Survey_BW_2007

Abstract

Data availability narrows the list considerably. Sentiment may vary daily, but major episodes occur over years, and the most convincing tests of the effects of sentiment are those in which it is used to actually predict long-horizon returns—tests which suggest a contrarian trading strategy. This rules out using data that do not go back as far as our stock returns data (that is, to the 1960s), which would exclude, for example, data on insider trading, micro-level data on trading behavior, and implied volatility series. Instead, we construct an index based on the six proxies we use in Baker and Wurgler (2006): trading volume as measured by NYSE turnover, the dividend premium, the closed-end fund discount, the number and first-day returns on IPOs, and the equity share in new issues. All these data are available at http://www.stern. nyu.edu/ jwurgler . Later on, we will also consider some mutual fund series. Although these six proxies are highly correlated in the expected directions, some of them also contain idiosyncratic components that are unrelated to sentiment. For example, the 1975 deregulation of brokerage commissions and the subsequent long decline in trading costs has led to a decades-long upward trend in turnover, so we use the log of turnover minus a five-year moving average. With respect to closed-end fund discounts, if the majority of individual investors have come to prefer open-end funds in recent years, the discount provides a less useful summary of the opinion of the marginal investor than it once did. And the evolution of public debt markets has made the equity share less useful in recent years. But no obvious patch suggests itself in these cases.