Company Valuation and Bankruptcy Prediction

Contents

1 Introduction

2 To Buy or Not to Buy? The Value of Contradictory Analyst Signals
2.1 Introduction
2.2 Data and Methodology
2.2.1 Data, Variables, and Descriptive Statistics
2.2.2 Methodology
2.3 Empirical Results
2.3.1 Calendar Time Portfolios
2.3.2 Factor Loadings, Market Capitalization, and Transaction Costs
2.3.3 Potential Explanations
2.4 Conclusion

3 Valuing High Technology Growth Firms
3.1 Introduction
3.2 Related Literature: Firm Growth and Valuation
3.3 Valuation Models
3.3.1 Fundamental Pricing: The Schwartz-Moon Mode
3.3.2 Introducing a Benchmark: Enterprise-Value-Sales Multiple
3.4 Data and Methodology
3.4.1 Data Collection
3.4.2 Model Implementation
3.4.2.1 Revenue Dynamics
3.4.2.2 Cost Dynamics
3.4.2.3 Balance Sheet and Remaining Firm Parameters
3.4.2.4 Environmental and Risk Parameters
3.4.2.5 Simulation Parameters
3.4.3 Summary Statistics
3.5 Main Empirical Results
3.5.1 Feasibility and Deviations from Market Values
3.5.2 Detecting Over- and Undervaluation: The Trading Strategy
3.6 Robustness Checks
3.7 Discussion and Conclusion

4 Bankruptcy Prediction Based on Stochastic Processes: A New Model Class to Predict Corporate Bankruptcies?
4.1 Introduction
4.2 Prior Research
4.3 The Mode
4.3.1 Sales and Costs
4.3.2 The Accounting Volatilities
4.3.3 The Change in Net Working Capita
4.4 Data and Model Implementation
4.4.1 The Data
4.4.2 Parameter Estimation
4.5 Empirical Analyses
4.5.1 Summary Statistics and Correlations
4.5.2 Accuracy
4.5.3 Test of Information Conten
4.6 Discussion and Conclusion

5 Summary and Conclusion

References

List of Tables

Table 2.1: Descriptive Statistics for the Target Price Changes

Table 2.2: Average Change of Target Prices, Prices, and Implicit Return Estimates Within Recommendation Changing Categories

Table 2.3: Distributions of Recommendation Changing Categories

Table 2.4: Calendar Time Portfolios, Recommendation Reiteration

Table 2.5: Factor Loadings, Market Capitalization and Transaction Costs

Table 2.6: Characteristics per Reiteration Category

Table 3.1: Data Sources

Table 3.2: Sample Selection Procedure

Table 3.3: Variable Definitions

Table 3.4: Summary Statistics

Table 3.5: Deviations from Market Values

Table 3.6: Deviations by Industry Classification and Firm Size

Table 3.7: Trading Strategy

Table 3.8: Model Implied Default Probability

Table 3.9: Sensitivity Analysis

Table 3.10: Regression Analysis

Table 4.1: Distress-Related Delisting Reasons and Frequency

Table 4.2: Estimation of Parameters

Table 4.3: Parameter Estimates

Table 4.4: Descriptive Statistics

Table 4.5: Explanatory Power for S&P Credit Ratings

Table 4.6: Forecasting Accuracy

Table 4.7: Logit Model Estimation Results

List of Figures

Figure 3.1: Income Statement and Balance Sheet Illustration

Figure 3.2: Proportion of Loss Making Firms Over Time

Figure 3.3: Quarterly Median Absolute Deviations

Figure 3.4: Quarterly Median Non-Absolute Deviations

Figure 3.5: Median Quarterly Defaults

Figure 4.1: Median Estimated Default Probability per S&P Rating Class

Figure 4.2: Default Probabilities Over Time

Figure 4.3: Evolvement of Default Probabilities Before Delisting

Figure 4.4: Accuracy Ratio for Different Horizons

Figure 4.5: Accuracy Ratio Over Time

List of Symbols and Abbreviations

illustration not visible in this excerpt

1 Introduction

“If a company’s future cannot be predicted, it cannot be valued.” - Warren Buffett

Not only since the recent financial crisis there is evidence for a deviation of market prices from intrinsic values for companies. Indeed, it is well established in the literature that financial markets are not necessarily efficient (Sloan 1996; Baker and Wurgler 2007; Stambaugh et al. 2012). Consequently, observed market prices do not always reflect the company’s “true” fundamental value (Lee et al. 1999). The results are periods of overvaluation and financial bubbles, which can destroy a large amount of wealth when they burst. Naturally, when markets misprice companies, several questions arise: given the less efficient markets, can specialists as financial analysts provide additional information, which contain investment value? How can the true value of a company be determined with publicly available data and can discrepancies between fundamental and market values be exploited? Finally, is it possible to assess the firm’s financial health and its likelihood of failure several years into the future?

Valuing assets generally requires looking into the future and predicting trends and outcomes. The basic idea of most valuation methods, the so called present value techniques, is to consider the asset’s expected future pay-offs as basis for its value. Given the uncertainty of the future, forecasting is a challenging task, particularly when the horizons are long. This is the case for companies, which are assumed to act as going concern but might default at any random time in the future. Consequently, uncertainty is high as various factors determine the future states of a firm. One particularly important future state is bankruptcy, which is highly relevant for valuation and one focus of this dissertation. Therefore, when forecasting a company’s future performance one should explicitly take different future states into account. This doctoral thesis contributes to forecasting firm performance in the context of valuation and bankruptcy prediction. It offers two alternative approaches. The first part of this thesis relies on experts’ predictions, namely financial analysts’ assessments. Valuing firms is their profession, so that their experience might result in better forecasts of future firm performances as, for example, measured by the evolvement of their market values. More importantly, the second part offers comprehensive stochastic modeling which allows for a whole distribution of future outcomes to estimate firms’ fundamental values and predict bankruptcies.

Overall, company valuation is highly relevant for academics and practitioners such as investors, banks and regulators worldwide for several reasons. First of all, obtaining a precise estimate for the underlying value of a company is important to prevent misallocation of resources in the economy. Moreover, it can help to detect market misvaluation and prevent the evolvement of speculative bubbles. Benchmarking market prices against theoretically derived company values therefore offers early warning signals for regulators or monetary policies of central banks in case of substantial deviations. From an investor’s perspective it is the basis for private investing in businesses, for mergers and acquisitions of companies or for tax purposes. Finally, it plays an important role for banks in the process of lending where the firm’s assets serve as collateral.

At the same time, company valuation is closely related to bankruptcy prediction as failure is a special state of a firm that affects its value negatively, i.e., a financially distressed firm is normally worth less. On the other hand, a deteriorating company value, ceteris paribus, generally increases the likelihood of the firm defaulting. After the burst of a bubble for example, when company values drop substantially, the number of bankruptcies rises due to tightening financing conditions (Balcaen and Ooghe 2006). Eventually the link between bankruptcy and valuation has been established in the recent academic literature (Vassalou and Xing 2004; Campbell et al. 2008; Kapadia 2011; Garlappi et al. 2011), which shows that distress risk imposes a discount on the firm value. Hence, valuation and bankruptcy prediction should be considered jointly, which is why this work offers two similar stochastic frameworks for both. Beside its effect on the company value, an early prediction of bankruptcy helps to prevent financial losses to investors because bankruptcy is costly (Warner 1977; Altman 1984).

Making statements about a company’s value and its future performance can rely on various sets of information. There is quantitative information from the markets or the firm’s financial statements as well as more qualitative information as costumers’ beliefs or expert knowledge from analysts. This study concentrates on two important sets of information: analysts’ expertise and accounting data. Especially accounting data has important advantages. First, compared to market data, relying on accounting data from financial statements does not require well-functioning markets. During times of severe mispricing and financial bubbles prices are usually not reliable and therefore do not offer a solid basis for forecasts (Das et al. 2009). Second, while the majority of firms are not exchange listed and hence do not offer market values (Kapadia 2011), they publish financial statements. Consequently, in this context it is essential to employ accounting data for forecasting reasons.

Due to their importance, there are large bodies of literature on company valuation (e.g., McGrath 1997; Liu et al. 2002; Bhojraj and Lee 2002; Heinrichs et al. 2013) and bankruptcy prediction (e.g., Altman 1968; Ohlson 1980). However, there are still several aspects where further research is necessary (Balcaen and Ooghe 2006). While there are many studies on analysts and their recommendations (e.g., Barber et al. 2001; Brav and Lehavy 2003; Jegadeesh 2004), there is comparably little research on the analysts’ quantitative value estimates, namely the target prices. Further, the literature still does not provide a standard method to value high technology growth firms. Finally, many studies on bankruptcy prediction still rely on models developed 30 years ago with theoretical shortcomings (Balcaen and Ooghe 2006). Therefore this dissertation sheds light on predicting future performance of a company with three distinct chapters. Each of the studies offers a comprehensive literature review, the corresponding theoretical framework and extensive empirical evidence from large samples.

The first study (“To buy or not to buy? The value of contradictory analyst signals”) presented in chapter 2 is a joint work with Stefan Kanne, Daniel Kreutzmann and Sönke Sievers (Kanne et al. 2012). It is published in the Financial Markets and Portfolio Management Journal and awarded as best professional paper 2012 by Swisscanto. The study employs expert knowledge, in particular financial analyst announcements to forecast future company performance measured as the change in market value. This goes beyond fundamental valuation and also captures market sentiment and qualitative information as analysts’ experience. Generally, sell-side financial analysts work for banks or brokerage firms and publish research reports which serve as guidance for their clients’ investment decisions. The analyst reports are usually publicly available and their most popular piece of information is the recommendation which summarizes the results in an advice for the client. For example, if an analyst thinks that a company is undervalued by the market and expects a favorable future performance, she will announce a “buy” recommendation. While this measure offers rather broad categories, the reports also include target prices, which provide a more detailed quantitative scaling. The target price indicates the analyst’s expectation of the stock price usually 12 months into the future. To arrive at the target price, analysts apply various valuation methods and therefore represent another perspective on company valuation (Bonini et al. 2010). While the literature in this field has already shown that recommendations have investment value, chapter 2 of this dissertation focuses on target price signals and their relation to recommendations. In particular, the research question is how the market interprets the valuation signals when they are contradictory. Thus, chapter 2 studies the predictive ability of individual analyst target price changes for post-event abnormal stock returns within each recommendation category. Although prior studies generally demonstrate the investment value of target prices, this dissertation finds that target price changes do not cause abnormal returns within each recommendation level. Instead, contradictory analyst signals (e.g., strong buy reiterations with large target price decreases) neutralize each other, whereas confirmatory signals reinforce each other. Further, the analysis reveals that large target price downgrades can be explained by preceding stock price decreases. However, upgrades are not preceded by stock price increases, thereby demonstrating an asymmetric analyst behavior when adjusting target prices to stock prices. The results suggest that investors should treat recommendations with caution when they are issued with large contradictory target price changes. Thus, instead of blindly following a recommendation, investors might put more weight on the change in the corresponding target price and consider transaction costs.

Turning to fundamental analysis, there is a whole range of different methods to value firms where one common categorization of models is into absolute value models and relative value models. The basic idea of the first method is to discount future pay-offs as in the discounted cash flow models (DCF) or the residual income models (RIM). The second method relies on market prices and uses multiples as the price-earnings or the price-sales multiple to value firms. While there are various methods to value established firms, there is no standard approach to value fast growing high technology firms. This is of particular interest as these firms are hard to value because they deviate from basic assumptions as positive earnings or analyst coverage. Besides, their businesses are more volatile so that their future performance is more challenging to predict. In the third chapter I present a study (“Valuing high technology growth firms”) which is co-authored by Sönke Sievers (Klobucnik and Sievers 2013) to address this topic. It is published in the Journal of Business Economics and awarded as best practice paper 2012 by the German Academic Association for Business Research. The study presents a method to estimate a fundamental fair value based on accounting data, hence belonging to the absolute value models. In particular, the chapter offers an implementation of the Schwartz and Moon (2000, 2001) model. For the valuation of fast growing innovative firms Schwartz and Moon (2000, 2001) develop a fundamental valuation model where key parameters follow stochastic processes. The firm’s possible paths of development are simulated to obtain future balance sheets and income statements. Compared to standard valuation approaches this technique not only yields a point estimate of the value but a whole distribution of possible outcomes. While prior research shows promising potential for this model, it has never been tested on a large scale dataset. Thus, guided by economic theory, this study is the first to design a large-scale applicable implementation on around 30,000 technology firm quarter observations from 1992 to 2009 for the US to assess this model. Evaluating the feasibility and performance of the Schwartz-Moon model reveals that it is comparably accurate to the traditional sales multiple with key advantages in valuing small and potentially non-listed firms. Most importantly, however, the model is able to indicate severe market over- or undervaluation from a fundamental perspective in the U.S. high technology market of the last 20 years. We demonstrate that a trading strategy based on our implementation has significant investment value. Consequently, the model seems suitable for detecting misvaluations as the dot-com bubble.

Finally, one major aspect that is often neglected in valuing companies is the possibility of firms to go bankrupt. In chapter 4 I address this issue with my third study (“Bankruptcy prediction based on stochastic processes: a new model class to predict corporate bankruptcies?”, Klobucnik 2013). While there is a large body of literature on bankruptcy prediction, the standard today still seem to be statistical models developed more than 30 years ago (Aziz and Dar 2006; Balcaen and Ooghe 2006). These are the well known Z-score and O-score by Altman (1968) and Ohlson (1980). However, they have deficiencies as the lack of theoretical underpinning and their sample dependency. Inspired by the promising results of stochastic modeling in chapter 3, I evaluate a novel and distinct approach in chapter 4 to estimate the probability of default. In particular, I offer an implementation of a new and theoretically grounded accounting-based approach for the estimation of default probabilities using stochastic processes. It simulates the firm’s cash flow from operations into the future by employing recent research on company valuation. Additionally, as volatilities have been found relevant for bankruptcy prediction using implied volatilities from the stock markets in previous studies (e.g., Merton 1974), this chapter includes three sources of accounting volatility in the comprehensive framework. Thus, it has key advantages over traditional accounting models which neglect the explicit integration of volatility. Next, I benchmark the model on a large sample of more than 200,000 firm quarters against the Z- and O-score along several dimensions. First, the analysis reveals that the bankruptcy prediction based on stochastic processes approach fits the distribution of historic default rates reasonably well. Second, the stochastic bankruptcy prediction approach is more accurate in distinguishing between non-delisting and delisting firms than the prominent O- or Z-score models. Third, it significantly outperforms the two statistical models for longer horizons in predicting bankruptcies providing important early warning signals for regulators and investors. Fourth, while the individual models perform comparably, a combination of all three models has a significantly higher explanatory power for bankruptcies. Consequently, these findings support the usefulness of stochastic modeling for forecasting purposes as found in chapter 3.

To sum up, the contribution of this dissertation is manifold and relevant for academics and practitioners alike. It adds to the literature in the fields of corporate finance, financial accounting and stochastic modeling. In particular, this dissertation provides answers to the questions raised in the first paragraph. First, it illustrates the company valuation assessment by financial analysts as summarized in their target prices and the information processing by analysts and investors in detail. Second, this thesis offers a novel empirical implementation of a model for fundamental company valuation that employs accounting data. In this context it demonstrates severe over- and undervaluation from a fundamental perspective in the U.S. technology sector over the last 20 years. Both the analysts’ company valuation captured by their target prices and the implementation of the fundamental company valuation model translate into significant investment value before and after transaction costs, which supports the notion of non-efficient markets. Finally, one major contribution is to evaluate a new approach for bankruptcy prediction that is based on stochastic processes. It is theoretically appealing and performs better especially for longer forecast horizons than standard methods.

The structure of this doctoral thesis is as follows. The next chapter demonstrates the investment value of analysts’ recommendations and target price announcements. In particular it looks at target price changes conditioned on the corresponding recommendation levels and investigates the case when they contradict each other. Through a trading strategy this approach illustrates that expert opinions can be exploited for profitable strategies. Thereafter in chapter 3 I turn to the valuation of technology firms with the help of a stochastic framework. It offers an implementation of the Schwartz-Moon model and an application to the US technology market. By finding severe market misevaluations, which can be exploited through a trading strategy, the model’s value for forecasting future performance becomes apparent. Chapter 4 covers the topic of bankruptcy prediction and evaluates a novel approach to estimate probabilities of default based on stochastic simulations. Subsequent tests reveal promising results compared to standard models, especially for longer horizons. Consequently this chapter suggests the use of stochastic modeling for predicting a particular future state of a firm, namely bankruptcy. Finally, chapter 5 summarizes and concludes.

2 To Buy or Not to Buy? The Value of Contradictory Analyst Signals

The following chapter investigates the ability of financial analyst reports, especially target prices, to forecast firm future value performance. The analysis is largely based on a research study which won the Swisscanto Award for the Best Professional Paper 2012 and which we published as a double-peer-reviewed journal article in the fourth issue 2012 of Financial Markets and Portfolio Management.^[1]

2.1 Introduction

It is well established in the academic literature that analysts’ stock recommendations can predict post-event abnormal returns.^[2]In contrast, the performance of analysts’ target prices has only recently received attention.^[3]In an influential study, Brav and Lehavy (2003) show that target price changes have considerable information value. These authors investigate the performance of target price changes conditional on the direction of the recommendation change (upgrades, reiterations, downgrades) issued by the same broker. Sorting stocks according to their target price change within each category, they show, for both the upgrade and reiteration categories, that the extreme portfolios have abnormal returns that are remarkably different from those of the collective portfolios within their respective categories. Following this approach, Gleason et al. (2012), Huang et al. (2009), and Da and Schaumberg (2011) expand this line of research. Because most target prices are associated with a specific time horizon, they represent an implicit return estimate. Gleason et al. (2012) sort stocks by the implicit return estimates of target prices. This sorting conveys information about future abnormal returns if the analyst also issued relatively accurate earnings estimates. Huang et al. (2009) reveal that target price changes combined with recommendation revisions yield higher abnormal returns than each signal alone. Elaborating on trading returns, Da and Schaumberg (2011) find even higher abnormal returns when taking industry affiliation into account. All three studies are based on datasets from First Call over the period 1997 to 2004 and use consensus analyst signals. Most recently, Feldman et al. (2012) study the relative importance of analysts’ earnings, target price, and recommendation level revisions using short event windows and one-month hedge return analysis. They find that target price revisions and recommendation changes have the largest impact.

Although these studies focus on the change in recommendations, they do not consider the level of recommendations. This has two main implications. First, it is unclear whether target price changes contain valuable information for each recommendation level. For example, large target price increases (reductions) for reiterated strong buy (sell) recommendations may not provide valuable information to the market because the recommendation already provides a trading signal. In particular, the positive performance of the portfolios with the most favorable target price revisions reported in Brav and Lehavy (2003) might be driven by buy and strong buy recommendations, and the negative performance of the portfolio with the least favorable target price revisions might be driven by hold, sell, and strong sell recommendations. In contrast, because recommendations are bounded from above (strong buy) and below (strong sell), analysts must resort to target price increases (decreases) to signal private information about an increase in the undervaluation (overvaluation) if the stock has already been given a strong buy (sell) recommendation.

Second, analyzing the performance of target price changes conditional on the recommendation level enables the researcher to examine whether observed abnormal returns are consistent with the recommendations. For example, assuming that, on average, analysts interpret and process information correctly, large target price reductions in combination with reiterated strong buy recommendations should not be followed by average negative abnormal returns. Although it might initially seem unreasonable, such contradictory analyst announcements are fairly common. For example, JPMorgan Chase lowered the target price for Bank of America substantially from $18 to $13 in September 2011 but retained an “overweight” rating.^[4]For both researchers and practitioners, the question arises how to interpret the conflicting signals. Our approach builds on the research design of prior studies, such as Brav and Lehavy (2003) or Huang et al. (2009). However, our approach focuses on individual analyst announcements. By sorting target price changes within reiteration categories, we calculate abnormal returns using the calendar time portfolio approach.

Consequently, we make several contributions to the literature. Although several studies employ target prices and recommendations in combination, we are the first to look in detail at the target price changes within each recommendation category. Doing so sheds light on contradictory analyst signals as discussed above and contributes to the research on analyst signaling behavior and market information processing. Additionally, we demonstrate asymmetric behavior by analysts in target price upgrades and downgrades, which has consequences for their informational value and offers a starting point for future research. From a methodological perspective, we focus on individual analyst signals, instead of the more frequently used consensus forecasts, and consider different signals derived from the target price changes as robustness tests. We also extend previous research such as Feldman et al. (2012) by investigating whether profitable trading strategies also hold after transaction costs. Finally, compared to the studies mentioned above (i.e., Gleason et al. 2012; Huang et al. 2009; Da and Schaumberg 2011), we use analyst data for the period after the regulatory environment changed in 2000. The shift in regulation had a significant impact on analyst announcements, as shown in Bradshaw (2011), and hence it is unclear whether their findings hold in the current regulatory framework.

By expanding prior literature in this way, we identify the following results. First, we confirm previous findings, such as those by Brav and Lehavy (2003), in a larger and more recent sample (2001–2007) that the information value in target price changes is incremental to that contained in stock recommendation levels. Second, there is evidence that the information in target price changes is even more important than the information in recommendation levels for predicting future abnormal returns. However, in accord with our arguments above, we do not find abnormal returns for target price changes in each recommendation level. Hence, the investment value depends on the combination of the two signals, the recommendation level and the target price change. Specifically, we provide robust evidence that contradictory analyst signals neutralize each other, particularly in the month after the announcement. Therefore, third, abnormal returns follow the direction of the target price change when it is in line with the recommendation level and do not clearly follow the direction of the target price change when it contradicts the other signal (e.g., buy combined with a large target price reduction). Finally, our findings indicate that large target price downgrades can mainly be explained by preceding stock price decreases, whereas upgrades contain information other than historical stock price changes. Consequently, analysts are unaware of the contradictory character of certain signals as they adjust target prices to stock price decreases while maintaining a buy recommendation, for example.

Our findings have implications for researchers and investors. Future research on the investment value of target price changes should consider recommendation levels because our study reveals asymmetric reactions among different recommendation levels. At the same time, investors should be highly cautious about recommendations when they are issued with large contradictory target price changes. Rather than blindly following an analyst’s recommendation, investors may put more weight on the change of the corresponding target price.

The paper proceeds as follows. In the next section, we introduce the data, provide descriptive statistics, and illustrate the methodology. In Section 2.3, we present and interpret the empirical findings, provide further analyses, and provide potential explanations for our findings. Finally, we conclude by summarizing the results.

2.2 Data and Methodology

2.2.1 Data, Variables, and Descriptive Statistics

Target prices and recommendations come from the Institutional Brokers Estimates System (I/B/E/S), which was integrated with the widely used First Call Database in 2000. Analyst reports contain different outputs, such as recommendations, target prices, and earnings forecasts, which are stored in different tables. I/B/E/S claims to offer access to more than 6.5 million research documents from more than 850 brokerage firms.^[5]Among other information, the database contains the name of the company covered, the name of the analyst, a target price, and a recommendation between one and five. A recommendation of one represents a strong buy; two, a buy; three, a hold; four, a sell; and five, a strong sell. If a broker uses another scale, I/B/E/S converts the broker’s recommendation to its five-point scale. Returns are obtained from the CRSP database.

The sample comprising the target prices covers the period from January 2001 to December 2007. There are two main reasons for this choice. First, we want to avoid the influence of heavy market turbulences during the high tech bubble and the financial crisis on our results. Second, there were several regulatory changes around 2000, including Regulation FD, NYSE 472, and NASD rule 2711, which caused a decrease in overall analyst optimism, as presented in Bradshaw (2011). Beginning our analysis in 2001 prevents these changes from affecting our analysis. We restrict the sample to 12-month target prices and to companies listed in the United States by using the indicator variables HORIZON and USFIRM from I/B/E/S. Although we concentrate on the U.S. market, the analyses can generally be expanded to non-U.S. markets, as shown in Wallmeier (2005), Kerl and Walter (2008), or Bonini et al. (2010). We require the name of the analyst and a previous target price from the same analyst for the firm not to be older than one year. Target prices for which no recommendation is available from the same analyst on the date of the announcement or whose stock price was below one dollar at the time of the announcement are dropped from the sample. Taking these conditions into account, we are left with 253,756 target price change observations.

Table 2.1: Descriptive Statistics for the Target Price Changes

illustration not visible in this excerpt

Target price changes and recommendations for U.S. firms are obtained from the I/B/E/S database. The sample consists of all target price changes available between January 2001 and December 2007 for which a recommendation reiteration or change is available from the same analyst at the time of the target price change announcement. Further, only considered are target prices for firms that have an available return in the CRSP database for at least one month after the target price announcement. Observations for stocks with a stock price below $1 at the announcement date of the target price are excluded.

Because our study builds on research by Brav and Lehavy (2003), Gleason et al. (2012), Huang et al. (2009), and Da and Schaumburg (2011), a brief comparison of their samples and ours is in order. The first main difference is that our sample spans the years 2001–2007, whereas the previous studies are based on target price changes from around 1997 to 2004. The Brav and Lehavy (2003) sample covers only the bull years of 1997–1999. Due to the significant differences in overall market returns for these two periods^[6]and the fact that the value of stock recommendations is known to depend on the overall market condition (see Barber et al. 2003), it is not clear ex-ante whether target price changes in our sample are also correlated with future abnormal returns. The change in the regulatory environment around 2000 led to more private information being disclosed to the public and, therefore, possibly less informative target prices. Consequently, we expect smaller abnormal returns, as documented in Da and Schaumburg (2011). Furthermore, Huang et al. (2009) and Da and Schaumburg (2011) use consensus recommendations and target prices at fixed dates, whereas our study is conducted on the individual analyst level on an ongoing basis. We treat all analysts the same because Bradshaw and Brown (2012) and Bonini et al. (2010) do not find systematic differences in the target price forecasting abilities of analysts. Finally, our sample covers more than 5,200 companies, whereas the analyses by Huang et al. (2009) and Da and Schaumburg (2011) include approximately 3,000 entities.

For our trading strategies, we sort stocks according to their scaled target price changes (∆TP/P). ∆TP/P is the difference between the current (t) and prior target price (TP) issued by the same analyst, deflated by the closing stock price (P) outstanding at the current date.

illustration not visible in this excerpt

(2.1)

Given our research question, we focus on changes in target prices, in contrast to Da and Schaumburg (2011), who consider the levels of the implicit target price return. Studies by Womack (1996) and Jegadeesh et al. (2004) provide evidence that changes are more informative than levels. Moreover, we believe that this measure might have a higher informational value than the target price change scaled by the previous target price as in Feldman et al. (2012). The reason for this measure, which is in line with Brav and Lehavy (2003), is that investors consider target prices relative to current prices when making investment decisions. The measure therefore corrects for preceding price changes, which can forestall investment value from analysts. Table 2.1, Panel A shows the number of target price changes, analysts and equities per year for our sample. Although target price change observations are relatively few in 2001, with only 18,560 observations, target price coverage in I/B/E/S has been increasing steadily and was 47,095 in 2007.

The distribution across the recommendation changing categories in Table 2.1, Panel B shows that 85% of the target price changes are not accompanied by a recommendation change, and most of these recommendation reiterations are strong buy, buy, or hold recommendations. This bias toward positive recommendations is well documented in the literature. Explanations can be found, for example, in Bradshaw (2011).

The sample contains remarkably more hold, sell, and strong sell recommendations issued with target prices (104,939) than the sample (20,881) used by Brav and Lehavy (2003), presumably due to the observed time period. As mentioned above, the major part of the target prices in our sample were announced during the period after the regulatory changes became effective, which led to a generally greater share of negative recommendations.^[7]

Table 2.2 shows the mean scaled target price change ( , the mean scaled price change , the mean implicit return change , and the implicit return estimate itself within the extreme target price change groups. We winsorize the values at the 1% and 99% levels, as in Brav and Lehavy (2003) and Huang et al. (2009), to prevent outliers from driving our results. For every month, target price change quintile breakpoints are calculated using all data up to the preceding month for every recommendation changing category.^[8]This growing window approach allows us to use all available information up to the actual month, which is also possible for an investor. Target price changes are classified as most favorable (least favorable) if they exceed (fall below) the highest (lowest) quintile breakpoint. This research design is in line with prior literature, such as Brav and Lehavy (2003). In general, large price changes tend to precede large target price changes in Table 2.2 (e.g., –46.69% and –64.19% for the least favorable target price changes in the strong buy reiteration), keeping the change of the implicit return estimate fairly small (1.36%). One could argue that the change of the implicit return estimate might be a superior investment signal because it better represents the shift in the analysts’ opinion. However, to be in line with Brav and Lehavy (2003), we use target price changes as the primary investment signal. When we consider the change in the implicit return estimate as an investment signal, the abnormal returns in the next section are smaller, but the results remain qualitatively the same. In general, Table 2.2 demonstrates that the target price investment signals from the most and least favorable portfolios contradict, in their respective categories, the signals from the corresponding recommendations. In the case of the two positive buy recommendation signals, for example, the least favorable quintile portfolios display large negative target price cutbacks of more than 50%.

It is interesting that the actual implicit return estimates (fourth row of Table 2.2) for the portfolios with the least favorable target price changes are, in several cases, larger than the estimates for the portfolios with the most favorable target price changes. This finding particularly holds for the buy and strong buy categories, indicating that analysts continue to believe in the stock’s performance even if they decrease the target price. Because these observations constitute a large fraction of the overall sample (55%), it is not surprising that Gleason et al. (2012), who sort stocks by the target price’s implicit return estimates, do not find a strong relation between the information in target prices and future abnormal returns.

Table 2.2: Average Change of Target Prices, Prices, and Implicit Return Estimates Within Recommendation Changing Categories

illustration not visible in this excerpt

For recommendation changing categories quintile breakpoints are calculated from the beginning of the sample up to the previous month using the scaled target price changes ∆TP/P ((TP_t-TP_t-1)/P_t). A target price change ∆TP/P is defined as most favorable (least favorable) if it exceeds (falls below) the highest (lowest) quintile breakpoint of the previous month. For these extreme target price changes this table shows the average scaled target price change ∆TP/P, the average scaled price change ∆P/P ((P_t-P_t-1)/P_t), the average change of the implicit return estimate ∆(TP/P) (TP_t/P_t-TP_t-1/P_t-1), and the actual implicit return estimate itself TP/P (TP_t/P_t-1). The values are winsorized at the 1st and 99th percentiles to mitigate the possible effect of extreme observations.

Finally, it is worthwhile to take a closer look at the distribution of recommendation levels within the extreme portfolios. Although the results in Brav and Lehavy (2003) show that abnormal returns are correlated with past target price changes, the results do not directly imply that target price changes convey valuable information incremental to that contained in recommendations. A study by Barber et al. (2009) shows that abnormal returns to analysts’ stock recommendations stem from both the recommendation levels assigned and the changes in those recommendations. Because abnormal returns vary across recommendation levels and target price changes are correlated with recommendation levels, the investment value for target prices found by Brav and Lehavy (2003) might be due to the missing control for the recommendation level. That is, the portfolios with the most favorable target price revisions are biased toward more favorable recommendation levels, whereas the portfolios with the least favorable target price revisions include less favorable recommendations. Table 2.3 presents the overall distribution and the most and least favorable target price changes for each recommendation changing category. The results provide descriptive evidence of a correlation between the target price change and the recommendation level. Interestingly, even within the reiterations in Table 2.3, Panel B, there is an asymmetry in the distribution of the recommendation levels. The portfolio with the most favorable target price revisions contains more buy (35.72%) and strong buy recommendations (25.32%) than the portfolio with the least favorable target price revisions (30.88% and 22.41%, respectively). The opposite applies to the hold and sell categories. Because more favorable recommendations within the recommendation reiteration category have higher abnormal returns (see Barber et al. 2009), this asymmetry might drive some of the results in Brav and Lehavy (2003). However, given the size of the average abnormal returns reported by Barber et al. (2009), this asymmetry does not seem strong enough to explain the predictive value of the most favorable target price revisions. Nevertheless, this relation should be taken into account when drawing conclusions about the information value of target price changes.

Table 2.3: Distributions of Recommendation Changing Categories

illustration not visible in this excerpt

For recommendation changing categories quintile breakpoints are calculated from the beginning of the sample up to the previous month using the scaled target price changes ∆TP/P. A target price change ∆TP/P is defined as most favorable (least favorable) if it exceeds (falls below) the highest (lowest) quintile breakpoint of the previous month. This table shows the percentage of changing categories for the highest (most fav.) and the lowest (least fav.) target price changes in comparison to the overall distribution of the recommendation changes within the downgrades (Panel A), reiterations (Panel B), and upgrades (Panel C). The recommendations are encoded as follows: 1 = strong buy, 2 = buy, 3 = hold, 4 = sell, 5 = strong sell.

2.2.2 Methodology

To study the investment value of target prices more closely, we analyze the predictive value of extreme target price changes conditional on the following recommendation levels: strong buy, buy, hold, sell/strong sell. The sell and strong sell recommendations are combined because the number of observations in these categories is very low (4.5% of all reiterations). If not otherwise noted, the sample is confined to recommendation reiterations, which constitute the bulk of observations (see Table 2.1, Panel B).

Disaggregating the class of recommendation reiterations into the respective recommendation levels allows us to investigate our research questions, which are, first, whether target price changes, not only on average but also for each recommendation level, provide valuable information, and second, whether target price changes provide more valuable information than the recommendation level. Further, we can examine whether the observed abnormal returns are consistent with the recommendations. Assuming that analysts interpret the information conveyed by their target price changes correctly, on average, large reductions of target prices in combination with strong buy recommendations, for example, should not be followed by average negative abnormal returns.

Analyzing the predictive value of target price changes is especially interesting for the extreme recommendation levels. On the one hand, large target price increases (reductions) for strong buy (sell) recommendations may not provide valuable information to the market because the recommendation already provides a trading signal. On the other hand, because recommendations are bounded from above (strong buy) and below (strong sell), analysts must resort to target price increases (decreases) to signal private information about an increase in the undervaluation (overvaluation) if the stock already has a strong buy (sell) recommendation.

We use calendar time regressions and calculate post-event abnormal returns to test for abnormal performance, as in Brav and Lehavy (2003). In the calendar time regression approach, for each recommendation category, quintile breakpoints on scaled target price changes ∆TP/P are calculated, including all data up to the preceding month. A target price change ∆TP/P is defined as most favorable (least favorable) if it exceeds (falls below) the highest (lowest) quintile breakpoint. These stocks enter the respective portfolio at the close of trading on the first trading day following the date of an individual analyst’s announcement to change her target price and remain in that portfolio for a predefined time span. Waiting one trading day ensures that the portfolios are based on available information. Although most target prices typically are issued before the close of trading, we prefer to avoid a potential bias caused by the possible inclusion of event returns, which Brav and Lehavy (2003) have shown to be large for high target price changes. It is plausible to assume a similar time structure for the abnormal returns after target price changes and after recommendation changes. Given the evidence from Green (2006), waiting one trading day should lead to a high reduction of abnormal returns, which, in principle, can be achieved by an investor because subscribers can access analyst reports in real time. Hence, the abnormal returns we find are a conservative estimate.

We assume a one-dollar investment in every stock entering the portfolio. The return for a portfolio is

illustration not visible in this excerpt

(2.2)

where R_it is the return for stock i on day t, n_jt is the number of stocks in portfolio j on day t, and x_it is the value of the investment in stock i on day t -1. Computing portfolio returns in such a buy-and-hold manner avoids the upward bias in equal weighting documented by Canina et al. (1998). Note that a stock can enter a portfolio even if it is already in the portfolio because different analysts can cover the same stock. Because the calendar time approach eliminates the problems of cross-sectional dependencies, this will not result in misleading conclusions.^[9]

We test the abnormal performance of each extreme quintile portfolio using the three-factor model developed by Fama and French (1993) with an additional momentum factor following Carhart (1997):^[10]

illustration not visible in this excerpt

(2.3)

In this model, represents the average abnormal return of portfolio j.

Note that our approach ensures that the abnormal returns presented in this paper are properly adjusted for momentum effects, which clearly play an important role because extreme target price changes are preceded by large price changes (see Table 2.2).

Finally, we examine whether the reported abnormal returns imply significant trading profits. To estimate total round-trip transaction costs for buying and selling, we use the results of Keim and Madhavan (1998), who provide an estimation procedure for the costs incurred by institutions in trading exchange-listed and NASDAQ stocks depending on their market capitalization. Similar to Liu and Strong (2008), we impose an upper bound of 2% for the half-way transaction costs to eliminate unreasonable estimates. Liu and Strong (2008) argue that transaction costs decline over time; in particular, decimalization in 2001 increased liquidity such that it lowered costs for buying and selling, as in Da and Schaumburg (2011). Therefore, the transaction costs used in this paper can be interpreted as an upper bound for the actual transaction costs and ensure that the abnormal returns after transaction costs present a lower bound of the profit that could have been realized by an institutional investor. This conservative perspective of identifying abnormal returns after costs ensures that it would be profitable for investors to follow the trading strategies. These returns cannot be attributed to market imperfections.

2.3 Empirical Results

2.3.1 Calendar Time Portfolios

Table 2.4 presents the results for the calendar time regressions. Given a holding period of one month, Panel A of Table 2.4 shows that the portfolios with the most favorable target price changes produce significant abnormal monthly returns of approximately 1% for the strong buy and buy recommendations and approximately 0.6% for the hold recommendation, which is higher than the 0.4% in Huang et al. (2009). For the least favorable portfolios of the reiteration categories, only the abnormal returns in the sell recommendation are significantly different from zero, at –0.82%. This finding demonstrates that the investment value of target price changes depends on the recommendation level for reiterations. Further, this detailed look at the results reveals an interesting pattern. Recalling the contradictory character of buy (sell) recommendations in combination with the least (most) favorable target price changes, it appears that confirmatory analyst signals reinforce each other, whereas contradictory signals weaken each other. The abnormal returns are more pronounced when the signals go in the same direction, as with (strong) buy and most favorable, which leads to monthly abnormal returns that are significantly different from zero. In contrast, the reaction for the sell recommendation is insignificantly different from zero. For the least favorable portfolios, there is a significant return for the sell category, in which the two analyst signals are confirmatory. From the overall perspective of the last column, the risk-adjusted returns of 0.70% (–0.21%) for the most (least) favorable target price changes are consistent in magnitude and direction with those found by Huang et al. (2009) and Da and Schaumburg (2011), acknowledging that they form their portfolios on consensus signals.

Row five of Table 2.4, Panel A, sets out the t-statistics for the difference in abnormal returns between the portfolios with the most and least favorable target price revisions. The spreads are clearly significant within each recommendation category. Hence, target prices provide additional information to the markets.

[...]

---

^[1] For this chapter see Kanne et al. (2012). For an early version of this research project please see Kreutzmann (2010, Chapter 4).

^[2] See, among others, Womack (1996), Barber et al. (2001), Jegadeesh et al. (2004), Green (2006), and Barber et al. (2009).

^[3] See Brav and Lehavy (2003), Gleason et al. (2012), Huang et al. (2009), and Da and Schaumberg (2011). For an exhaustive literature review, we refer to Ramnath et al. (2008) and Bradshaw (2011).

^[4] The stock price of Bank of America was approximately $7 in September 2011. Source of the analyst announcement: The Street (2011).

^[5] See Thomson Reuters (2011).

^[6] The average yearly return of the S&P 500 was 2.73% from 2001–2007, 5.76% from 1997–2004, and 25.73% from 1997–1999.

^[7] See Barber et al. (2006).

^[8] In addition to this growing window approach, we calculate the results on a rolling window basis for the preceding year. The results are qualitatively the same but are slightly weaker.

^[9] The heteroscedasticity problem, which might arise from the changing composition of the portfolios, as described in Mitchell and Stafford (2000), does not affect our conclusions for two reasons. For the portfolios from which we draw the main conclusions, heteroscedasticity seems implausible because the number of stocks in these portfolios is always high in terms of diversification (e.g., for the one-month holding period, the average number of stocks per portfolio is approximately 170). Furthermore, we use heteroscedastic robust estimates.

^[10] R_j,tis the return of portfolio j on day t, R_mis the return on a value-weighted market portfolio, R_fis the one-month Treasury bill rate, SMB is the return on a zero-investment portfolio calculated by the return on a portfolio consisting of small market capitalization stocks minus a portfolio of stocks with high market capitalization, HML is calculated by subtracting the return of a portfolio of low book-to-market stocks from a portfolio of high book-to-market stocks, and UMD is the return on a portfolio of stocks with high returns in the preceding year minus the return on a portfolio of stocks with low returns in the preceding year on day t. The factor-portfolio data are obtained from Kenneth French’s website.