Loss sensitive investors and positively biased analysts in Hong Kong stock market

By offering high quality and up-to-date investment analyzes and recommendations to individual and institutional clients, brokerage houses and investment banks can charge significant fees for the access to accurate research reports which are of benefit to the customers. Beside this, one can expect these investment recommendations to generate transaction profits for the brokerage houses as clients will implement the suggested trading strategies and therefore buy or sell securities.

Unsurprisingly, the value and performance of analysts’ recommendations are constantly subjected to criticisms. A critical problem is the analysts’ constant exposure to potential conflicts of interests and hence the possible publication of over-optimistic stock recommendations that are not in the best interest of the investor. Michaely and Womack (1999), and Agrawal and Chen (2008) point towards this issue within financial institutions and suggest that frequent conflicts of interest between the research and the corporate finance department may result in analysts’ compromising their recommendations in order to maximize their company’s profits. This conflict of interest is especially high and leads to positively biased recommendations when they concern existing and potential corporate clients (Michaely and Womack 1999, 2005). And Chan et al (2018) find that analysts’ stock ownership induces analysts to bias upwards their target price forecasts. This study empirically investigates the relationship between equity performance, stock recommendations and market sentiment in the Hong Kong equity market.

It is not surprising that sell-side analysts are criticized for having a greater incentive to issue buy rather than sell recommendations because the former usually generates greater trading volume from clients (Lin and McNichols 1998). Firth et al. (2013) find that analysts’ recommendations on a stock are significantly higher if the stock is held by mutual fund clients of the analysts’ brokerage firm; however, their study focuses on the relationship between mutual funds and sell-side analyst recommendations. Michaely and Womack (1999) go even further and show that a positive recommendation issued after a company’s IPO may be necessary to increase the likelihood that the institution will be chosen as the next offering underwriter. Other studies argue that generally, analysts are more reluctant to provide negative outlooks for companies (Jegadeesh and Kim 2006; Michaely and Womack 2005; Barber et al. 2001). Aggarwal et al. (2018) document that when forming recommendations, analysts underestimate the role of the cost of equity.

As a result, the stock recommendations provided by analysts may be of lesser value than expected by the clients and might therefore result in unprofitable opportunities for investors. More specifically, it is questionable if investors can realize profits (or avoid losses) by purchasing upgraded securities (selling downgraded securities). Apart from that, market sentiment also plays an important role in the performance of an analyst’s stock recommendations. Baker and Wurgler (2007) give two definitions of investor sentiment: first as a driver of relative demand for speculative investments, and second as the investors’ collective optimism (or pessimism) about stocks in general. Furthermore, they also find that stock prices are subject not only to fundamental factors but also to market sentiment.

It is therefore understandable that analysts might consider current market sentiments when conducting their analysis and making recommendations. When issuing stock ratings based on the comparison between the firm’s expected future stock price and the firm’s current stock price, analysts will consider both the fundamental underlying factors, such as earnings, assets and liabilities, as well as investor sentiment such as past stock returns (Brown and Cliff 2004); the same applies when making recommendations (Bagnoli et al. 2009). In a more recent study, Li et al. (2015) show that analysts could facilitate price discovery to corporate news by issuing trending and contrarian revisions. Even though their study applies intraday data in the US, Li et al. (2015) do not consider if the market environment (such as bullish and bearish conditions) affect price discovery and reversal of recommendations.

As stated above, this study addresses the shortcomings of previous research and explores the empirical relationship between equity performance, stock recommendations and market sentiment between January 2009 and May 2014 in the Hong Kong equity market. The Hang Seng Index, Hong Kong’s major stock index, is known to be highly volatile which makes professional investors eager to obtain more timely information on corporate earnings. However, Hong Kong analysts tend to make drastic revisions since their interim reports are not frequent enough compared to the quarterly reports generated by US analysts with richer information (Cheng et al. 2003). Therefore, analyzing the effect of stock recommendations on equity prices based on Hong Kong equity market can provide a helpful environment to understand and seek for the answers the investors may have about the analyst recommendation effects. The Hong Kong Stock Exchange (SEHK) is Asia’s third largest stock exchange in terms of market capitalization and is placed seventh largest in the world; it is therefore of great significance to investors. SEHK operates two markets on which companies may choose to list their shares: the Main Board and the Growth Enterprise Market (GEM). The Hang Seng Index (HSI) is widely regarded as the market benchmark monitoring daily movements of the largest companies listed on the Hong Kong stock market and is the main indicator of the overall market performance in Hong Kong. The HSI is a free float-adjusted market capitalization-weighted stock market index and comprises the 50 major constituent stocks, therefore including 60% of the total market share of the Hong Kong Stocks Exchange (Bloomberg.com). By considering the market environment, we specifically test whether stock recommendation upgrades and downgrades would generate higher and lower profits, respectively under the given market sentiment. Answering these questions is important for decision makers in order to understand if the performance of potentially biased recommendations might be affected by short-term market atmosphere. Previous studies ignore these crucial considerations allowing us to fill a gap and provide vital insights into Asia’s third largest stock exchange in terms of market capitalization.

This study examines the short-term impact of stock recommendations on the 26 largest constituent weighted stocks of the Hong Kong stock market, representing together a market capitalization of more than three trillion US Dollars. Market conditions of the global financial context affect investors’ confidence and the quality of their recommendations; we therefore argue that analysts factor market sentiment into their forecasts and issue over-optimistic and/or over-pessimistic stock recommendations during different periods. In more bullish periods, analysts and investors may be over-optimistic about the future of a company, hence issuing buy recommendations leading to higher short-term returns as investors are more inclined to purchase the given security.

We investigate the short-term impact of those recommendations in bullish and bearish phases¹ by bootstrapping the stock market data and evaluating whether the analysts’ recommendations affect the overall market sentiment under different market conditions. We gauge investor confidence firstly by relying on surveys reflecting what investors and key actors in the Hong Kong stock market believe, and second, by considering the actual state of the economy, by examining daily news coverages. With this data in hand, two subperiods are generated: a bullish phase and a bearish phase. Papakroni (2018) finds strong evidence that the impact of analyst forecast dispersion is more pronounced in the group of stocks that receive the least favorable recommendations in each period.

This study makes significant contributions to the literature on recommendations of stock analysts. First, our study investigates whether variations in the market conditions affect the recommendations issued by analysts. This is done by investigating the performance of stock recommendations through different periods of time, considering bull and bear markets, in order to determine whether analysts’ recommendations are influenced by stock market atmosphere. If indeed analyst recommendations are influenced by market sentiment, stock forecasts will be cyclical and correlated with the economic climate. Previous studies, such as Jegadeesh and Kim (2006), Bagnoli et al (2009) and Womack (1996) ignore this issue, making their conclusions less comparable to our work. A second contribution is that our study examines all major recommendations by analysts, including downgrades, upgrades, and initiations to sell or to buy respectively. Most related studies are narrow and focus on specific recommendations, such as upgrade and downgrade (Ang et al. 1998); initiations of coverage (Irvine 2003); and secondhand analyst recommendations (Chan and Fong 1996). A third contribution is the robustness of our paper which analyzes the period before and after a recommendation. Previous studies focus on posterior effects of analysts’ recommendation (Moshirian et al 2009) and pay little attention to the market situation before the recommendation. Although Petaibanlue et al. (2015) examine the period before the recommendations; they only do it based on impact on IFRS adoption. Our study makes a vital contribution to market participants and researchers by analyzing both the pre and post recommendation period and use standardized abnormal returns. Finally, and perhaps the most original contribution stems from the fact that ours is the first study to examine the conditional effect of market sentiment on the analysts’ recommendations with and without look-ahead biases. We extend this line of research by taking into consideration the effect of the look ahead bias (hindsight bias) and provide an approximation for the error associated by this look ahead bias.

The rest of the paper is organized as follows: Sect. 2 presents the relevant literature on the performance and value of stock recommendations in different time frames and target markets. Section 3 illustrates the methodology of this study while Sect. 4 describes the data and the sample selection. Section 5 presents and analyzes the major empirical results with robustness tests. Then Sect. 6 concludes.

The extant literature on analysts’ recommendations related to investment information has developed in various aspects. Based on the comprehensive stock recommendations on U.S. stocks provided by fourteen major U.S. brokerage firms, Womack (1996) finds that the new added-to-sell recommendations are less frequent but more predictive than new added-to-buy recommendations. In other words, the analysts tend to perceive substantial risk in issuing sell recommendations since these are less frequent, more visible and contain more information compared to buy recommendations (Pratt 1993). Other recent studies (He et al 2020; Arif and De George 2020) have shown that analyst coverage has a negative impact on tax risk; and that loss sensitive investors are unable to completely eliminate the information loss arising from infrequent financial reporting which can affect the valuation of firms and the quality of financial markets. Consequently, analysts tend to herd more when the firm has negative media sentiment (Frijns and Huynh 2018). Barber et al. (2001) find results consistent with Womack (1996) and show that investors can generate significant positive (negative) abnormal returns by following analyst’ upgrades (downgrades) on securities, though abnormal loss provision generally reduces analyst coverage particularly in the banking sector (Hong et al 2020). Loh (2010) examines why the information conveyed by stock recommendations doesn’t get fully incorporated in the stock price when recommendations are released. He finds that the investor inattention contributes to a larger stock recommendation drift and that the subsequent drift reflects investors’ gradual realization of the true stock price implied by the published recommendation.

Green (2006), Chen and Cheng (2006) and Kadan et al. (2018) find that market participants with early access to recommendation information tend to earn higher profits than their peers without early access. Green (2006) argues that the trading profits increase with early access to both buy and sell recommendations while the latter shows larger increase by selling short following downgrades. Chen and Cheng (2006) maintain that stock recommendations contain valuable information and institutional investors under soft dollar arrangements can have timely access to these recommendations which contribute to their superior performance. Kadan et al. (2018) find that the early-informed news-driven institutions tend to accumulate wealth transferred from program traders who have less early information access advantages. Francis and Soffer (2006) find that both earnings forecast revisions and stock recommendations are significantly related to the market reactions to analysts’ reports and neither of these signals about the share values subsumes the other. According to Lee et al. (2018), however, the firms voluntarily issuing CSR-related reports tend to have richer information environment which reduces the value of analyst’s recommendations.

Investor sentiment serves as another important factor when analyzing stock recommendations. Bagnoli et al. (2009) and Corredor et al. (2013) find that analysts tend to issue more favorable stock recommendations when recent or future investor sentiment are optimistic. Bagnoli et al. (2009) also show that investor sentiment sensitive analysts tend to issue relatively less profitable stock recommendations while analysts who issue investor sentiment sensitive stocks offer more profitable recommendations than their peers. Chen and Matsumoto (2006) find that analysts issuing more favorable recommendations experience a greater increase in their relative forecast accuracy compared with analysts with less favorable recommendations. Furthermore, Barber et al. (2010) argue that the abnormal stock returns are the greatest conditional on the ratings change to buy and strong buy recommendations than to holds, sells and strong sells. Conditional on the ratings level, they find upgrades earn the highest returns and downgrades the lowest.

Evidence of impact from stock recommendations outside the US market has also been established. Focusing on the United Kingdom, Ryan and Taffler (2006) find that the price reaction in the UK market was similar to that of the US market in which the price impact of new sell recommendations was greater than that of new buy recommendations (Barber et al. 2007) but less susceptible to potential conflicts of interest than in the US. They find that the abnormal returns are associated with a firm’s information environment and analyst incentives.

Offering insights into an Asian country, Ang et al. (1998) evaluate the ability of four individual anonymous brokerage houses in Singapore to make buy recommendations from 1990 to 1992. They find that out of the four brokerage houses under research, only the buy recommendations of one firm had significant impact on the date of the recommendations. Surprisingly, by accounting for the effect of transactions costs, all positive abnormal returns were eliminated; implying that investors can’t earn positive abnormal return in excess of transactions costs by following the brokerage analyst’s recommendations.

Chan and Fong (1996) focus on whether the publication of an individual investors’ sentiment survey in Securities Firms’ Investment Analysis column (SFIA) of the Hong Kong Economic Journal (HKEJ) newspaper temporarily affects stock returns in the Hong Kong stock market. They suggest that the reporting of the recommendations in SFIA does induce trading, and there is also a heavy trading in the day just before the publication of SFIA. However, their study only focuses on the predictive ability for stock returns. Later, Chan and Fong (2004) investigate how stock prices are temporarily affected by SFIA. They find that the publication does not really predict the coming week’s return on large, medium, or small stocks while the daily closing prices for medium and small stocks are affected but not the large stocks. However, the SFIA publishes only the second-hand analysts’ recommendations not the first-hand ones which are usually provided to securities firms’ clients. Our paper uses the J.P. Morgan Hong Kong Investor Confidence Index, a quarterly investors’ sentiment index, and use the top 26 major stocks listed in the Hang Seng index to analyze the impact of analysts’ recommendations on the Hong Kong stock market. We then investigate the investors’ reactions to analysts’ recommendations based on investment sentiment with conditional probability applications and consider the effect of look-ahead bias.

In order to analyze the performance of stock recommendations, we measure the abnormal return of the stocks linked to the release of recommendation upgrades or downgrades initiations (see Liu et al. 1990). The abnormal return for security i on event day t is defined as:where R_i,t is the daily total return (including dividends) of security i for day t and R_m,t is the daily total return (including dividends) of the benchmark index in the Hong Kong market: the Hang Seng index.

In addition, the normal return of security i is calculated, while \({\widehat{\alpha }}_{i}\) and \({\widehat{\beta }}_{i}\) are the ordinary least square (OLS) estimates for the model parameters of security i obtained by regressing the total return of security i against the total return of the index. Moreover, the market model is estimated over the observations for 100 days,² while t = 0 is the release of the stock recommendation upgrades or downgrades initiations obtained from Bloomberg. Since the abnormal return of each security is defined, the average daily abnormal return of each day t within the event window, AAR_t, is calculated as:where t =  − 5, − 4, − 3 \(\cdots\), 0,\(\cdots\) 8, 9, 10.

The average cumulative daily abnormal return between t = t₁ and t = t₂, defined as \({ACAR}_{{t}_{1},{t}_{2}}\), is calculated as:

We test the statistical significance of the abnormal return for security i on day t by standardizing the returns as:where \({S}_{i}\) is the residual standard deviation for stock i by using the estimation of the market model from t =  − 5 to t = 10.

The average standardized daily abnormal return for each day of the event window (t =  − 5, − 4, − 3 \(\cdots\), 0,\(\cdots\) 8, 9, 10), \({ASAR}_{t}\), is calculated by averaging the \({SAR}_{i,t}\) of all securities:where t =  − 5, − 4, − 3 \(\cdots\), 0,\(\cdots\) 8, 9, 10.

Furthermore, the t-statistics of average abnormal return for each date t is computed by:

We apply the 2-tailed t-statistics test to determine whether or not the abnormal return and the cumulative return are statistically significant for each day t. Specifically, if the t-statistics calculated lie outside the range of the distribution depending on a particular level of statistical significance (e.g. 5% significance level), then we should reject the hypothesis that the return is equal to 0 (i.e. the return is statistically significant).

There are other recent studies of analyst recommendations that also used a single factor model to estimate abnormal returns (see Amiram et al. 2016; Du et al. 2017; Kudryavtsev 2019; Kim et al. 2019). But, in a later section of our paper, we also estimate abnormal returns using the Fama & French 3-factor model (Eq. (7)) which has been used in some papers in the area (e.g., Rubanov and Nnadi 2018; Su et al. 2019; Li et al. 2020).

\({R}_{f,t}\) is the risk-free rate, \({\widehat{\alpha }}_{i}\) is the intercept and \({\widehat{\beta }}_{i}\) is the coefficient estimate of the excess market return \(\left({R}_{m,t}-{R}_{f,t}\right)\).\(\widehat{{s}_{p}}\) and \(\widehat{{h}_{p}}\) are the coefficients for the difference in our total sample Hong Kong stock portfolio p between small and big stocks (SMB_t) and high and low book-to-market stocks (HML_t) at time t, respectively. We produce the size (SMB_t) and book-to-market equity (HML_t) factors as follows. We sort our sample stocks into two size groups based on the median market capitalization (i.e., median size), denoted by S (small) and B (big). We then make another group by sorting our sample stocks into three groups of book-to-market equity values based on the thresholds for the bottom 30 percent (L, for “low”), middle 40 percent (M, for “medium”) and top 30 percent (H, for “high”). SMB_t is the difference between the average returns on the three small stock portfolios (S/L, S/M and S/H) and the average returns on the three big stock portfolios (B/L, B/M and B/H) at time t. HML_t is the difference between the average returns on the two high book-to-market equity portfolios (S/H and B/H) and the two low book-to-market equity portfolios (S/L and B/L) at time t. We repeat the steps from Eqs. (2) to (6) using our Eq. (7) instead of (1) to perform our event study.

Finally, complementary to the above analysis, we study the impact of the previous quarter’s market sentiment when an analyst makes an ‘upgrade’, ‘downgrade’ or a ‘hold’ recommendation by employing a conditional probability model based on a binomial framework. In particular, we seek to find out what would be the conditional probability of an upgrade/hold/downgrade of a given stock, if the investor sentiment is positive/negative in the last quarter. We denote \({rec}_{i,t}\) as the analyst recommendation i = {upgrade, downgrade, hold} at time t. \({sen}_{j,t}\) as the investor sentiment j = {positive, negative} at time t. Then we express these in terms of probabilities, and use the cumulative data from one quarter to the next: The analyst recommendation probability is the cumulative recommendations from one quarter to the next, for example as from time s to time t. Similarly, the investor sentiment probability is the change in investor sentiment indices from one quarter to the next, also from time s to time t. (Please refer Panel D of Table 8 for a detailed methodology of the estimation). This investment sentiment probability is the normalized percentage change in investor sentiment index relative to the average sentiment. We express the combined effect between these analyst recommendation \({rec}_{i,t}\) and investor sentiment \({sen}_{j,t}\) in the following conditional form. We use the expected (i.e., average) value form for this as below.

We then find the expected stock return from the analyst’s recommendation conditional on the investor sentiment by multiplying Eq. (8) by the corresponding stock return (\({r}_{t})\).

For example, when investigating an analyst upgrade recommendation with positive investors sentiment at time t = 2, the following Eq. (10) expresses this case.

Then the corresponding expected stock return can be calculated by multiplying Eq. (10) by the stock return at time t = 2.

However, this analysis in this current from suffers from a look-ahead bias as the investor sentiment in the current quarter would only be available in the following quarter. To avoid this bias, we assume that the best proxy for the current quarters’ sentiment value is the previous quarters sentiment value (which is a known figure in the current quarter) thus implicitly assuming the investor sentiment process is a pure martingale.³

In the financial market, stock recommendations issued by different brokerage houses and investment firms may take different forms. Generally, based on their independent analysis and predictions on the covered firms and companies, sell-side analysts would usually issue one of the following statements: 1. strong buy; 2. buy; 3. hold/neutral; 4. sell and 5. strong sell.⁴

Our sample comprises 50 actively traded constituent stocks in the Hong Kong stock market. The Hang Seng Index is the free-adjusted market capitalization-weighted stock market index in Hong Kong which serves as the main indicator of the overall market performance of the local stock market. Although the constituent stocks of the index have been regularly reclassified, our analysis focuses on the top 26 major stocks in terms of index weighting which covers more than 80% of market capitalization of the Hang Seng Index.

The sample period is defined as the period between the January 1, 2009 and the May 1, 2014 when the market experienced significant waves in trading. All sample recommendations of those 26 stocks are selected using Bloomberg. Companies that are listed within the sample period are considered to be stocks that have proven to be actively traded with significant market capitalization and turnover. In addition, all stocks have been covered by at least one analyst throughout the whole sample period. Delisted firms are excluded from the sample.

For those 26 target stocks, a total of 1751 recommendations from 402 analysts have been obtained within the sample period, consisting of 884 recommendation upgrades and 867 recommendation downgrades. Details of the data are shown in Table 1. However, in order to ensure consistency in the comparison of the effects of different stock recommendations, we exclude all recommendations that were released on weekends or holidays.

To derive the conditional probability statements based on the investor sentiment we use the J.P. Morgan Hong Kong Investor Confidence Index which is a survey-based index reflecting the investors’ sentiment over the next six months. Over 500 investors aged between 30 and 60 participate in the survey and have a minimum of HK$100,000 in liquid assets and at least 5 years of continuous investment experience.

The J.P. Morgan Investor Confidence Index (JPMICI) score is derived by asking survey respondents six questions to assess the confidence of investors about (Q1) the Hang Seng Index, (Q2) HK economic environment, (Q3) HK investment environment and sentiment, (Q4) global economic environment, (Q5) the possibility of personal asset appreciation, and (Q6) the possibility of increasing their investment. These six questions form the sub-indices of the J.P. Morgan Investor Confidence Index. The index and all sub-indices have a range between 0 and 200. A number greater than 100 represents a positive outlook and vice versa. The Index is designed to reflect local investor sentiment towards the Hong Kong market over the next 6 months following its last date of publication. It is fair to assume that the JPMICI index is a representation of investor confidence in the HK market. As such, the semi-annual changes of the index value will reflect the perceived changes in investor sentiment. We have attached associated probability values of these changes in order to facilitate a richer analysis using a Bayesian framework. Assessing changes through probability values within a Bayesian framework has been practiced in the finance literature before (see Beaver 1966, 1968). We are certainly not the first to work with sentiment indices (see Huang et al. 2015; Lee et al. 2002, where the authors use the Baker and Wurgler 2006 Index). However, the Baker and Wurgler (2006) Index does not cover Hong Kong during our period of interest, so we opted to use the JPMICI Index.

This study examines the short-term price impact of stock recommendations issued by sell-side analysts on 26 major constituent stocks of the Hang Seng Index, the market benchmark monitoring daily movement of the largest companies listed on the Hong Kong Stocks Exchange. Numerous academic studies show that stock analysts play an important role in the stock market in which their recommendations deliver substantial information to the investors, resulting in different levels of price impact.

We analyze the Hong Kong stock market reactions from analysts’ stock recommendations (1696 upgrades and downgrades, and 504 initiation coverages) and consider the ongoing market sentiments. We find that the analysts’ upgrade recommendations result in significantly positive market reactions around the announcement day since these deliver newly favorable information to the investors. Similarly, we also find significantly negative market reactions to analysts’ downgrade recommendations, which shows even larger reactions in absolute figure compared to the market reactions to analysts’ upgrade recommendations. Therefore, the investors are more sensitive to losses than to gains consistent with prospect theory by Kahneman and Tversky (1979). On the other hand, the market reactions to initiation coverage with buy and sell recommendations are insignificant since these do not deliver enough information to the market or the market may reflect all this related information too quickly, contrary to Branson et al. (1998) and Irvine (2003). Then we assess the short-term impact of analysts’ recommendations under two bullish and bearish market conditions. We find that the investors react more optimistically to upgrade recommendations during bullish phase but react even more pessimistically to downgrades during bearish phase confirming more loss sensitive investors’ behaviors.

We apply the conditional probability framework to the analyst recommendations conditioned on the investor sentiment and estimate the corresponding expected stock returns. Our finding shows that with look ahead bias the differential returns when one is in a positive sentiment as opposed to a negative one, is significant, and that the differential returns between the two corresponding positive sentiments with and without look ahead bias also statistically significant. Our findings have implications to capital market participants as they suggest that exploitable trading strategies can be developed for the Hong Kong market with the information of the market sentiment in the past.

This study provides evidence in support to the fact that analysts’ recommendation upgrades and downgrades deliver permanent and significant information to the Hong Kong stock market. However, analysts’ initiation coverages convey little information and bring about limited impact to the stock market. In addition, analysts’ upgrades and downgrades result in significant price impact to different extents in the bullish and the bearish phases.