Between the Lines: Investor Responses to Analyst Bias, Firm Guidance, and Credibility Cues Authors Elias Johansson & Simon Reinbro Supervisor: Viktor Elliot Master’s Thesis in Accounting and Financial Management Spring 2025 Graduate School, School of Business, Economics and Law, University of Gothenburg, Sweden I Abstract While earnings surprises in relation to analyst forecasts are well studied, less is known about how investors react when a firm’s own earnings guidance proves inaccurate. This thesis addresses this by investigating whether the source of a forecast error, firm issued guidance versus analyst consensus, differentially influences investor reactions. Using a sample of annual earnings announcements from publicly listed companies between 2010 and 2019, the study employs an event study methodology to capture short term abnormal stock price reactions to earnings surprises. Panel regression analysis is used to examine patterns in stock price volatility related to persistent guidance behaviors, such as underpromising or overpromising over multiple event windows. Short term price reactions are significantly stronger when analyst forecasts turn out to be wrong compared to the actual EPS, in relation to similar errors in firm guidance. Firms that Underpromise & Beat their own guidance experience increased stock price volatility, which suggest that repeated forecast failures erode management’s credibility. This is reinforced by the extended event windows showing increasing differentiation between guidance behaviors over time. Although the models lack statistical significance, the directional consistency of results aligns with signaling and disclosure theory, suggesting that trust and transparency influence investor responses more than the immediate reaction window. This thesis contributes to finance literature by showing that analyst forecasts have a stronger impact in the short term but also the effects of guidance credibility. Key words: Firm Guidance; Analyst Consensus; Investor Trust; Earnings per share (EPS); Event Study; Credibility; Underpromise; Overpromise II Acknowledgement We would like to take this opportunity to express our sincere gratitude to all those who generously contributed their time, knowledge and insights to this study. We are also profoundly grateful to our supervisor, Viktor Elliot, for his expert guidance, constructive feedback and continuous encouragement throughout the research process of the study. Simon Reinbro & Elias Johansson 2025 Gothenburg, 2025 Gothenburg, 2025 Simon Reinbro Elias Johansson III 1. Introduction ............................................................................................................................................... 1 1.1 Background ........................................................................................................................................ 1 1.2 Problematization ................................................................................................................................. 2 1.3 Purpose & Research Question ........................................................................................................... 2 2. Theoretical Framework ............................................................................................................................. 4 2.1 Foundational Theories ........................................................................................................................ 4 2.1.1 Signaling Theory ....................................................................................................................... 4 2.1.2 Efficient Market Hypothesis ...................................................................................................... 5 2.1.3 Disclosure Theory ..................................................................................................................... 5 2.2 Previous Research & Findings ........................................................................................................... 6 2.2.1 Firm Guidance & Analyst Forecasts ......................................................................................... 6 2.2.2 The Impact of Credibility on Investor Trust ............................................................................... 8 2.2.3 Short-Term Market Reactions to Underpromising & Overpromising ......................................... 9 3. Method ..................................................................................................................................................... 12 3.1 Methodology ..................................................................................................................................... 12 3.1.1 Event Study ............................................................................................................................ 12 3.1.2 Regression Models ................................................................................................................. 12 3.2 Data Collection ................................................................................................................................. 16 3.3 Descriptive Statistics ........................................................................................................................ 19 3.4 Error Term Calculations .................................................................................................................... 20 3.5 Ethical Considerations ...................................................................................................................... 21 3.5.1 Ethical Implications ................................................................................................................. 21 3.5.2 Use of Artificial Intelligence ..................................................................................................... 21 3.6 Limitations ........................................................................................................................................ 21 3.6.1 Methodological Limitations ..................................................................................................... 21 3.6.3 Empirical Limitations ............................................................................................................... 22 3.6.4 Practical & External Limitations .............................................................................................. 22 3.7 Robustness Check ........................................................................................................................... 22 4. Results & Robustness Checks .............................................................................................................. 24 4.1 Results - Hypothesis 1...................................................................................................................... 24 4.2 Results - Hypothesis 2...................................................................................................................... 26 4.3 Results - Hypothesis 3...................................................................................................................... 29 4.4 Robustness Checks ......................................................................................................................... 33 5. Analysis & Discussion ............................................................................................................................ 36 5.1 Findings, Theory & Prior Research .................................................................................................. 36 5.1.1 Hypothesis 1 ........................................................................................................................... 36 5.1.2 Hypothesis 2 ........................................................................................................................... 37 5.1.3 Hypothesis 3 ........................................................................................................................... 39 5.4 Future Research ............................................................................................................................... 42 6. Conclusions & Final Discussion ............................................................................................................ 43 References ................................................................................................................................................... 45 Appendix ...................................................................................................................................................... 49 IV List of Tables Method and research design Table 1 - Overview of Hypotheses, Variables & Regression Models .......................................................... 15 Table 2 - Variable Description..................................................................................................................... 18 Table 3 - Consort Table ............................................................................................................................... 19 Table 4 - Summary Descriptive Statistic. .................................................................................................... 20 Results Table 4.1 – RE Regression with clustered SE. ......................................................................................... 25 Table 4.2 - Panel Regression of post-announcement Stock Volatility ........................................................ 27 Table 4.3 - Panel Regression of Post-announcement Stock Volatility by Guidance Strategy .................... 29 Table 4.4 - Panel regression of CAR between groups ............................................................................... 32 Table 4.5 - Panel Regression of CAR against neutral Base. ...................................................................... 33 Table 4.6 - Random Effects GLS Regression with Firm-Level Clustered Standard Errors ...................... 34 V 1. Introduction 1.1 Background Earnings guidance has a crucial role in shaping expectations for investors and market behavior. Firms will frequently release forward looking earnings per share (EPS) targets to provide insight into their expected future financial performance. Meanwhile, financial analysts separately evaluate EPS using a combination of market conditions and firm-specific data (Alowski, Feng & Skinner, 2006). Analysts predict earnings management strategies with the purpose of minimizing minor declines which affect how earnings projections match the reported earnings. This alignment implies that companies may manage earnings to meet or slightly exceed analyst expectations, lowering the probability of negative stock price reactions (Burgstahler & Eames, 2003). Based on existing research it is not clear whether investors react more strongly to firms missing their own guidance, or the analyst consensus. While both firm guidance and analyst consensus influence investment decisions, we do not know whether investors rely more on one or the other. Prior research indicates that investors regularly overestimate analyst consensus, even when they contain systematic inaccuracies, such as consistent and predictable errors in analyst earnings consensus (So, 2013). In contrast, the credibility of firm guidance may influence investor responses, with firms that regularly provide correct information benefiting from higher investor trust (Hutton & Stocken, 2021). Credibility management is also an important part of earnings guidance. Firms might underpromise to increase the long-term investor trust and to mitigate the negative consequences of missing expectations (DeFond & Park, 1997; Hutton et al, 2003). Firms that overpromise by setting high EPS targets risk losing credibility if they fail to meet expectations on a continuous basis. In a symmetric world, investor reactions to earnings guidance would align with expectations. Companies with a strong track record of accurate forecasting would see a positive initial market reaction, as investors trust their credibility. If these firms deliver an earnings surprise, the market response would be even stronger, boosting their reputation and stock prices. Conversely, firms with a history of overpromising would face a weak initial market reaction due to investor skepticism. If these firms fail to meet expectations, the market reaction would be proportionally weak, reflecting a consistent response. However, current research indicates asymmetry in investor reactions to underpromising and overpromising strategies. Firms that consistently overpromise face stronger market penalties when they fail to meet expectations. Research suggests that investors value analyst estimates, as firms often overestimate their recommendations (So, 2013; Hollander, Pronk, & Roelofsen, 2010; Matsumoto, 2002; Graham, Harvey, & Rajgopal, 2005). There is ongoing debate in financial research about the short- and long- term effects of firm guidance errors versus analyst projection errors on the overall market. 1 1.2 Problematization Although earlier research has studied investor reactions to earnings surprises to a large extent, it has mostly overlooked the strategic dimension of guidance credibility. In particular, how market responses to firms that frequently exceed conservative guidance versus those that consistently miss their optimistic guidance. Credibility building strategies such as Underpromising & Beat may shape investor trust and stock price stability over time. Traditional finance theories, such as the Efficient Market Hypothesis (Fama, 1970), suggest that all publicly available information should be incorporated into stock prices. In contrast to this, literature in behavioral finance (Kahneman & Tversky, 1979) emphasizes that investors might depart from rational expectations due to cognitive biases such as loss aversion, overconfidence or their trust level in the information sources. Fewer studies have researched if investors place different weight on earnings guidance from firms guidance compared to analyst consensus. Current research does not fully consider how a firm’s track record with guidance shapes how investors respond to new information. This is especially important as earning forecast statements have an increasingly central role in how companies communicate with the market. This master dissertation examines how the credibility and accuracy of earnings guidance shape investor reactions in the short term. It also offers insight into how company communication influences investor decisions and how information is priced in the market. 1.3 Purpose & Research Question This study aims to investigate how investors respond to differences between analyst consensus- and firm guidance-error. Our specific goal is to find out if market reactions are differently affected when a firm- estimated EPS and actual earnings diverge compared to when analyst consensus diverge, which measure the difference between analyst consensus EPS projections and actual earnings. This study offers insights into how investor reactions are influenced by the credibility of company guidance by reviewing the weight of these two sources of earnings forecasts. This study will investigate if investors respond differently to strategies that are overpromising versus underpromising and how this affects stock price volatility and abnormal returns over time. Alongside confirming already theorized hypotheses within economic theory, the marginal contribution of this thesis are mainly in extending signaling and disclosure theory by showing how firm guidance behaviors function as a credibility signal that influences investor trust and stock price volatility. Recent literature has mainly analyzed short-term market reactions to earnings surprises and the signaling value of voluntary disclosure while this study connects persistent guidance behavior with market stability. Showing that under promising firms faced reduced volatility because of accumulated credibility while overpromising face some negative effects. The study also provides new evidence for limits to market efficiency, showing that even publicly available, credible guidance is often underutilized by the market, leading to temporary mispricing and return anomalies. The thesis adds empirical depth to various aspects in ongoing discussions about how trust, signaling, and strategic disclosure influence investor behavior and asset pricing over time. 2 Based on the problem discussion and stated purpose of the study, the following research question have been formulated as follows: - How do investors react to earnings forecast errors issued by firms versus analysts and how do these reactions affect stock price volatility, abnormal returns and market trust in the firm. By addressing this question, we can learn how investors respond to earnings guidance and which source of expectation carries the greatest weight in shaping market reactions. The findings offer a deeper understanding of how investors prioritize and process information, and whether firm credibility and investor trust are influenced by a firm’s guidance behavior. The results allow us to explore whether firms possess a strategic advantage in EPS targeting by using guidance not only to inform but to shape investor expectations. 3 2. Theoretical Framework 2.1 Foundational Theories The theoretical foundation of this thesis is composed of the signaling theory, the Efficient Market Hypothesis (EMH) and disclosure theory, where each theory gives a different perspective on how information is communicated and processed in the financial markets. Signaling theory explains the role of credible communication in the sense of reducing informational asymmetries where firms use voluntary disclosures to signal financial health and reliability (Skinner, 1994). EMH proposes that stock prices reflect all available information by suggesting that investors cannot achieve abnormal returns through their own analysis (Fama, 1970). Disclosure theory complements both of these perspectives by examining how companies strategically disclose or withhold information to form the investors perspective, balance transparency and manage market expectations. In combination, these theories offer a framework to analyze the link between corporate disclosures, investor decision- making, and market efficiency. 2.1.1 Signaling Theory Signaling theory was developed from studies on information asymmetry where one party of a transaction has more information than the other party, which leads to potential inefficiencies in the decision-making process (Spence, 1974). The theory is relevant in markets where buyers and sellers have different levels of information by creating uncertainty about the actual quality of a product or service. This concept was first depicted in labor markets where job applicants use educational credentials as signals to employers about their skills and capabilities, even if the direct productivity impact of education was unclear (Spence, 1973). A classic example of signaling theory in product markets is the differentiation between search, experience and trust properties. Nelson (1970) argues that search properties such as price and brand reputation, can be evaluated before purchase while experience properties such as the durability of a jacket or the reliability of an automobile, can only be assessed during consumption. Lastly, trust properties such as the long-term safety of pharmaceuticals may never be fully verifiable by the consumer. In instances where experience and trust properties dominate, buyers face significant uncertainty, which increases the risk of market failures like adverse selection (Akerlof, 1970). Similar uncertainty arises in capital markets when investors must evaluate a firm’s future performance based on incomplete or asymmetric information. Firms attempt to mitigate this by engaging in what could be called “credible signaling”, including financial disclosures, earnings guidance or consistent reporting practices (Skinner, 1994). However, for such signals to be effective they must be viewed as credible and consistent over a longer timespan. In the context of this study, guidance strategies such as under promising and beating can act as credibility-building mechanisms. Firms that repeatedly issue conservative forecasts and subsequently exceed them may be perceived as more trustworthy, while those that overpromise & miss risk destroying investor confidence (Hutton & Stocken, 2021; Matsumoto, 2002). This thesis tests these dynamics by analyzing how firms guidance behavior impacts investor responses. 4 2.1.2 Efficient Market Hypothesis Furthermore, Fama (1970) research on market efficiency aligns with earlier findings on how the market does react to deviations from stakeholders expectations. In accordance with EMH, stock prices do fully reflect all available information on the market, which means that investors cannot systematically achieve abnormal returns through fundamental or technical analysis (Fama, 1970). However, Fama (1970) acknowledges that market anomalies such as under- and over-reaction to earnings announcements do happen. These variations from a supposed, instant price efficiency are not necessarily a breach of the EMH but rather a result of how new information is gradually absorbed and understood by the investors. The occurrence of short-term mispricing suggests that investors do not always react instantly or proportionally to financial disclosures, which leads to temporary inefficiencies that are gradually corrected as more information is absorbed into prices on the market. In this thesis, EMH is used as a reference point for analyzing whether different types of earnings guidance are absorbed efficiently by the market as the theory assumes. Post-announcement volatility serves as an indicator of potential pricing inefficiencies or delayed investor reactions. If stock prices do not immediately adjust to certain guidance behaviors this may lead to elevated volatility in the days and weeks following the announcement, thereby suggesting temporary deviations from informational efficiency. 2.1.3 Disclosure Theory The foundation of disclosure theory goes back to Akerlof’s (1970) earlier work on information asymmetry which demonstrated that when one party in a transaction has more information than another, the markets become inefficient. This paved the foundation for research on corporate disclosures which highlighted the need for transparency to maintain investor confidence. Healy and Palepu (2001) further discussed this by explaining that firms voluntarily disclose financial information further than what regulatory policy requires to reduce uncertainty and increase market efficiency. The credibility of these disclosures is important since investors consider the reliability of firm guidance against external analyst forecasts (Graham, Harvey & Rajgopal, 2005). Building on signaling theory (Spence, 1974) firms use voluntary disclosures as signals of financial health to stand out from what are known as lower quality firms since disclosures are known as a costly method. However, strategic disclosure management such as withholding negative news (Skinner, 1994) or purposely underpromising (DeFond & Park, 1997) can create market inefficiencies by manipulating investor expectations. In addition to this, So (2013) argues that investors may overweight analyst forecasts when firm disclosures are viewed as biased, which reinforce the role of a third-party intermediaries in financial markets. This asymmetry shows the impact of behavioral finance, where trust and past disclosure credibility shape the investor responses to earnings guidance and analyst projections. As a result, the disclosure theory 5 provide a framework for understanding how firms navigate the trade-off between transparency and strategic communication in capital markets. This study uses disclosure theory to examine the strategic implications of whether firms that consistently underpromise and beat their targets gain investor trust whilst those that overpromise & miss face prolonged market penalties. By observing post-announcement stock return volatility, this thesis explores how disclosure credibility affects investor reactions and whether deviations from expected performance increase uncertainty. In the context of this thesis disclosure theory provides a valuable lens to interpret the market consequences of strategic communication in financial reporting. 2.2 Previous Research & Findings The area of research examining how managerial communication and disclosure strategies influence investor behavior and stock prices is well established in financial academics. Foundational work from Skinner (1994) and Graham et al (2005) has researched the motivations behind earnings guidance while (Bartov, 1993; Kasznik, 1999) discussed the basis around the strategic methods managers use to influence market expectations and reported outcomes. This literature review is structured around the key behavioral mechanisms and informational dynamics that underpin each hypothesis of the thesis. It first explores how firm-issued guidance interacts with analyst consensus and investor reliance. Followed by the relationship between guidance strategy and post- announcement volatility. And lastly the impact of guidance credibility on stock performance. These sections combine both classic and recent studies that inform the theoretical framing of this research. 2.2.1 Firm Guidance & Analyst Forecasts Investor expectations are shaped by all of the information released from companies but also information about the company from external professionals. In this case we focus on two forward-looking informational sources; firm guidance and analyst consensus. - Firm guidance: refers to a firm’s management’s voluntary disclosures about expected future performance, disclosed during earnings calls, under press releases or investor presentations. These forecasts are based on internal knowledge and strategic planning. - Analyst consensus: refers to independent earnings predictions made by financial analysts, that base their predictions on publicly available information, firm disclosures and proprietary valuation models. Investor views and market responses are significantly influenced by guidance in profit expectations as humans tend to react to positive and negative news as a whole. So (2013) investigates the differences between model-based projections based on firm fundamentals and how investors react to analyst forecasts. 6 According to So (2013), there is so called "systematic mispricing" when stock prices fail to account for predictable inaccuracies in analyst estimates. The model which the author presents, forecasts earnings by taking into account firm-level factors. He finds that investors over-rely on possibly biased analyst projections. And that these fundamental-based forecasts made by the companies themselves predict actual earnings more correctly than professional analyst estimates. Despite this, investors seem to rely more on analyst forecasts even when they contain systematic biases. A trading strategy that exploits the gap between firm fundamentals and analyst forecasts generates greater abnormal returns of 5.8%–9.4% per year. This indicates that investors undervalue firm guidance and that while firm fundamentals provide greater predictive power, investor behavior continues to be linked to analyst expectations, potentially leading to mispricing and valuation errors (So, 2013). Furthermore, Hollander et al (2010) offer insights into investor reliance on the topic of analyst consensus and firm guidance. Their findings show that managers often withhold certain information during earnings conference calls, leaving investors uncertain about the firm’s performance. The authors find that the market interprets managerial silence as a negative signal, often leading to stock price declines. This suggests that investors turn to analysts as an alternative source of information when firm disclosures are incomplete, which therefore further reinforces their impact in shaping the market’s expectations. Matsumoto (2002) highlights that firm guidance is often influenced strategically when companies issue optimistic forecasts to manage investor sentiment. In contrast to this, analysts adjust their forecasts downward over time which therefore leads to a more realistic earning expectation. This suggests that investors discount firm-issued guidance and favor analyst consensus regardless of their limitations. Graham, Harvey, & Rajgopal (2005) reinforced this statement by further emphasizing the importance of analyst expectations in investor decision-making by revealing that firm management prioritizes meeting the analysts forecasts to keep stock price stability, sometimes even by adjusting financial reporting decisions to align with analyst expectations. This indicates that investors rely more on analysts as firms actively shape their disclosures to meet analysts and thereby also market demands. More recent studies have discussed the relationship between firm guidance, analyst forecasts, and stock prices. Veenman and Verwijmeren (2018) show that markets do not fully account for analyst consistent pessimism, which can result in stock prices jumping more than expected when firms report earnings that beat forecasts. In a similar way, an overly positive analyst consensus can drive stock prices in the short term, only for them to fall later as the market corrects itself. This gives indications on how investor behavior and analyst biases can contribute to mispricing. Managers also tend to adjust the tone of the announced firm guidance based on “market mood”. According to Hurwitz (2017), when investor sentiment is high, managers present more optimistic earnings forecasts and when sentiment is low they take a more cautious stand. Hurwitz (2017) also finds that sentiment influences the bias in the earnings forecasts, where managers become more optimistic during high- 7 sentiment periods and more pessimistic during low-sentiment periods which are likely unintentionally. Analysts on the other hand are more likely to align their forecasts with firm guidance if managers have a strong track record. However, the market tends to react more to analyst consensus than the firm guidance. Previous literature shows that investors tend to favor analyst forecasts compared to firm guidance when deciding upon which disclosure to take into account when making their decisions. This behavioral tendency should lead to stronger market reactions to analyst forecast errors than to guidance errors and therefore, based on the previous literature on this topic, our first hypothesis concludes: - H1: Investors react more strongly to analyst forecast errors than to firm guidance errors in the short term. 2.2.2 The Impact of Credibility on Investor Trust. Hutton & Stocken (2021) discuss the importance of accuracy in their historical forecast when identifying the investors reaction to earnings guidance. Their study shows that companies with a track record of more accurate earnings forecasts will result in a higher stock price response to the new guidance since investors will place more trust in companies that have repeatedly delivered trustworthy forecasts. Firms that have a history of providing inaccurate or unfairly optimistic recommendations on the other hand, tend to lose credibility, causing investors to discount future estimates. The link between firm reputation and market reactions suggests that credibility in earnings guidance is not static but will instead build up over a period of time, depending on firms previous forecasting strategies. Firms that have a historical background of making accurate forecasts will gain a reputation for trustworthiness which will boost market reactions to their projections Hutton et al (2003). Firms that consistently overpromise experience stronger market punishments when they fail to meet expectations which leads to the asymmetry in investor reactions between underpromising and overpromising strategies (Matsumoto, 2002; Miao & Yeo 2009). DeFond & Park (1997) discussed the purpose of signaling in corporate earnings announcements and the potential effect it has on the investors and their trust in a firm. Their findings showcase how signaling theory can help to explain why a reserved guidance strategy might result in stronger market response due to a reduced level of uncertainty and improved credibility. The study further finds that firms that signaling with more conservative guidance can mitigate adverse selection and the problems that might occur which aligns with the strategic underpromising method and offers a theoretical basis for evaluating the effectiveness. Healy & Palepu (2001) further examine this with what incentives exist for firms to disclose information beyond the regulatory requirements by focusing on the issue of information asymmetry between managers and investors. Their study suggests that managers voluntarily release data such as future guidance to reduce uncertainty and help investors understand a firm’s performance more accurately. This can improve the 8 credibility and improve the transparency. Voluntary disclosures also carry risks and misleading statements may lead to litigation and firms may unintentionally expose strategic insights to competitors. Matsumoto (2002) supports the strategy of signaling through managed expectations by demonstrating how firms use deliberate guidance to influence investor opinions. Matsumoto (2002) analyzed the strategic motives behind managing expectations over multiple reporting cycles. The findings differ from earlier studies since he identified how excessive guidance manipulation could destroy investor trust over time which contradicts the historical findings of increased trust associated with careful and consistent signaling strategies. Hutton, Miller & Skinner (2003) focused on how companies with a history of underpromising benefit from increased investor confidence as previously found by DeFond & Park (1997) and Bartov et al (2002). Unlike previous studies such as Bartov et al (2002) and Skinner (1994) which focused on immediate stock price reactions to surprises or bad news pre-announcements, Hutton, Miller & Skinner’s (2003) study emphasized the reputational capital gained through consistent conservative guidance over time. Their findings based on signaling theory, suggesting that firms use conservative guidance to build a reputation for reliability and trust. This aligns with the strategy of Underpromising & Beat, showcasing the advantages of investor confidence. The study’s focus on maintaining consistent conservative guidance assumes stable market conditions and predictable investor behavior. Although in reality, financial markets are often volatile, which raises questions about how well these findings apply across different economic contexts. These findings support the idea that firms that consistently underpromise & beat expectations are rewarded with investor trust and reduced volatility, while firms that regularly overpromise & miss face ongoing credibility issues and stronger negative market reactions. Based on the previous literature on this topic, our second hypothesis concludes: - H2: Firms that consistently Underpromise & beat EPS targets experience lower stock price volatility over time compared to firms that frequently Overpromise & Miss. 2.2.3 Short-Term Market Reactions to Underpromising & Overpromising Bartov, Givoly & Hayn (2002) discuss the rewards linked with meeting or beating analysts earnings expectations (MBE) and the role of managerial discretion in achieving such outcomes. Their study found that firms that meet or exceed the analyst consensus can often achieve significantly higher cumulative abnormal returns (CAR) compared to those that miss expectations. They further find that the "premium" is maintained even when MBE is most likely achieved through earnings or expectations management. The study further found that while these practices generate short-term gains, they might have a risk of damaging the investor trust if such patterns are to be found over time. This can be linked to the signaling theory since the act of meeting or beating expectations can signal better future performance to investors, even if achieved through managed guidance. (Bartov et al, 2002) 9 Miao & Yeo (2009) Researched the efficiency of market reactions to earnings announcements, examining whether stock prices respond rationally and thus fully incorporate new information as opposed by Fama (1970) or if mispricing occurs due to behavioral biases. One of the main findings of Miao & Yeo (2009) is that firms that report better than expected earnings seem to experience a gradual stock price increase in the weeks following the announcement event, while firms that present negative surprises experience prolonged decrease on stock prices. This suggests that the market does not instantly adjust to earnings news which creates opportunities for traders to exploit post-earnings drift (PED) which is an anomaly contradicting the semi-strong form of the Efficient Market Hypothesis (Fama, 1970). The study offers an understanding into the timing and magnitude of market reactions to earnings guidance credibility. Firms that have a track record of under promising and beating their targets may therefore experience muted initial reactions but benefit from a gradual upward price trend as investors recalculate their expectations over time. However, firms that overpromise and miss will experience an initial selloff, but the findings of Miao & Yeo (2009) suggest that these declines can persist longer than fundamental value changes would justify, which suggest an extended negative sentiment effect in the market. Alwathnani et al (2017) discuss how markets respond to earnings surprises by studying whether investor reactions are rational, delayed or exaggerated. By using an event study methodology the authors examine stock price behavior following earnings announcements and analyses whether price movements align with the fundamental values or if they show a pattern of either under reaction or overreaction. The study found that market reactions to earnings surprises are asymmetric by identifying an underreaction to positive surprises and an overreaction to negative surprises. Stocks that exceed earnings expectations tend to experience a delayed price increase which indicates that investors do not immediately adjust to the full effects of the new information. This will therefore lead to a post earnings announcement drift (PEAD) where positive returns continue to accumulate over time. Similar to Miao & Yeo (2009), stocks of firms that report worse than expected earnings experience an exaggerated initial drop with price declines often overshooting fundamental valuation adjustments. This suggests that investors are more likely to over penalize firms for missing their expectations in the short term, especially if the company has a history of overpromising. Kwon & Tang (2023) have a more neutral stance towards events such as earning announcements. They find that markets overreact to “extreme corporate events” such as M&A announcements and leadership changes, leading to short-term price spikes which are followed by a reversal as investors correct their earlier reactions. Less extreme events, such as routine earnings disclosures, seem to experience underreaction, causing gradual price adjustments over time. This pattern is found to be driven by behavioral biases where investors outweigh the significance of extreme news, resulting in higher trading volume and short-term market inefficiencies. Previous literature research shows that the market’s reactions to underpromising versus overpromising are asymmetrical. More specifically, companies that overpromise and then fall short of expectations experience more severe and quick unfavorable reactions than those that underpromise and meet expectations (Miao & 10 Yeo, 2009; Alwathnani et al 2017). There is also evidence of PEAD, which suggests that price changes happen gradually over time (Bernard & Thomas, 1989; Hirshleifer et al., 2009; Miao & Yeo, 2009; Alwathnani & Al-Zoubi, 2017). Lastly, studies on guidance behavior and managerial signaling show that meeting or exceeding expectations, even through profits management, produces immediate benefits but can harm credibility if caught (Bartov et al., 2002). All of these findings point to the short-term punishment of too optimistic corporations being more severe than the reward of underpromising firms. Thus, the third hypothesis concludes: - H3: Overpromising firms that miss earnings targets experience stronger immediate stock price declines compared to the positive reactions received by underpromising firms beating expectations. 11 3. Method 3.1 Methodology This study employs an empirical research approach to analyze investor reactions to earnings guidance credibility, analyst consensus and firm guidance. The research follows an event study methodology combined with regression analysis to assess short-term stock price reactions. 3.1.1 Event Study An event study is appropriate for this research since it allows us to measure the CARs for each year and therefore identify whether and how investors react to new information being presented. The method follows the foundational framework outlined by Brown & Warner (1985) and MacKinlay (1997), where Brown and Warner (1985) showed how using daily returns in event studies can capture short-term market reactions and MacKinlay (1997) presented a framework for examining how stock prices respond to significant events like earnings announcements. The main advantage of using this method is that it depicts how much of the price movement is due to the event itself, rather than overall market fluctuation or economic trends, also know as noise. By establishing a baseline of expected returns through the estimation window any unusual returns around the announcement date to the new information disclosed (MacKinlay, 1997). However, MacKinlay (1997) also states that the method relies on accurate model specification and the assumption of market efficiency. If the model is misspecified or if investors do not react efficiently to new information, it can lead to biased or misleading results. Choosing the right event window is also important since longer windowsmight introduce noise and unrelated influences such as external factors, which makes it harder to isolate the real effect of the event from the rest. 3.1.2 Regression Models Multiple regression models were employed in this study to analyze investor reactions to earnings guidance credibility, analyst consensus errors and firm guidance behaviors. To address the distinct research questions in each hypothesis, different regression approaches were applied to different hypotheses. Following the event study frameworks of MacKinlay (1997) and the empirical approach of So (2013), an event study regression model with CAR as the dependent variable and the absolute Firm and Analyst EPS target error terms was deployed to analyze short-term investor reactions to forecast errors arising from analyst projections and firm issued guidance. To test H1, “Investors react more strongly to analyst forecast errors than to firm guidance errors in the short term” The following regression was used: CARi,t= α + β1 * FirmErrori,t + β2*AnalystErrori,t+ γ Xi,t + μi + εi,t 12 Where CARi,t represents the Cumulative Abnormal Return for firm i at time t. FirmErrori,t represents the Firm Error Guidance Term explained below. AnalystErrori,t represents the Analyst Error Guidance Term also explained below. Xi, represents control variables such as ROA, ROE, VIX and market capitalization. ui represents firm-specific random effects and εi,t represents the idiosyncratic error term. The dependent variable CAR is calculated using the cumulative abnormal returns before, during and after the event. For this hypothesis CAR 1day, 2day, 5day and 21day was used. These CAR windows and CAR 3day used in hypothesis 2 is calculated through the following: CAR 1day = Abnormal returns on the day of the event CAR 2day = Abnormal returns on the day of the event + 1 day after the event day. CAR 3day = Sum of the abnormal returns from day t - 1 to t + 1 with t being the event day. CAR 5day = Sum of the abnormal returns from day t - 2 to t + 2 with t being the event day. CAR 21day = Sum of the abnormal returns from day t - 10 to t + 10 with t being the Event Day. And the independent variables Firm and Analyst error guidance terms were calculated by the following formulas: 𝐹𝑖𝑟𝑚 𝐺𝑢𝑖𝑑𝑎𝑛𝑐𝑒 𝐸𝑟𝑟𝑜𝑟 𝑇𝑒𝑟𝑚 = 𝐹𝑖𝑟𝑚 𝐸𝑃𝑆 𝑇𝑎𝑟𝑔𝑒𝑡 − 𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 & 𝐴𝑛𝑎𝑙𝑦𝑠𝑡 𝐺𝑢𝑖𝑑𝑎𝑛𝑐𝑒 𝐸𝑟𝑟𝑜𝑟 𝑇𝑒𝑟𝑚 = 𝐴𝑛𝑎𝑙𝑦𝑠𝑡 𝐸𝑃𝑆 𝐶𝑜𝑛𝑠𝑒𝑛𝑠𝑢𝑠 − 𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 Firm Guidance Error Term measures the difference between the firm expected EPS set by the firms themselves one year ahead of time and the actual EPS reported at the end of the year in absolute terms. It assesses the accuracy of the company earnings guidance and lays basis for whether companies consistently overestimate or underestimate their financial success in the upcoming year. The Firm Guidance Error Term is based on Wei & Zhang (2023) similar usage of the terms. A positive value indicates that the firm overestimated its earnings, while a negative value suggests an underestimation. Understanding this variable helps analyze whether firms strategically manage their guidance for reasons that this thesis aims at discovering. The Analyst Guidance Error Term measures the difference between what analysts have predicted the firm EPS to be gathered from the database of AlphaVantage one year prior to the announcement date and the firm’s actual EPS. Analysts base their EPS estimates on company specific details and broader economic market trends but their estimates often come with a margin of error (Hope & Kang, 2005). The Analyst Guidance Error Term is therefore calculated in the same way as the Firm Guidance Error Term, again following the methodology outlined by Wei & Zhang (2023). A positive value suggests that analysts were 13 overly optimistic in their earnings forecasts, whereas a negative value indicates a conservative estimate. Comparing this metric with the Firm Guidance Error Term allows for an analysis of whether investors react more strongly to errors in firm guidance or analyst consensus. Building on Brown & Warner (1985) and MacKinlay (1997) approach, we implement a multivariate regression design with time-varying controls to account for simultaneous market and firm-specific dynamics. This extended framework allows for a transparent and comparable measurement of stock price responses across multiple event windows which helps to assess whether investors weigh analyst or firm forecast deviations more heavily. For H2, “Firms that consistently underpromise and beat EPS targets experience lower stock price volatility over time compared to firms that frequently overpromise and miss.” the following regression model was used: Volatilityi,t= α+β1 * UnderPromise&Beati,t + β2 * Overpromise&Missi,t+ γ Xi,t + ui + εi,t Where Volatilityi,t Represents the realized stock return volatility for firm i at time t. UnderPromiseBeati,t represents a dummy variable for the category of the Underpromise & Beat group. OverPromiseMissi,t represents a dummy variable for the category of the Overpromise & Miss group. Xi, represents control variables such as ROA, ROE, VIX and market capitalization. ui represents firm-specific random effects and εi,t represents the idiosyncratic error term. We used a panel regression model with random effects. The random effects model adjusts for variation across firms by including firm-specific error terms and a transformation that combine within-firm and between-firm information, all under the assumption that firm-specific effects are uncorrelated with the regressors. The regressions allow us to assess longitudinal patterns in stock price fluctuations. The rolling standard deviation of stock returns is used as the dependent variable, while independent variables capture persistent guidance behaviors. To mitigate the challenge of endogeneity in panel regressions we used a random effects model which was supported by the Hausman test (p = 0.37). Furthermore, we control for observable firm-level variation by including time-varying variables such as ROE, ROA and market capitalization . For H3, “Overpromising firms that miss earnings targets experience stronger immediate stock price declines compared to the positive reactions received by Underpromising firms beating expectations.” the following regression model was used: CARi,t = α + β1*OverpromiseMissi,t + β2 * UnderpromiseBeati,t+ γ Xi,t + εi,t Where once again, CARi,t represents the Cumulative Abnormal Return for firm i at time t. OverpromiseMissi,t represents a dummy variable for the category of overpromising firms. UnderpromiseBeati,t represents a dummy variable for the category of underpromising firms. Xi, represents 14 control variables such as ROA, ROE, VIX and market capitalization. ui represents firm- specific random effects and εi,t represents the idiosyncratic error term. We again use an event study regression model. Given the need to measure investor reactions to earnings announcements, CAR over short-term event windows function as the dependent variable, while the main independent variables differentiate between overpromising firms missing their estimates and under promising firms exceeding the expectations and separate the firm into two groups: Firms that over promise but fail to meet their target EPS and Firms that underpromise but exceed their targets EPS. For hypothesis 2 and 3 we used a variable gathered from Chen et al (2023) representing the percentage deviation between firm-issued earnings guidance and actual reported EPS, calculated as: 𝐸𝑃𝑆 𝑡𝑎𝑟𝑔𝑒𝑡 𝐹𝑖𝑟𝑚 − 𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 𝐹𝑖𝑟𝑚 𝐸𝑟𝑟𝑜𝑟 𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑇𝑒𝑟𝑚: 𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 This variable captures the relative size of each firm’s forecast deviation, relative to its actual EPS. As the outliers remain large even after winsorization in the dataset, we then calculated the median absolute deviation, which equaled approximately 29.7%. This threshold was then used to classify the firm behavior as: Firms whose guidance were more than 29,7% below actual EPS were classified as underpromising and firms whose guidance exceeded actual EPS by more than 29.7% were classified as overpromising. Firms with deviations within the ±29.7% interval were excluded from the analysis to focus only on economically significant guidance behaviors, rather than minor or unintentional deviations likely attributable to noise or rounding. By using an event study method for H1 and a panel regressions for H2 and H3, our approach maintains a comprehensive analysis of how investors respond to forecast errors and firm guidance strategies. The combination of firm fixed effects, robust standard errors and alternative event windows increases the validity and reliability of our results while also recognizing the trade-offs that come with each choice of method. A mean comparison table by firm behavior category can be found in Appendix 1A. 15 Hypothesis Dependent Variable Independent Variables Model Type H1 CAR over different event Analyst Forecast Error, Firm Event study regression windows Guidance Error H2 Volatility measured as Underpromise & Beat, Panel regression with firm rolling standard deviation Overpromise & Miss random effects of stock returns H3 CAR over different event Overpromise & Miss, Panel regression with firm windows Underpromise & Beat random effects Table 1 - Overview of Hypotheses, Variables & Regression Models 3.2 Data Collection The data used in this study was sourced from several financial databases, including Eikon Refinitiv, Capital IQ, Alpha Vantage and Yahoo Finance. Although these databases are reputable and commonly used in financial research, minor differences in their coverage, methodologies for data collection, or definitions of certain variables may exist. To mitigate potential inconsistencies, cross-validation between databases was performed, ensuring accuracy and robustness of the dataset. Combining multiple sources enhanced the reliability of the dataset, reducing the risk that findings might reflect database-specific biases or omissions. The first sample included variables such as analyst consensus EPS estimates, firm-issued EPS guidance, actual reported EPS, return on equity, return on assets and the VIX index (Table 2). The raw dataset consisted of 16,734 observations across 16 variables, covering a 10-year period from 2010 to 2019. The period of 2010–2019 is chosen in order to capture a neutral market environment between the global financial crisis of 2008 and pandemic in 2020. A neutral market environment was sought to avoid skewed measures in abnormal returns and volatility amongst other important variables. This is under the assumption that economic neutrality is considered times where extremely large anomalies in the market appear. This period is used in order to increase generability throughout this paper, as it aims at targeting the general market. 16 Regression Variables Description Firm level variables EPS Target Firm Guidance Earnings per share guidance issued by the firm prior to the earnings announcement. Analyst EPS Consensus Forecasted EPS by analysts before the earnings release. Actual EPS Reported earnings per share at the time of the earnings announcement. ROA Return on Assets, net income divided by total assets. ROE Return on Equity, net income divided by shareholders’ equity. Market Capitalization (000) Firm size measured as market value of equity in thousands. CAR1 (0) Cumulative abnormal return on the earnings announcement day. CAR2 (0,+1) Cumulative abnormal return over a two-day window starting on the announcement day. CAR3 (-1,+1) Cumulative abnormal return over a three-day window centered on the announcement. CAR5 (-2,+2) Cumulative abnormal return over a five-day window surrounding the announcement. CAR21 (-10,+10) Cumulative abnormal return over a 21-day window around the earnings event. Market level variables VIX Index Market-implied volatility based on S&P 500 options. Table 2 - Variable description 17 To prepare the raw dataset for regressions in STATA we employed a data cleaning process. The first filtering step ensured variable completeness by excluding all observations with missing values for any of the key variables identified in Table 2 shown in Table 3. This included dropping firms that lacked analyst estimates, EPS guidance, actual EPS or return metrics. These steps were essential to accurately calculate both firm guidance and analyst consensus error terms. Further cleaning made sure that each firm and year combination was correct and not repeated. Duplicated entries and inconsistencies in firm or fiscal year labeling were identified and removed which ensured that each observation represented a unique and valid firm-year combination. Furthermore, observations containing extreme or implausible financial ratios, such as negative EPS values or ROE/ROA values outside reasonable range, were reviewed and discarded to mitigate the risk of data errors originating from reporting or collection of the data. Lastly, the cleaned datasets from the stages were merged into a consolidated sheet. This final merging process also included a round of validation checks to confirm that each retained observation had complete data across all required accounting and market based variables. After the completion of this process, the resulting dataset consisted of 2 457 firm-year observations (Table 3). Steps Observations Excluded Remaining Initial dataset 16 734 – Excluded due to missing Firm Guidance 11 611 5 123 Excluded due to missing key variables (EPS estimates, 6 764 4 847 Analyst guidance, Returns) Excluded due to incomplete event window data 3 457 3 307 Excluded due to duplicate or inconsistent firm-year entries 2 860 597 Excluded due to implausible financial ratios (ROA, ROE, 2 472 388 EPS) Final Cleaned Dataset 2 457 – Table 3 - Consort Table 18 3.3 Descriptive Statistics Variables Standard Frequency Mean Median Deviation Min Max N Firm level variables EPS Target Firm Guidance Yearly 2.880 2.200 2.877 -172.100 53.820 2457 Analyst EPS Consensus Yearly 1.471 1.200 7.492 -172.200 52.500 2457 Actual EPS Yearly 2.189 1.780 5.052 -172.400 55.770 2457 Firm Guidance Error Term (Absolute) Yearly 2.759 1.735 5.245 0 72.423 2457 Firm Guidance Error Term (Percentage) Yearly 61.79% -7.69% 78.20% -5565.54% 41032.69% 2457 Analyst Guidance Error Term (Absolute) Yearly -0.717 -0.117 2.912 -170.53 50.28 2457 ROA Yearly 5.11% 4.22% 58.36% -232.13% 2827.92% 2457 ROE Yearly -59.40% 11.16% 143.00% -1562318.2% 48390.68% 2457 Market Capitalization (000) Yearly $9,951,000 $13,409,610 $41,979,351 $9,951 $1,155,433,928 2457 CAR1 (0) Yearly 0.24% -0.02% 8.03% -26.90% 30.36% 2457 CAR2 (0,+1) Yearly 0.47% 0.14% 9.72% -32.95% 41.32% 2457 CAR3 (-1,+1) Yearly 0.45% 0.12% 11.85% -37.41% 41.39% 2457 CAR5 (-2,+2) Yearly 0.41% 0.08% 16.24% -43.09% 48.16% 2457 CAR21 (-10,+10) Yearly 0.00% -0.53% 48.31% -86.54% 129.03% 2457 Market level variables VIX Index Yearly 24.993 25.177 8.739 10.047 39.992 2457 Table 4 - Summary Descriptive Statistic Table 4 provides an overview of the key variables used in the analysis. The average firm guidance error term is 2,76, but the values vary from 0 to 72,43 which show the difference in how accurately firms predict their earnings. Analyst consensus also shows a significant variation, with an average error of -0.717 and a standard deviation of 2.91. Return based variables like CAR3 and CAR5 are on average close to zero, but again, there is a lot of variation, especially in CAR 21 which has a standard deviation of 48,31% which reflects the high volatility around earning announcements. Both ROA and ROE differ across firms, with negative outliers in ROE which indicates extreme losses for some observations. Market capitalizations range from under $10.000 to more than $1.1 trillion which show a highly skewed distribution. The VIX index show an average of 25 and a MAX value of 40, which is consistent with periods of high market uncertainty. A frequency table of the distribution between the overpromise and underpromise category when the neutral category has been omitted can be found in Appendix 3H for clarity and information regarding the normal distribution of that subsample. A mean comparison Table by firm behvaiour category can be found in Appendix A1. 19 3.4 Error Term Calculations When determining whether to use the absolute values or percentual values when calculating the differences between analyst/firm targets and actual targets, it is important to consider the context of the sample and the general objective of the analysis. Both methods offer advantages and limitations that can be applied to this context. Using absolute values allows for a more straightforward measure of the deviation between the actual EPS and the projections. Absolute values simplify the interpretation of the error terms when the nominal magnitude between the projections and the actuals matters more than the relative size between the two. For example, consider a forecasted income being $10 000 and the actual outcome turned out to be $9800, the absolute error is $200. The difference can be seen as small percentually, only 2%, but in reality the $200 is highly relevant and can be the decider for future projections. Contrary to this, if the forecasted income would be $0.05 per share and the actual income would turn out to be $0.00, the difference would be small in absolute terms of $0.05, but in percentage terms the difference is large at 100%. Here the relative size of the miss is more informative than the nominal value if the sample set would consist of values that are so small that a small difference in absolute terms represents a large difference in percentage terms. On the other hand, the variance in relation to the magnitude of the actual value or forecast is captured when percentual numbers are used. This method provides a normalized view of forecasting accuracy by allowing comparability across companies of various sizes or profit levels. It is especially helpful when comparing performance across industries with different scales or in cross-sectional analyses. Additionally, a percentage-based error draws attention to the proportionate importance of variances, showcasing if a firm or analyst consistently overestimates or underestimates performance. Moreover, when displaying differences in percentage terms, a critical limitation is that when the forecast errors are normally distributed between negative and positive values, the differences can become inherently skewed, especially in this case when comparing under- and overestimations. For example, if a forecast overestimates the actual result by 50%, it would result in a +50% percentage error. But if the forecast underestimates the actual result by the same amount but, the percentage error could be -100% or more in some cases depending on the denominator. This is because the percentage changes are non- linear and bounded differently on each side. This distortion means that even if the differences might be equally large in absolute terms, the percentage error distribution will be negatively skewed by overstating downside misses and understating upside ones, ultimately making the interpretation a lot more complex. Given the characteristics of the data in our sample (Table 4), absolute values were chosen as the measure for hypothesis 1. Mainly due to the fact that the overall purpose is to look at the raw deviations effect on the abnormal returns, instead of looking at the comparative differences. Furthermore the standard deviations 20 for lots of the dependent variables such as EPS, firm guidance and analyst forecasts are relatively high and the range of values includes extreme outliers even after winsorization. This large dispersion implies that percentage errors would be highly unstable and prone to distortion, particularly when EPS values are near zero or negative as displayed by the mean and medians in Table 4. For hypothesis 2 and 3 absolute percentual values will be used, in order to be able to categorize the sample set into those needed for the purpose of the study. Categorizing by magnitude would severely damage the statistical models as the variances are too large in the error term categories. 3.5 Ethical Considerations 3.5.1 Ethical Implications The ethical implications of the research approach are small considering that this study is based on publicly available data. All processes required to get secondary data, run statistical tests and assess the results are disclosed. This study has no aim of causing harm to any of the parties participating and no confidential inquiries will be addressed or discussed during data collecting. Furthermore, the study has no intention of inflicting legal harm considering all data is obtained openly. 3.5.2 Use of Artificial Intelligence In line with the guidance from the School of Business, Economics and Law at the University of Gothenburg regarding the use of generative AI in higher education (School of Business, Economics and Law, 2024), we confirm that the content of this thesis reflects our own original work. Generative AI tools have not been used to produce or generate any part of the submitted text, and no AI-generated material is presented as our own writing. All AI-assisted explorations were critically reviewed, independently verified and served only as informal guidance to support our own understanding. ChatGPT (Open AI) has been used for programming support, specifically to identify and correct coding errors and improve the scripts written in Python, Stata as well as to enhance the structure and language of the text, without altering its content or meaning. 3.6 Limitations 3.6.1 Methodological Limitations Regarding the event windows used in this study, even a 3-day window can be impacted by concurrent firm- or market-level news, especially in larger windows like ±10 days. Meaning that even though the event study employs short windows to isolate investor reactions, this does mean that the event windows are free from external effects that cause market reactions. As previously discussed, although using absolute forecast was deemed more fitting, it still ignores relative company size and profits levels, which could obscure economic significance in tiny businesses. Limitations connected to the methodology of H3 include that percentage errors may be inflated for firms with very small actual EPS as mentioned in 3.4.2, potentially overstating 21 deviations. The findings connected to H3 are also restricted to short-term investor responses and may not capture delayed reactions. 3.6.2 Theoretical Limitations The use of the signaling and disclosure theory is predicated on consistent information interpretation and logical investor behavior. In reality, past expectations, perceived managerial credibility and cognitive biases all probably have an impact on investor responses. Noticeable, this study does not explicitly include analyst function as mediators, who may dispute or reinforce instruction. 3.6.3 Empirical Limitations The sample primarily consists of U.S. large cap firms issuing annual EPS guidance between 2010–2019 which limits generalizability to other markets, smaller firms or quarterly announcements. Importantly, using realized volatility as a proxy for “credibility loss” captures general investor uncertainty but can lack direct measurement of investor sentiment or belief revision and arguably only captures a segment of the term “credibility loss”. 3.6.4 Practical & External Limitations Findings are most applicable in markets where both management guidance and analyst forecasts coexist and are trusted. In other markets such as emerging or less regulated markets, these dynamics may differ. Evolving disclosure norms, including reduced use of short-term guidance or increased reliance on non- GAAP metrics may also affect how investors react. Finally, we do not distinguish between investor types, whose sensitivity to forecast errors may vary depending on their time horizons, access to management, and reliance on analysts. 3.7 Robustness Check Several robustness tests were used to increase the empirical findings reliability across all three hypotheses. The robustness procedures were modified to fit each analytical framework based on the variations in model structures among the hypotheses. Firstly, datasets were winsorized at the first and 99th percentiles for important variables like forecast errors and abnormal returns to estimate all regression models. Winsorization is used to enhance the robustness of statistical analyses by reducing the influence of extreme outliers, and thereby avoiding eventual false negatives in the results caused by the outliers. All analyses were repeated on raw, non- winsorized data to determine how sensitive the results were to this decision. If there was small or no difference between the results, it indicated that outliers were not the driving factors in the results (Yang et al, 2024). 22 Given the structure of one observation per firm per event, a straightforward OLS regression with firm- specific dummies was carried out for Hypothesis 1 to approximate fixed effects estimates. The outcomes of random effects estimators and these models were contrasted. The acceptability of random versus fixed effects were formally evaluated for Hypothesis 2 and 3 using the Hausman specification test. For Hypothesis 2, fixed effects estimations were performed as an extra check to look at the stability of coefficient magnitudes and signs (Wooldridge, 2010) Standard errors were grouped at the firm level to account for any serial correlation and residual heteroskedasticity. Some additional specifications were applied such as clustering by event time, in this case year (Cameron & Miller, 2015). If the significance levels of important coefficients held true for various error architectures, robustness would be verified. To assess the sensitivity of aberrant return patterns to the selected window length the analysis for Hypothesis 1 were carried out over a number of event windows. Windows of 1, 3, 5, and 21 trading days were examined. Extending the window to 21 days enables the discovery of any longer-term consequences that still can be captured by investors that take the target errors into consideration, even though the main focus was on short-term investor reactions (MacKinlay, 1997). To investigate whether firm qualities impact the magnitude of investor reactions, interaction terms between analyst forecast errors and firm characteristics were added. Including such interaction terms is critical when the effect of one variable may depend on the level of another, as failing to account for these relationships can lead to biased or misleading conclusions (Burks et al, 2019). Placebo tests were used to ensure that the connections that were seen were event-driven. Stock returns and earnings surprises were randomized for Hypothesis 1 to assess whether the analyst error effects persisted under random assignment. This approach helps in verifying that the observed effects are not due to random chance but are indeed linked to the specific events being studied (Kothari & Warner, 2007). Similarly, Hypotheses 2 and 3 return patterns during non-announcement times were also examined to verify the lack of systematic disparities or pre-existing trends. In order to identify any possible multicollinearity issues, Variance Inflation Factors (VIF) will be computed for each independent variable (O’brien, 2007) We took VIF values below conventional thresholds as proof that coefficient estimates were not skewed by multicollinearity 23 4. Results & Robustness Checks 4.1 Results - Hypothesis 1 Investors react more strongly to analyst forecast errors than to firm guidance errors in the short term. Our regression analysis supports H1, as seen in Table 4.1 which summarizes the results of random effects regressions (with firm-level clustered standard errors) for the event windows around earnings announcements. The independent variables in this regression are the analyst consensus error and the firm guidance error measured as the difference between actual earnings and the consensus/guidance. The surprise value is adjusted based on the stock price to make comparisons between companies fair. Random Effects Regression with Firm-Level Clustered Standard Errors (1) (2) (3) (4) CAR_1day CAR_3day CAR_5day CAR_21day Firm Guidance Error -0.000** -0.001*** -0.001*** -0.005*** (0.000) (0.000) (0.000) (0.001) Analyst Error 0.002*** 0.006*** 0.008*** 0.025** (0.001) (0.013) (0.017) (0.027) ROA -0.001 -0.002 -0.003 -0.015 (0.001) (0.002) (0.003) (0.012) ROE -0.000** 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) (0.000) VIX -0.005 -0.019* -0.033* 0.0129* (0.004) (0.011) (0.018) (0.067) Large Cap 0.000 0.000 0.000 0.000 (.) (.) (.) (.) Constant 0.003 0.012 0.022 0.093 (0.003) (0.09) (0.015) (0.058) Observations 2457 2457 2457 2457 R2 0.010 0.011 0.009 0.007 24 Adjusted R2 0.008 0.009 0.007 0.005 Standard errors in parentheses. Standard errors Clustered at firm level. * p < 0.10, ** p < 0.05, *** p < 0.01 Table 4.1 Regression of CAR windows on Guidance Error Terms As seen in Table 4.1, the regression results for short-term reactions under the different event windows. All models include the same set of controls being ROE, ROA and VIX with winsorized data (1% tails) to mitigate outliers. Observations used in the regressions are N = 2 452. Stars indicate significance levels (*** p<0.01: **p<0.05: *p<0.10). Across all event windows shown in table 4.1, analyst forecast error has a positive and significant coefficient. In the 1-day window, a larger analyst forecast error is associated with a positive AR on the announcement day (β = 0.002, p < 0.01). This suggests that when the actual EPS is announced and then exceeds the analyst prediction the stock price rises as a result. Even though the 1-day window explains only about 1% of the variation, the market reaction suggests that investors heavily factor in analyst forecasts when pricing stocks before earnings announcements. When actual earnings are revealed and differ from expectations, stock prices adjust quickly, correcting earlier potential mispricings. In the 3- day window (announcement day and the subsequent two days), the coefficient on analyst forecast error rises to β = 0.006 (p < 0.01) and remains strongly positive in the 5-day window (β = 0.008, p < 0.01). For the 21-day window, the cumulative effect of analyst surprises becomes more pronounced (β = 0.026, p < 0.05). This pattern indicates that investors continue to react to analyst-related surprises beyond the immediate announcement day. In contrast to analyst surprises, the market’s reaction to Firm guidance error is less pronounced initially. On the announcement day, the coefficient on firm guidance error is close to zero and slightly negative (β = -0.000, p < 0.05). This indicates that a negative guidance surprise produces a small, negative abnormal return on day 1 which means that there is a low penalty for Underpromising & Miss the guidance. However, as the window is extended, the impact of guidance errors becomes more evident. By the 3-day window, the negative coefficient on firm guidance error grows in magnitude and is statistically significant at the 1% level, implying that investors do start to incorporate the implications of firm guidance miss over a couple of days. In the 5-day window, the effect is slightly larger (β =- 0.001, p< 0.01) and by 21 days the coefficient reaches β = -0.005 (p < 0.01). The consistently negative sign indicates that when firms miss their own guidance, stock returns tend to be lower in the days and weeks following the announcement. In other words, management guidance errors do matter for valuation, but their impact unfolds more slowly and remains smaller in absolute magnitude than analyst related surprises. A Wald test comparing the coefficients of firm forecast error and analyst consensus error yielded an F- statistic of 11.32 (p = 0.0008) which supported that the two effects are statistically different (Appendix 1B). 25 4.2 Results - Hypothesis 2 Firms that consistently underpromise and beat their EPS targets experience lower stock price volatility over time compared to firms that frequently overpromise & miss. We test H2 using panel regressions that categorize firm-year observations by their guidance strategy outcomes. Specifically, we identify each earnings announcement in our sample as one of three categories: ● Underpromise & Beat - The firm’s guidance was conservative and the firm largely exceeded its target. ● Overpromise & Miss - The firm’s guidance was overly optimistic, resulting in a larger miss. ● Neutral - The firm’s guidance was very close to actual earnings, effectively meeting expectations As previously mentioned, the dependent variable is a measure of stock return volatility in the period following the earnings announcement, operationalized as the variance of abnormal returns over different post-event windows. We examine volatility in the short term, 3-day and 5-day windows, aiming at capture immediate announcement volatility and on a slightly longer horizon, 21 trading days, aiming to capture more prolonged volatility that can be caused by the investors reacting to the announcement. All models include firm random effects as justified by a Hausman test that indicated no significant fixed- effect bias and are run on a panel of firm-year observations. Control variables include indicators for ROA, ROE, VIX - index, and firm size. Panel Regression of Post-Announcement Stock Volatility. 3-Day 5-Day 21-Day Overpromise & Miss -0.028*** -0.040*** -0.171*** (0.039) (0.010) (0.039) Neutral 0.019** 0.037*** 0.113*** (0.036) (0.040) (0.081) ROA -0.026** -0.042*** -0.181*** (0.044) (0.011) (0.044) ROE -0.003 -0.005 0.009 (0.050) (0.012) (0.050) VIX -0.001 -0.004 -0.009 (0.036) (0.009) (0.036) 26 Large Cap 0.000 0.000 0.000 (0.003) (0.002) (0.002) Observations 1230 1230 1230 Wald Chi2 4.124 4.231 4.682 R2 0.022 0.017 0.024 Root MSE 0.042 0.024 0.0 Standard errors in parentheses. Clustered by firm, random effects. * p < 0.10, ** p < 0.05, *** p < 0.01 Table 4.2 Panel Regression of Post-announcement Stock Volatility. Table 4.2 displays the results from a panel regression of post-announcement stock volatility on guidance strategy. The dependent variable consists of the standard deviation of daily abnormal returns in the windows of 3-days 5-days and 21-days. The coefficients for strategy dummies are relative to the Underpromise & Beat base group. A negative β coefficient indicates lower volatility than the base group while a positive β coefficient indicates higher volatility than base. Firm random-effects have been used as the Hausman test shows p=0.37 when checking for Firm fixed-effects. The model looks at 2 452 firm-level observations. In the immediate post-announcement window of 3 days the coefficient for Overpromise & Miss firms is negative and statistically significant (β = -0.028, p < 0.01). This indicates that Overpromise & Miss firms experienced lower volatility than the Underpromise & Beat base group. Similarly, for the 5-day window, the coefficient for Overpromise & Miss firms is slightly more negative and significant (β = - 0.040, p < 0.01). Contrary to initial expectations, these results suggest that Overpromise & Miss firms had less volatile stock price reactions than Underpromise & Beat firms in the days immediately following the announcement. For neutral firms, the coefficients are positive and statistically significant in both the 3-day window (β = +0.019, p < 0.05) and the 5-day window (β = +0.037, p < 0.01), indicating slightly higher volatility relative to Underpromise & Beat firms during the short-term windows. In the 21-day window post-announcement the coefficient for Overpromise & Miss remains negative and highly significant (β = -0.171, p < 0.01). Again, the neutral firms exhibit significantly higher volatility relative to Underpromise & Beat firms (β = +0.113, p < 0.01) over the 21-day window. The control variable ROA is consistently negative and significant across all event windows, indicating that firms with higher return on assets experienced lower abnormal return volatility following earnings announcements. Other controls, including ROE and Volatility Index were statistically insignificant in all models. Market capitalization was omitted from the model due to collinearity. 27 Panel Regression of Post-Announcement Realized Volatility by Guidance Strategy 3-Day 5-Day 21-Day Overpromise & Miss 0.003*** 0.016** 0.010*** (.) (.) (.) 0.012** 0.041** 0.034** Underpromise & Beat (0.002) (0.002) (0.002) -0.007** -0.006*** -0.004*** ROA (0.003) (0.002) (0.001) ROE -0.003 -0.002 -0.001 (0.003) (0.002) (0.002) -0.003** VIX -0.004* -0.001 (0.002) (0.002) (0.001) 0.038*** 0.034*** 0.023*** Constant (0.003) (0.002) (0.002) Observations 1230 1230 1230 Wald Chi2 4.233 4.815 4.802 R2 0.016 0.020 0.028 Root MSE 0.030 0.024 0.016 Standard errors in parentheses. Clustered by firm, random effects. * p < 0.10, ** p < 0.05, *** p < 0.01 Table 4.3 Panel Regression of Post-announcement Stock Volatility by Guidance Strategy To ensure robustness an additional regression was conducted where we now include both guidance strategy variables, Overprimse & miss and underpromise & beat, directly in the model seen in table 4.3. This change lets us assess the absolute impact of each strategy on post announcement volatility, instead of interpreting the result relative to a reference group as seen in table 4.2. The results show that overpromise & miss are associated with a significantly lower volatility across all event windows, with a coefficient ranging from 0.003-0.016 while statistically significant at the 1% level. Underpromise & beat firms show a significant higher volatility, especially over longer windows, where β = 0.041 for 5-day and β = 0.034 for 21-day. The results provided in table 4.2 and 4.3 contradicts Hypothesis 2. The Underpromise & Beat group are associated with higher post-announcement stock volatility while Overpromise & miss behaviors are associated with lower volatility. A formal coefficient test (Lincom) comparing the two behavioural types shows that the Underpromise & Beat shows significantly higher post event volatility (β = 0.0063, p < 0.001) (Appendix 2D). This further reimburses that the results challenge the assumption that Underpromise & Beat strategies necessarily lead to more stable investor responses. 28 4.3 Results - Hypothesis 3 Overpromising firms that miss earnings targets experience stronger immediate stock price declines compared to the positive reactions received by underpromising firms beating expectations. For the third Hypothesis 3 it is assumed that investor reactions to earnings surprises would be asymmetric, specifically, that the market would punish firms that overpromise & miss more severely than it rewards an underpromise & beat. This hypothesis is rooted in behavioral theories like loss aversion. To test H3 in the context of our dataset, we performed a focused analysis comparing two groups of firm-events: ● Underpromise & Beat - Representing positive surprises. ● Overpromise & Miss - Representing negative surprises. The variables were categorized as “Underpromise & Beat” if their percentage guidance was below - 29.7% and as “Overpromise & Miss” if their error exceeded 29.7%. We isolated these two groups and examined their immediate stock price reactions. We ran a regression and mean comparison of the short- term CAR on a dummy variable that equals “1” for Underpromise & Beat firms and “0” for Overpromise & Miss firms, including firm-level controls and clustering standard errors by firm. The coefficient on this dummy captures the difference in market reaction between positive and negative surprise scenarios. A significantly positive coefficient would indicate the positive surprises have a larger impact than negative surprises, contrary to H3, whereas a significantly negative coefficient would support H3. We also looked at the within-group CARs: the average abnormal return for each group around the announcement. 29 Panel Regression of CAR – Surprise Threshold ±29.7% Panel regression of CAR (1) (2) CAR_1day CAR_2day Overpromise & Miss 0.000 0.000 (.) (.) Underpromise & Beat 0.013** 0.030** (0.006) (0.014) ROA -0.001 -0.001 (0.005) (0.012) ROE 0.009 0.018 (0.007) (0.016) VIX -0.009 -0.021 (0.007) (0.013) Constant -0.006 -0.010 (0.007) (0.015) Observations 1230 1230 Wald Chi2 1.297 1.699 R2 0.007 0.009 Root MSE 0.091 0.189 Standard errors in parentheses. Clustered by Firm, Random effects. * p < 0.10, ** p < 0.05, *** p < 0.01 Table 4.4 Panel Regression of CAR Between Groups The results from the panel regression in table 4.4 provide support for Hypothesis 3. Specifically, firms that underpromise and thereafter exceeded their earnings guidance faced a significantly higher CAR over the 2- day window around the earnings announcement. The difference in CAR between Underpromise & Beat firms and overpromise & miss firms was 3.05 percentage points, significant at the 5% level (p = 0.029). This suggests that investors respond more positively to firms that Underpromise & Beat than they penalize firms that overpromise & miss. The result supports the hypothesis that market participants place greater weight on positive forecast surprises that signal caution and credibility, rather than overreactions to negative deviations. This shows a notable improvement compared to earlier results where the difference was not statistically significant. When using broader or fixed thresholds in the previous version, the gap in CAR between the two groups was smaller around 0,81 percentage points and not significant with p=0.462. This made it difficult to show any meaningful difference in how the market responded. Although the pattern stayed the same, with underpromising firms showing positive AR and overpromising firms saw negative 30 returns, only the underpromise group had statistically strong results, the difference was thus not enough to reject the idea that the market reacts symmetrically. CAR Response by Guidance Behavior (vs. Neutral) (1) (2) (3) (4) CAR_1day CAR_2day CAR_5day CAR_21day Overpromise & Miss -0.006* -0.013* -0.017* -0.030** (0.003) (0.007) (0.09) (0.015) Underpromise & Beat 0.007 0.017 0.020 0.031 (0.006) (0.013) (0.017) (0.027) Neutral (Base) 0.000 0.000 0.000 0.000 (.) (.) (.) (.) ROA -0.004 -0.008 -0.011 -0.019 (0.004) (0.009) (0.012) (0.019) ROE 0.004 0.006 0.007 0.011 (0.004) (0.010) (0.013) (0.021) VIX -0.005 -0.014* -0.019* -0.033* (0.004) (0.008) (0.011) (0.018) Constant 0.002 0.010 0.013 0.024 (0.004) (0.010) (0.012) (0.020) Observations 1230 1230 1230 1230 Wald Chi2 1.578 2.030 1.953 2.203 R2 0.004 0.006 0.005 0.005 Root MSE 0.080 0.168 0.225 0.368 Standard errors in parentheses. Neutral Group +- 29.7% in base. Standard errors Clustered at firm level. * p < 0.10, ** p < 0.05, *** p < 0.01 Table 4.5 - Panel Regression of CAR Against Neutral Base The findings of the linear regressions that looked at cumulative abnormal returns in response to firm earnings guidance behavior are shown in Table 4.5. These regressions were compared to a neutral base group, which is made up of firms with guidance variances within the threshold of ±29.7%. With and without control variables, the regressions cover event windows between one and twenty-one days. Firms categorized as Overpromise & Miss have negative but not considered statistically significant CARs across all criteria, with the effects becoming stronger as the event window becomes larger over time. 31 Overpromisers show a -0.030 CAR at the 21-day horizon (p < 0.05) suggesting that investors penalize companies that continuously overestimate predicted performance and then fall short of it. In every model companies’ classified as Underpromise & Beat show positive but statistically insignificant CARs. VIX is one of the control variables that has a negative correlation with CARs. This relationship is statistically marginally significant in the windows from 2- day to 21-days, and the effects increase across longer windows. This suggests that the price response to earnings guidance is tempered by increased market uncertainty. In every model, neither ROE nor ROA achieve statistical significance. Random Effects GLS Regression with Firm-Level Clustered Standard Errors (1) (2) (3) (4) CAR_1day CAR_3day CAR_5day CAR_21day Underpromise & Beat 0.011 0.038 0.071 0.363* 2 Years Consecutively (0.011) (0.028) (0.050) (0.203) Overpromise & Miss -0.006 -0.008 -0.017 -0.024 2 Years Consecutively (0.005) (0.010) (0.021) (0.083) ROA -0.001 -0.001 -0.003 -0.013 (0.005) (0.011) (0.025) (0.099) ROE 0.009 0.018 0.035 0.094 (0.007) (0.013) (0.035) (0.140) VIX -0.009 -0.021 -0.050* -0.229** (0.007) (0.013) (0.029) (0.104) Large Cap 0.000 0.000 0.000 0.000 (.) (.) (.) (.) Constant -0.002 -0.003 -0.005 -0.005 (0.007) (0.016) (0.036) (0.142) Observations 1230 1230 1230 1230 R2 0.010 0.011 0.009 0.007 Wald Chi2 5.53 5.82 5.69 6.89 Standard errors in parentheses. Standard errors Clustered at firm level. p < 0.10*, ** p < 0.05, *** p < 0.01 32 Table 4.6 - Random Effects GLS Regression with Firm-Level Clustered Standard Errors The results in table 4.6 reveal limited explanatory power with an overall R-squared values ranging from 0.008 to 0.013. None of the models are significant at conventional levels based on the Wald chi-squared tests, with p-values ranging from 0.35 to 0.23 in the overall models. While no coefficients reach conventional levels of significance in the shorter windows the underpromise category shows a positive effect throughout, increasing from 0.012 in the 1-day window to 0.363 in the 21- day window. In the 21-day model, this effect approaches marginal significance (p = 0.074), suggesting that sustained underpromising behavior may yield positive abnormal returns over longer horizons. Conversely, the overpromise category remains negative in all windows but is not statistically significant. Notably the volatility index control variable is significantly negative in the 21-day window (p = 0.028) and marginally significant in the 5-day window (p = 0.082) indicating that heightened market uncertainty tends to suppress post-guidance abnormal returns. Due to collinearity, the large cap dummy was omitted from all regressions. A formal coefficient comparison (Lincom) revealed once again that Underpromise & Beta firms exhibit higher CARs than Overpromise & Miss firms, but the difference is still not statistically significant (β = 0.197, p = 0.218) (Appendix 3I). Overall, these findings indicate no statistical evidence for the effect of guidance behavior and firm characteristics on cumulative abnormal returns. Although the direction and magnitude of the underpromising coefficients may signal delayed investor reactions to consistent underpromising behavior. 4.4 Robustness Checks A series of robustness checks were conducted to ensure the reliability of the results for all three hypotheses. As each hypothesis was tested with a somewhat different model, we tailored the robustness tests accordingly. Overall, our main findings hold under these alternative specifications and assumptions. All regressions were run on datasets that were winsorized at the 1st and 99th percentiles for key variables such as forecast errors and abnormal returns to mitigate the influence of extreme outliers. Our reported results already reflect this winsorization. To confirm robustness within this area, we also ran the models on the raw (non-winsorized) data. The results were qualitatively unchanged and no hypothesis conclusion was altered. This gives confidence that outliers are not driving the significance of H1. Similarly, H2 volatility outcomes were stable whether we winsorized the volatility measures. The small changes that did occur in the Overpromise & Miss coefficient in the 21-day window model moving from -0.171 to -0.165 were well within the range of sampling error. 33 We tested fixed effects vs. random effects in our panel models. For H1 we also conducted a simple OLS with dummies for each firm which equivalates to fixed effects, given one observation per firm per event and found virtually identical results to the random-effects model. For H2’s panel, we formally conducted a Hausman test which supported the use of random effects by displaying p = 0.37 which indicates no systematic difference versus fixed effects estimators (Appendix 2B). Similarly, another Hausman test was performed to justify using random effects as fixed effects did not provide a good fit in H3 (Appendix 3F) The H2 model was ran with firm fixed effects as a robustness check and signs and relative magnitudes for categories Overpromise & Miss and Neutral remained the same. The fixed effect version showed slightly smaller coefficients but still significant for the volatility differences in the 5-day and 21-day windows (Appendix 2A). The core inference, that underpromisers have lower volatility and overpromisers eventually higher volatility held true. Therefore, unobserved firm heterogeneity is not biasing our conclusions. Clustered standard errors at the firm level were used for all models to address any serial correlation or heteroskedasticity. Clustering by time was used to ensure this did not under- or over-state significance for the H1 event regressions. The significance levels of main effects were robust under all of these alternatives. For instance, analyst error in H1 was p<0.01 with firm clustering and remained p<0.01 with time clustering. Similarly, the difference in H3 remained non-significant regardless of error structure. The H3 and H2 model was also re-estimated with standard errors clustered at the year level instead of the firm level. The coefficient for underpromise & beat firms in H3 remained consistent around +3.05 percentage points and statistical significance increased by lowering the p-value to 0.001 on the 2-day, 3-day and 5-day interval, while remaining low at the 1-day regression as well (Appendix 3A:3D). Indicating strong support for Hypothesis 3. However, this result should be interpreted with caution as the amount of clusters was capped at 10 which is noted by Cameron and Miller (2015) as a low. Clustering with fewer than 30 - 50 groups may lead to underestimated standard errors and therefore also inflated significance. Conclusively, while the year-clustered result confirms our initial results, the firm-level clustering (p = 0.029) is retained as the primary inference. Another robustness check for H1 was pre-built into the analysis by examining multiple event windows. The consistency of findings across these windows strengthens H1. Additionally, although Hypothesis 1 is about short-term reaction, we extended the window to 21 days to see if any stark differences emerge in a longer horizon. Seeing how analyst vs. firm guidance effects persisted suggests that our conclusions are not an effect of a particular cutoff window. We tested whether the H1 might be moderated by firm characteristics like reputation or size by including interaction terms such as AnalystError × LargeCap and AnalystError × PastAccuracy. These interactions were generally not significant, indicating that the propensity to react to analyst surprises over guidance errors was broad-based across firm types. The dominant effect of analyst error remained in all subgroups. Importantly, across all CAR windows the core results remained directionally and statistically consistent 34 with those reported under the pooled OLS specification. The coefficient on the analyst forecast error remained positive and significant, while the firm guidance error was smaller in magnitude and generally insignificant. These findings suggest that our primary conclusion holds even after controlling for unobserved firm heterogeneity. Instead of raw return standard deviation, the models looked at idiosyncratic volatility for the 21-day window. The pattern of higher idiosyncratic volatility for Overpromise & Miss vs Underpromise & Beat remained. For H1, a placebo test was conducted by randomly shuffling the pairing of earnings surprises and stock returns to verify that the relationship would be lost, which it was. This helps confirm that the correlation we see in H1 is due to the earnings event and not a pattern or trend. Similarly, for H2 and H3, an implicit placebo check was done by comparing groups during no announcement periods, by seeing no systematic return differences and finding no significance further extends the claim that our results are not randomized (Appendix 2C; Appendix 3G). Lastly, we calculated Variance Inflation Factors (VIF) for our regression models (Appendix 3E). All VIF values were low (<2) indicating that multicollinearity is not a concern. The analyst error and firm error in H1 are uncorrelated by construction as they measure different gaps and in H2 the dummies are exclusive categories. This should indicate that the regression coefficients are well-identified. The robustness checks confirm that the results are stable and credible. Differences between original and robustness results were very minor, often under 5-10% in coefficient terms and no significant changes of coefficients occurred. 35 5. Analysis & Discussion 5.1 Findings, Theory & Prior Research This study has examined whether investors respond to earnings news more favorably toward analyst forecasts or company guidance, effectively a matter of credibility and trust in financial information. The low R² values seen across our regression models align with EHM (Fama, 1970), which suggests that the market is efficient and that stock prices rapidly integrate all publicly available information. The modest explanatory power in our results indicates that firm guidance and analyst forecasts, while influential, represent only a portion of the factors shaping investor expectations and stock valuations. In efficient markets, individual pieces of information, such as guidance or analyst consensus, typically have limited explanatory power, reflecting rapid market adjustments and the complexity of investor decision making processes. Thus, even small statistically significant impacts are economically meaningful, as they highlight consistent deviations from absolute market efficiency. This explanation of the low R² is consistent across all regression models and hypotheses tested, strengthening the interpretability of our results. 5.1.1 Hypothesis 1 The results for H1 confirms the hypothesis that investors react more strongly to analyst forecast errors than to firm guidance errors in the short term, which means that investors mainly rely on analyst expectations when they respond to earnings surprises. This is in line with previous research that highlights the impact of analyst forecasts in driving investor behavior. As Graham et al (2005) found, managers do often feel considerable pressure to meet analyst expectations due to fear that an Overpromise & Miss could trigger negative stock price reactions. So (2013) further states that investors do overweight analyst estimates and that this can cause mispricing which gets corrected at the earnings announcement. The results seen in chapter four support this claim, when there is a deviation from the analyst consensus vs actual earnings, a stock price response was found which is in line with investors adjusting their earlier mispricing or surprise due to analyst optimism/pessimism. Veenman & Verwijmeren (2018) finding state that investors often underreact to analyst predictable pessimism in short term forecasts. As a result, firms that exceed these conservative estimates tend to show a steeper increase than what is expected, even when earlier guidance had already pointed towards favorable performance. This underreaction suggests that investors do not completely account for systematic biases in analyst consensus. Therefore, the strong market reactions to analyst consensus error may be due not only to new information, but also how investors tend to interpret and react to these forecasts based on prior experience and biases. On the other hand, the relatively minor reaction to firms own guidance errors could suggest that a significant amount of firm guidance data is already absorbed by the market at the time of the earnings announcement. That is consistent with the efficient markets theory (Fama, 1970) since firm issued guidance is most often out in public well before the earnings announcement the investors and analysts have 36 plenty of time to absorb it. By earnings day, any expected variance of reported results from previous guidance will already have been anticipated and factored in. Analysts, for their part, incorporate firm guidance into their forecasts (along with other information), so the analyst consensus effectively represents the market’s expectations given all available data. Therefore, when the earnings announcement comes, the “news” is defined mostly by the gap between actual earnings and the analyst forecast, which our results are consistent with and reinforce the idea from information intermediary theory (Healy & Palepu, 2001) that analysts play a crucial role in interpreting and disseminating information; investors appear to trust the analyst consensus as a baseline, reacting strongly if that baseline is wrong. The findings also support Hollander et al (2010) shows that when firms withhold or limit disclosure, investors tend to lean more on analysts´ consensus. This may explain the dominant role analyst forecasts play in market reaction, as is reflected in the regression results. It is notable that the market’s immediate response to firm guidance errors was smaller and slower. One of the interpretations is related to management credibility, if investors have learned that some firms tend to underpromise or overpromise, they may already discount those biases ahead of time. For example, a firm with a reputation for underpromising might not see a stock price increase simply for beating its guidance, since investors expected them to guide conservatively. Likewise, a firm known for overpromising might not see a huge decline solely based on the earnings announcement if investors had already been skeptical of its guidance. Thus, analyst forecasts capture the net effect of all such anticipations. This perspective is supported by behavioral finance findings that market participants learn and adapt to patterns. The H1 result, in combination with H2, suggests a learning effect. The credibility that management builds or loses may show up more in how the market reacts over time, through increased or decreased volatility, rather than in an immediate price jump. That initial reaction is often already priced in by investors who anticipated the predictable parts of the announcement. Furthermore, the robustness test using percentual error terms validated these conclusions. The proportional level of reactions was consistent with what could be observed in absolute terms: investors are more responsive to errors from analyst estimates than to firm-issued guidance errors. The percentual-based analysis, on the other hand, demonstrated a significantly less pronounced difference in reaction sizes, suggesting that, while the pattern persists, the stock price’s sensitivity to absolute versus relative misses may differ slightly in terms of severity. Nonetheless, the overall result that analyst estimates dominate investor reactions in the short term and are consistent across both measuring methodologies. 5.1.2 Hypothesis 2 The second hypothesis, which proposed that a firm’s guidance strategy influences its post- announcement stock performance, was not supported by the empirical results. It was hypothesized that firms who consistently underpromise and then beat the EPS prediction by exceeding their guidance experience lower 37 stock price volatility. In contrast, companies that set overpromise targets and then fail to meet them may face longer lasting negative effects, such as higher volatility or a prolonged weaker performance. However our panel regression shows the opposite, firms that consistently underpromise and then beat the expectations face a significantly higher volatility, particularly over longer time windows (21 days). This contradicts the theoretical expectations that investors have greater confidence in firms that regularly exceed their guidance. This result may suggest that positive surprises may trigger greater investor reactions than negative ones, potentially due to their unexpected nature. Hutton and Stocken (2021) argue that firms with a strong forecasting track record benefit from increased investor confidence and information credibility which leads investors to view such firms as a less risky alternative, resulting in more stable stock prices. This is consistent with prior research indicating credible and transparent disclosures reduce information risks. For instance, Kothari et al. (2009) found that increased management transparency correlates with lower stock volatility and a reduced cost of capital. Similarly, DeFond & Park (1997) argues that firms engage in earnings smoothing to reduce volatility which our findings contradict, with an empirical result that shows underpromising firms experience less stable price movements post announcement. Kasznik (1999) also finds that voluntary disclosures are used strategically when firms anticipate a miss earning in contrast to earning expectations. This further supports our view that investors may adjust their expectations prior, in response to disclosure tone and timing. However, the results show that this relationship is consistent across event windows. Firms that Overpromise & Miss demonstrate significantly lower post-announcement stock price volatility than those that Underpromise & Beat, across 3-day, 5-day and 21-day windows. This contradicts our theoretical expectation that conservative guidance would reduce volatility. One possible explanation is that positive surprises following underpromising may generate more investor reaction and trading activity than negative surprises following overpromising, particularly when disappointment is already anticipated. It is important to state, however, that longer event windows such as 21 days are more susceptible to external noise, including unrelated market events, macroeconomic changes, or firm-specific developments. Therefore, the elevated volatility observed over this horizon should be interpreted cautiously. This could appear consistent with theories of credibility risk, but the results suggest that underpromising firms were met with greater, not reduced volatility. Underpromise is often seen as credible signaling practice but the results from our analysis do not support the notion that it reduces post announcement volatility. This contradicts with how Hutton et al. (2003) discussed the role of management forecast and that when accompanied by credible supplementary information, firms enhance the information environment and thereby also financial results. By underpromising, management may be indirectly providing an indication that assures investors the firm can at least meet stated goals, which in itself inherently should thereby reduce downside uncertainty. The idea that habitual overpromisers suffer prolonged negative stock effects was only weakly supported in the data, yet worth discussing in light of theory. Initial expectations were that if a firm frequently 38 overpromises, investors would penalize the firm with higher perceived risk or persistent stock price underperformance. While our volatility measure trended higher for such firms, the effect was not statistically strong. One possible reason is that investors may not wait for repeated misses to adjust their expectations, but instead they might punish an overpromise the first time it happens or grow skeptical quickly and thereby incorporate that skepticism into the stock price early on. Hurwitz (2018) findings show that managers tend to adjust their tone of the guidance based on investor market expectations, which may explain some of the asymmetric investor responses we observe in different guidance scenarios. Matsumoto (2002) emphasizes that firms often issue overly optimistic guidance to influence investor sentiment. However, when behavior such as this becomes predictable, it can undermine credibility. This helps explain why frequent overpromisers may gradually lose investor trust, leading to more weak price movements. If a firm gains a reputation for missing its own targets, investors and analysts could start discounting the firm’s optimistic guidance in advance. In that scenario, by the time an earnings announcement occurs the stock price may have already drifted downward in anticipation of a likely miss as a form of gradual information absorption. This would mean less pronounced volatility at and after the announcement as the damage might be done gradually rather than in one big shock. This reasoning is in line with Fama (1970) efficient market argument that apparent longer-term effects can be hard to detect if markets adjust continuously and accounts for everything that could be accounted for. It also echoes what Skinner (1994) noted where firms choose to voluntarily disclose bad news through warnings to reduce litigation risk and avoid these massive stock price shocks. Firms that realize they will miss may guide expectations down before the official announcement to try and mitigate the impact. 5.1.3 Hypothesis 3 It was hypothesized in H3 that investors would punish firms that Overpromise & Miss more severely than they reward firms that Underpromise & Beat. Interestingly, our results did not confirm this asymmetry. The Underpromise & Beat firms did see a significant positive reaction, but the Overpromise & Miss firms did not suffer a proportional statistically significant negative reaction. This outcome has therefore contrasted with theoretical expectations and prior empirical findings, such as the Prospect Theory (Kahneman & Tversky, 1979). Which suggests that investors perceive losses more intensely than equivalent gains. As such, a negative earnings surprise is likely to trigger a stronger adverse reaction in stock prices than the corresponding positive surprise would generate in returns. Moreover, other studies such as Bartov et al. (2002) have documented that failing to meet expectations often leads to stock price declines, sometimes sharper than the increases for upside surprises. Additional support for this asymmetric expectation can be found in Miao & Yeo (2009) and Alwathnani & AlZoubi (2017), who both found that investors tend to overreact to negative earnings surprises while underreacting to positive ones. These studies reinforce the notion that losses carry more psychological weight, therefore boosting negative stock responses. Our findings, however, suggest a more symmetric or even modestly asymmetric outcome in the opposite direction. There are a few potential explanations for this unexpected result. One explanation, 39 briefly touched above, is the role of expectation management and pre-positioning by investors. If a firm overpromised, investors might anticipate the miss and adjust their positions ahead of the announcement. By contrast, when a firm underpromises, investors might be unsure how much the firm will beat by since there is more uncertainty on the upside. Thus, the positive surprise can genuinely surprise the market, whereas the negative surprise from an overpromise may be halfway expected which leads to a smaller relative drop in stock price. In this sample, it could be that many of the overpromise-miss firms had already seen their stock price decline in the weeks leading up to the earnings release, thereby smoothing out the announcement day effect. This conjecture is consistent with the idea of information leakage and gradual adjustment even if companies do not formally warn, other indicators such as order slowdowns, industry news, etc. might alert the market that an optimistic guidance will not be met. Kwon & Tang (2023) suggested that investor reactions are more exaggerated for extreme corporate events. While routine earnings misses may not trigger large swings, overpromising in a high-stakes context could provoke sharper selloffs. The lack of significant negative reaction in our sample may partly reflect the nature of those events that are being studied. Arguing that if most misses were perceived as routine or low stakes the investor response may have been lowered as the announcement is not seen as extreme. Further reinforcing the idea that not all guidance failures are equal in the eyes of the market but instead context, credibility, and perceived materiality matters perhaps equally as much. Another factor to account for could be sample biases. Firms that massively overpromise might be relatively fewer since managers try to avoid such scenarios, or those events might coincide with broader bad news that is hard to cleanly attribute to the miss itself. Despite not finding the expected asymmetry, our H3 result still reinforces some observations. It confirms that markets do reward positive surprises, underpromising & beat saw a significant positive CAR. This aligns with the literature on earnings surprises, firms that beat expectations tend to see abnormal returns and other benefits (Bartov et al, 2002). Another model was using the neutral category as base group as supplemental evidence to the initial regression that compared the two categories against each other. Instead of focusing only on the relative differences between extreme methods, this model allows for a more thorough analysis of the absolute effects of each action. In that model the Overpromise & Miss had negative returns displayed in Table 4.5. The most extreme case was the 21-day CAR difference of –3.0 percentage points statistically significant at the 5% level. This could imply that when optimistic guidance is overlooked, the market penalizes significant negative surprises. On the other hand, Underpromise & Beat firms have positive CARs in comparison to neutral firms. These results cannot conclusively prove the idea that underpromising is rewarded more than overpromising is punished, but they do point to an imbalance in investor responses that is compatible with behavioral theories like loss aversion. It should be mentioned that the model ran on 21 days, yields the only statistically significant result for Overpromise & Miss. The reported significance however cannot be fully recognized due to the potential noise rather than pure guidance reaction that becomes increased by the wider 40 window, which increases the likelihood of capturing confounding news or market movements unrelated to the earnings event. Therefore, the result should be regarded cautiously and might not be a precise assessment of the market penalty for overpromising alone at that interval of time. Table 4.6 offers further detail by analyzing the CAR outcomes of guidance behaviors against a neutral base group when categorized by consistent patterns in overpromising and underpromising. The findings reinforce the earlier results by illustrating that while firms categorized as Overpromise & Miss experienced significantly negative CARs, the Underpromise & Beat group consistently showed positive but statistically insignificant abnormal returns across all event windows. These patterns continuously suggest asymmetric market reactions that become more apparent with time, emphasizing that investors penalize overpromising behavior more heavily as the realization of the miss unfolds. These results align with disclosure theory in general and prior research that displays the role of strategic communication in shaping investor trust. Healy & Palepu (2001) mention that voluntary disclosures are assessed not just for their content, but for their credibility. The punishment for overpromising firms is consistent with the view that the credibility of firms becomes worse when management overstates and fails to deliver. Moreover, the longer interval of 21-days CARs resonates with findings by Dichev & Tang (2009) and Graham et al. (2005) who show that forecast inaccuracies and manipulative signaling strategies that repeatedly occur eventually do erode investor confidence. The lack of statistical significance for Underpromise & Beat firms challenges the assumption that caution is directly rewarded. This nuance contrasts with some behavioral finance literature and Prospect Theory (Kahneman & Tversky, 1979), which posits that losses are larger than equivalent gains, yet in this case, negative surprises provoke stronger reactions than positive ones, even if they were anticipated. However, it cannot be overstated that even though this analysis allows for speculation, it displays no conclusive evidence that both short term and/or consistent communication strategies of over- or underpromising EPS targets from a firm’s perspective do have an effect on the abnormal returns of US large cap listed firms. While we did not measure how large and sudden the market’s negative response is for misses, it would be too early to conclude that missing targets has no consequence. The consequences might simply be more long-term or subtle, such as the increase in perceived risk or a hit to management’s reputation. Instead, these findings might suggest that management’s credibility entering the announcement moderates the outcome if investors already distrust the management. Therefore, from a theoretical standpoint H3 ties back to the importance of expectation conditioning. The market’s reaction to news depends not just on the news itself, but on what investors expected about the expectation and their prior level of trust in the firm. 41 5.4 Future Research It would be interesting to analyse differences such as if certain types of companies are more prone to overpromise or underpromise and do investors respond differently by sector? For example: High- growth tech firms might be given more flexibility for optimistic promises, whereas utility companies or others in stable industries might be penalized more severe for failing to deliver on guidance because their predictability is valued. Another area could be to study the role of media. How does social media or other platforms shape the investors view? Incorporating measures of media sentiment or Twitter activity around earnings could reveal how public sentiment might amplify or dampen reactions to guidance vs analyst forecasts. If overpromising firms face negative media coverage or bad commentary from influencers, that could be an alternate part that is affecting their stock. 42 6. Conclusions & Final Discussion The first result observed from hypothesis 1 shows that short-term investor reactions are mainly driven by analyst forecast errors rather than firm guidance errors. Across all CAR windows, the stock price response was observed to be stronger when analyst expectations differed from the actual earnings reported on the announcement day which supported the first hypothesis. This implies that investors continue to rely on the analyst consensus as their main benchmark for decision making even when firm guidance may offer a more accurate forecast. Firm-issued guidance appeared to be priced in earlier, which is consistent with the efficient market hypothesis (Fama, 1970). Investors possibly absorb and factor in guidance information before the earnings announcement and therefore leaving little room for further reaction when results are finally released in the earnings announcement. The second hypothesis examined whether consistent guidance strategies influence post-announcement stock price volatility. Firms that had a record of regularly underpromise and then exceeding expectations were associated with higher post-announcement stock price volatility. This finding challenges the assumption that credibility leads to more stable price movements. Firms that Overpromise & Miss their targets experienced lower volatility, signaling investor distrust. These results extend the understanding of how trust is built over time in financial markets. Although the market did not react much to firm guidance errors in the short term, the increased volatility over time shows that credibility still plays a big role in how investors view a company. Table 4.6 offers further evidence of this pattern. While short-term abnormal returns to Underpromise & Beat firms were not statistically significant, the 21-day CAR approached significance suggesting that investor recognition of cautious guidance behavior may unfold over a longer horizon. The strength of the signal over time strengthens the idea that credibility is not only a volatility reducing factor but potentially a long term contributor to improved returns. The third hypothesis focused on the assumption that investors would penalize firms that Overpromise & Miss more severely than they reward firms that Underpromise & Beat, our findings did not support this asymmetry. Instead we found that Underpromise & Beat firms faced a significant, yet small positive response, while Overpromise & is firms did not see an equivalent negative reaction to the stock price. This result has therefore challenged prior theories which state that negative surprises typically result in stronger reactions. The pattern of coefficients observed in Table 4.6 aligns with theoretical expectations that Overpromise & Miss firms show consistently negative CAR across all windows. This pattern suggests that even if statistical power is limited, the market still interprets these behaviors differently. One possible explanation to the outcome in hypothesis 3 may be that negative surprises were anticipated by the investor and therefore partially reflected in prices before the announcement. If expectations are already adjusted, the actual miss may carry less impact. Opposed to a positive surprise, which may catch the investors off guard and trigger more immediate response. The results show, even though not what was expected, show how investor 43 reactions are affected not only by the news itself but by the expectations built in the lead-up. Trust, timing and context continue to shape the market’s interpretation of earnings guidance. 44 References Anilowski, C., Feng, M., & Skinner, D. J. (2007). Does earnings guidance affect market returns? The nature and information content of aggregate earnings guidance. Journal of Accounting and Economics, 44(1-2), 36–63. https://doi.org/10.1016/j.jacceco.2006.09.002 Akerlof, G. A. (1970). 4. The market for ‘lemons’: quality uncertainty and the market mechanism. Market Failure or Success, 66. Alwathnani, A. M., Dubofsky, D. A., & Al-Zoubi, H. A. (2017). Under-or-overreaction: Market responses to announcements of earnings surprises. International Review of Financial Analysis, 52, 160-171. Bartov, E. (1993). The Timing of Asset Sales and Earnings Manipulation. Source: The Accounting Review, 68(4), 840– 855. Bartov, E., Givoly, D. and Hayn, C. (2002). The rewards to meeting or beating earnings expectations. Journal of Accounting and Economics, 33(2), pp.173–204. https://doi.org/10.1016/s0165-4101(02)00045-9. Bernard, V. L., & Thomas, J. K. (1989). Post-Earnings-Announcement Drift: Delayed Price Response or Risk Premium? Journal of Accounting Research, 27, 1–36. https://doi.org/10.2307/2491062 Brown, S.J. and Warner, J.B. (1985). Using daily stock returns: the case of event studies. Journal of Financial Economics, 14(1), pp.3–31. https://www.sciencedirect.com/science/article/abs/pii/0304405X8590042X Burgstahler, D.C. & Eames, M.J. (2003). “Earnings Management to Avoid Losses and Earnings Decreases: Are Analysts Fooled?”. Contemporary Accounting Research, 20(2), pp.253–294. doi:https://doi.org/10.1506/bxxp- rgtd-h0pm-9xal. Burks, J. J., Randolph, D. W., & Seida, J. A. (2019). Modeling and interpreting regressions with interactions. Journal of Accounting Literature, 42(1), 61-79. Cameron, A. C., & Miller, D. L. (2015). A practitioner’s guide to cluster-robust inference. Journal of human resources, 50(2), 317-372. Chen, J., Jung, M. J., & Tang, M. (2023). Does lowball guidance work? An analysis of firms that consistently beat their guidance by large margins. The British Accounting Review, 55(6), 101219. Costola, M., Hinz, O., Nofer, M., & Pelizzon, L. (2023). Machine learning sentiment analysis, COVID-19 news and stock market reactions. Research in International Business and Finance, 64, 101881. Daniel, M., Neves, R. F., & Horta, N. (2017). Company event popularity for financial markets using Twitter and sentiment analysis. Expert Systems with Applications, 71, 111-124. DeFond, M. L., & Park, C. W. (1997). Smoothing income in anticipation of future earnings. Journal of accounting and economics, 23(2), 115-139. 45 Dichev, I. D. and Tang, V. W. (2009) ‘Earnings volatility and earnings predictability’, Journal of Accounting and Economics, 47(1-2), pp. 160–181. https://doi.org/10.1016/j.jacceco.2008.09.005. Fama, E. F. (1970). Efficient capital markets: A review of theory and empirical work. The Journal of Finance, 25(2), 383–417. https://doi.org/10.2307/2325486 Graham, J. R., Harvey, C. R. and Rajgopal, S. (2005) ‘The economic implications of corporate financial reporting’, Journal of Accounting and Economics, 40(1–3), pp. 3–73. https://doi.org/10.1016/j.jacceco.2005.01.002. Healy, P.M. and Palepu, K.G. (2001). Information asymmetry, corporate disclosure, and the capital markets: A review of the empirical disclosure literature. Journal of Accounting and Economics, 31(1-3), pp.405– 440. Hirshleifer, D., Lim, S. S., & Teoh, S. H. (2009). Driven to Distraction: Extraneous Events and Underreaction to Earnings News. The Journal of Finance, 64(5), 2289–2325. JSTOR. https://doi.org/10.2307/27735172 Hollander, S., Pronk, M., & Roelofsen, E. (2010). Does silence speak? An empirical analysis of disclosure choices during conference calls. Journal of Accounting Research, 48(3), 531-563. Hope, O. and Kang, T. (2005). The Association between Macroeconomic Uncertainty and Analysts’ Forecast Accuracy. Journal of International Accounting Research, 4(1), pp.23–38. doi:https://doi.org/10.2308/jiar.2005.4.1.23. Huang, X., Teoh, S.H. and Zhang, Y. (2014). Tone Management. The Accounting Review, 89(3), pp.1083–1113. https://www.jstor.org/stable/24468328. Hutton, A. P., Miller, G. S., & Skinner, D. J. (2003). The role of supplementary statements with management earnings forecasts. Journal of Accounting Research, 41(5), 867-890. Hutton, A., & Stocken, P. C. (2021). Prior Forecasting Accuracy and Investor Reaction to Management Earnings Forecasts. Journal of Financial Reporting, 6(1). https://doi.org/10.2308/jfr-2020-005 Kahneman, D. and Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), pp.263–292. https://doi.org/10.2307/1914185. Kasznik, R. (1999) ‘On the association between voluntary disclosure and earnings management’, Journal of Accounting Research, 37(1), pp. 57–81. https://doi.org/10.2307/2491396 . Kothari, S.P., Li, X. and Short, J.E. (2009). The Effect of Disclosures by Management, Analysts, and Business Press on Cost of Capital, Return Volatility, and Analyst Forecasts: A Study Using Content Analysis. The Accounting Review, 84(5), pp.1639–1670. Kothari, S. P., & Warner, J. B. (2007). Econometrics of event studies. In Handbook of empirical corporate 46 finance (pp. 3-36). Elsevier. Kwon, S. Y., & Tang, J. (2020). Extreme Categories and Overreaction to News. SSRN Electronic Journal, 92(3). https://doi.org/10.2139/ssrn.3724420 Lennox, C. S., Francis, J. R., & Wang, Z. (2012). Selection models in accounting research. The accounting review, 87(2), 589-616. MacKinlay, A.C. (1997). Event Studies in Economics and Finance. Journal of Economic Literature, 35(1), pp.13– 39. https://www.jstor.org/stable/2729691. Matsumoto, D. A. (2002). Management’s incentives to avoid negative earnings surprises. The accounting review, 77(3), 483-514. Miao, B., & Yeo, G. H. (2009). The efficiency of market reactions to earnings news. National University of Singapore. Nelson, P. (1970). Information and consumer behavior. Journal of political economy, 78(2), 311-329. O’brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality & quantity, 41, 673-690. Skinner, D. J. (1994) ‘Why firms voluntarily disclose bad news’, Journal of Accounting Research, 32(1), pp. 38–60. doi:https://doi.org/10.2307/2491386. So, E.C., (2013). A new approach to predicting analyst forecast errors: Do investors overweight analyst forecasts? Journal of Financial Economics, 108(3), pp.615–640. https://doi.org/10.1016/j.jfineco.2013.02.002 Spence, M. (1974). Competitive and optimal responses to signals: An analysis of efficiency and distribution. Journal of Economic Theory, 7(3), 296-332. Valle-Cruz, D., Fernandez-Cortez, V., López-Chau, A., & Sandoval-Almazán, R. (2022). Does twitter affect stock market decisions? Financial sentiment analysis during pandemics: A comparative study of the h1n1 and the covid-19 periods. Cognitive computation, 14(1), 372-387. Veenman, D. Verwijmeren, P. (2025). Do Investors Fully Unravel Persistent Pessimism in Analysts' Earnings Forecasts? https://www.jstor.org/stable/26551361 Hurwitz, H. (2017). Investor sentiment and management earnings forecast bias. Journal of Business Finance & Accounting, 45(1-2), pp.166–183. School of Business, Economics and Law, 2024. Generative AI guidance to staff and students. University of Gothenburg. Wei, C., & Zhang, L. (2023). Trust in Financial Markets: Evidence from Reactions to Earnings News. 47 Management Science, 69(10), 6393–6415. https://doi.org/10.1287/mnsc.2022.4569 Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press. Yang,L., Zhang, X., & Chen, J. (2024). Winsorization greatly reduced false positives by popolar differential expression methods when analysing human population samples Genome Biology, 25(1), 282. 48 Appendix Appendix 1A: Mean Comparison Table by Firm Behavior Category Appendix 1B: Wald Test between Absolute Firm Error and Absolute Analyst Error Appendix 2A: Fixed effect regression of realized volatility 5-day interval 49 Appendix 2B: Hausman Test Results for Hypothesis 2 Appendix 2C: Placebo test for Hypothesis 2 Appendix 2D:Lincom test between behavioral types on post volatility 50 Appendix 3A: Clustering by Year 1-day Interval Appendix 3B: Clustering by Year 2-day Interval Appendix 3C: Clustering by Year 3-day Interval 51 Appendix 3D: Clustering by Year 5-day Interval Appendix 3E: Variance Inflation Factor Appendix 3F: Hausman Test Results for Hypothesis 3 52 Appendix 3G: Placebo test for Hypothesis 3 Appendix 3H: Frequency Table for Hypothesis 3 Appendix 3I: Formal Coefficient Comparison test between behavioral types 21day 53