ECONOMIC FORECASTING EXPERIMENTAL EVIDENCE ON INDIVIDUAL BIASES, GROUP COOPERATION AND HERD BEHAVIOR Dissertation zur Erlangung des Doktorgrades der Wirtschaftswissenschaftlichen Fakultät der Georg-August-Universität Göttingen vorgelegt von Lukas Meub geboren in Lich Göttingen, 2014 Erstgutachter: Prof. Dr. Kilian Bizer Zweitgutachter: Prof. Dr. Markus Spiwoks Weiteres Mitglied der Prüfungskommission: Prof. Dr. Claudia Keser CONTENTS 1 INTRODUCTION AND SUMMARY 1 2 ANCHORING: A VALID EXPLANATION 21 FOR BIASED FORECASTS WHEN RATIONAL PREDICTIONS ARE EASILY ACCESSIBLE AND WELL INCENTIVIZED? (with Till Proeger and Kilian Bizer) 3 AN EXPERIMENTAL STUDY ON SOCIAL ANCHORING 49 (with Till Proeger) 4 ARE GROUPS LESS BEHAVIORAL? 75 THE CASE OF ANCHORING (with Till Proeger) 5 OVERCONFIDENCE AS A SOCIAL BIAS: 109 EXPERIMENTAL EVIDENCE (with Till Proeger) 6 STRATEGIC COORDINATION IN FORECASTING: 123 AN EXPERIMENTAL STUDY (with Kilian Bizer, Till Proeger and Markus Spiwoks) 7 A COMPARISON OF ENDOGENOUS AND EXOGENOUS 169 TIMING IN A SOCIAL LEARNING EXPERIMENT (with Till Proeger and Hendrik Hüning) 8 THE IMPACT OF COMMUNICATION REGIMES 201 ON GROUP RATIONALITY: EXPERIMENTAL EVIDENCE (with Till Proeger) V VI Chapter I INTRODUCTION AND SUMMARY - 1 - - 2 - “Much has been written about the doubtful accuracy of economists’ predictions. [...] They are better at predicting the direction than the actual magnitude of events. [...] This is disappointing, but it does not mean that economics is not a science.” (‘Economics’, Encyclopædia Britannica1) The forecasting of economic developments guides decisions in all areas of public interest, whether in governments, finance or corporations.2 Accurate predictions hold an essential role for private and public actors’ ability to enable the efficient planning and execution of their strategic objectives. Accordingly, the development and evaluation of models to enhance accurate forecasting has always been among the central areas of economic research interest. The rapid progress in information technology in recent times has intensified the search for the empirically sound modelling of economic developments. However, this “virtual revolution in how economists compute, apply, and evaluate forecasts” (Elliot and Timmermann, 2008, p.3) has not altered the deplorable finding that naïve “no change” forecasts regularly outperform professional predictions in terms of their accuracy (Spiwoks, 2004). The basic neoclassical explanation for this empirical finding is markets’ information efficiency. Following the efficient market hypothesis, forecasting is futile by definition, given that asset prices incorporate new information perfectly as soon as it becomes accessible. With current prices mirroring all available information, forecasting – necessarily based on private information – is either obsolete or a gamble on future events independent of actual developments. Despite being a constant companion during the development of neoclassical economics, the evident criticism of the axioms underlying market efficiency has received additional attention lately.3 Following the recent macroeconomic turmoil, neoclassical assumptions have been criticized in favor of a behavioral understanding of decision-making. Criticism on the deterministic perspective of rational expectations has opened up a broad field of research that enables a more realistic understanding regarding agents’ incentives, as well as the institutions and processes involved in economic forecasting.4 Consequently, behavioral evidence from laboratory studies has become influential for the interpretation of potentially 1 Quoted from Ottaviano and Sorensen (2006a, p. 120). 2 Please note that this introduction summarizes the studies that constitute the dissertations of both Till Proeger and Lukas Meub. Thus, while the shared topic of research is discussed from different angles in the respective introductions, the chapters’ summaries are necessarily closely related. 3 Comprehensive overviews of the development of discussions on the efficient market hypothesis are provided by Fama (1970; 1991) and Spiwoks (2002); a critical account on its influence in the years leading up to the recent financial crisis is given by Krugman (2009). 4 For a general overview on the development of behavioral economics and its connections to neoclassical economics, please refer to Berg and Gigerenzer (2010). - 3 - dysfunctional markets for forecasts, as it yields a central advantage, in that analyzing actual time series always leaves open the theoretical possibility that low forecasting accuracy is due to markets’ information efficiency – an argument that can logically never be dismissed. At the same time, there are distinct behavioral approaches equally suitable to explain the status quo bias in forecasting, that also cannot be tested explicitly using empirical forecasting data. Introducing laboratory experiments to forecasting research firstly enables the construction of decision situations in which achievable outcomes and optimal behavior can be unambiguously benchmarked. Changing distinct parameters subsequently allows the identification and weighing of determinants for forecasting quality compared to the predefined benchmark. Thus, experiments enable a clean ceteris paribus analysis of the determinants of behavior in forecasting, which otherwise remains inaccessible. In this book, the authors pursue particular aspect of this broad research program. It has been shown that many forecasting time series share the characteristic tendency of predictions being biased towards current values. This implies that a major share of forecasters rely on the strategy of making (quasi-) naïve predictions. Denominated as a “status quo bias in forecasting” (Gubaydullina et al., 2011), this finding has been confirmed in numerous empirical studies.5 Naturally, merely forecasting present states precludes the revelation of private expectations on future states and accordingly disqualifies the respective predictions as a useful means for planning economic activities. To investigate the factors determining this finding, the authors build on two of the major behavioral explanations for poor forecasting quality: individual heuristics and biases, as well as rational herding. Heuristics and Biases The first paradigm drawn upon in this book is the psychological research on biases and heuristics and its application to forecasting. While psychological research considers numerous heuristics,6 the anchoring bias is most prominent when behavioral anomalies in forecasting are discussed.7 Tversky and Kahnemann (1974) were the first to show that individuals’ 5 Among the recent examples are Welch (2000), Gallo et al. (2002), Bofinger und Schmidt (2003), Spiwoks (2004), Clement and Tse (2005), Batchelor (2007), Spiwoks und Hein (2007), Spiwoks et al. (2008; 2010), Ager et al. (2009), Jegadeesh and Kim (2009), as well as Gubaydullina et al. (2011). 6 See Harvey (2007) for a comprehensive review of forecasting research in psychology. 7 For applications see e.g. real estate price forecasts (Bucchianeri and Minson 2013), financial forecasts (Fujiwara et al. 2013), sports betting (Johnson et al. 2009; McAlvanah and Moul 2013), earnings forecasts (Cen et al. 2013), macroeconomic forecasts (Campbell and Sharpe 2009; Hess and Orbe 2013), as well as sales forecasting (Lawrence and O'Connor 2000). - 4 - assessments can be systematically influenced by random numbers, which are completely irrelevant for the respective tasks. A large body of psychological experiments has since addressed the validity of this result across specific settings, leading to the conclusion that anchoring is “exceptionally robust, pervasive and ubiquitous” (Furnham and Boo, 2011, p. 41). Transferring this finding to forecasting, it is claimed that forecasters “use the last data point in the series as a mental anchor and then adjust away from that anchor to take account of the major feature(s) of the series. However, as adjustment is typically insufficient, their forecasts are biased.” (Harvey, 2007, p.17). Accordingly, it is argued that forecasts clustered around current values do not necessarily reflect strategic decisions, but rather a subconscious, irrational bias towards the current state, which cannot be alleviated through contradictory incentives or alternative market designs. A second behavioral anomaly that has regularly been considered, particularly in the context of financial forecasting, is the overconfidence bias. The term refers to individuals’ systematical inability to realistically evaluate their capabilities and the resulting tendency of overly optimistic self-assessments across a wide variety of decision situations. As with anchoring, the bias has been comprehensively investigated in experimental psychology and has also been applied to economic contexts for some time.8 While assuming robust overconfidence among analysts is not considered as a reason for biased forecasts, it can serve to explain the continuation of forecasting despite their poor accuracy. Self-denial of the obvious failure to produce correct predictions, fueled by overconfidence, might hold considerable influence on the persistence of forecasts’ uniformity.9 Overconfidence can thus be interpreted as a psychological mechanism that individually reinforces and justifies uninformative predictions. Rational Herding Behavior: Reputation and Cascades The second paradigm built upon in this book is forecasters’ herding on public information as another explanation for uninformative homogenous
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages213 Page
-
File Size-