The Return of Survey Expectations.

• Author: Shleifer, Andrei
• Position: Research Summary - Report

The use of survey expectations data was a key feature of macroeconomics in the 1950s and 1960s, and an important part of research at the NBER during that period. Yet this work slowly ground to a halt in the aftermath of the rational expectations revolution. Under rational expectations, economic agents forecast the future by optimally using the true structure of the economy they operate in. This means that the structure of the economy itself dictates what beliefs they should hold. From the viewpoint of economic research, this implies that expectations data are redundant as long as the econometrician knows the model that economic agents rely on and can compute their statistically optimal expectations of future variables from that model. In financial economics as well as in macroeconomics, the rise of the efficient markets hypothesis rendered expectations data largely irrelevant for addressing key questions.

Although the last 30 to 40 years have seen occasional studies using survey expectations data, this line of work has picked up pace significantly in the last several years. Part of the reason is that we now have much better data, and that survey expectations are actually quite useful in distinguishing alternative models, but part is undoubtedly the fact that rational expectations models in both macroeconomics and finance have increasingly reached dead ends. As a result, survey expectations are staging a remarkable comeback.

Rapidly growing evidence shows that, far from being random noise, measured expectations are highly consistent across surveys that are conducted with different methodologies and using somewhat different questions. Furthermore, actual behavior of survey respondents is predicted more successfully by their survey responses than by some model-based predictors from a rational expectations model. People literally put their money where their mouth is--not where it ought to be in rational expectations models. Last but not least, the evidence shows that forecast errors can be predicted from the information that the decision maker has at the time of making the forecast. This is inconsistent with the rational expectations hypothesis, and points to more realistic economic models of expectation formation and actual behavior. In this summary, I review some of my research that contributes to these findings, conducted jointly with Pedro Bordalo, Nicola Gennaioli, Robin Greenwood, Rafael La Porta, and Yueran Ma. Many other researchers have generated closely related results.

Expectations of Aggregate Stock Returns

Perhaps because the movements of the stock market engage so many people, from individual investors to managers to professional forecasters, there are multiple sources of data on expectations of aggregate stock market returns. Robin Greenwood and I put together data on such expectations, some quantitative and some qualitative, from six sources, with very diverse surveyed populations and different survey questions. (1)

The first source is the Gallup survey of individual investors, with data from 1996 to 2012. For most of this period, this survey asked respondents about their beliefs about stock market returns over the next year, with possible answers ranging from "very optimistic" to "optimistic" to "neutral" to "pessimistic" to "very pessimistic." One can construct a qualitative indicator of return expected by Gallup respondents as the difference between the percentages of bullish and bearish investors...