A sizeable literature reports that financial market analysts and forecasters herd for reputational reasons (See e.g. Devenow and Welch (1996) and Hirshleifer and Teoh (2003) for surveys). As an example, Lamont (2002) argues that if forecasters are "punished" for being wrong, forecasters could have an incentive to mimic other forecasters, i.e. to herd. Using new data from a large survey of professional forecasters' expectations about stock market movements, we find strong evidence that the expected average of all forecasters' forecasts (the expected consensus forecast) in uences an individual forecaster's own forecast. This looks like herding. In our survey, forecasters do not herd for reputational reasons, however. Instead of herding, we suggest that forecasters form higher-order expectations in the spirit of Keynes (1936) who compared the setting of prices in financial markets with a newspaper beauty contest. In this contest, competitors were invited to pick the six prettiest faces from among one hundred photographs. The winner of the contest was the person whose choice most closely corresponded to the average preference of all competitors. Hence, a competitor should not only pick the prettiest faces, but those he thought the other competitors would also view as the prettiest, and -at the same time- take into account that all other competitors would form expectations in a similar manner. Keynes' idea was that financial markets work in this way, too. Keynes' description of asset pricing as a beauty contest has recently received renewed attention in the theoretical literature examining the theoretical implications of higherorder expectations for asset prices (e.g. Allen, Morris, and Shin (2006), Bacchetta and van Wincoop (2006, 2009), Nimark (2007), Banerjee, Kaniel, and Kremer (2009), and Makarov and Rytchkov (2008)). We follow up on those papers by providing some first empirical evidence on this form of expectation formation in this paper. Our first result is that forecasters are in uenced by the expected consensus forecast. This basic result is highly statistically significant, remains significant when adding different kinds of control variables, shows up in all our robustness tests, and, most importantly, is also of substantial economic significance. Finding that forecasters are in uenced by the consensus forecast in this dataset of German forecasters both confirms findings from the U.S. to other countries, and, at the same time, thereby provid "out-of-sample"evidence on the results from the U.S. studies, as reported in, e.g., Graham (1999), Welch (2000), and Lamont (2002). We find that young forecasters and portfolio managers, who in previous studies have been reported to be those who in particular herd, rely more on the expected consensus forecasts than other forecasters. Given that forecasters have no incentive to herd in our study, we conclude that our results indicate that the incorporation of the expected consensus forecast into individual forecasts is most likely due to higher-order expectations.
Rangvid, Jesper, Maik Schmeling and Andreas Schrimpf (2009), Higher-Order Beliefs Among Professional Stock Market Forecasters: Some First Empirical Tests , ZEW Discussion Paper No. 09-042, Mannheim. Download