Impact of analyst’s target prices and stock
recommendations on the returns of the stocks
traded on the Stockholm Stock Exchange
Author: Mattias Holm
(930705)
Spring 2020
Independent Project II, Master Thesis, 15 credits Statistics
Örebro University, School of Business Supervisor: Olha Bodnar
Abstract
The study provides an interpretation of how target prices affect the day ahead asset returns. By using prices of the Large Cap company stocks on the Nasdaq Stockholm Stock Exchange, the return is predicted with the IV regression estimator, where the difference between the target price and stock price, denoted DrTP, is used as the primary exogenous variable whereby company information as instrumental variables. When comparing the IV models with regular OLS method and subset regression models, the performance is similar, and the IV approach does not seem to provide any additional explanatory degree. DrTP and an upgrade in the stock recommendation are though positively significant in several models, controlling for the Tokyo index Nikkei 225 and the US based index DJIA. Mentionable is that the stock recommendation seems to provide higher predictive accuracy than DrTP. This is indicated mainly by the subset regression result, where an upgrade in the recommendation on average implies an increase in the day-ahead stock return with 1.77 percent, controlling for Nikkei 225.
Keywords: Target price, stock recommendation, stock market, instrumental variables, subset regression
Acknowledgement: I want to thank my supervisor Olha Bodnar for guidance on the methodology process, data processing, and the dissemination of knowledge, together with being both very flexible and helpful.
Table of Contents
1. INTRODUCTION ... 2
1.1.PREVIOUS RESEARCH ... 3
1.2.TARGET PRICE AND STOCK RECOMMENDATION ... 4
2. METHODOLOGY APPROACH ... 5
2.2.ECONOMIC CONTROL FACTORS ... 7
3. SAMPLE AND DATA ... 9
4. MODELING ... 12
4.1.IVREGRESSION ... 12
4.2.EMPIRICAL MODELING ... 13
4.3. Subset Regression Approach ... 14
5. RESULTS ... 15
5.1.OLS REGRESSION ... 15
5.2.IV REGRESSION ... 16
5.3.SUBSET REGRESSION BY STOCK RECOMMENDATION ... 17
6. DISCUSSION ... 18 7. CONCLUSIONS ... 20 REFERENCES ... 21 APPENDIX 1 ... 22 APPENDIX 2 ... 25 RESULT IVCONDITIONS ... 25 ROBUST CHECK ... 26
1. Introduction
The stock market is known to be affected by future expectations of the market itself, as a consequence of human subjectivity, and not only as a result of quantitative, measurable statistics. These latent aspects, partly steering the way the market participants invest as one entity, is by nature hard to evaluate and fit into a relative perspective compared to more available measures such as revenue, profits, and growth. To what extent future expectations affect the stock price are naturally different for different events and market segments (Da & Schuamburg, 2011) and will probably shift over time. One way to understand the fluctuations in the market can be to predict these effects with future expectations as a function of available economic information.
One approach can be to consider target prices as future expectations, since it reflects a financial analysts' subjective estimate of the stock price level for some time in the future, usually twelve months (Bilinski et al., 2013). Consequently, the target price is not built only on quantitative measurements but also on the analysts' individual thoughts of the quantitative measurements and other possible external effects. Further, studies have through the years tried both to evaluate how accurate the target price works as a predictor, both for the intended time perspective (Brav & Lehavy, 2003; Bradshaw et al., 2012; Bilinski et al., 2013) and to model portfolio asset return (Da & Schaumburg, 2011). Few have though examined the day ahead effect of the target price on stock returns (although some has seen it as a subpart in previous research papers, e.g., Brav and Lehavy (2003), Asquith et al. (2005)) that might cause a direct reaction on the market through changed future expectations. One could assume this will affect the price, as the price is a function of future expectations, or as Harrison & Kreps (1978) describe, that investors (analysts) exhibit “speculative behavior” and foresee that other parties are willing to pay a higher price for a stock in the future. In other words, if a target price - seen as new economic information - is available for actors in the market, this can affect the stock price in the near future.
Additionally, the globalisation, primarily with the created availability to live update, is certainly a market predictability disturbing factor, changing both short and long-term expectations. On proof that can be considered as a consequence is the effect changes in larger markets have on changes in smaller ones, as Wu and Su (1998) proclaims. This can be considered an explanatory factor for the daily change of return. Thus, this study aims to examine the one day-ahead effect of stock returns of analyst announcements of target prices, together with stock recommendation, where quantitative economic information both believed affecting the return direct and indirect through the target price will be considered.
In this study, price data of the Large Cap company stocks listed at the Nasdaq Stockholm Stock Exchange and the available target prices at the Swedish bank Avanza’s web page has been used. The return at time t is regressed on the difference between a target price at time t – 1 and the price at t – 1, denoted DrTP, using the IV regressor. Stock and company information is used as instrumental variables to predict the values of DrTP, whereas N225 and DJIA - index
representing large markets - are used as exogenous variables. This is compared with the regular OLS method and the subset regression, classified by the change in stock recommendation. The results demonstrate that the IV models provide similar prediction capacity as the regular OLS and the subset regression, with the stock and company information as instruments. However, DrTP and an upgraded stock recommendation (REC.UP) both provide significantly positive estimates of the day ahead return, whereby the indexes also provide significant results for the estimated parameters. Additionally, an upgrade in stock recommendation seems to be a better explanatory factor than DrTP, especially visualized as the result of the subset regressions.
The thesis continues with a presentation of previous research on the topic of target price and stock recommendation effects on the return, and a brief explanation of what the target price and stock recommendation are. This is followed by explaining the methodology approach and mapping of the economic factors that will be used to control for in the estimation process in section II. In section III, the data and data processing are thoroughly explained, supplemented by tables in Appendix 1. The models used in the empirical analysis is then presented in section IV, followed by the results of these models in section V, whereas the conditions for conducting IV regression is illustrated in Appendix 2. A summation of the results, drawbacks, and essential considerations for future similar approaches is presented in section VI, followed by conclusions in section VII.
1.1. Previous Research
One of the first comprehensive research papers within the field of target price is written by Brav and Lehavy (2003) and examined the short-term effect using analysts’ target price, stock recommendation, and earning forecast. They used over 200 000 released targets price together with over 100 000 stock recommendations within the period 1997 to 1999 The authors found that the target price can explain the future market return, both conditionally and unconditionally on stock recommendation and earning forecast. The authors also conclude through there results that the revision in target price influences the returns up to a six-month subsequent period. Even though this might not be direct relevant aspects to consider when setting a model for the return one day-ahead a target price announcement, the results highlight that the target price is overall considered by the market. As a complementary result, the authors regress the return of the release day on target price and earnings forecast, controlling for change in stock recommendation between three categorical level (hold, buy, strongly buy). They conclude that the intercept, which can be interpreted as the effect of the change in recommendation, is significant on a ten percent level for all six possible combinations of change (hold to buy; buy to hold, etc).
By collected information from equity analyst reports, Asquith et al. (2005) filter out the earnings forecast, together with target price and stock recommendation revision information from the period 1997 to 1999. The authors discovered that the change in target price (from the last prediction) is greater than an equal percentage change in earnings forecasts. Further, the
target price also contained information outside the stock recommendation in terms of increased adjusted R2 when the target price revision is added to the different tested models. The authors
also evaluate the change in stock recombination change in a similar way as Brav and Lehavy (2003), but examine the revision as an upgrade, downgrade, or reiteration, and find significant positive effect on the five days adjusted abnormal return. However, the target price change, used as an explanatory variable among others in the applied OLS regression, show no significant result for the estimated parameter. For the downgrade and reiteration, a significant result was presented for the target price parameter, but not for the categorical stock recommendation. These results indicate that the target price and stock recommendation contribute to similar information for the investors at the market, where an upgrade seems more effective in signalizing a bright future relative to a reiteration and downgrade, and contrary to the target price.
Da and Schaumburg’s (2011) study the short-term effect of a target price announcement in monthly periods and locates significant evidence for abnormal return both economically and statistically. With the so-called “TPER strategy” (target price implied expected return in their paper), which involve exploitation off mean reversion (i.e., buy recent loser and sell recent winners), the authors use the monthly average target price and the market price from the day of the portfolio formation date. Also, they sort the stocks by the sector division of GICS (Global Industry Classification Standards) into nine portfolios using different classification variables, there among TPER. The main conclusion is that the explanatory factor of the analyst target prices is the relative sector valuation and not the individual forecasts themselves. Also, the target price is statistically significant predicting the subsequent monthly return. Other included explanatory variables, such as previous monthly return, book-to-price ratio, earning-top-price ratio, and log market capitalization did not end up significant, identified through estimation with linear regression. Compared to Da & Schaumburg, where TPER is used as a classifier for portfolios to control for the subsequent monthly return, the percentage difference between the price and the target price (DrTP) is used as an explanatory variable of the return in this study.
1.2. Target Price and Stock Recommendation
As a price evaluation of stock, analysts in the financial business set a target price, which is an estimate of the future valuation. This evaluation is made for the value of the stock somewhere around 12 months forward (Bilinski et al., 2013), where the analyst set the price often using both quantitative measures together with unmeasurable information and future expectations. Due to this, both human assessment and actual data are included factors of the target prices and assumptions drawn from this can be many. One possible assumption is that the short-term (one day-ahead in this study) implication on the stock price change is unequal zero as a new target price for the specific stock is released. Additionally, the target price is almost always announced together with a categorical stock recommendation from the analyst of either buy, hold, or sell1. How these two measures might affect the stock price, and additionally how they
1Strongly buy and strongly sell at both ends of the categorical scale are often possible recommendations as well but will be left out in this study due to that the three categories buy, hold, and sell often are the ones available.
and the stock price can be affected by other economic information, are presented and unified into models of the return in the next section.
2. Methodology Approach
A walkthrough of the theoretical approach for using target price as an explanatory factor of short-term stock price change is demonstrated in this section, starting with explaining the return followed by fitting the target price into modeling together with additional considerable economic information.
The return, denoted r, of a stock price will be used to evaluate the short-term stock price changes, as the return of stock is comparable between investment/trading opportunities and easier to handle from a statistical perspective (Tsay, 2010). Thus, let 𝑃! be equal to the adjusted closing price at time t and 𝑃!"# be equal to the adjusted closing price2 from the previous day. Then, the change in the stock price (the percentage return) is expressed as:
𝑟! =𝑃!− 𝑃!"#
𝑃!"# (1)
Further, we can assume that a price of a stock in time t can be explained by economic information at time t – 1. Notable is, as Harrison & Kreps (1978) discuss, that different information can be available for different investors at the market, but the decision to sell, hold or buy at a certain price will be based on information at time t - 1. Following this principle, a return at some point in the future, let say t + k, where k is a positive integer, depends on the available information before time t (i.e. at time t – 1). This information can, among others, be a function of previous returns and can, for example, be modelled as a time series (Tsay, 2010). However, the return in time t does not need to be modelled as a time series but still depends on time-relevant aspects relative to a specific observed return - which will be the approach in this study. The simplest model considered in this thesis will be whether a target price announced by an analyst will influence the corresponding stock return the following day. The target price set by analyst at time t – 1 can be assumed as a consequence of both available economic information and future expectations from the analyst. If the analyst changes the target price from their previous one, it is either since the stock price has changed or that the analyst has changed the valuation of the stock, or both. Further, a target price adjustment might affect the stock price due to that the receiver of this information (the market participants) consequently changes their own valuations. Using the percentage difference between the target price and stock price, DrTP, therefore seems to be reasonable both for the same statistical advantages as using the return instead of price, as well as being more informative than the absolute value (or absolute difference) since the analysts often rely on relative valuation models (Da & Schaumburg, 2011).
To consider is that the announcements of target prices are irregular, and the construction of the data matrix is therefore relatively more challenging. For following in (2), by the index 𝑖 we denote the 𝑖-th line in the data matrix, and let 𝑇𝑃$!,&!,!!"# be the target price announced by analyst 𝑎' for stock 𝑠' at time 𝑡' − 1. Further, let 𝑃&!,!! denote the observed price of stock 𝑠' at time 𝑡' on the stock exchange market. Because the price processes are not stationary, the corresponding returns are defined by:
𝑟' =𝑃&!,!! − 𝑃&!,!!"#
𝑃&!,!!"# 𝑎𝑛𝑑 𝐷𝑟𝑇𝑃' =
𝑇𝑃$!,&!,!!"#− 𝑃&!,!!"#
𝑃&!,!!"# (2)
To be more specific, the indexes t when comparing, for example, 𝑟# and 𝑟( do not imply that the return 1 and 2 are observed at the same day. However, the same stock, let’s say AAK, can receive several target price announcements the same day, but in that case from different analysts. As such, 𝑟# and 𝑟( can both be returns of AAK from the same day, i.e. the same observed return, but the target price attached to it will thus come from different analysts. For an explicit example in the data matrix, see date 13/02/2020 in Table A1.1 in Appendix 1. Since a target price releases together with new economic information, the market participants will change their future expectations (or keep it the same) of what other participants are willing to pay for the stock. If this effect is sufficiently significant, the target price relative to the price should be a valid predictor of the future stock returns. For the empirical estimation, the base in the modeling of the return will be DrTP of the previous day of an observed return, and can in the simplest form without any other explanatory factors be written as:
𝑟' = 𝛽)+ 𝛽# 𝐷𝑟𝑇𝑃'+ 𝜀' (3)
where 𝛽) is an intercept, 𝛽# is the slope coefficient for 𝐷𝑟𝑇𝑃', and 𝜀' is the error term assumed to be a white noise process with 𝑁(0, σ(). Model (3) can be extended with other/different analyst information that can affect the return. Asquith et al. (2005) demonstrate that including the change of the analysts target price increase the fit of their regression model, compared to when using the discrete recommendation alone (in Asquith et al. (2005): strongly buy, buy, hold, sell, and strongly sell). Thus, the stock recommendation (REC) can be included in the model to control for whether the analysts has changed opinion on the trading action, as the target price might not include that information (at least not directly), and rather measure the magnitude that the recommendation have and consequently provide different information to a model (Bradshaw et al. 2013, Bray and Lehavy, 2003). REC will be as a categorical variable and is included to support the estimation of how “strong” the new valuation is of the stock price from the analyst perspective. For clarification, suppose that the target price is significantly higher than the actual stock price, then the probability that the stock recommendation is buy can be assumed higher than when the difference is small between the target and the stock price, ceteris paribus. However, if the previous recommendation (least the same analyst released a target price and stock recommendation) also was buy, the perceived valuation change might be
lower than if the previous stock recommendation was hold or sell, ceteris paribus. A model for the return can be expressed as followed:
𝑟' = 𝛽)+ 𝛽#𝐷𝑟𝑇𝑃'+ 𝛽(𝑅𝐸𝐶' + 𝜀' (4) where the 𝛽’s are the slope coefficients for the respectively variable and REC is a categorical variable of either, Buy, Hold, or Sell. The two variables DrTP and REC will in this study be viewed and used as the analyst information, that both can affect and be affected by economic information.
2.2. Economic Control Factors
As the target price and stock recommendation built on economic information and future expectations are addressed above, the next step is to incorporate the economic factors (information) into the modelling process. These factors are divided into two segments - those variables that are assumed to influence and those assumed not to influence the target price and stock recommendation.
Starting with the former, analyst information can be influenced by variables available before the target price is released. As previous mentioned, diverse time series models can be used to explain the variation of the stock market, where one well known method is to use the negative first-order serial correlation of monthly returns as Jegadeesh (1990), where the “TREP”-strategy by Da & Schaumburg (2011) relates to this terms of time (monthly evaluation) and the usage of mean reversion. As such, the average return 𝑟̅' of stock 𝑠' is computed for a month prior to time 𝑡'. As Da & Schaumburg (2011) also discuss, there are problems with the TPER measure denoted by DrTP in the thesis, where the value is estimated and consequently there is a systematic forward-looking risk factor. Therefore, as described in Da & Schaumburg (2011), all DrTP cannot be compared with each other because of different risk characteristics. Thus, the sample variance 𝑣(𝑟') of stock 𝑠' computed for three months prior to time 𝑡' will be used to control the risk factor.
Further, Brad & Lehavy (2003) follows the classical Fama & French (1993) three-factor model, by controlling their portfolios for size risk (SMB, small minus big) and value risk (HML, high
minus row), based on the market capitalization3 and the book-to-market equity4, respectively.
Even though in this thesis the intents is to predict the day ahead return, and not use a portfolio approach, an effect on the on the actual return when a target price is released might be different due to risk factor, whereby the market capitalization (MCap) and Price/Book value (PB) (inverse of book-to-market), will be used as instruments. Fama and French (1993) also express that these two aspects relate to essential economic factors where companies with a low PB tend to have low earnings on their assets. As the low earnings seem to persist over time, one can
3 Stock price multiplied by the number of shares in the market.
4 Book-to-market is the ratio between the market capitalization and the book value, with is the total assets subtracted by liabilities, preferred shares, and intangible assets:
imagine this will be a factor to consider when setting the target price, and there through the return. Further, when controlling for the PB effect, the size, viewed as MCap, tends to have a positive effect on the asset earnings. As well as for the analyst information, these two factors will be used as explanatory variables when estimating the return. As the stock price is a component of the MCap and the PB, the latest available price, 𝑃&!,!!"#, will be used when
measuring the variables 𝑀𝐶𝑎𝑝' and 𝑃𝐵'. Due to the skewness of the observed market capitalization of the companies included, the MCap will be transformed into logarithm scale. Since the four variables discussed so far in section 2.2 are expected to influence the return and the target price, one can suspect that there might be correlation between the target price and the error term 𝜀' in the model (4). Under these circumstances, the OLS estimator will be biased and inconsistent whereby the IV (instrumental variable) regression estimator can be applied (Stock and Watson, 2012). The IV estimation process is a two-stage approach where the endogenous variable, DrTP, is regressed on all exogenous variables which are both the instrument variables, 𝑟̅, 𝑣(𝑟), MCap, and PB, and the assumed covariates in the model of the return, so far discussed REC. This is explained further in section VI, where the empirical models also are presented. Moreover, the fundamental model of the DrTP can be expressed as:
𝐷𝑟𝑇𝑃' = 𝜋#𝑟̅' + 𝜋(𝑣(𝑟') + 𝜋*𝑀𝐶𝑎𝑝' + 𝜋+𝑃𝐵'+ 𝜐' (5)
Where the 𝜋’s are the parameters for their respective variable and 𝜐' is the error term of the model and assumed to be a white noise process with 𝑁(0, γ().
As the factors that potentially can influence the DrTP and there through the daily return now are addressed, there are still other economic information that probably have an impact as well. Since the market has become more globalized with time, one can believe that the outer effects such as stock market fluctuations can affect each other. One consequence is that larger markets can lead smaller market’s variabilities, which Wu & Su (1998) proclaim in their study, where the authors focus on dynamics between stock markets. The authors illustrate that the cross-correlation is greater between past returns of larger stock markets and returns of smaller stock markets than vice versa; that the cross-correlation between past returns of small stock markets and returns of larger stock markets are smaller. In this thesis, the Nasdaq Stockholm Stock Exchange (SSE) can be viewed as a small market, whereas the US and Asian market can be viewed as large according to Wu & Su (1998). Therefore, Dow Jones Industrial Average (DJIA) will be used to control for the US stock markets (New York Stock Exchange (NYSE) and NASDAQ), whereas Nikkei 225 (N225) will be used to control for the Japanese market (Tokyo Stock Exchange, (TSE)). As these stock markets have different opening hours compared to SSE5, the latest available price information before SSE opens at day 𝑡
', will for NYSE and NASDAQ be 𝑝!!"#, while for TSE be 𝑝!!. Hence, percentage change of the indexes DJIA and N225 will be used to control for the change in the foreign markets; NYSE,
NASDAQ, and TSE, for observation i at time 𝑡'. Adding these explanatory factors into (3), the model is expressed as:
𝑟' = 𝛽)+ 𝛽#𝐷𝑟𝑇𝑃'+ 𝛽(𝑅𝐸𝐶' + 𝛽*𝑁225' + 𝛽+𝐷𝐽𝐼𝐴' + 𝜀' (6) where 𝑁225',! is the index percentage change of N225 at time 𝑡' for observation i, and 𝐷𝐽𝐼𝐴' at time 𝑡' – 1 for observation i. One issue that arises is when groups in the data, in this case
different stocks, has different characteristics regarding variation and consequently are affected differently by the two indexes. This would imply that different stock should have different 𝛽* and 𝛽+. Further, by observing the beta coefficient in the classic CAPM model, one can identify those differences in variation. However, when evaluating these group factors (random effects, RE) with a robust check, the effect on the model is small and can be neglected and assumed constant6. The empirical models will be illustrated in section VI together with an explanation
of how the IV regression approach works.
3. Sample and Data
Most of the included data take its standpoint from the available target price data at the web page of Avanza for each stock listed as the Large Cap at SSE7. The target prices are located at
each of the individual stock’s pages at Avanza8 and collected manual, where the target price
data stretch from February-06 2020 until May-06 2020. There is a total of 398 observations of new and previous target prices, stock recommendations, as well as the analyst releasing the target price and stock recommendation. There are 25 missing values of either the new or previous target price (or both), whereby 373 values of DrTP are calculated over a 62 days period (open stock market days). The release of the target price is available before the Nasdaq Stockholm opens on several web sites (e.g., Avanza, Dagens Industri (DI)), and is therefore treated, notation wise, as released the day before the actual release date. The total number of stocks at the Large Cap list is 1349, whereas 87 stocks are included in this study, which is due
to the necessary and practical restrictions of that: no new stock recommendation or target price is released at Avanza’s web page during the time period; and that one stock per company is used. More explicit, when more than one stock is purchasable for one company, e.g. A, B or preference stocks, the recommendation and target price are set for the company in almost every case, and not for the specific stock. Hence, only one stock will be used per company, to avoid target price duplicates and the under/overestimation effect it might entail. A list of the included stocks is compiled in Table A1.3 in Appendix 1.
6 See Appendix 2 for the results of the robust check, controlling for random effects.
7 The criteria for being listed as the Large Cap is if the market capitalization value is more than one billion Euro during the previous trading day closing price: https://www.avanza.se/aktier/aktiegeneratorn/aktiegeneratorn-foerdjupning/borsvarde.html. The stock listed as Large Cap at SSE are found at:
https://www.avanza.se/aktier/lista.html
8 Example for AAK is: https://www.avanza.se/aktier/om-aktien.html/26268/aak 9 At date: 2020-05-16
To each of the observed target prices on the day of the announcement, the adjusted closing price to calculate the return, together with other relevant economic information, is entailed to these dates. These data are collected from Yahoo Finance web site, both manually and through
R Studio with the package tidyquant. From each specific stock page at Yahoo Finance, the
valuations measures Market Capitalization (MCap) and price-to-book values (PB) are gathered10. Since both these measures are a function of the stock price and only are assembled
for one date, MCap and PB for all necessary days are calculated by adjusting with the quotient of the opening stock prices and the “base” stock price (i.e., the price for the one date the MCap and PB where measured).
Other economic factors included in this study are indexes, which measure the average price for a selection of stocks. As discussed in the previous section, with respect to that large market’s changes can affect smaller market return (Wu and Su, 1998), the Dow Jones Industrial Average (DJIA) and Nikkei 225 (N22) indexes will be used to control for foreign daily changes. The DJIA index covers 30 large publicly owned companies11 listed on the NYSE or NASDAQ,
while Nikkei 225 covers 225 large publicity owned companies12 listed on the TSE. Descriptive
statistics for the return, target price, and economic information are compiled in the table below.
Table 1. Descriptive statistics for stocks in the Large Cap list at Nasdaq Stockholm and the indexes DJIA and N225, from 2020-02-06 to 2020-05-06, for the observed data.
Statistic Mean St. Dev. Median Min Max Num. Obs.
daily return, r 0.51 4.67 0.16 -17.96 16.55 398 DrTP, lag 1 19.79 26.96 16.30 -40.04 210.43 373 N225 -0.26 2.68 -0.52 -6.08 8.04 357 DJIA, lag 1 0.53 3.97 0.17 -12.93 11.37 394 MCap, lag 1 92.48 109.14 51.53 3.94 1 054.88 398 PB, lag 1 2.65 2.68 1.7 0.27 23.26 386
Average return one
month back, 𝑟̅ -0.17 1.21 0.10 -6.07 2.11 398
Variance three months
back, v(r) 11.96 8.80 10.34 0.98 53.35 398
Note: MCap in billions SEK, PB is a price SEK ratio, and the rest of the statistics are in percent. Variables denoted with lag 1 are the observed values one day before the return are observed. This is just for clarity, and the “lag 1” notation will not be used throughout the continuation of the text. For further clarification, this table does not consider that the observations can be the same for the variables, whereby the variable return statistics does not represent time series data from 2020-02-06 to 2020-05-06. The mean and median for the log MCap are 3.92 and 3.94, respectively.
10 Example of the valuations data page for AAK: https://finance.yahoo.com/quote/AAK.ST/key-statistics?p=AAK.ST
11 The included companies https://markets.businessinsider.com/index/components/dow_jones 12 The included companies https://markets.businessinsider.com/index/components/nikkei_225
As seen in the table above, the daily return is positive during the three-month period, February-20 to May-February-20, and the largest change upwards for one day is 16.55 percent. Focusing on DrTP, we see that the average difference between the target price and the actual price is almost 20 percent. The average return one-month back is slightly negative, -0.17, whereby the average change for N225 is positive and for DJIA negative. These measurements can be compared with the actual historical data of the index changes between the observed dates, and not the statistics for the observed data, as in Table 1 above. Namely, during the period between 2020-02-06 and 2020-05-06, the mean change for both N225, DJIA (lagged one), and the OMXS3013 are
negative, respectively -0.26, -0.25, and -0,27 percent. Comparing this with the mean return when removing duplicates of dates and companies14, 0.45, this indicates that the return is higher
for a company when a target price is released at the day before the observed return. This can also be an effect of the next up considered change in stock recommendation.
Looking closer into the stock recommendation data, it can take on three different categorical values; BUY, HOLD, or SELL. Due to the relative low number of observations, the change in stock recommendation will take three sorts of values, REC.UP, REC.RET, or REC.DOWN. Consequently, a change from sell to hold or buy will be treated the same, whereas a retained recombination of all three possibilities will be observed as RET. In the table below, statistics of the stock recommendations are presented, where most of the recommendations are repeated buy, 164. Overall, we can see that there is higher representation of upgraded than downgraded recommendations, more precise, a difference of 77.
Table 2. Stock recommendation and previous stock recommendation for stocks in the Large Cap list at Nasdaq Stockholm, from 2020-02-06 to 2020-05-06, for the observed data.
Previous recommendation
BUY HOLD SELL total
Recommendation
BUY 164 67 9 240
HOLD 30 43 48 121
SELL 0 17 20 37
total 194 127 77 398
REC.UP REC.RET REC. DOWN
total 124 227 47 398
Finally, together with the companies are their respectively Industry they mainly operate within observed and collected from Avanza’s web page. As Boni and Womack indicate in their study in 2006, the GICS are in line with the expertise of the most analysts, whereby some analysts can be better at estimating the price in a specific industry and, therefore, be more trusted by the market. Since the data covers the Large Cap list at SSE, there are eight of these industries included, but as there are overall few observations where the same analyst is making estimate
13 Index of the thirty most traded shares on the SSE.
14 With duplicates of date and company removed, 330 observations remain (as some several target prices have been released for the same company the same day.
announcements in the same industry, this factor will be left out in this study. However, a summary of the analysts and industries are compiled in Table A1.2 in Appendix 1.
4. Modeling
This section covers the statistical modeling and why it is suitable for the data and theoretical approach. This is followed by a presentation of the models applied to estimate the return.
4.1. IV Regression
For the modeling of the return, the Instrument variable (IV) regression, supplemented by the OLS, will be applied to several models to illustrate how the future expectations, measured as the target prices, can affect the estimations.
As for many extensions of a miss specified regression modeling of a dependent variable, Y (a vector of size 𝑛 x 1), the intension with IV regression is to obtain a consistent estimator when the explanatory variables, X (a matrix of size 𝑛 x (𝑘 + 1)), are correlated with the error term, 𝜺 (a vector of size 𝑛 x 1) (Stock and Watson, 2012). The methodology behind enabling the mentioned property with the IV regression is to implement instrument variables, Z (a matrix of size 𝑛 x 𝑚), satisfying the following two conditions:
1. Instrument relevance: 𝑐𝑜𝑟𝑟(𝑍', 𝑋') ≠ 0 2. Instrument exogeneity: 𝑐𝑜𝑟𝑟(𝑍', 𝜀') = 0
In words, the instruments need to be correlated, thereby be relevant, with the explanatory variable it intends to be “instrumented”, as well as be uncorrelated, thereby be exogenous, with the error term of the regression model. Additionally to the informative variables X and Z, the included exogenous variables, W (a matrix of size 𝑛 x 𝑟), will be utilized in the empirical models, demonstrated further down in this section, which needs to meet the requirements of being: uncorrelated with both the error term and the explanatory variable. This implies that condition one above needs to be supplemented by, when there is one X (as it will be in this study), that at least one instrument must have a non-zero coefficient when X is regressed on Z and W. If these conditions are meet, and Z thereby is both relevant and exogenous, the variation in Z can be utilized to estimate the coefficient of X through the two-stage least square estimator (TSLS).
In the first stage of the TSLS, the explanatory variable, X, is regressed with OLS on the instruments, Z, and the included exogenous variables, W. The predicted value of the regression, 𝑋Q, are thereafter compute. Further, these predicted values are used in the second step of the TSLS, where we regress Y on 𝑋Q and W using OLS. The equation of the first stage estimator in the TSLS can be written as (7):
Where 𝝅𝟏 is a vector of coefficient from 1, … , 𝑚 and 𝝅𝟐 is a vector of coefficient from 𝑚 + 1, … , 𝑚 + 𝑟. Further, the second stage estimator can be formulized as:
𝑌Q = 𝛽X)./0/+ 𝛽X
#./0/𝑋Q + 𝜷Z𝑾𝑻𝑺𝑳𝑺𝑾 (8) Here, 𝛽X)./0/ is the estimated intercept and 𝛽X
#./0/ is the estimated coefficient for predicted values 𝑋Q, and 𝜷Z𝑾𝑻𝑺𝑳𝑺 is an estimated coefficient vector going from 2, … , 1 + 𝑟.
4.2. Empirical Modeling
In this sub-section, the models fitted for the return will be presented, both for the OLS and IV models and additionally for the subset regression.
First, six OLS models of the return are presented in different constellations with DrTP, stock recommendation, and the indexes, together with the other economic information, to see how this affects the estimations compared to when using them as IV regressors. As such, the first model (M1) displayed below includes DrTP and the indexes, whereas in the second model (M2), DJIA is excluded while the dummy variables REC.UP and REC.RET are includes. For (M3), REC.RET is excluded, as shown below.
𝑟̂' = 𝛽X)+ 𝛽X#𝐷𝑟𝑇𝑃'+ 𝛽X(𝑁225' + 𝛽X*𝐷𝐽𝐼𝐴' (M1)
𝑟̂' = 𝛽X)+ 𝛽X#𝐷𝑟𝑇𝑃'+ 𝛽X(𝑁225' + 𝐷Z#𝑅𝐸𝐶. 𝑈𝑃'+ 𝐷Z(𝑅𝐸𝐶. 𝑅𝐸𝑇' (M2)
𝑟̂' = 𝛽X)+ 𝛽X#𝐷𝑟𝑇𝑃'+ 𝛽X(𝑁225' + 𝐷Z#𝑅𝐸𝐶. 𝑈𝑃' (M3)
In the three models above, the 𝛽X’s are estimated parameters for the corresponding variables. The two REC variables are binary, where REC.UP takes on the value one if the change in recommendation is upgrade, zero otherwise, whereas REC.RET and REC.DOWN take the value one if there is no change in recommendation, or the change in recommendation is to downgrade, respectively (REC.DOWN not included due to mitigate perfect multicollinearity). 𝐷Z5, where 𝑗 = 1,2, are the coefficients for the corresponding binary recommendation variables. Model (M4) – (M6) include the variables that are suspected to influence, and thereby correlate, with DrTP. (M4) and (M5) holds all variables that will be the base for the first stage of the IV regression but with REC.RET omitted in (M5). The last IV model, (M6), all but PB of the four variables that are expected to affect DrTP are incorporated.
𝑟̂' = 𝛽X)+ 𝛽X#𝐷𝑟𝑇𝑃'+ 𝛽X(𝑁225' + 𝐷Z#𝑅𝐸𝐶. 𝑈𝑃'+ 𝐷Z(𝑅𝐸𝐶. 𝑅𝐸𝑇'+ 𝛽X*𝑟̅'
+ 𝛽X+𝑣(𝑟') + 𝛽X6𝑀𝐶𝑎𝑝' + 𝛽X7𝑃𝐵' (M4)
𝑟̂' = 𝛽X)+ 𝛽X#𝐷𝑟𝑇𝑃'+ 𝛽X(𝑁225' + 𝐷Z#𝑅𝐸𝐶. 𝑈𝑃'+ 𝛽X*𝑟̅' + 𝛽X+𝑣(𝑟') + 𝛽X6𝑀𝐶𝑎𝑝' + 𝛽X7𝑃𝐵'
𝑟̂' = 𝛽X)+ 𝛽X#𝐷𝑟𝑇𝑃'+ 𝛽X(𝑁225' + 𝐷Z#𝑅𝐸𝐶. 𝑈𝑃'+ +𝛽X*𝑃𝐵' (M6)
As the two conditions for IV regression require that the instruments need to be correlated with the explanatory variables X and be uncorrelated with 𝜀', the first stage in the TSLS scheme of the models are performed separately to assure that at least one estimated instrumental parameter is significantly different from zero (Stock and Watson, 2012). Two models are thus estimated and illustrated below, where A.M1 includes both the instruments and the assumed endogenous variables, whereas in (A.M2), the endogenous variables are omitted.
𝐷𝑟𝑇𝑃_' = 𝜋`)+ 𝜋`#𝑁225' + 𝜋`(𝑅𝐸𝐶. 𝑈𝑃'+ 𝜋`*𝑟̅' + 𝜋`+𝑣(𝑟') + 𝜋`6𝑀𝐶𝑎𝑝'
+ 𝜋`7𝑃𝐵' (A.M1)
𝐷𝑟𝑇𝑃_' = 𝜋`)+ 𝜋`#𝑟̅'+ 𝜋`(𝑣(𝑟') + 𝜋`*𝑀𝐶𝑎𝑝'+ 𝜋`+𝑃𝐵' (A.M2)
Alongside these model results, the residuals of the IV models are regressed on the instrument and endogenous variables in two models, (A.M3) and (A.M4) to validate condition two of
exogeneity (Stock and Watson, 2012). All empirical results for these four models are reconciled
in Appendix 2. As for the second stage when modelling the return, might be different characteristics regarding the variation, whereby a robust check is considered for the DrTP model as well. These results can also be found in Appendix 2 where the effect on the model is small and can be neglected and assumed constant.
The IV regression model is executed in six different shapes where the first two, (M7) and (M8), comprises all four instruments in the first stage, while 𝑣(𝑟) and PB are left out in (M9) and (M10), followed by also leaving out MCap in the two last models. For recall, the instruments are 𝑟̅, v(r), MCap, and PB. The outline of model (M7) to (M9) are seen below, where (M10) and (M12) are similar to (M8) while (M11) is similar to (M9), but with different instruments.
𝑟̂' = 𝛽X)./0/+ 𝛽X#./0/𝐷𝑟𝑇𝑃_' + 𝛽X(./0/𝑁225'+ 𝛽X*./0/𝐷𝐽𝐼𝐴'+ 𝛽X+./0/𝑅𝐸𝐶. 𝑈𝑃' (M7)
𝑟̂' = 𝛽X)./0/+ 𝛽X
#./0/𝐷𝑟𝑇𝑃_' + 𝛽X(./0/𝑁225' (M8)
𝑟̂' = 𝛽X)./0/+ 𝛽X
#./0/𝐷𝑟𝑇𝑃_' + 𝛽X(./0/𝑁225'+ 𝛽X+./0/𝑅𝐸𝐶. 𝑈𝑃' (M9) The 𝛽X./0/’s are the two stage least squared estimated parameters for the included variables, and 𝐷𝑟𝑇𝑃_' is the estimated variable in the first stage of the two-stage process.
4.3. Subset Regression Approach
To control for the stock recommendation with a similar approach as performed by Brav and Lehavy (2003), OLS regression on the subset categorized by the change in recommendation will be applied. The model will be predicted with the index N225, with and without DrTP. This implies a total of six models where the structure of three of them are the same but fitted to the three categories of change in stock recommendation. (M13) presented below is the model with
DrTP applied to the observations where an upgrade in stock recommendation has been taken,
whereas DrTP is left out in (M16).
𝑟̂',89 = 𝛾`)+ 𝛾`#𝐷𝑟𝑇𝑃',89+ 𝛾`(𝑁225',89 (M13)
𝑟̂',89 = 𝛾`)+ +𝛾`#𝑁225',89 (M16)
In (M13) and (M16) above, 𝛾`) is the intercept and 𝛾`5where 𝑗 = 1,2 are the estimated slope parameter values for the variables but only with observations where REC.UP is equal to one. Furthermore, (M14) and (M17) is similar to the models above but based on the observation where there is no change in stock recommendation (REC.RET = 1), whereas (M15) and (M18) is applied on the observations with recommendation down (REC.DOWN = 1).
5. Results
This section contains both results of the OLS models, the IV models, and finally the results using the subset approach. The two conditions discussed and tested empirically for the IV regression and the instruments are located in Appendix 2.
5.1. OLS regression
In Table 3, the result for the different OLS models are displayed, where the first three models exclude the variables indenting to affect DrTP while added for (M4) to (M6).
Table 3. Six OLS models where the return is regressed on DrTP, the indexes, stock recommendations, and stock and company information.
(M1) (M2) (M3) (M4) (M5) (M6) (Intercept) 0.1053 -0.9361 -0.4755 -2.5395* -1.8083 -1.1028** (0.3105) (0.7149) (0.3434) (1.3560) (1.1704) (0.4396) DrTP, lag 1 0.0176* 0.0187* 0.0209** 0.0208* 0.0249** 0.0243*** (0.0093) (0.0096) (0.0091) (0.0120) (0.0114) (0.0093) N225 1.0667*** 0.5774*** 0.5649*** 0.6276*** 0.6094*** 0.5958*** (0.2591) (0.2172) (0.2163) (0.2276) (0.2271) (0.2189) DJIA, lag 1 -0.2385*** (0.0713) REC.UP 2.2172** 1.7176*** 2.7814*** 2.0238*** 1.7370*** (0.8655) (0.5351) (0.9089) (0.5678) (0.5420) REC.RET 0.6043 0.9151 (0.8223) (0.8574) 𝑟̅ 0.1911 0.2144 (0.2551) (0.2543) v(r) 0.0528 0.0487 (0.0346) (0.0344) MCap, lag 1 0.0077 0.0153 (0.2421) (0.2420)
PB, lag 1 0.2079** 0.1988** 0.1988** (0.0980) (0.0977) (0.0927) AIC 1921.6883 1944.2619 1942.8096 1884.4401 1883.6099 1887.1149 BIC 1940.6686 1967.1108 1961.8504 1922.1856 1917.5809 1909.7808 Log Likelihood -955.8441 -966.1310 -966.4048 -932.2200 -932.8050 -937.5575 Adj. R2 0.0591 0.0534 0.0547 0.0732 0.0728 0.0688 RMSE 4.4481 4.4367 4.4336 4.4386 4.4395 4.4438 Num. obs. 329 333 333 322 322 323
***p < 0.01, **p < 0.05, *p < 0.1. Standard error in parenthesis.
As seen in the table above, the estimated parameters for DrTP is positively significant on at least a ten percent level in all models, where the value is slightly higher when the variables, r, v(r), MCap, and PB are integrated. Both the indexes are significant in (M1) on a one percent level, where N225 is positive, and DJIA is negative. However, these indexes correlate15, and
DJIA is therefore left out in the rest of the model due to that N225 is more significant and relatively more reasonable to include due to the opening hours of the stock markets. More specifically, As NASDAQ and NYSE open before the SSE close, the day before the observed returns might capture a portion of the that effect, in contrast to TSE where the opening hour do not interrupt with SSE. The most prominent recommendation factor, REC.UP, suggests a positive return between 1.72 and 2.78 percent compared to when the recommendation is not upgrade, ceteris paribus. Further, a downgrade in a recommendation, incorporated in the intercept in (M2) and (M4) show negative affect but are only significant in model four. Overall, the recommendation parameters are in line with previous research (Brav and Lehavy, 2003; Asquith et al., 2005) where the estimated parameter signs are expected. The contribution of the variables that are expected to influence both the r and DrTP, as seen in the model (M4) to (M6), does affect the other estimated parameters notable, but the only significant estimate of these parameters regards to PB.
Taking the goodness-fit-test and error measure into consideration, the models including the stock and company information is the best model looking at the AIC and BIC score, whereas (M2) and (M3) has the lowest RMSE. However, the results are somewhat similar overall, and the main factors in evaluating are the significant level and their respectably estimated parameter’s sign, which, as seen in Table 3, is well in line with the expectations and previous research, even though the estimated parameters for DrTP seem smaller.
5.2. IV regression
Together with the estimate results using the IV approach, a comparison between the models is done through different goodness-of-fit and error measures.
Table 4. Six IV regression models where the return is regressed on DrTP, the indexes, stock recommendations, and stock and company information.
(M7) (M8) (M9) (M10) (M11) (M12)
(Intercept) -0.2444 0.0245 -0.1162 0.1738 -0.0056 0.1814 15 Correlation between N225 and DJIA is 0.54.
(0.4468) (0.4070) (0.4604) (0.4369) (0.5003) (0.4838) DrTP, lag 1 0.0100 0.0192 0.0031 0.0132 -0.0023 0.0128 (0.0157) (0.0158) (0.0176) (0.0182) (0.0201) (0.0211) N225 1.0717*** 0.6065*** 0.5618** 0.5785*** 0.5609** 0.5784*** (0.2618) (0.2236) (0.2176) (0.2195) (0.2185) (0.2195) DJIA, lag 1 -0.2386*** (0.0724) REC.UP 1.6015*** 1.6850*** 1.6749*** (0.5474) (0.5389) (0.5414) Adj. R2 0.0837 0.0303 0.0438 0.0265 0.0360 0.0263 RMSE 4.4321 4.5401 4.4592 4.4993 4.4772 4.4998 Num. obs. 319 322 333 333 333 333
***p < 0.01, **p < 0.05, *p < 0.1. Standard error in parenthesis. (M7) and (M8) includes all four as instrumental variables; (M9) and (M10) includes 𝑟̅ as instrumental variables and MCap; (M11) and (M12) includes 𝑟̅ as instrumental variable.
Compared with the regular OLS models in Table 3, the validity in terms of significance drops for the estimated parameters of DrTP in all models, even though the IV conditions are met, as seen in Appendix 2. Analysing the other variables, the parameter estimates are approximately the same as in Table 3, indicating low relationship with the instruments. Analysing the RMSE and adjusted R2, the best IV model seems to be (M7), where all instruments are included, and
are the IV model that are most similar in these aspects when comparing with the OLS models in Table 3, even though (M7) seems slightly better.
5.3. Subset Regression by Stock Recommendation
In the table below, the empirical result of six different models with the three different subsets of change recommendation – upgraded, reiterated, and downgraded – are presented.
Table 5. Six OLS regress of return on DrTP, N225, and DJIA, with subsets of the change in stock recommendation.
UP (M13) RET (M14) DOWN (M15) UP (M16) RET (M17) DOWN (M18) (Intercept) 0.9477 -0.2823 -0.9360* 1.7680*** 0.0808 -1.0376* (0.6858) (0.4189) (0.5242) (0.4486) (0.3227) (0.5202) DrTP, lag 1 0.0446 0.0149 -0.0153 (0.0275) (0.0107) (0.0387) N225 -0.2519 1.1583*** 0.3489 0.0222 1.1698*** 0.2558 (0.3818) (0.3030) (0.4225) (0.3678) (0.3014) (0.4309) AIC 571.2480 1161.2368 203.2866 698.4696 1170.9481 215.8886 BIC 581.5469 1174.3696 209.9409 706.7561 1180.8280 221.0293 Log Likelihood -281.6240 -576.6184 -97.6433 -346.2348 -582.4741 -104.9443 Adj. R2 0.0121 0.0708 -0.0326 -0.0087 0.0663 -0.0165 RMSE 4.4822 4.5527 3.0795 4.7065 4.5409 3.2081 Num. obs. 97 197 39 117 199 41
***p < 0.01, **p < 0.05, *p < 0.1. Standard error in parenthesis.
There are significant- and sign-wise similar results as in previous studies (Brav and Lehavy, 2003; Asquith et al., 2005), whereby an upgraded recommendation is illustrated, here though
only in (M16), through the significantly positive estimated intercept with a value of 1.77 percent. This is followed by an insignificant intercept for both models of subsets reiterate and downgrade, but still with expected signs even though they are not valid. The estimates for DrTP is not significance which, more illustratively than in Table 3 and 4, signalizing that the stock recommendation and DrTP partly explains similar factors of the day ahead return. Examining the unchanged stock recommendation subset results (model (M14) and (M17)), the opposite is the case, as the estimated parameters for N225 are significant at a one percent level. Even though the estimated parameters are not significant for the other four models, it is notable that the values are different between three various change in stock recommendation, disposing inequalities in the risk level.
6. Discussion
The results of a positive effect of DrTP and REC.UP on the return, and large markets' effect on small markets, follow previous research results (Brav and Lehavy, 2003; Asquith et al. ,2005; Wu and Su, 1998;). Looking closer into the models, the difference between the estimated parameters does not seem too large compared with or without additional economic information, which probably is the most prominent reason the IV model in Table 4 are fairly equal to the OLS models in Table 3. However, taking the stock recommendation into consideration and analysing the models in Tables 3, 4, and 5, this seems to be a better indicator of a higher return when there is a change up in the recommendation (REC.UP) than a change in target price (DrTP). Considering the subset regressions, this also seems to be the case, where we can suspect that the relationship between DrTP and the REC-variables disturb the predictability. This is similar to Asquith et al. (2005) conclusions that the target price and stock recommendation partly contain the same information. Further, the stock recommendation results where REC.UP is significant whereas REC.RET is not, are also in line with both Brav and Lehavy (2003) and Asquith et al. (2005). To conclude from this thesis result is that the stock recommendation seems to be a better predictor of the day ahead return than the target price.
Uniquely from this study is the control of globally relatively large indexes, representing other stock market changes, which seems reasonable and a usable explanatory factor of the stock return, despite that some considerations is needed. As all large market indexes are not considered (e.g., S&P 500 at NYSE or NASDAQ and SSE composite at Shanghai SE) and the indexes of large markets seems to be a valid predictor of a small market such as Stockholm exchange market, additional indexes might produce even better predictable. This could be accomplished by creating an indicator combining several market indexes, where a higher level of explanatory factor to the variation in return can, for example, be used as a weight function. This approach would also omit the correlation problem occurring when several indexes are included separately in the same model (as in (M1) and (M7)).
Additionally, to the analyse of the result in comparison to previous studies, there are several drawbacks and other factors that need to be discussed and deliberated. First, Brav and Lehavy (2003) write and refer to that the price forecast (target price in this case) is often based on earning forecast, which has been controlled for in their research and might be a valid predictor
of the target price. As such, the earnings forecast, the price/earnings (PE), and maybe the price/sales (PS) could have been utilized in this study to improve the predictability of the target price and, consequently, the return. There are furthermore few observations both in general and compared to previous research. This is overall unfavourable per se, but it also prevents comparability and control for various effects from different analysts and industries, as Boni & Womack (2006) consider has an effect on the return. Moreover, a comparison would enable the control for how a specific analyst might have great insight as well as influence in a certain sector, where Boni & Womack (2006) discuss it from the aspect that controlling for sector noise is associated with a forecast of the future price. Consequently, if a certain analyst has high trustworthiness, overall or within a branch, this will probably contribute to a greater effect when they announce a change target price or stock recommendation. As such, the analyst could also be integrated with the change in recommendation, but as said, this demands a larger dataset.
Other company factors that can affect the market participant's future expectations should, in the best scenario, also be considered. These can be new board decisions and reactions on annual/quarterly reports but might be complex to incorporate in a model. One can assume that these effects can be equalized with an announcement of a target price, and if this happened on the same day as a target price resale, there could be a ripple effect in the return. However, as the analysts consider these events short before (same day), this information can be assumed included in the target price, but exactly how these information events affect the return is out of the scope of this thesis whereas, for example, Asquith et al. (2005) discuss this.
One aspect that should be examined to gain further information out of the target price is that the price difference from the previous release relative to the target price change might have an impact on the return. This is as Da & Schaumburg (2011) find evidence that the DrTP (TPER in their research paper) does have an effect on the average return the coming month when controlling for the previous. As the target price is a valuation of the stock price 12 months ahead (generally), it is reasonable to consider the price change relative to when the last target price was announced (the previous one to measure the change to the new one). Let say that the price changes up over a three-month period of time, and at the end of this three-month period, the target price is updated. This can, partly, be a consequence of the natural increase of the valuation, but also an effect of the increase in price relative to the target price change. This can be one reason for the negative estimated parameter of r in (A.M1) and (A.M2). However, this cannot certainly be concluded as a drawback when measuring the effect one day ahead, and as the previous monthly average shows a significant effect, this parameter might cover this effect. One explanation as to why the three-month variance does not seem to be a good predictor could be the large current fluctuation at the stock market, which also can be seen in the descriptive statistics in Table 1, where the mean of the variance is over 11.96 and the standard deviation 8.80. Another factor to consider when evaluating the significance level of the estimated parameters and models overall are the specific circumstances that the market has been within during the observed period February-2020 to May-2020. Moreover, the fact that the within time periods might have different average return are not considered. Hence, it cannot be
excluded that the more positive returns and REC.UP values lean towards the second part of the period, which is more positive. However, the overall negative index values and over the time period indicate that this should not be too big of an issue.
As well, the change in market participants’ expectations might also not be fully accounted for in one day, whereby instead, as Brav and Lehavy (2003) approach it, a five-day average return can be used. To use several days ahead can be further investigated to assess how sensitive the declining effect, in time, of change in target price or stock recommendation is. This comes with several new complications such that additional new information (e.g. board decision, press releases) can imply an impact, and how changes in other larger markets affects smaller market over several days. One aspect that this way of measure the impact of target price and stock recommendation might capture better, is a levelled risk factor, as a “buyer” might be faster to the buy button, than a “seller” is to the sell button. This can be one reason that the absolute value of the estimated parameters overall is higher for REC.UP than REC.DOWN.
7. Conclusions
Changed expectations of future returns together with daily fluctuation effects, viewed as analyst’s information and stock market indexes in this study, seems overall to co-work as model predictors of the day-ahead stock return. However, usual relevant economic information such as previous mean returns, book-to-market value, and Market Cap does not seem to work as indirect predictors using the IV modeling approach. The results indicate that the applied regular OLS regressions, both including these just mentioned economic factors or not, seems comparable if not as better models than the IV models. Comparing the models in Tables 3 and 4, the IV model (M7) seems to be, marginally, the best of the estimated models, with the lowest RMSE and highest adjusted R2. The easy and possible the most reasonable explanation to the
small differences in the models is that the day ahead return depends on multiple complex factors not captured by the all control variables, at least not when elaborated on altogether. To conclude, a large part of the daily change does seem to be controlled for when including analyst’s information in terms of the target price and the stock recommendation, both when examining the OLS, IV, and subset regression. Additionally, the indexes controlling for the large markets changes, is reasonable to include under the observed circumstances and validate alongside the analyst information, to some extent, that indented mid long-term changed expectation (target price, 12 months horizon) and daily events both should be considered in the day-to-day estimation of stock return.
References
Asquith, P., Mikhail, M. B., Au, A. S. 2005. Information content of equity analyst reports.
Journal of Financial Economics. 75, 245-282.
Brav, A., Lehavy, R. 2003. An Empirical Analysis of Analyst’s’ Target Prices: Short-term Informativeness and Long-term Dynamics. The Journal of Finance. 58 (5).
Bilinski, P., Lyssimachou, D., Walker, M. 2013. Target Price Accuracy: International Evidence. The accounting Review. 88 (3), 825-851.
Boni, L., Womack, K. L. 2006. Analysts, Industries, and Price Momentum. Journal of
Financial and Quantitative Analysis. 41 (1).
Bradshaw, M. T., Brown, L. D., Huang, K. 2012. Do sell-side analysts exhibit differential target price forecasting ability? Rev Account Stud. 18, 930-955. DOI 10.1007/s11142-012-9216-5
Da, Z., Schuamburg, E. 2011. Relative valuation and analyst target price forecasts. Journal of
Financial Markets. 14, 161-192.
Fama, E.F., French, K.R. 1993. Common risk factors in the return on stock and bonds. Journal
of financial economics. 33, 3-56.
Harrison, J. M., Kreps, D. M. 1978. Speculative Investor Behavior in a Stock Market with Heterogeneous Expectations. Oxford Journals. 92 (2), 323-336.
Jegadeesh, N. 1990. Evidence of Predictable Behavior of Security Returns. The Journal of
Finance. 45 (3), 881-898.
Stock, J., Watson, M. 2012. Introduction to Econometrics. 3. ed. Harlow: Pearson.
Tsay, R. S. 2010. Analysis of Financial Time Series. 3 ed. New Jersey: John Wiley & Sons, Inc.
Wu, C., Su, Y-C. 1998. Dynamic Relations among International Stock Markets. International
Appendix 1
Table A1.1. Example of the dataset structure
Date Stock Recommendation Previous Stock Recommendation Stock Analyst Return DrTP
... ... ... ... ... ... ...
... ... ... ... ... ... ...
10/02/2020 INDU-A BUY HOLD Nordea 1.7171 13.6549
10/02/2020 SWMA HOLD BUY ABG 0.3610 -1.3152
10/02/2020 SKA-B HOLD SELL SHB 2.2865 7.8033
11/02/2020 BOL BUY BUY RBC 1.0287 22.0967
12/02/2020 KLED BUY HOLD ABG 0.6616 9.7316
12/02/2020 WIHL HOLD HOLD Pareto 1.1022 -8.3798
13/02/2020 KLED BUY HOLD SHB 3.0047 18.4895
13/02/2020 SWMA HOLD BUY SHB -0.4093 9.7875
13/02/2020 SWMA BUY BUY Nordea -0.4093 12.4015
13/02/2020 TREL-B HOLD HOLD SHB 0.4476 5.5935
13/02/2020 TREL-B BUY BUY Nordea 0.4476 22.7168
13/02/2020
KIND-SDB BUY BUY Nordea 1.1490 32.7160
13/02/2020 KLED BUY BUY Nordea 3.0047 9.9583
14/02/2020 BOL BUY HOLD UBS 4.0107 24.6312
14/02/2020 BETS-B BUY BUY DnB
Market 8.3572 43.3839
14/02/2020 ATT BUY HOLD Carnegie 8.4619 21.7532
14/02/2020 SOBI BUY HOLD SHB 3.8760 13.6951
14/02/2020 NIBE-B HOLD HOLD DnB
Market -0.6643 -0.3598
14/02/2020 SOBI BUY BUY Danske
Bank 3.8760 19.3798
14/02/2020 BOL BUY BUY JP
Morgan 4.0107 24.6312
14/02/2020 FPAR-A HOLD SELL Carnegie 2.8302 -4.3882
14/02/2020 NIBE-B HOLD SELL ABG -0.6643 -4.7883
... ... ... ... ... ... ...
... ... ... ... ... ... ...
Table A1.2. Number of target prices released for companies on the Large Cap list at Nasdaq Stockholm stock market between 2020-02-06 to 2020-05-06, published on Avanza’s Web page, divided by Branch and Analyst.
Branch Real
Estate Financials Industrials
Tele-communication Consumer Basic Materials Health Care Technology # 40 46 110 26 90 35 21 30
Analyst ABG Barclays Berenberg Carnegie CITI Commerz-bank Credit Suisse Danske Bank # 36 3 3 20 5 2 2 15 Analyst Deutsche Bank DnB Market Exane Goldman Sachs HSBC Jefferies JP Morgan JYB # 5 58 4 7 2 3 11 1 Analyst Kepler Cheuvreux Morgan
Stanley Nordea OP Equity Pareto
Raymond
James RBC Redburn
# 55 10 52 1 28 1 2 2
Analyst SEB SHB Soc Gen UBS
Table A1.3. Stocks listed on Large Cap at Nasdaq Stockholm Stock Exchange included in this study.
Stock
AAK Elekta B Kindred Group Sandvik
ABB Ltd Epiroc A Kinnevik A SCA A
Ahlstrom-Munksjö
Oyj Ericsson A Klövern A SEB A
Alfa Laval Essity A Kungsleden Securitas B
Arjo B Evolution Gaming Group Latour B Skanska B
ASSA ABLOY B Fabege Lifco B SKF A
AstraZeneca Fast. Balder B Loomis B SSAB A
Atlas Copco A Fastpartner A Lundbergföretagen B Stora Enso A Atrium Ljungberg B
Fenix Outdoor International B
Lundin Mining
Corporation SWECO A
Attendo Getinge B Millicom Int. Cellular SDB Swedbank A Autoliv SDB Handelsbanken A Modern Times Group A Swedish Match Avanza Bank Holding Hennes & Mauritz B NCC A
Swedish Orphan Biovitrum
Axfood Hexagon B NIBE Industrier B Tele2 A
Beijer Ref B HEXPOL B Nobia Telia Company
Betsson B Holmen A Nolato B Thule Group
BillerudKorsnäs Hufvudstaden A Nordea Bank Abp Trelleborg B
Boliden Husqvarna A Pandox B Veoneer SDB
Bonava A ICA Gruppen Peab B Vitrolife
Bravida Holding Industrivärden A Ratos A Volvo A
Castellum Indutrade Resurs Holding Wallenstam B
Dometic Group Investor A SAAB B Wihlborgs Fastigheter
Electrolux A JM Sagax A
