Search results

Macroeconomic expectation data are of great interest to different agents due to their importance as central input factors in various applications. To name but a few, politicians, capital market participants, as well as academics, incorporate these forecast data into their decision processes. Consequently, a sound understanding of the quality properties of macroeconomic forecast data, their quality determinants, as well as potential ways to improve macroeconomic predictions is desirable. This thesis consists of three essays on the quality of analysts' forecasts. The first essay deals with macroeconomic forecast quality on the consensus level, while the second one investigates individual analysts' predictions and their quality determinants. In the third essay a bottom-up approach is introduced to derive macroeconomic forecasts from analysts' predictions at the microeconomic level. It is generally assumed that macroeconomic consensus forecasts provide a reasonable approximation of market participants' expectations regarding upcoming macroeconomic releases. Research areas in which these expectation data are a central input to isolate the unanticipated news component of a given announcement include studies analyzing the price impact of macroeconomic news in bond markets (e.g., Balduzzi et al., 2001; Gilbert et al., 2010), stock markets (e.g., Boyd et al., 2005; Cenesizoglu, 2011) as well as in foreign exchange markets (e.g., Andersen et al., 2003; Evans and Lyons, 2008). Furthermore, these forecast data are used to study market co-movement (e.g., Albuquerque and Vega, 2009), market volatility (e.g., Beber and Brandt, 2008; Brenner et al., 2009), changes in market liquidity (e.g., Brandt and Kavajecz, 2004; Pasquariello and Vega, 2007, 2009) as well as bond and equity risk premiums (e.g., Savor and Wilson, 2012; Dicke and Hess, 2012). It appears reasonable to assume that macroeconomic consensus forecasts represent market participants' expectations properly. So far available studies on forecast rationality at the consensus level largely test for general quality properties. They commonly find no evidence of systematic or persistent inefficiencies. In contrast to these previous studies, Campbell and Sharpe (2009) test for a specific behavioral inefficiency, the anchoring bias, first documented by Tversky and Kahneman (1974) in psychological experiments. Transferred to the context of macroeconomic forecasts, anchoring means that analysts put too much importance on last months' data and therefore underweight meanwhile released relevant information. This behavior implies a false incorporation of all available information into their forecasts. Consequently, a correction, i.e., the efficient use of the entire available information set would yield forecast improvements. Our analysis reveals a counter-intuitive result: We find strong statistical significance for anchoring in most macroeconomic forecast series, but applying a look-ahead bias free estimation and adjustment procedure leads to no systematic forecast improvements. Therefore, our results question the economical significance of the anchoring bias. To provide an explanation for the disconnection of statistical and economical significance, we decompose the anchoring bias test statistic and find that the test is biased itself. While the test assumes a univariate information environment, it neglects the possibility that analysts may provide superior forecasts by using a more comprehensive information set than just the univariate time series itself. Our empirical as well as our simulation results strongly support this explanation for a broad range of macroeconomic series. Our analysis contributes to different strands of literature. First, our results directly add to the scarce literature analyzing the efficiency of macroeconomic survey forecasts by showing that informational advantages of analysts, i.e., the incorporation of related macroeconomic data, enable them to outperform mechanically generated time series forecasts. Furthermore, our results provide motivation for other research areas, such as studies analyzing equity analysts' outputs, to control for a larger information set, for instance by including earnings information of related companies or information about overall business conditions. Second, our findings strongly support the assumption that macroeconomic survey forecasts represent a reasonable proxy measure for the anticipated information component in macroeconomic releases and consequently justify their use in the above mentioned research areas. Furthermore, our results highlight the danger to test for cognitive biases in a time series context which were previously only tested in controlled experiments. Especially when experiments are conducted in a highly regulated informational setting, i.e., when information given to test participants has to be strictly controlled for, as in anchoring bias experiments, it is questionable whether a direct transfer in a time series setting is possible at all. Future studies analyzing cognitive biases in time series frameworks have to consider carefully whether informational constraints might drive the results and lead to false conclusions. The first essay provides strong evidence for the quality of macroeconomic forecasts at the consensus level, the second essay deals with individual macroeconomic forecasts and analyzes why certain analysts provide better forecasts then others. In particular, we focus on the association between the idiosyncratic predictability of a given macroeconomic indicator and the relation between analyst characteristics and macroeconomic forecast accuracy. Obviously, there might be quality differences on the individual analyst level, i.e., there are more and less precise macroeconomic analysts. Exploiting these quality differences is a desirable task, because academics would obtain better proxy measures for market participants' expectations, and for investors an information advantage should translate into higher profits. We argue that if an indicator's idiosyncratic predictability is low, i.e., the series is almost not predictable, for instance due to information constrains and very volatile processes, then analysts' forecast performance is rather random than systematic because skills cannot take effect. In contrast, if a macroeconomic indicator has a high idiosyncratic predictability, then analysts with certain characteristics benefit from their abilities and skills, and generate more precise forecasts than less skilled analysts. Accordingly, for the unpredictable indicators the relation between analyst characteristics and forecast accuracy should be less pronounced than for the predictable ones. Consequently, we hypothesize that the idiosyncratic predictability of a certain macroeconomic indicator has to be taken into account whenever the relation between analyst characteristics and forecast accuracy is analyzed. So far there is only contradictory evidence concerning differences in individual forecast quality of macroeconomic analysts. While some studies provide evidence for different forecast quality among individual macroeconomic analysts (e.g. Zarnowitz, 1984; McNees, 1987; Zarnowitz and Braun, 1993; Kolb and Stekler, 1996; Brown et al., 2008) other articles come to the opposite conclusion (e.g. Stekler, 1987; Ashiya, 2006). Despite this disagreement, the relation between macroeconomic forecast accuracy differences and analyst characteristics has not been analyzed so far, although the extensive strand of literature analyzing the association of equity analyst characteristics and earnings per share forecast accuracy (e.g. Clement, 1999; Clement and Tse, 2005; Brown and Mohammad, 2010) provides a sound framework for an analysis. Most importantly, we find that model performance heavily depends on the idiosyncratic predictability of macroeconomic indicators. With decreasing idiosyncratic predictability the relevance of analyst characteristics for forecast accuracy diminishes for some characteristics and disappears for others. In terms of economic significance we find substantial differences between macroeconomic indicators with high and low idiosyncratic predictability. Consequently, our results show that the idiosyncratic predictability of a given forecast target has to be taken into account when the association between analyst characteristics and forecast accuracy is analyzed. Our findings have implications for different research areas. Most importantly we directly add to the literature analyzing individual macroeconomic analysts' forecast performance. We provide evidence that the idiosyncratic predictability of an indicator has to be taken into account if the relation between analyst characteristics and forecast accuracy is analyzed. Differentiation among analysts is only very limited if the figure to be forecasted is virtually unpredictable, because analysts do not benefit from their abilities and experiences. Systematic forecast accuracy differences arise if the forecast target is predictable at all and more skilled analysts have the opportunity to differentiate themselves form less skilled ones based on superior skills. Since there are differences in the predictability of company earnings our framework is transferable. Analogous to our findings for macroeconomic analysts, we expect that idiosyncratic predictability plays an equally important role analyzing the association between equity analysts' characteristics and their earnings per share forecast performance, i.e., for company earnings with higher idiosyncratic predictability we expect higher heterogeneity in forecast accuracy which can be explained by analyst characteristics. The first two essays provide evidence that macroeconomic predictions are in general of high quality as they incorporate rationally information from various sources. Besides the previously analyzed macroeconomic forecasts, agents such as politicians and employers, also heavily rely on other information, for example, on coincident and leading macroeconomic indicators. Determining the current state of the economy and obtaining sound projections about future overall macroeconomic developments plays an important role in their decision processes. Coincident and leading macroeconomic indicators incorporate a large set of macroeconomic variables as well as stock and bond market measures, e.g., returns and interest rate spreads. However, there is no evidence about how expectations at the microeconomic level relate to expectations at the macroeconomic level. Consequently, an aggregate of microeconomic expectation data, i.e., individual company expectations, are not included in coincident and leading macroeconomic indicators so far. To overcome this shortcoming we introduce a bottom-up approach that aggregates individual company expectations to derive macroeconomic content. Since the development of the entire economy is closely related to the development of its individual parts, among them individual companies, aggregated company information must contain macroeconomic information. Unfortunately, there is no database containing managements' expectations, however, we use equity analysts' outputs as proxy measure. Equity analysts' information sets comprise public macroeconomic-, industry- and company-specific content as well as non-public company-specific information (Grossman and Stiglitz, 1980) and is therefore arguably the best available proxy for managements' expectations. Regarding the choice of the best analyst's output we use recommendation changes instead of earnings per share (EPS) changes, because recommendations comprise more information. Besides the one year earnings estimate, recommendations also contain a series of future earnings expectations as well as interest rate and risk premium expectations. We show that aggregated recommendation changes as proxy measure for changing company outlooks have predictive power for overall economic developments. Our results provide evidence that aggregated recommendation changes, which approximate changing expectations about individual companies' economic prospects, have predictive power for future macroeconomic developments of about one year. Controlling for other well established macroeconomic predictors our results remain robust indicating that our measure contains additional independent information. Consequently, it seems promising to include our new predictor into the set of macroeconomic predictors in future applications. Additionally, we find that EPS changes have no predictive power lending support to our assumption that more forward looking information, as included in recommendation changes, is required if one attempts to forecast future macroeconomic developments. Furthermore, our findings provide the missing link between previous studies showing that aggregated analyst outputs have predictive power for overall stock market developments (Howe et al., 2009) and those showing that the stock market leads the real economy (Stock and Watson, 1998). Our results support the notion that changes in expectations about future company performance rationally determine asset values in advance of overall economic activity changes providing the explanation why stock markets lead the real economy. Overall, the three essays in this thesis advance different strands of literature. We show that macroeconomic consensus forecasts are a reliable proxy measure for market participants' expectations. Furthermore, our results provide strong evidence that it is dangerous to transfer psychological experiments into time series frameworks without appropriately controlling the informational environment. Additionally, we show that the idiosyncratic predictability of a given forecast objective, i.e. whether a forecast task is satisfyingly feasible at all, has to be taken into account whenever the association between analyst characteristics and forecast accuracy is analyzed. Macroeconomic analysts do only benefit from their superior skills compared to their competitors if the macroeconomic series is idiosyncratically predictable. For unpredictable series, forecast accuracy is rather random than systematic, because superior skills do not systematically translate in better forecasts. Finally, we show that the aggregation of forecasts on the microeconomic level, i.e., company expectations, is a promising approach to extract macroeconomic information. Overall, we conclude that macroeconomic analysts are very efficient information processors and play an important role as intermediaries in financial markets.