Boni & Womack, 2006 and Jegadeesh & Kim, 2006). We provide a better understanding of comparative analyses of stock analyst in emerging stock market. The remainder of the paper is structured as follows: Section 1 reviews the literature; Section 2 describes the data collections; Section 3 presents empirical results; Section 4 concludes the paper.

1.1. Related literature Most analysts' reports highlight three key summary measures: short-term forecasts of earnings, the target price, and a buy/sell stock recommendation. However, analysts' target prices have received limited attention in academic research probably because it was only after the mid-1990s that equity analysts began to disclose price targets in their published stock research reports. Investors perceive analyst price targets as informative signals regarding a firm's value. Brav and Lehavy (2003), using an extensive database of price targets, find a significant market reaction to price targets. Moreover, the reaction was robust both unconditionally, and conditional on simultaneous recommendation and earnings forecast revisions. Asquith et al. (2005) find significant effects on stock prices due to target prices not attributable to earnings forecasts and stock recommendations, providing evidence that investors consider target price forecasts to be valuable.

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

235

However, prior research indicates that analysts' estimates of target price vary in accuracy and often display optimistic bias. In general, only about 50% of analysts' price targets are actually achieved or exceeded in 12 months—the most common horizon specified by analysts (Asquith et al., 2005; Bradshaw et al., 2013). Bradshaw et al. (2013) find no evidence that analysts have persistent ability to forecast accurate target prices. Given that target prices are not subjected to media scrutiny, and there is no evidence that analyst compensation is tied to their target price forecast, Bradshaw et al. (2013) attribute the unequal abilities of earnings forecasts and target prices to the fact that analysts have few incentives to set accurate price targets. Bonini et al. (2010) find that the accuracy of target price is relatively limited and that prediction errors are consistent, auto-correlated, nonmean reverting and large in the Italian stock market. Bradshaw (2002) suggests that analysts sometimes concoct price targets to justify ex-post their buy–sell recommendations. Asquith et al. (2005) argue that analysts are likely to issue highly favorable recommendations due to concerns over relationships with the analyzed firms' management, or their own firms underwriting business. Lim (2001) indicates that analysts may be more willing to issue biased forecasts to please managers from large firms. Using a sample of 16 countries, Bilinski, Lyssimachou, and Walker (2013) find that better target price forecasters tend to be analysts with higher forecasting experience. Also, analysts who follow more firms, are country-specialized and employed by a large broker issue more accurate target prices. In the German stock market, Kerl (2011) finds that target price accuracy correlates negatively with the size of the target price and the firmspecific risk. Concerning the derivation of target prices, Asquith et al. (2005) report that the majority of analysts use simple heuristic and do not exhibit significantly ability in forecasting future stock prices. Demirakos, Strong, and Walker (2010) find that target prices derived from discounted cash flow valuation models are relatively more accurate than produced from price-to-earnings multiples. Gleason, Johnson, and Li (2013) find the superior target-price investment performance when analysts use proper valuation techniques rather than simple heuristics. Recent studies also provide links between analysts' forecasts and market sentiment. Baker and Wurgler (2004a, 2004b) develop a catering theory, in which the decision to pay dividends is driven by investor demand. They argue that managers will tend to initiate dividends when investors put a stock price premium on dividend payers, and tend to omit dividends when investors prefer non-payers. Since Baker and Wurgler's (2004a, 2004b) investigations on the relationship between manager decision and investor demand, other scholars began to study the relation between investor demand and behavior of other market participants. For instance, Rajgopal, Shivakumar, and Simpson (2007) find that catering incentive plays an important role in explaining earnings management. They argue that managers cater to investor sentiment through managing earnings. Greater earnings optimism in the market increases managers' incentives to inflate earnings in order to temporarily avoid stock price drops. On the other hand, greater earnings pessimism in the market increases managers' incentives to decrease earnings partly to avoid getting prosecuted for overstating earnings. Baker et al. (2009) find strong empirical support for the catering predictions that managers maintain nominal share prices toward the price range that investors currently prefer. Ertimur, Sunder, and Sunder (2007) provide empirical evidence that analysts initiate recommendation following high returns and prior growth. Ke and Yu (2009) show that the translation of analysts' forecasts into profitable stock recommendations is adversely affected by periods of extreme investor sentiment and a high reliance on trading commissions. Lai (2005) and Hribar and McInnis (2012) suggest that analysts cater to investor sentiment through earnings forecast announcements. Hribar and McInnis (2012) examine whether or not time-series fluctuation in security analysts' earnings forecast errors aligns with time-series fluctuation in sentiment. They find that sentiment affects analyst expectations across all stocks. Specifically, earnings forecasts become relatively more optimistic across the board when sentiment is high, and less optimistic when sentiment is low. Lai (2005) finds that analysts are influenced by the beliefs of investors as well. Bagnoli, Clement, Crawley, and Watts (2009) argue that some analysts appear to recommend a stock based on “investor sentiment”, but not based on “fundamentals,” such as earnings, cash flows, and discount rates. They find that analysts whose stock recommendations are positively correlated with recent or future investor sentiments tend to issue relatively less profitable recommendations. Clarkson, Nekrasov, Simon, and Tutticci (2015) indicate that analysts not only use forecasts of firm fundamentals but appeal to the recent high stock price and market sentiment when determining their target price forecasts. They show that the roles of the non-fundamental factors in target price bias are greater when they place a greater role in the target price formation process. Brown, Christensen, Elliott, and Mergenthaler (2009) argue that managers tend to disclose more favorable pro forma earnings metrics when investors hold optimistic beliefs about future firm performance. When investor sentiment increases, managers will exclude higher levels of expenses in calculating the pro forma earnings and will put the pro forma figures more prominently within the earnings press release. Using the BW (Baker & Wurgler, 2006) investor sentiment index, they show that managers' catering behavior is more pronounced for companies whose stock valuations are sensitive to market sentiment. However, since the BW index varies by a particular period of time not by individual stock, Kumar and Lee (2006) form a BSI to measure changes in retail sentiment for a certain basket of stocks. Based on over 1.85 million transactions made by retail investors from 1991 to 1996, Kumar and Lee (2006) show two related phenomena. One of them is a strong positive correlation in their BSI across non-overlapping portfolios of different stocks. This finding indicates that when one portfolio of stocks is bought (sold) by individual investors, other non-overlapping portfolios tend to be bought (sold) as well. Another phenomenon is that correlated trading behavior holds across different individuals. This finding means that when one group of investors buy (sell) stocks, a different group of retail investors also tend to buy (sell) stocks. Based on these findings, there exists a systematic component in the trading activities of investors—i.e., individuals are likely to buy (sell) stocks in concert with each other.

236

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

2. Data and methodology 2.1. Data In this study, the target price data are collected directly from cnYES. Com. cnYES is currently the most trusted and consulted financial website used by Chinese finance professionals. The sample period spans from September 1st, 2003 through July 31th, 2008. During our sample period of 2003–2008, there were 720 stocks listed on TWSE. Our sample contains 5184 analystcompany observations across 421 companies. The market value of sample firms accounts for 66.51% of the total market value. IPO data and other financial data are collected from the Taiwan Economics Journal (TEJ) database. Table 1 provides a summary of the distribution of companies, industries, and month and year when reports are announced. Panel A of Table 1 indicates that 2236 reports of 192 firms are issued by foreign analysts and 2948 reports of 412 firms are issued by domestic analysts from September 1st, 2003 through July 31th, 2008. Panel B of Table 1 shows summary statistics for reports distribution at the industry level. Consistent with the overall distribution of the general population of firms and trading volume represented in the Taiwan Stock Exchange (TWSE), our sample firms is most concentrated in the electronics industry with

Table 1 Distribution of analyst targets. This table reports aggregate, year-level, and industry-level target price statistics. There are 5184 analyst reports in our sample across eight industries during the 2003–08 sample period. Panel A reports numbers of firms covered by analysts, numbers of analysts who announced reports in each year, and numbers of targets announced. Panel B reports numbers of targets in various industries. Panel C reports yearly and monthly distribution of target price for the period Sep 2003 to Jul 2008. Panel A: Distribution of sample firms across analysts Foreign analysts

Domestic analysts

Total

Annual period #Firms covered

#Targets

#Firms covered

#Targets

#Firms covered

#Targets

2003 (Sep.–Dec.) 2004 2005 2006 2007 2008 (Jan.–Jul.)

59 85 104 81 121 97

216 403 420 250 542 405

130 253 211 181 206 133

258 891 631 474 444 250

150 264 236 203 243 178

474 1294 1051 724 986 655

All periods

192

2236

412

2948

421

5184

Panel B: Distribution of sample firms across industries Firms Covered

Total Targets

Industry Electronics Industry Plastic, Rubber, and Chemical Industry Financial & Insurance Automobile and Iron & Steel Electric Machinery Cement, Building Material & Construction Food and Textile Transportation Other Industry

N. 216 29 29 19 26 22 22 12 46

% 51.31% 6.89% 6.89% 4.51% 6.18% 5.23% 5.23% 2.85% 10.93%

N. 3411 297 415 200 145 192 92 180 252

% 65.80% 5.73% 8.01% 3.86% 2.80% 3.70% 1.77% 3.47% 4.86%

Total

421

100.00%

5184

100.00%

Panel C: Yearly and monthly report distribution Month January February March April May June July August September October November December

F – – – – – – – – 84 88 27 17

2003 D – – – – – – – – 65 70 72 51

T – – – – – – – – 149 158 99 68

F 24 41 51 65 32 28 39 38 19 37 22 7

2004 D 48 82 102 138 85 33 40 74 103 105 47 34

T 72 123 153 203 117 61 79 112 122 142 69 41

F 10 3 3 13 7 7 75 107 70 59 44 22

2005 D 40 41 49 63 55 48 49 67 55 58 41 65

T 50 44 52 76 62 55 124 174 125 117 85 87

F 41 51 11 7 7 18 26 39 27 17 5 1

2006 D 45 45 33 65 60 39 34 40 22 22 31 38

T 86 96 44 72 67 57 60 79 49 39 36 39

F 17 1 12 15 30 45 57 103 52 106 68 36

2007 D 38 27 37 36 42 30 41 71 44 35 21 22

T 55 28 49 51 72 75 98 174 96 141 89 58

F 67 33 46 56 42 19 65 77 – – – –

2008 D 26 12 35 45 32 31 23 46 – – – –

T 93 45 81 101 74 50 88 123 – – – –

F 159 129 123 156 118 117 262 364 252 307 166 83

Total D 197 207 256 347 274 181 187 298 289 290 212 210

T 356 336 379 503 392 298 449 662 541 597 378 293

Total

216

258

474

403

891

1294

420

631

1051

250

474

724

542

444

986

405

250

655

2236

2948

5184

Note: “F” in the second row denotes number of targets announced by foreign analysts, “D” denotes targets reported by domestic analysts, and “T” denotes all targets announced.

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

237

216 firms (51.31%) and 3411 reports (65.80%). Panel C of Table 1 provides evidence on yearly and monthly report distribution. Within each year, three accumulation points exist around the months of April, August, September and October which typically host the major corporate event such as shareholder's meeting, dividend announcement, or capital expenditure for future fiscal year. This pattern is consistent with the notion that analysts usually update their target prices when new information arrives. 2.2. Measurement of target price bias The ratio of target price to actual trading price (TP/P) is broadly used when evaluating performance of analysts' price targets (Asquith et al., 2005; Bradshaw et al., 2013). Asquith et al. (2005) examine accuracy by a simple method which considers accurate a target if the underlying share price reaches or exceeds the target at the end of time horizon. Bradshaw et al. (2013) extend the analysis by testing whether the price is met also at any time during the report time horizon. However, these variables cannot evaluate the intensity of analyst target price biases. Therefore, to quantify the intensity of analyst target price bias, we follow Bonini et al. (2010) by adopting two proxies to measure target price bias. One of those proxies is AB_ANY, which measures the “Ideal Strategy (IS)” bias for target price. IS assumes investors can earn maximum return during the analyst prediction period if investors follow analysts' announcements of target prices strategically. This means that after buying shares on the announcement date of target price, investors can sell the underlying shares at the highest price in the analyst prediction period (or after short selling the underlying shares on the announcement date, investors can buy back shares at the cheapest price in the prediction period). Therefore, this measure can quantify target price prediction error when investors can follow IS to earn the maximum return in the prediction period. Computationally, we first calculate the difference between the issued target price and the maximum market price (or minimum market price if the target price is smaller than the market price at announcement date) in the relevant prediction time-horizon. Then, we divide the difference by the maximum market price (or minimum price), which is termed AB_ANY. Another proxy is AB_END, which measures the “Feasible Strategy (FS)” bias for target price. FS assumes that investors buy and hold underlying shares from the announcement date to the end of the prediction period (or short sell underlying stocks at the announcement date and buyback at the end of the subsequent period). Thus, this measure can quantify analyst target price prediction error when investors can follow FS to earn the holding period return in the prediction period. The difference between the issued target price and the stock market price at the end of the relevant prediction time-horizon is divided by the market price at the end of the relevant period, which is termed the AB_END. Both proxies express ex-post analyst target price bias compared with stock market price and show how much analyst target price has overshot or undershot in a quantitative way.2 The explicit formulas are:

AB ANY i; j;t ¼

8 max TP i; j;t −P j;t TP i; j;t > > > ¼ max −1; if TP i; j;t NP j;t > max < P j;t P j;t min > P j;t −TP i; j;t TP i; j;t > > ¼ 1− min ; if TP i; j;t bP j;t > : P min P j;t

AB ENDi; j;t

ð1Þ

j;t

8 TP i; j;t −P j;tþn TP i; j;t > > > ¼ −1; if TP i; j;t NP j;t < P j;tþn P j;tþn ¼ P j;tþn −TP i; j;t TP i; j;t > > > ¼ 1− ; if TP i; j;t bP j;t : P j;tþn P j;tþn

ð2Þ

In the above, TPi,j,t is a specific target price for company j at the announcement date t, issued by analyst i and Pj,t is the actual closing stock price of company j at the announcement date t. However, if the announcement date is not a trading day in the stock is the maximum actual closing market, Pj,t would be the actual closing stock price of company j on the last trading day prior t. Pmax j,t min is the minimum acprice of company j during the relevant prediction time horizon starting from the announcement date t. Pj,t tual closing price of company j during the relevant prediction time horizon starting from the announcement date t. The term “t + n” denotes the end of the relevant prediction time-horizon and Pj,t + n is the stock market price of company j at time t + n. When the target price is greater than maximum market price realized during the period, we define a positive difference between TPi,j,t and Pmax as “upside overshooting”. In contrast, a negative difference is considered to be a “conservative” prediction. When the tarj,t min and TPi,j,t get price is smaller than minimum market price realized during the period, we define a positive difference between Pj,t as “downside overshooting”. This means that the analysts predict greater downside than the real price downside observed ex-post on the stock market. Thus, if the sign of AB_ANY or AB_END is positive (negative), the price target is overshot (conservative). To simplify our analysis, the relevant prediction time-horizon is assumed to be one year, or 250 trading days, given that most analyst reports anticipate corporation value in the current year (Brav & Lehavy, 2003). However, if any target is updated within

2 The original sample consists of 8256 analyst-company observations across 493 companies. We exclude samples where analysts' biases are equal or less than zero to focus on why analysts overshoot the actual prices if their targets are not touched. After dropping observations where analyst biases are equal to or less than zero, our sample contains 5184 analyst-company observations across 421 companies and 3406 and 5158 observations for AB_ANY and AB_END, respectively.

238

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

the year, the relevant prediction time-horizon is shortened to be the period between the initial announcement date and three days before the subsequent announcement date.3 Second, if any price target is constructed as a range between two prices, we simply extract the average as the target price. 2.3. Investor belief measures in the aggregate market level and company level Bagnoli et al. (2009) provide partial evidence that analysts' recommendations may be influenced by market sentiment. Here, we suggest that high market sentiment may cause significant analyst target price bias. However, the absence of the composite market sentiment index like the BW index in the Taiwan equity market compelled us to measure aggregate market sentiment in an alternative way. Following Baker and Stein (2004) and Baker and Wurgler (2000, 2006), we adopt the monthly share turnover rate (TURN), the yearly number of IPOs (NIPO) and yearly average first-day returns of IPOs (RIPO) to compute this measure. Given that analysts' targets are sometimes biased, we want to know whether or not the bias is affected by investor beliefs in the company and aggregate market level, analyst-specific characteristics, and company-specific characteristics. We argue that in addition to market consensus, investor beliefs of company level may influence analyst price targets. Following Kumar and Lee (2006), we use the trading activities of all investors of the particular stock to construct the buy-sell imbalances (BSI) to measure investor beliefs for the underlying stock. The BSI for any stock prior to the announcement date could be shown as follows: D X

BSI V j;t−3 ¼

k¼1 D X

VB j;k;t−3 −VS j;k;t−3 ð3Þ VB j;k;t−3 þ VS j;k;t−3

k¼1

where t is the announcement date. t − 3 is three days before t. D is the number of days in the month prior to date t − 3. VBj,k,t − 3 is the buy orders measured in dollars offered for stock j on day k; and VSj,k,t − 3 is the sell orders measured in dollars offered for stock j on day k. An alternative way to compute BSI is to utilize trade orders measured in numbers of shares. The specific formula for the second definition of BSI is: D X

BSI N j;t−3 ¼

k¼1 D X

NB j;k;t−3 −NS j;k;t−3 ð4Þ NB j;k;t−3 þ NS j;k;t−3

k¼1

where NBj,k,t − 3 is the buy orders measured in numbers of shares offered for stock j on day k, and NSj,k,-3t is the sell orders measured in numbers of shares offered for stock j on day k. In addition to the original BSI, we constructed the residual BSI to remove the common component in investor net demand that is due to overall market movements. Precisely, we performed the following regressions:   BSI V j;t−3 ¼ β0 þ β1 Rm −R f þ ε V j;t−3

ð5Þ

  BSI N j;t−3 ¼ β0 þ β1 Rm −R f þ ε N j;t−3

ð6Þ

where Rm is the market return during the month prior to date t; Rf is the risk free rate in the month before date t; and ε_Vj,t − 3 and ε_Nj,t-3 are the residual BSI for stock j at date t − 3 measured in dollars and shares, respectively (hereafter denoted as BSI_V_R and BSI_N_R). The stock-level BSI indicates whether retail investors are net buyers or net sellers of a given stock during a given month. In other words, the stock-level BSI measure is a directional indicator of net retail demand for that stock in a given month. Finally, since trades by retail investors and trades by institutional investors may influence each other, we run the two stage least square model to take account of the endogeneity. We first regress BSI_V (BSI_N) on percentage of institutional investor holdings and then use the coefficient to predict BSI_V (BSI_N). 3

It is presumed that the serviceableness of former target price announcements range from the issuance date up to three days before updating the price target.

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

239

Table 2 Analyst target price bias. This table reports basic statistics of analysts' target price bias. AB_ANY and AB_END are our measures of analyst target price bias which are defined as: 8 max TP i; j;t −P j;t TP i; j;t > > > ¼ max −1; if TP i; j;t NP j;t ; > max < P j;t P j;t

AB ANY i; j;t ¼ min > P j;t −TP i; j;t TP i; j;t > > ¼ 1− min ; if TP i; j;t bP j;t ; > : P min P j;t

AB ENDi; j;t

j;t

8 TP −P j;tþn TP i; j;t > > > i; j;t ¼ −1; if TP i; j;t NP j;t < P j;tþn P j;tþn ¼ TP i; j;t > P j;tþn −TP i; j;t > > ¼ 1− ; if TP i; j;t bP j;t : P j;tþn P j;tþn

Panel A shows the statistics in different years; Panel B reports it in different industries. Panel A: Analyst bias across time Foreign analysts

Domestic analysts

Total

Period

Items

AB_ANY

AB_END

AB_ANY

AB_END

AB_ANY

AB_END

2003 (Sep.–Dec.)

Mean Std. Dev. Max Min N. Mean Std. Dev. Max Min N. Mean Std. Dev. Max Min N. Mean Std. Dev. Max Min N. Mean Std. Dev. Max Min N. Mean Std. Dev. Max Min N. Mean Std. Dev. Max Min N.

13.24 8.64 44.2 0.45 138 15.47 10.82 50.5 0.17 324 11.14 8.7 41.29 0.12 276 11.03 10.01 49.17 0.24 155 15.2 10 50.79 0.17 396 13.66 11.33 50.73 0.19 266 13.68 10.23 50.79 0.12 1555

34.9 32.33 142.11 0.61 214 38.85 34.33 190 0.22 403 29.46 27.64 174.16 0.18 413 27.46 25.92 180.37 0.19 249 65.08 62.68 272.55 0.33 539 57.83 53.98 269.82 1.01 402 45.25 47.26 272.55 0.18 2220

13.22 10.38 46.07 0.28 144 14.04 10.94 49.73 0.21 595 10.29 8.35 41.84 0.12 360 12.15 9.22 49.84 0.14 309 12.56 9.87 50.43 0.11 261 11.83 8.92 47.49 0.17 182 12.51 9.89 50.43 0.11 1851

41.52 39.5 207.69 0.48 258 47.81 44.03 262.32 0.15 887 29.85 30.26 238.03 0.16 627 29.84 27.02 247.22 0.17 474 60.19 56.58 258.35 0.28 443 59.24 51.87 266.77 1.45 249 43.36 43.38 266.77 0.15 2938

13.23 9.55 46.07 0.28 282 14.55 10.91 50.5 0.17 919 10.66 8.51 41.84 0.12 636 11.78 9.49 49.84 0.14 464 14.16 10.02 50.79 0.11 657 12.92 10.45 50.73 0.17 448 13.05 10.06 50.79 0.11 3406

38.52 36.54 207.69 0.48 472 45.01 41.44 262.32 0.15 1,290 29.7 29.23 238.03 0.16 1,040 29.02 26.65 247.22 0.17 723 62.87 60.02 272.55 0.28 982 58.37 53.14 269.82 1.01 651 44.17 45.1 272.55 0.15 5158

2004

2005

2006

2007

2008 (Jan.–Jul.)

All periods

Panel B: Analyst bias across industries Foreing analysts

Domestic analysts

Total

Mean of Industry

AB_ANY

AB_END

AB_ANY

AB_END

AB_ANY

AB_END

Electronics Industry Mean Std. Dev. Max Min N.

14.29 10.37 50.79 0.15 1090

48.06 48.82 272.55 0.19 1489

12.45 9.8 50.43 0.11 1217

45.3 44.89 266.77 0.15 1899

13.32 10.11 50.79 0.11 2307

46.52 46.67 272.55 0.15 3388

Plastic, Rubber, and Chemical Industry Mean 13.53 Std. Dev. 9.76 Max 44.64 Min 0.12 N. 101

33.55 35.83 269.82 0.18 139

12.16 8.98 33.33 0.18 102

37.5 38.09 218.58 0.3 156

12.84 9.38 44.64 0.12 203

35.63 37.03 269.82 0.18 295

34.87 37.3

10.68 8.87

26.46 25.78

11.22 9.37

30.96 32.69

Financial & Insurance Mean Std. Dev.

11.72 9.82

(continued on next page)

240

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

Table 2 (continued) Panel B: Analyst bias across industries Foreing analysts Mean of Industry

Domestic analysts

Total

AB_ANY

AB_END

AB_ANY

AB_END

AB_ANY

AB_END

44.03 0.17 135

189.86 1.16 222

49.84 0.28 122

187.43 0.54 193

49.84 0.17 257

189.86 0.54 415

Automobile and Iron & Steel Mean Std. Dev. Max Min N.

8.39 7.26 32.14 0.47 44

26.53 28.13 163.16 0.28 69

10.8 9.49 41.77 0.31 78

44.86 45.39 198.51 1.12 131

9.93 8.8 41.77 0.31 122

38.53 41.14 198.51 0.28 200

Electric Machinery Mean Std. Dev. Max Min N.

16.88 11.61 41.87 1.82 21

105.9 77.95 272.55 2.37 26

12.22 10.04 44.65 0.17 84

51.56 51.62 247.68 0.67 119

13.15 10.48 44.65 0.17 105

61.3 60.63 272.55 0.67 145

Cement, Building Material & Construction Mean 13.07 Std. Dev. 10.95 Max 49.17 Min 0.17 N. 44

54.26 53.7 229.71 0.5 71

15.37 10.96 47.19 1.12 63

45.49 47.1 242.86 0.48 121

14.43 10.96 49.17 0.17 107

48.73 49.68 242.86 0.48 192

Food and Textile Mean Std. Dev. Max Min N.

12.15 11.92 48.53 0.9 15

47.19 45.1 216.67 4.31 29

16.47 12.22 45.08 0.22 37

51.52 44.56 227.33 0.22 63

15.23 12.18 48.53 0.22 52

50.15 44.53 227.33 0.22 92

Transportation Mean Std. Dev. Max Min N.

12.43 8.8 38.89 0.67 53

40.94 35.8 159.02 1.1 82

15.09 11.43 48.61 0.28 61

40.41 34.95 153.73 0.28 98

13.85 10.33 48.61 0.28 114

40.65 35.24 159.02 0.28 180

Other Industry Mean Std. Dev. Max Min N.

11.81 9.18 42.09 0.23 52

35.77 42.47 224.34 0.7 93

12.59 9.8 47.34 0.23 87

35.96 32.87 143.44 0.17 158

12.3 9.55 47.34 0.23 139

35.89 36.63 224.34 0.17 251

Max Min N.

2.4. Regression analysis In this section, we set several equations to detect the influence of various categories of variables, including investor beliefs in the company level, investor beliefs in the aggregate market level, analyst separate characteristics, and company-specified characteristics on analysts' bias separately. We employ two variables to control analyst characteristics. One is the COMPETEi,t , denoting the number of analysts issuing targets toward the same company within the month prior to the announcement date t issued by analyst i, and the other is DorFi denoting the nationalities of analysts i who issue a target. If the target is issued by a domestic analyst, DorF is set to be one. If the target is issued by a foreign one, the dummy is equal to zero. The dummy DorF, allows us to examine whether catering effects are unequal between foreign and domestic analysts. If the intensity of domestic analysts catering to investors is stronger than the intensity of foreign analysts, we expect that the coefficient for DorF would be positive and statistically significant. We also include an interaction term, which is the product of the two variables, investor beliefs (BSI_V and BSI_N) and DorF, in the regression. If domestic analysts know investors better than foreign analysts do and then cater more to investor beliefs, the coefficient of interaction term should be positive. According to Lai (2005), the more volatile a stock price is, the more apparent the opportunity for analysts to cater to investors will be. We use volatility of daily return of each stock (VOL) as the proxy of volatility. When volatility of individual stock is high, analysts' catering behavior would be conspicuous, and analyst target price bias would be significant. The correlation between volatility and analyst target price bias should be positive. In addition, to control for characteristics of specific corporations, we include company size (natural log of total assets) at the announcement date (SIZE). We include market-to-book ratio (MB) and price-to-

Table 3 Investor beliefs measures in the company level. This table reports the original, residual, and 2-stage-least-square investor beliefs by year and industry. BSI_V are calculated in Panel A and Panel C, and BSI_N are calculated in Panel B and Panel D to quantify investor beliefs one month prior to the announcement of target price, and they are defined as:

BSI V j;t−3 ¼

k¼1 D X

D X

VB j;k;t−3 −VS j;k;t−3 ;

BSI N j;t−3 ¼

VB j;k;t−3 þ VS j;k;t−3

k¼1

k¼1 D X

NB j;k;t−3 −NS j;k;t−3 NB j;k;t−3 þ NS j;k;t−3

k¼1

In addition, we run the 2 stage least square model to take account of the endogeneity. Panel A: Statistics for BSI_V of AB_ANY sample BSI_V (%)

Electronics Industry Plastic, Rubber, and Chemical Industry Financial & Insurance Automobile and Iron & Steel Electric Machinery Cement, Building Material & Construction Food and Textile Transportation Other Industry All Stocks

2003 (Sep.–Dec.)

2004

2008 (Jan.–Aug.)

All period

Ori

Res

2SLS

Ori

Res

2SLS

Ori

2005 Res

2SLS

Ori

2006 Res

2SLS

Ori

2007 Res

2SLS

Ori

Res

2SLS

Ori

Res

2SLS

12.23 12.05 8.59 6.78 13.21 20.92 25.02 14.99 12.04 12.40

14.27 16.27 14.86 7.59 10.14 18.50 16.27 14.95 9.76 14.22

8.69 9.22 8.11 9.43 9.34 12.71 13.22 8.57 9.75 8.90

12.11 13.23 10.66 12.94 13.65 18.55 19.90 16.65 8.89 12.65

13.41 15.19 15.18 11.36 12.88 20.04 17.60 17.61 10.33 13.99

8.86 9.40 8.63 10.43 9.20 12.64 12.65 8.90 10.58 9.23

14.30 11.24 9.34 12.12 10.69 12.68 6.94 11.96 15.78 13.07

14.33 10.28 10.54 12.19 9.62 5.99 7.46 9.63 15.27 12.99

8.06 9.02 8.64 10.45 8.91 10.55 11.39 7.93 11.83 8.57

15.72 20.13 12.92 26.42 20.70 24.02 22.41 12.35 17.42 16.67

15.26 18.28 12.03 25.17 17.26 24.08 22.16 9.42 13.85 15.94

8.19 10.43 7.86 11.16 11.14 9.96 10.65 9.71 8.68 8.58

17.39 25.07 8.10 12.46 22.57 26.79 17.87 15.62 21.77 17.94

16.82 24.23 8.70 16.30 21.53 22.18 19.76 15.48 17.33 17.22

6.52 10.07 8.19 9.11 10.30 9.87 7.17 5.59 9.66 7.17

19.41 17.93 24.69 11.95 13.02 18.71 28.10 16.20 15.80 19.10

20.24 16.34 22.16 16.42 15.82 18.35 26.03 13.05 15.50 19.34

6.93 7.34 6.77 8.60 11.28 7.04 9.11 5.93 10.04 7.34

15.04 15.42 11.71 14.41 16.99 20.87 20.16 14.65 15.69 15.12

15.47 16.01 13.24 14.53 15.88 20.28 18.60 13.77 14.44 15.41

7.88 9.08 8.27 10.18 9.98 10.10 10.78 7.95 10.41 8.34

Panel B: Statistics for BSI_N of AB_ANY sample BSI_N (%) 2004

2008 (Jan.–Aug.)

All period

Ori

Res

2SLS

Ori

Res

2SLS

Ori

2005 Res

2SLS

Ori

2006 Res

2SLS

Ori

2007 Res

2SLS

Ori

Res

2SLS

Ori

Res

2SLS

12.17 11.81 8.63 6.99 13.17 20.73 25.65 14.71 12.29 12.34

14.08 15.95 14.81 7.35 10.21 18.23 16.83 14.66 9.79 14.04

8.77 9.30 8.17 9.52 9.43 12.84 13.35 8.65 9.84 8.98

12.27 13.39 10.70 13.25 13.89 18.93 19.98 16.69 9.01 12.80

13.42 15.07 15.12 11.31 13.10 20.39 17.55 17.55 10.02 13.98

8.94 9.49 8.70 10.53 9.28 12.77 12.78 8.98 10.68 9.31

14.27 11.27 9.29 12.15 10.93 12.92 7.10 12.11 15.79 13.06

14.15 10.21 10.37 12.18 9.59 6.44 7.36 9.63 15.19 12.85

8.12 9.10 8.71 10.55 8.99 10.65 11.51 8.00 11.95 8.64

15.88 20.01 13.13 26.10 20.93 23.80 22.51 12.28 17.37 16.78

15.27 18.15 11.86 24.85 17.19 23.80 22.12 9.26 13.50 15.89

8.26 10.53 7.92 11.27 11.24 10.05 10.76 9.80 8.75 8.66

17.38 24.97 8.14 12.37 22.55 26.60 18.37 15.95 21.58 17.92

16.67 24.16 8.59 16.25 21.56 21.88 20.53 15.57 17.09 17.09

6.58 10.16 8.26 9.19 10.40 9.96 7.22 5.62 9.75 7.23

19.43 18.05 24.56 12.41 12.93 18.71 28.15 16.67 15.63 19.13

20.20 16.45 21.98 16.50 15.64 18.24 25.92 13.33 15.17 19.28

6.99 7.40 6.82 8.67 11.39 7.09 9.19 5.97 10.13 7.40

15.09 15.46 11.72 14.52 17.10 20.92 20.36 14.77 15.65 15.17

15.39 15.92 13.11 14.45 15.90 20.26 18.70 13.75 14.22 15.33

7.94 9.16 8.34 10.27 10.07 10.19 10.88 8.01 10.51 8.42

241

Electronics Industry Plastic, Rubber, and Chemical Industry Financial & Insurance Automobile and Iron & Steel Electric Machinery Cement, Building Material & Construction Food and Textile Transportation Other Industry All Stocks

2003 (Sep.–Dec.)

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

D X

242

Panel C: Statistics for BSI_V of AB_END sample BSI_V (%) 2004

2005

2006

2007

2008 (Jan.–Aug.)

All Period

Ori

Res

2SLS

Ori

Res

2SLS

Ori

Res

2SLS

Ori

Res

2SLS

Ori

Res

2SLS

Ori

Res

2SLS

Ori

Res

2SLS

11.32 15.92 8.78 7.44 13.62 17.60 18.82 11.46 15.59 11.95

13.30 20.17 12.73 6.07 14.37 18.58 13.91 11.88 11.95 13.70

9.17 9.23 9.08 9.34 8.23 12.47 12.24 8.29 10.06 9.30

11.68 13.33 10.35 13.49 12.61 18.07 16.48 14.88 11.27 12.29

12.80 15.06 14.23 12.22 12.15 18.51 14.88 13.94 12.58 13.34

9.06 9.65 8.64 10.71 9.61 13.03 12.66 8.96 10.74 9.44

15.06 13.10 9.48 9.95 11.41 9.53 6.94 13.14 14.99 13.72

14.51 10.44 10.62 10.07 10.78 9.86 7.46 10.92 16.11 13.33

8.33 9.43 8.79 10.26 9.41 10.13 11.39 8.30 10.80 8.74

16.40 23.16 11.39 27.95 18.94 29.00 19.59 10.85 17.97 17.30

15.26 21.25 9.88 26.73 16.14 28.39 17.04 9.58 13.80 16.00

8.11 10.25 7.96 11.25 10.97 10.40 9.63 8.87 9.25 8.52

18.04 17.83 9.70 16.59 23.19 19.19 26.36 19.25 19.17 18.17

17.40 18.83 8.80 17.33 20.48 17.56 24.55 18.76 15.71 17.29

6.86 9.16 8.53 9.30 10.60 9.44 8.13 6.16 9.53 7.53

20.60 17.95 27.92 10.35 14.44 21.94 26.92 16.31 16.17 20.36

21.14 17.15 26.53 13.83 15.06 20.13 25.19 12.76 17.12 20.43

6.89 7.40 6.78 9.21 11.89 6.92 9.77 6.01 9.73 7.39

15.34 15.62 11.94 14.53 16.88 20.83 20.84 14.68 15.65 15.39

15.45 16.17 13.01 14.25 15.53 20.12 18.85 13.33 15.11 15.39

8.09 9.14 8.40 10.29 10.14 10.11 10.62 8.06 10.13 8.54

Panel D: Statistics for BSI_N of AB_END sample BSI_N (%)

Electronics Industry Plastic, Rubber, and Chemical Industry Financial & Insurance Automobile and Iron & Steel Electric Machinery Cement, Building Material & Construction Food and Textile Transportation Other Industry All Stocks

2003 (Sep.–Dec.)

2004

2008 (Jan.–Aug.)

All Period

Ori

Res

2SLS

Ori

Res

2SLS

Ori

2005 Res

2SLS

Ori

2006 Res

2SLS

Ori

2007 Res

2SLS

Ori

Res

2SLS

Ori

Res

2SLS

11.30 15.73 8.80 7.68 13.75 17.72 19.21 11.33 15.96 11.95

13.15 19.92 12.67 5.90 14.42 18.59 14.04 11.59 12.11 13.57

9.25 9.31 9.16 9.42 8.30 12.59 12.37 8.36 10.15 9.38

11.83 13.45 10.36 13.68 12.88 18.45 16.64 14.99 11.34 12.43

12.78 14.97 14.12 12.08 12.35 18.84 14.80 13.84 12.26 13.30

9.14 9.74 8.71 10.82 9.69 13.16 12.79 9.04 10.84 9.53

15.05 13.14 9.44 10.01 11.75 9.66 7.10 13.30 14.93 13.73

14.39 10.35 10.47 10.03 10.89 9.88 7.36 10.95 15.97 13.23

8.40 9.51 8.86 10.36 9.49 10.22 11.51 8.37 10.90 8.82

16.59 23.05 11.58 27.68 19.17 28.73 19.60 10.74 17.85 17.43

15.29 21.09 9.79 26.47 16.09 28.08 16.96 9.40 13.58 15.98

8.18 10.35 8.03 11.36 11.08 10.50 9.72 8.95 9.34 8.60

18.01 17.70 9.81 16.63 23.16 19.17 26.69 19.46 19.06 18.16

17.25 18.68 8.77 17.34 20.46 17.45 24.89 18.82 15.47 17.16

6.91 9.24 8.60 9.38 10.70 9.52 8.20 6.20 9.62 7.59

20.65 18.21 27.63 10.88 14.22 21.73 26.99 16.81 16.12 20.39

21.15 17.39 26.23 13.96 14.82 19.81 25.11 12.99 16.96 20.40

6.95 7.46 6.83 9.29 12.01 6.97 9.86 6.05 9.82 7.45

15.40 15.66 11.94 14.63 17.01 20.83 21.03 14.82 15.63 15.45

15.38 16.11 12.88 14.16 15.56 20.04 18.88 13.30 14.92 15.32

8.16 9.23 8.47 10.39 10.23 10.20 10.72 8.12 10.23 8.61

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

Electronics Industry Plastic, Rubber, and Chemical Industry Financial & Insurance Automobile and Iron & Steel Electric Machinery Cement, Building Material & Construction Food and Textile Transportation Other Industry All Stocks

2003 (Sep.–Dec.)

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

243

Table 4 Summary statistics. This table reports summary statistics for our controlling variables, including investor beliefs variables in aggregate market level, analyst specific variables, and company specific variables. Investor beliefs measures in aggregate market level contain monthly share turnover rate prior to any announcement of targets (TURN), the yearly number of IPOs (NIPO), and the yearly average first-day returns of IPOs (RIPO). COMPETE is the number of analysts issuing targets toward the same company when any target price is issued. Company specific variables include standard deviation of daily return of each stock (VOL), natural log of total assets (SIZE), market-to-book ratio at the announcement date (MB), price-to-earnings ratio at the announcement date (PE). Panel A: AB_ANY Categories of variables

Mean

Std. Dev.

Min

Max

Median

3406 3406 3406

15.2780 12.8538 33.6562

6.5852 6.4711 26.5936

5.4500 5.0000 0.6959

31.6600 24.0000 107.2300

13.5600 9.0000 25.0746

Analyst specific variables COMPETE

3406

0.8831

1.3810

0.0000

12.0000

0.0000

Company specific variables VOL SIZE MB PE

3406 3406 3406 3406

2.4337 10.5002 2.8693 17.2998

0.5635 1.7685 1.7855 10.7747

1.0600 7.2800 0.7100 3.4600

4.2200 14.6600 11.5400 89.7100

2.4000 10.2600 2.3200 14.7900

Mean

Std. Dev.

Min

Max

Median

Market sentiment TURN(%) NIPO RIPO(%)

Observations

Panel B: AB_END Categories of variables

Observations

Market sentiment TURN(%) NIPO RIPO(%)

5158 5158 5158

14.5712 12.5832 32.5796

6.1389 6.4556 25.0651

5.4500 5.0000 0.6959

31.6600 24.0000 107.2300

12.9600 9.0000 25.0746

Analyst specific variables COMPETE

5158

0.8292

1.3134

0.0000

12.0000

0.0000

Company specific variables VOL SIZE MB PE

5158 5158 5158 5158

2.4506 10.4237 2.8411 17.2407

0.5778 1.7929 1.8252 10.6644

1.0600 7.2800 0.7000 3.4600

4.2200 14.6600 11.5400 89.7100

2.4200 10.1400 2.2700 14.6550

earnings ratio (PE) at announcement date of each company in our model to control firm characteristics. Finally, we add firm and year dummy variables to control fixed effect. The regression models are shown below:

ABi; j;t ¼ β0 þ β1 InvestorBelief si; j;t þ β2 COMPETEi; j;t þ β3 InvestorBelief si; j;t  Dor F i; j;t þ β4 VOLi; j;t þ β5 SIZEi; j;t þ β6 MBi; j;t þ β7 PEi; j;t þ Firm dummies þ Yeardummies þ εi; j;t

ð7Þ

ABi; j;t ¼ β0 þ β1 InvestorBelief si; j;t þ β2 TURNt þ β3 NIPOt þ β4 RIPOt þ β5 COMPETEi; j;t þ β6 InvestorBelief si; j;t  Dor F i; j;t þ β7 VOLi; j;t þ β8 SIZEi; j;t þ β9 MBi; j;t þ β10 PEi; j;t þ Firm dummies þ Year dummies þ εi; j;t

ð8Þ

where ABi,j,t denotes analyst target price bias containing AB_ANYi,j,t and AB_ENDi,j,t; The independent variable BSI_V and BSI_N are investor beliefs measures in the company level, computed by the two-stage-least-squares model. Market sentiment variables are defined as those in previous section. The t-values are computed with autocorrelation- and heteroskedasticity-consistent standard errors. 3. Empirical results 3.1. Analyst target price bias Panel A of Table 2 shows yearly prediction errors for the full sample. During the five sample years, prediction errors are quite stable across time. At the aggregate level, analysts' targets that were not touched overshot by 13.05% during the relevant period,

244

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

Table 5 Influence of investor belief on analyst target price bias. The regression model is shown below:

ABi; j;t ¼ β 0 þ β 1 InvestorBelief si; j;t þ β 2 COMPETEi; j;t þ β 3 InvestorBelief si; j;t  Dor F i; j;t þ β 4 VOLi; j;t þ β 5 SIZEi; j;t þ β 6 MBi; j;t þ β 7 PEi; j;t þ Firmdummies þ Yeardummies þ ε i; j;t

The dependent variables AB_ANY and AB_END are our measures of analyst target price bias. The independent variable BSI_V and BSI_N are investor beliefs measures in the company level, computed by the two-stage-least-squares model. Analyst specific measures contain two variables: the number of analysts issuing targets toward the same company when any target price is issued (COMPETE) and the interaction terms of BSI_V (BSI_N) and dummy variable of domestic analyst (DorF). Company specific variables include standard deviation of daily return of each stock (VOL), natural log of total assets (SIZE), market-to-book ratio at the announcement date (MB), price-to-earnings ratio at the announcement date (PE). Firm dummies and Year dummies denote firm- and industry-specific dummy variables. Newey and West (1987) autocorrelation- and heteroskedasticity-consistent standard errors are reported in parentheses. AB_ANY

AB_END

Equations

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Intercept

27.2556*** (7.1935) 0.7437** (0.3080)

27.2701*** (7.1946)

35.1029** (14.4662) 0.5881** (0.2562)

35.1645** (14.4766)

89.7215*** (7.1332) 1.2310*** (0.3050)

89.7954*** (7.1299)

109.3511*** (11.5813) 1.0337*** (0.3212)

109.5123*** (11.5760)

BSI_V BSI_N COMPETE BSI_V × DorF

0.8346* (0.4272) −0.7010** (0.2876)

BSI_N × DorF VOL SIZE MB PE Firm dummies Year dummies Adjusted R2 F-value Observations

0.0771 (0.0649) −1.1009** (0.4787) −0.9305*** (0.2736) 0.0127 (0.0085) No Yes 0.0121 4.79*** 3406

0.7360** (0.3053) 0.8349* (0.4272)

−0.6947** (0.2854) 0.0771 (0.0649) −1.1011** (0.4788) −0.9302*** (0.2735) 0.0127 (0.0085) No Yes 0.0121 4.79*** 3406

0.4446 (0.3595) −0.4845** (0.2446)

0.3949 (0.2713) −1.9571 (1.2489) −0.3149 (0.2259) −0.0005 (0.0066) Yes Yes 0.3863 9.08*** 3406

0.5784** (0.2530) 0.4447 (0.3595)

−0.4798** (0.2427) 0.3951 (0.2714) −1.9588 (1.2491) −0.3167 (0.2260) −0.0005 (0.0066) Yes Yes 0.3863 9.08*** 3406

−1.0806* (0.6335) −0.6365*** (0.2368)

1.0370*** (0.2676) −3.3322*** (0.4720) 0.7520* (0.4157) 0.0073 (0.0079) No Yes 0.0929 49.03*** 5158

1.2141*** (0.3017) −1.0810* (0.6335)

−0.6300*** (0.2347) 1.0371*** (0.2676) −3.3341*** (0.4720) 0.7514* (0.4158) 0.0073 (0.0079) No Yes 0.0929 49.02*** 5158

−0.2556 (0.6150) −0.7780*** (0.1905)

−0.8671 (0.8568) −5.2764*** (0.8541) 4.0756*** (0.6251) −0.0018 (0.0060) Yes Yes 0.2199 7.51*** 5158

1.0156*** (0.3174) −0.2556 (0.6150)

−0.7700*** (0.1888) −0.8666 (0.8570) −5.2838*** (0.8541) 4.0717*** (0.6250) −0.0018 (0.0060) Yes Yes 0.2199 7.51*** 5158

and 44.17% at the end of the period. As expected, analyst bias measured by AB_END was larger than that measured by AB_ANY. The maximum and minimum target price biases were 50.79 (272.55) and 0.11 (0.15) measured by AB_ANY (AB_END). Furthermore, the prediction errors of foreign analyst were larger than those of domestic analyst for both metrics, indicating that foreign analysts produced more biased target prices compared to domestic peers. Panel B of Table 2 reports prediction errors across industries. Panel B shows that average target price biases (AB_ANY and AB_END) in electric machinery and food and textile industry are larger than other industries. 3.2. Investor belief measures in the company level Table 3 reports three alternative measures for investor beliefs in company level, including original BSI, residual BSI, and twostage-least-squares BSI. For AB_ANY sample, BSI_V and BSI_N are calculated in Panel A and Panel B and for AB_END sample BSI_V and BSI_N are calculated in Panel C and Panel D to quantify investor beliefs one month prior to the announcement of target price. Table 3 shows that all original BSIs are positive across all industries for BSI_V and BSI_N metrics, indicating that during our sample period, investors tend to buy stocks rather than sell them. Further, after removing the common component in investor net demand, which is due to overall market movements, residual BSIs are still positive, indicating that investors tend to buy stocks rather than sell them after removing the common effect in the whole market. The results of BSI computed by the two-stage-leastsquares model are similar to those of original BSI and residual BSI. 3.3. Market sentiment Summary statistics on market sentiment variables are provided in Table 4. This table also reports summary statistics for our controlling variables, including, analyst specific variables, and company specific variables for AB_ANY and AB_END sample.

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

245

Table 6 Influence of investor belief and market sentiment on analyst target price bias. The regression is shown here below:

ABi; j;t ¼ β 0 þ β 1 InvestorBelief si; j;t þ β 2 TURNt þ β 3 NIPOt þ β 4 RIPOt þ β 5 COMPETEi; j;t þ β 6 InvestorBelief si; j;t  Dor F i; j;t þ β 7 VOLi; j;t þ β 8 SIZEi; j;t þ β 9 MBi; j;t þ β 10 PEi; j;t þ Firmdummies þ Yeardummies þ ε i; j;t

The dependent variables AB_ANY and AB_END are our measures of analyst target price bias. The independent variable BSI_V and BSI_N are investor beliefs measures in the company level, computed by the two-stage-least-squares model. The independent variable TURN, NIPO, and RIPO are investor beliefs measures in the aggregate market level. TURN is the monthly share turnover rate prior to any announcement of targets, NIPO is the yearly number of IPOs, and RIPO is the yearly average first-day returns of IPOs. Analyst specific measures contain two variables: the number of analysts issuing targets toward the same company when any target price is issued (COMPETE) and the interaction terms of BSI_V (BSI_N) and dummy variable of domestic analyst (DorF). Company specific variables include standard deviation of daily return of each stock (VOL), natural log of total assets (SIZE), market-to-book ratio at the announcement date (MB), price-to-earnings ratio at the announcement date (PE). Firm dummies and Year dummies denote firm- and industry-specific dummy variables. Newey and West (1987) autocorrelation- and heteroskedasticity-consistent standard errors are reported in parentheses. AB_ANY

AB_END

Equations

(1)

(2)

(3)

(4)

Intercept

24.4438*** (9.4101) 0.7293** (0.3091)

24.4547*** (9.4112)

31.5651** (14.7997) 0.5834** (0.2560)

31.6243** (14.8088)

BSI_V BSI_N Turn NIPO RIPO COMPETE BSI_V × DorF

0.1765** (0.0771) −0.0772 (0.2534) 0.0607* (0.0318) 0.7474* (0.4309) −0.6942** (0.2919)

BSI_N × DorF VOL SIZE MB PE Firm dummies Year dummies Adjusted R2 F-value Observations

0.0805 (0.0656) −1.0955** (0.4828) −0.9526*** (0.2725) 0.0126 (0.0083) No Yes 0.0127 4.23*** 3406

0.7218** (0.3063) 0.1765** (0.0771) −0.0772 (0.2534) 0.0607* (0.0318) 0.7477* (0.4309)

−0.6880** (0.2896) 0.0805 (0.0656) −1.0958** (0.4828) −0.9523*** (0.2723) 0.0126 (0.0083) No Yes 0.0127 4.23*** 3406

0.1502** (0.0638) −0.0434 (0.2325) 0.0660*** (0.0255) 0.3368 (0.3630) −0.4728* (0.2474)

0.4051 (0.2709) −1.9383 (1.2361) −0.3859* (0.2187) −0.0006 (0.0064) Yes Yes 0.3870 9.03*** 3406

0.5737** (0.2528) 0.1503** (0.0638) −0.0434 (0.2325) 0.0660*** (0.0255) 0.3369 (0.3630)

−0.4682* (0.2455) 0.4052 (0.2710) −1.9400 (1.2363) −0.3877* (0.2188) −0.0006 (0.0064) Yes Yes 0.3870 9.03*** 3406

(5) 30.5851** (13.1813) 1.1528*** (0.3022)

0.2166 (0.1530) 0.9062** (0.4383) 0.9065*** (0.0906) −2.1758*** (0.6267) −0.4751** (0.2378)

1.0455*** (0.2653) −3.3263*** (0.4603) 0.6947* (0.4093) 0.0041 (0.0077) No Yes 0.1254 55.64*** 5158

(6) 30.6592** (13.1801)

1.1364*** (0.2990) 0.2166 (0.1530) 0.9064** (0.4383) 0.9065*** (0.0906) −2.1764*** (0.6267)

−0.4701** (0.2358) 1.0457*** (0.2654) −3.3284*** (0.4603) 0.6939* (0.4093) 0.0041 (0.0077) No Yes 0.1254 55.63*** 5158

(7)

(8)

49.6957*** (15.4615) 0.9656*** (0.3157)

49.8554*** (15.4589)

0.0461 (0.1467) 1.0853*** (0.4181) 0.9411*** (0.0853) −1.3381** (0.6053) −0.5764*** (0.1884)

−0.9165 (0.8388) −5.5470*** (0.8341) 3.8595*** (0.6276) −0.0064 (0.0061) Yes Yes 0.2538 8.77*** 5158

0.9478*** (0.3119) 0.0461 (0.1467) 1.0856*** (0.4181) 0.9411*** (0.0853) −1.3383** (0.6053)

−0.5703*** (0.1867) −0.9159 (0.8391) −5.5546*** (0.8341) 3.8555*** (0.6276) −0.0064 (0.0061) Yes Yes 0.2538 8.77*** 5158

COMPETE is the number of analysts issuing targets toward the same company when any target price is issued. Company specific variables include standard deviation of daily return of each stock (VOL), natural log of total assets (SIZE), market-to-book ratio at the announcement date (MB), price-to-earnings ratio (PE) at the announcement date. Panel A of Table 4 reports summary statistics for AB-ANY sample. The average and median monthly share turnover rate prior to any announcement of target prices are 15.28% and 13.56%, respectively. The average (median) number and first-days returns of IPOs are 12.85% (9%) and 33.66% (25.07%), respectively. The average (median) number of analysts issuing targets is 0.88 (0), indicating most of the firms in the sample with only one analyst who issues target price report. The average (median) stock return volatility for firms is 2.43% (2.4%). The average (median) M/B ratio and P/E ratio are 2.86 (2.32) and 17.30 (14.79), respectively. Panel B of Table 4 reports summary statistics for AB_END sample. The average and median monthly share turnover rate prior to any announcement of target prices are 14.57% and 12.96%, respectively. The average (median) number and first-days returns of IPOs are 12.58% (9%) and 32.56% (25.07%), respectively. The average (median) number of analysts issuing targets is 0.83 (0). The average (median) stock return volatility for firms is 2.45% (2.42%). The average (median) M/B ratio and P/E ratio are 2.84 (2.27) and 17.24 (14.66), respectively.

246

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

3.4. Evidence on investor beliefs in analyst target price bias Table 5 presents the empirical results. For all twelve regression specifications shown on Table 5, the coefficients on BSI_V and BSI_N, computed by the two-stage-least-squares model, are positive and statistically significant, indicating that strong investor beliefs will lead to analyst target price bias. That is, when investors believe the price of underlying share is going to increase (or decrease), analysts will notice and follow the signal of investor beliefs then issue outright optimistic price targets for that company. For AB_ANY, the coefficients on COMPETE are positive but insignificant. For AB_END, they are negative and significant. The significant negative coefficient on COMPETE indicates that competition among analysts would narrow their target price bias. The negative coefficient on DorF in all regressions means that, on average, domestic analysts provide comparatively more accurate price targets than foreign ones. In addition, the coefficients on interaction term are significantly negative, implying that foreign analysts may be more willing to issue biased target prices to cater to investor belief. Orpurt (2006) finds that local analysts forecast earnings more accurately than non-local analysts in Europe. Bae, Stulz, and Tan (2008) find that analysts who reside in the country make more precise earnings forecasts in that country than non-resident analysts across 32 countries,4 even after controlling for firm and analyst characteristics. This result corresponds with Orpurt (2006) and Bae et al. (2008), but refutes the rationale arguing that the intensity by which domestic analysts cater to investors would be stronger than the intensity by which foreign analysts do. Lai and Teo (2008) find that in Asian emerging markets, domestic analyst recommendations are consistently more optimistic than those of foreign analysts. Our results show that foreign analysts produce more biased price targets, providing support for Bradshaw et al.'s (2013) notion that unlike earnings forecasts, target price forecasts are subject to no market examination and surveillance. Additionally, in the view of the catering theory, our results may be explained by the client base of the foreign analysts. The client base of foreign analysts consists mainly of investors in the global investment market. Only a small fraction of their client base consists of local investors. Given that a great number of foreign analysts are actually born locally and educated abroad, these analysts are familiar not only with local investors' demand but also with global investors' demand. In other words, foreign analysts may know their client's demand just as well as if not better than domestic analysts. They have better opportunity than domestic ones to access investor beliefs and then cater to those investors more easily. Accordingly, foreign analysts may be knowledgeable about their clients who may not necessarily be local Taiwanese investors but who may be international investors from outside Taiwan, and as such, the way they cater to them would be more severe. For characteristics of specific corporations, the coefficient on MB is negative and statistically significant and other coefficients are not significant when AB_ANY is the dependent variable. These results indicate that low market-to-book value firm increases analyst target price bias but volatility of stock daily return, company size, and price-to-earnings ratio do not affect it. On the contrary, when AB_END is the dependent variable, the coefficient on MB is not significant, whereas coefficient on SIZE is negative and significant. Table 6 provides corresponding results controlling for various factors that would affect analyst target price bias. The results confirm our prediction that strong investor beliefs will cause analyst target price bias regardless of whether the investor belief was defined by trade orders in dollars or in shares (i.e., BSI_V and BIS_N). The coefficients on Turn, NIPO and RIPO are significantly positive, indicating that when investor belief and market sentiment are high, analyst bias tends to widen. Coefficient on COMPETE is negative. This evidence is not consistent with the findings of Lai (2005) who suggested that with behavioral homogenous investors, analyst forecast slant (for EPS) in monopoly and duopoly industrial structures would be the same. DorF is negative, indicating that domestic analysts would provide comparatively more accuracy than foreign ones. Positive coefficient of VOL and negative coefficients of SIZE and MB provide evidence that analysts would cater to investors more possibly when the companies covered are more volatile, smaller, and possessing lower MB ratio. Finally, we also use the other two investor belief (original BSI and residual BSI) measured in the company level on analyst target price bias. The results are shown in Appendix A and Appendix B. These results are extremely similar to those of Table 65. 3.5. Robustness check To address the possibility that investors and analysts may be reacting to the same signals, we estimate the following system of simultaneous equations model using three Stage Least Square (3 SLS)6. ABi; j;t ¼ β0 þ β1 InvestorBelief si; j;t þ β2 TURNt þ β3 NIPOt þ β4 RIPOt þ β5 COMPETEi; j;t þ β6 InvestorBelief si; j;t  Dor F i; j;t þ β7 VOLi; j;t þ β8 SIZEi; j;t þ β9 MBi; j;t þ β10 PEi; j;t þ Firm dummies þ Year dummies þ εi; j;t

4

ð9Þ

Taiwan is one of the sample countries. We also use the original sample to repeat our empirical test. The signs and significance of investor beliefs in Appendix C are in line with the results of Table 6. Empirical results confirm a statistically significant relation between investor belief and target bias. 6 We thank an anonymous referee for suggesting this test. 5

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

247

Table 7 Influence of market sentiment on investor belief and analyst target price bias. To address the possibility that investors and analysts may be reacting to the same signals, we estimate the following system of simultaneous equations model using three Stage Least Square (3 SLS).

ABi; j;t ¼ β 0 þ β 1 InvestorBelief si; j;t þ β 2 TURNt þ β 3 NIPOt þ β 4 RIPOt þ β 5 COMPETEi; j;t þ β 6 InvestorBelief si; j;t  Dor F i; j;t þ β 7 VOLi; j;t þ β 8 SIZEi; j;t þ β 9 MBi; j;t þ β 10 PEi; j;t þ Firmdummies þ Yeardummies þ ε i; j;t

InvestorBelief si; j;t ¼ β 0 þ β 1 TURN t þ β 2 NIPOt þ β 3 RIPOt þ β 4 VOLi; j;t þ β 9 Returni; j;t þ β 5 SIZEi; j;t þ Firmdummies þ Yeardummies þ ε i; j;t

The dependent variables AB_ANY and AB_END are our measures of analyst target price bias. The independent variable BSI_V and BSI_N are investor beliefs measures in the company level, computed by the two-stage-least-squares model. The independent variable TURN, NIPO, and RIPO are investor beliefs measures in the aggregate market level. TURN is the monthly share turnover rate prior to any announcement of targets, NIPO is the yearly number of IPOs, and RIPO is the yearly average firstday returns of IPOs. Analyst specific measures contain two variables: the number of analysts issuing targets toward the same company when any target price is issued (COMPETE) and the interaction terms of BSI_V (BSI_N) and dummy variable of domestic analyst (DorF). Company specific variables include standard deviation of daily return of each stock (VOL), natural log of total assets (SIZE), market-to-book ratio at the announcement date (MB), price-to-earnings ratio at the announcement date (PE). Firm dummies and Year dummies denote firm- and industry-specific dummy variables. Newey and West (1987) autocorrelation- and heteroskedasticity-consistent standard errors are reported in parentheses. (1) Equations Intercept BSI_V

AB_ANY -30.7916* (18.0266) 2.9912*** (0.9640)

(2)

NIPO RIPO COMPETE BSI_V × DorF

0.1265** (0.0588) 0.1139 (0.1338) 0.0750*** (0.0243) 0.7336*** (0.2120) −0.8135*** (0.0958)

AB_ANY

BSI_N

AB_END

BSI_V

AB_END

BSI_N

24.3905*** (1.1428)

20.8011*** (7.5046)

29.9085*** (3.6932)

-164.3490*** (41.3726) 15.4417*** (2.1595)

24.1211*** (1.1567)

100.6813*** (17.8272)

28.2440** (11.8476)

−0.0013 (0.0014) −0.0016 (0.0029) −0.0005 (0.0007)

0.3044** (0.1468) 0.2035** (0.0898) −0.0132 (0.2122) 0.0734* (0.0386) 0.3311 (0.2753)

−0.0006 (0.0015) −0.0016 (0.0045) −0.0022 (0.0015)

0.7478** (0.3560) −0.2738 (0.2150) 1.4124*** (0.5084) 0.8134*** (0.0926) −1.0478 (0.6551)

0.0073 (0.0115) −0.0020 (0.0300) 0.0104 (0.0104)

BSI_N × DorF −0.0823 (0.2255) 1.0396 (0.8800) 2.0228*** (0.5522) 0.0090* (0.0055)

VOL SIZE MB PE Return Firm dummies Year dummies Adjusted R2 F-value Observations

Yes Yes 0.2118 4.41*** 3406

(4)

BSI_V

BSI_N Turn

(3)

−0.0209 (0.0205) −1.5820*** (0.1210)

0.8745 (0.9307) Yes Yes 0.8389 51.86*** 3406

−0.3045*** (0.1043) 0.3897 (0.3441) −0.9876** (0.4468) −0.4649* (0.2742) 0.0009 (0.0086)

Yes Yes 0.3173 9.63*** 3406

0.0134 (0.0109) 0.0013 (0.0247) 0.0029 (0.0046)

0.1310 (0.2095) −1.0397*** (0.3253)

−6.9630 (9.5649) Yes Yes 0.0980 2.55*** 3406

InvestorBelief si; j;t ¼ β0 þ β1 TURNt þ β2 NIPOt þ β3 RIPOt þ β4 VOLi; j;t þ β9 Returni; j;t þ β5 SIZEi; j;t þ Firm dummies þ Year dummies þ εi; j;t

−0.2733 (0.2124) 1.5431*** (0.4429) 0.8058*** (0.0835) 2.1105*** (0.8158) −3.0233*** (0.2262)

−2.8564*** (0.7649) 2.5489 (2.1823) 15.0910*** (1.1165) 0.0408*** (0.0128)

Yes Yes 0.2299 4.79*** 5158

−0.0206 (0.0205) −1.5488*** (0.1223)

0.4336 (0.9409) Yes Yes 0.8389 51.86*** 5158

−0.8263*** (0.2261) −0.7886 (0.8251) −9.0878*** (1.0483) 3.3950*** (0.5643) 0.0008 (0.0184)

Yes Yes 0.2312 6.58*** 5158

0.0365 (0.2097) −1.1986 (1.2566)

−9.7941 (9.6643) Yes Yes 0.1219 2.36*** 5158

ð10Þ

248

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

Under the above system of equations specification, the coefficients associated with the market sentiment variables (TURN, NIPO, and RIPO) should be statistically significant if the signals related to these market sentiment variables affects investor belief. Table 7 reports results of 3 SLS regressions. We find that the relation between investor belief and the market sentiment variables is not statistically significant for all specifications. On the other hand, the relation between analyst bias and these same market sentiment variables are statistically significant for most cases, implying that the investors are not reacting to the same information that analysts are reacting to.

4. Conclusion Although analyst bias has been broadly discussed in recent literature, only a few studies have dealt with the source of analyst bias. Our study extends the catering theory by applying it to analyze target price bias. We investigate whether or not analysts cater to investor beliefs through their target prices. We argue that strong investor beliefs will lead to target price bias. We adopt different proxies to quantify investor beliefs. Our results show that analysts do cater to investors via overshooting targets even after controlling for market sentiment, analyst and company characteristics. Overall results confirm our prediction that strong investor beliefs will cause analyst target price bias. This was true regardless of whether the investor belief was defined by trade orders in dollars or in shares. We find that when investor belief and market sentiment are high, analyst bias tends to widen. On the other hand, competition among analysts narrows target price bias. With respect to company specific characteristics, we find some evidence that analysts would cater to investors more when the companies covered are more volatile, smaller, and possess lower M/B ratio. Bradshaw et al. (2013) noted that unlike earnings forecasts, target price forecasts are subject to no market examination and surveillance. Our results show that regulation indeed can affect price target bias. We find that foreign analyst produced more biased target prices compared to domestic peers. Our results suggest the possibility that foreign analysts can cater more because they have clients outside Taiwan who may be less knowledgeable concerning the peculiarities of the Taiwan stock market and thus easier for these analysts to cater according to their client's demand. This study opens up a range of other questions. If the influence of investor beliefs on analysts' target prices is proven valid, how will readers of analysts' price targets be influenced by this fact? Will biased target prices enhance investor' beliefs or affect the behavior of institutional and/or individual investors? Finally, this study has some practical implications. If analysts do indeed cater to investors' demand, additional regulatory structures may desirable to reduce the catering.

Appendix A. Influence of investor belief (original investor belief) and market sentiment on analyst target price bias The regression model is shown below: ABi; j;t ¼ β0 þ β1 InvestorBelief si; j;t þ β2 TURNt þ β3 NIPOt þ β4 RIPOt þ β5 COMPETEi; j;t þ β6 InvestorBelief si; j;t  Dor F i; j;t þ β7 VOLi; j;t þ β8 SIZEi; j;t þ β9 MBi; j;t þ β10 PEi; j;t þ Firm dummies þ Year dummies þ εi; j;t

ð8Þ

The dependent variables AB_ANY and AB_END are our measures of analyst target price bias. The independent variable BSI_V and BSI_N are investor beliefs measures in the company level, computed by the original investor belief. The independent variable TURN, NIPO, and RIPO are investor beliefs measures in the aggregate market level. TURN is the monthly share turnover rate prior to any announcement of targets, NIPO is the yearly number of IPOs, and RIPO is the yearly average first-day returns of IPOs. Analyst specific measures contain two variables: the number of analysts issuing targets toward the same company when any target price is issued (COMPETE) and the interaction terms of BSI_V (BSI_N) and dummy variable of domestic analyst (DorF). Company specific variables include standard deviation of daily return of each stock (VOL), natural log of total assets (SIZE), market-tobook ratio at the announcement date (MB), price-to-earnings ratio at the announcement date (PE). Firm dummies and Year dummies denote firm- and industry-specific dummy variables. Newey and West (1987) autocorrelation- and heteroskedasticityconsistent standard errors are reported in parentheses.

AB_ANY Equations Intercept BSI_V BSI_N

(1) 24.3519** (10.0917) 0.1060*** (0.0282)

AB_END (2) 24.3294** (10.0835)

0.1136***

(3) 32.7894* (16.7917) 0.0992*** (0.0272)

(4) 32.7409* (16.7847)

0.1057***

(5)

(6)

(7)

(8)

37.5348*** (12.8238) 0.2515*** (0.0764)

37.5786*** (12.8215)

51.2315*** (16.0075) 0.2614*** (0.0729)

51.3123*** (16.0047)

0.2664***

0.2734***

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

249

Appendix A (continued) (continued) AB_ANY Equations

(1)

Turn

0.1695** (0.0786) −0.0712 (0.2518) 0.0586** (0.0294) 0.6815 (0.4339) −0.1206*** (0.0428)

NIPO RIPO COMPETE BSI_V × DorF BSI_N × DorF VOL SIZE MB PE Firm dummies Year dummies Industry dummies Adjusted R2 F-value Observations

0.1101* (0.0628) −0.8809** (0.4263) −1.0246*** (0.2967) 0.0112 (0.0081) No Yes Yes 0.0074 2.88*** 3406

AB_END (2) (0.0278) 0.1678** (0.0788) −0.0714 (0.2518) 0.0579** (0.0293) 0.6622 (0.4343)

−0.1244*** (0.0427) 0.1125* (0.0626) −0.8795** (0.4261) −1.0479*** (0.2954) 0.0111 (0.0081) No Yes Yes 0.0077 2.94*** 3406

(3) 0.1476** (0.0626) −0.0407 (0.2319) 0.0588** (0.0252) 0.2877 (0.3654) −0.1248*** (0.0426)

0.3483 (0.2497) −1.8041 (1.2091) −0.4989* (0.2779) −0.0013 (0.0063) Yes Yes Yes 0.3861 9.00*** 3406

(4) (0.0268) 0.1459** (0.0626) −0.0405 (0.2319) 0.0583** (0.0251) 0.2702 (0.3658)

−0.1274*** (0.0423) 0.3482 (0.2490) −1.8004 (1.2086) −0.5214* (0.2780) −0.0014 (0.0063) Yes Yes Yes 0.3863 9.00*** 3406

(5) 0.1936 (0.1536) 0.9333** (0.4371) 0.8824*** (0.0907) −2.5340*** (0.6271) −0.1084* (0.0652)

1.1400*** (0.2643) −3.4996*** (0.4234) 0.3091 (0.4049) 0.0023 (0.0074) No Yes Yes 0.1262 56.05*** 5158

(6) (0.0755) 0.1904 (0.1536) 0.9319** (0.4370) 0.8807*** (0.0906) −2.5680*** (0.6275)

−0.1179* (0.0650) 1.1446*** (0.2640) −3.5025*** (0.4232) 0.2647 (0.4030) 0.0022 (0.0074) No Yes Yes 0.1266 56.24*** 5158

(7)

(8)

0.0296 (0.1461) 0.9840** (0.4146) 0.9171*** (0.0858) −1.3788** (0.6146) −0.2054** (0.0855)

−1.2057 (0.7815) −5.0761*** (0.9591) 4.7265*** (0.6871) −0.0081 (0.0059) Yes Yes Yes 0.2704 8.13*** 5158

(0.0720) 0.0269 (0.1461) 0.9842** (0.4145) 0.9159*** (0.0857) −1.4110** (0.6149)

−0.2139** (0.0847) −1.2060 (0.7800) −5.0827*** (0.9589) 4.6824*** (0.6850) −0.0082 (0.0059) Yes Yes Yes 0.2706 8.14*** 5158

Appendix B. Influence of investor belief (residual investor belief) and market sentiment on analyst target price bias The regression model is shown below: ABi; j;t ¼ β0 þ β1 InvestorBelief si; j;t þ β2 TURNt þ β3 NIPOt þ β4 RIPOt þ β5 COMPETEi; j;t þ β6 InvestorBelief si; j;t  Dor F i; j;t þ β7 VOLi; j;t þ β8 SIZEi; j;t þ β9 MBi; j;t þ β10 PEi; j;t þ Firmdummies þ Yeardummies þ εi; j;t

ð8Þ

The dependent variables AB_ANY and AB_END are our measures of analyst target price bias. The independent variable BSI_V and BSI_N are investor beliefs measures in the company level, computed by the residual investor belief. The independent variable TURN, NIPO, and RIPO are investor beliefs measures in the aggregate market level. TURN is the monthly share turnover rate prior to any announcement of targets, NIPO is the yearly number of IPOs, and RIPO is the yearly average first-day returns of IPOs. Analyst specific measures contain two variables: the number of analysts issuing targets toward the same company when any target price is issued (COMPETE) and the interaction terms of BSI_V (BSI_N) and dummy variable of domestic analyst (DorF). Company specific variables include standard deviation of daily return of each stock (VOL), natural log of total assets (SIZE), market-to-book ratio at the announcement date (MB), price-to-earnings ratio at the announcement date (PE). Firm dummies and Year dummies denote firm- and industry-specific dummy variables. Newey and West (1987) autocorrelation- and heteroskedasticityconsistent standard errors are reported in parentheses.

AB_ANY Equations Intercept BSI_V

(1) 24.8441** (10.1254) 0.1001*** (0.0235)

BSI_N Turn NIPO

0.1733** (0.0780) −0.0675

AB_END (2) 24.7908** (10.1094)

0.1068*** (0.0231) 0.1713** (0.0781) −0.0670

(3) 32.9750** (16.6823) 0.0878*** (0.0219)

0.1497** (0.0617) −0.0386

(4) 32.9264** (16.6685)

0.0942*** (0.0216) 0.1478** (0.0618) −0.0378

(5)

(6)

(7)

(8)

38.9124*** (12.8295) 0.2633*** (0.0777)

39.0055*** (12.8255)

52.2380*** (15.9012) 0.2906*** (0.0731)

52.3611*** (15.8985)

0.1861 (0.1544) 0.9506**

0.2785*** (0.0766) 0.1820 (0.1545) 0.9493**

0.0210 (0.1465) 0.9960**

0.3008*** (0.0719) 0.0180 (0.1465) 0.9962**

(continued on next page)

250

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

Appendix B (continued) (continued) AB_ANY Equations RIPO COMPETE BSI_V × DorF

(1) (0.2518) 0.0619** (0.0293) 0.6626 (0.4351) −0.1287*** (0.0444)

BSI_N × DorF VOL SIZE MB PE Firm dummies Year dummies Adjusted R2 F-value Observations

0.1083* (0.0623) −0.9292** (0.4228) −1.0409*** (0.2968) 0.0116 (0.0082) No Yes 0.0074 2.88*** 3406

AB_END (2) (0.2517) 0.0617** (0.0292) 0.6448 (0.4356)

−0.1310*** (0.0442) 0.1100* (0.0621) −0.9275** (0.4218) −1.0632*** (0.2959) 0.0115 (0.0082) No Yes 0.0076 2.92*** 3406

(3) (0.2320) 0.0630*** (0.0244) 0.2866 (0.3663) −0.1222*** (0.0358)

0.3588 (0.2513) −1.8287 (1.1929) −0.5092* (0.2746) −0.0011 (0.0064) Yes Yes 0.3859 8.99*** 3406

(4) (0.2319) 0.0627*** (0.0243) 0.2701 (0.3666)

−0.1245*** (0.0355) 0.3592 (0.2507) −1.8277 (1.1922) −0.5303* (0.2729) −0.0012 (0.0064) Yes Yes 0.3860 8.99*** 3406

(5)

(6)

(7)

(8)

(0.4372) 0.8909*** (0.0909) −2.5805*** (0.6286) −0.1228* (0.0684)

(0.4371) 0.8899*** (0.0909) −2.6202*** (0.6290)

(0.4146) 0.9235*** (0.0858) −1.4334** (0.6153) −0.2331*** (0.0859)

(0.4145) 0.9230*** (0.0858) −1.4683** (0.6157)

1.1272*** (0.2660) −3.6684*** (0.4228) 0.2236 (0.4047) 0.0022 (0.0074) No Yes 0.1259 55.90*** 5158

−0.1325* (0.0682) 1.1307*** (0.2656) −3.6751*** (0.4223) 0.1728 (0.4033) 0.0022 (0.0074) No Yes 0.1263 56.10*** 5158

−1.1802 (0.7840) −5.2338*** (0.9531) 4.6640*** (0.6840) −0.0084 (0.0059) Yes Yes 0.2708 8.14*** 5158

−0.2412*** (0.0851) −1.1785 (0.7827) −5.2419*** (0.9527) 4.6107*** (0.6824) −0.0084 (0.0059) Yes Yes 0.2711 8.15*** 5158

Appendix C. Influence of investor belief and market sentiment on analyst target price bias: all sample The regression model is shown below: ABi; j;t ¼ β0 þ β1 InvestorBelief si; j;t þ β2 TURNt þ β3 NIPOt þ β4 RIPOt þ β5 COMPETEi; j;t þ β6 InvestorBelief si; j;t  Dor F i; j;t þ β7 VOLi; j;t þ β8 SIZEi; j;t þ β9 MBi; j;t þ β10 PEi; j;t þ Firmdummies þ Yeardummies þ εi; j;t

ð8Þ

The dependent variables AB_ANY and AB_END are our measures of analyst target price bias. The independent variable BSI_V and BSI_N are investor beliefs measures in the company level, computed by the two-stage-least-squares model. The independent variable TURN, NIPO, and RIPO are investor beliefs measures in the aggregate market level. TURN is the monthly share turnover rate prior to any announcement of targets, NIPO is the yearly number of IPOs, and RIPO is the yearly average first-day returns of IPOs. Analyst specific measures contain two variables: the number of analysts issuing targets toward the same company when any target price is issued (COMPETE) and the interaction terms of BSI_V (BSI_N) and dummy variable of domestic analyst (DorF). Company specific variables include standard deviation of daily return of each stock (VOL), natural log of total assets (SIZE), market-to-book ratio at the announcement date (MB), price-to-earnings ratio at the announcement date (PE). Firm dummies and Year dummies denote firm- and industry-specific dummy variables. Newey and West (1987) autocorrelation- and heteroskedasticity-consistent standard errors are reported in parentheses.

AB_ANY Equations Intercept BSI_V

(1) −2.7259 (6.7708) 0.7553** (0.3124)

BSI_N Turn NIPO RIPO COMPETE

0.1859*** (0.0451) 0.2726** (0.1332) 0.1099*** (0.0174) 0.2340

AB_END (2) −2.6997 (6.7756)

0.7452** (0.3090) 0.1858*** (0.0451) 0.2727** (0.1332) 0.1099*** (0.0174) 0.2340

(3) 23.6041 (18.5863) 1.6716*** (0.5409)

−0.1801 (0.1635) 1.0600** (0.4606) 0.8273*** (0.0838) −0.6304

(4) 23.6849 (18.5886)

1.6472*** (0.5344) −0.1802 (0.1635) 1.0603** (0.4606) 0.8273*** (0.0838) −0.6303

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

251

Appendix C (continued) (continued) AB_ANY Equations BSI_V × DorF BSI_N × DorF VOL SIZE MB PE Firm dummies Year dummies Adjusted R2 F-value Observations

AB_END

(1)

(2)

(3)

(4)

(0.2301) −0.2830*** (0.0780) 3.0519** (1.5414) −1.0288 (0.8350) 0.0230 (0.2297) 0.0246 (0.0374) −2.7259 (6.7708) Yes Yes 0.3143 4.09*** 8256

(0.2301) −0.2805*** (0.0774) 3.0530** (1.5420) −1.0287 (0.8351) 0.0221 (0.2299) 0.0247 (0.0374) −2.6997 (6.7756) Yes Yes 0.3143 4.09*** 8256

(0.6187) −0.6560** (0.2613) 7.3506*** (2.6633) −5.6342*** (1.3303) 4.1267*** (0.8738) 0.3433*** (0.1122) 23.6041 (18.5863) Yes Yes 0.2581 6.48*** 8256

(0.6187) −0.6495** (0.2593) 7.3529*** (2.6638) −5.6346*** (1.3304) 4.1236*** (0.8738) 0.3435*** (0.1122) 23.6849 (18.5886) Yes Yes 0.3143 4.09*** 8256

References Asquith, P., Mikhail, M. B., & Au, A. S. (2005). Information content of equity analyst reports. Journal of Financial Economics, 75, 245–282. Bacmann, J. F., & Bolliger, G. (2002). Who are the best? Local versus foreign analysts on the Latin American stock markets. European Financial Management Association 2002 London Meetings. Bagnoli, M., Clement, M., Crawley, M., & Watts, S. (2009). The profitability of analysts' stock recommendations: what role does investor sentiment play? Working paper. University of Texas at Austin. Baik, B., & Jiang, G. (2006). The use of management forecasts to dampen analysts' expectations. Journal of Accounting and Public Policy, 25, 531–553. Baker, M., & Stein, J. (2004). Market liquidity as a sentiment indicator. Journal of Financial Markets, 7, 271–299. Baker, M., & Wurgler, J. (2000). The equity share in new issues and aggregate stock returns. Journal of Finance, 55, 2219–2257. Baker, M., & Wurgler, J. (2004a). A catering theory of dividends. Journal of Finance, 59, 1125–1165. Baker, M., & Wurgler, J. (2004b). Appearing and disappearing dividends: The link to catering incentives. Journal of Financial Economics, 73, 271–288. Baker, M., & Wurgler, J. (2006). Investor sentiment and the cross-section of stock returns. Journal of Finance, 61, 1645–1680. Baker, M., Greenwood, R., & Wurgler, J. (2009). Catering through nominal share prices. Journal of Finance, 64, 2559–2590. Bernhardt, D., Campello, M., & Kutsoati, E. (2004). ). Analyst compensation and forecast bias. Working paper. Tufts University. Bilinski, P., Cumming, D., Hass, L., Stathopoulos, K., & Walker, M. (2015). Strategic distortions in analyst target prices in the presence of short-term institutional investors. Working paper,. City University London. Bilinski, P., Lyssimachou, D., & Walker, M. (2013). Target price accuracy: international evidence. Accounting Review, 88, 825–851. Boni, L., & Womack, K. L. (2006). Analysts, industries and price momentum. Journal of Financial and Quantitative Analysis, 41, 85–109. Bonini, S., Zanetti, L., Bianchini, R., & Salvi, A. (2010). Target price accuracy in equity research. Journal of Business, Finance and Accounting, 37, 1177–1217. Bradshaw, M. T. (2002). The use of target prices to justify sell-side analysts' stock recommendations. Accounting Horizons, 16, 27–41. Bradshaw, M. T., Brown, L. D., & Huang, K. (2013). Do sell-side analysts exhibit differential target price forecasting ability? Review of Accounting Studies, 18, 930–955. Brav, A., & Lehavy, R. (2003). An empirical analysis of analysts' target prices: Short-term informativeness and long-term dynamics. Journal of Finance, 58, 1933–1967. Brown, N. C., Christensen, T. E., Elliott, W. B., & Mergenthaler, R. D. (2009). Do managers use pro forma earnings disclosures to cater to investor sentiment? Working paper,. University of Southern California. Chang, C. (2006). Information footholds: Expatriate analysts in an emerging market. Working Paper. University of California at Berkeley. Clarkson, P., Nekrasov, A., Simon, A., & Tutticci, I. (2015). Target price forecasts: Fundamental and non-fundamental factors. Working paper. Simon Fraser University. Cooper, R. A., Day, T. E., & Lewis, C. M. (2001). Following the leader: A study of individual analysts' earnings forecasts. Journal of Financial Economics, 61, 383–416. Cowen, A., Groysberg, B., & Healy, P. M. (2006). Which types of analyst firms are more optimistic? Journal of Accounting and Economics, 41, 119–146. Demirakos, E. G., Strong, N. C., & Walker, M. (2010). Does valuation model choice affect target price accuracy? The European Accounting Review, 19, 35–72. Ertimur, Y., Sunder, J., & Sunder, S. (2007). Measure for measure: The relation between forecast accuracy and recommendation profitability of analysts. Journal of Accounting Research, 45, 567–606. Gleason, C., Johnson, W., & Li, H. (2013). Valuation model use and the price target performance of sell-side equity analysts. Contemporary Accounting Research, 30, 80–115. Hong, H., & Kubik, J. D. (2003). Analyzing the analysts: career concerns and biased earnings forecasts. Journal of Finance, 58, 313–351. Hribar, P., & McInnis, J. (2012). Investor sentiment and analysts' earnings forecast errors. Management Science, 58, 293–307. Irvine, P. (2004). Analysts' forecasts and brokerage-firm trading. Accounting Review, 79, 125–150. Jackson, A. (2005). Trade generation, reputation and sell-side analysts. Journal of Finance, 60, 673–717. Jegadeesh, N., & Kim, W. (2006). Value of analyst recommendations: International Evidence. Journal of Financial Markets, 9, 274–309. Ke, B., & Yu, Y. (2009). Why don't analysts use their earnings forecasts in generating stock recommendations? Working paper. Nanyang Technological University. Kerl, A. G. (2011). Target price optimism. BuR-Business Research, 1, 74–96. Kumar, A., & Lee, C. M. C. (2006). Retail investor sentiment and return comovements. Journal of Finance, 61, 2451–2486. Lai, R. (2005). A catering theory of analyst bias. Working Paper, the Wharton School. University of Pennsylvania. Lai, S., & Teo, M. (2008). Home biased analysts in emerging markets. Journal of Financial and Quantitative Analysis, 43, 685–716. Lim, T. (2001). Rationality and analysts' forecast bias. Journal of Finance, 56, 369–385. Mittendorf, B., & Zhang, Y. (2005). The role of biased earnings guidance in creating a healthy tension between managers and analysts. Accounting Review, 80, 1193–1209. Newey, W. K., & West, K. D. (1987). A simple positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55, 703–708. Orpurt, S. F. (2006). Local analyst earnings forecast advantages in Europe. Working paper. University of Chicago. Rajgopal, S., Shivakumar, L., & Simpson, A. (2007). A catering theory of earnings management. Working paper. University of Washington. Sonney, F. (2009). Financial analysts' performance: Sector versus country specialization. Review of Financial Studies, 22, 2087–2131.

252

A.-S. Chen et al. / International Review of Economics and Finance 44 (2016) 232–252

Stickel, S. E. (1990). Predicting individual analyst earnings forecasts. Journal of Accounting Research, 28, 409–417. Stickel, S. E. (1992). Reputation and performance among security analysts. Journal of Finance, 47, 1811–1836. Stickel, S. E. (1995). The anatomy of the performance of buy and sell recommendations. Financial Analyst Journal, 51, 25–39. Bae, K. H., Stulz, R. M., & Tan, H. (2008). Do local analysts know more? A cross-country study of the performance of local analysts and foreign analysts. Journal of Financial Economics, 88, 581–606.