Sentiment Metrics and Investor Demand

• DOI: http://doi.org/10.1111/jofi.12754
• Date: 01 April 2019
• Published date: 01 April 2019

THE JOURNAL OF FINANCE •VOL. LXXIV, NO. 2 •APRIL 2019

Sentiment Metrics and Investor Demand

LUKE DeVAULT, RICHARD SIAS, and LAURA STARKS∗

ABSTRACT

Recent work suggests that sentiment traders shift from safer to more specula-

tive stocks when sentiment increases. Exploiting these cross-sectional patterns and

changes in share ownership, we find that sentiment metrics capture institutional

rather than individual investors’ demand shocks. We investigate the underlying eco-

nomic mechanisms and find that common institutional investment styles (e.g., risk

management, momentum trading) explain a significant portion of the relation be-

tween institutions and sentiment.

THE INVESTOR SENTIMENT LITERATURE TYPICALLY assumes that irrational individ-

ual investors are the source of sentiment-based demand shocks that drive prices

from value. For instance, the earliest academic article focusing on this issue

uses odd-lot trading to identify demand shocks attributed to “public psychol-

ogy” relative to the more rational views of “New YorkStock Exchange members”

(Drew et al. (1950)). The assumption that individual investors are responsible

for sentiment-induced mispricing has been explicitly repeated in the nearly

70 years that have followed.1In fact, many of the traditional proxies for

∗Luke DeVault is with Clemson University. Richard Sias is with the Eller College of Manage-

ment, the University of Arizona. Laura Starks is with the McCombs School of Business, Univer-

sity of Texas at Austin. Wethank seminar participants at Boston College, Cambridge University,

Colorado State University, Southern Methodist University, the University of Arizona, UC River-

side, the University of Massachusetts Amherst, the University of New South Wales, University

of Technology Sydney, University of Waterloo, Vienna University of Economics and Business, VU

Amsterdam, the Wharton School, the 2013 European Finance Association Meetings, the 2014

Asian Bureau of Finance and Economic Research Annual Conference, the 2015 American Finance

Association Meetings, the 2015 UC Davis Symposium, and the 2017 UT Dallas Spring Finance

Conference as well as Gennaro Bernile, Doug Foster, Kelvin Law, Charles Lee, David McLean,

Harry Turtle, Kumar Venkataraman, and Jeff Wurgler for their helpful comments. We are also

grateful for helpful guidance from Ken Singleton (the Editor), an anonymous Associate Editor, and

two anonymous referees. We thank Brian Bushee, Ken French, TerryOdean, and Jeff Wurgler for

providing data. None of the authors received any funds for the production of this research. The

authors do not have any conflicts of interest as identified in the Journal of Finance Disclosure

policy. Sias sits on the board and investment committee for a nonprofit organization (unpaid posi-

tion). Starks sits on the board of a set of mutual funds and variable annuities. As a result of that

position, Starks also sits on the Board of Governors of the Investment Company Institute (ICI)

representing independent directors for mutual funds (an unpaid position).

1A small sample of the many studies that express this view include Zweig (1973), Shleifer

and Summers (1990), Lee, Shleifer, and Thaler (1991), Neal and Wheatley (1998), Nagel (2005),

Baker and Wurgler (2006,2007), Lemmon and Portniaguina (2006), Barberis and Xiong (2012),

Stambaugh (2014), and Da, Engelberg, and Gao (2015).

DOI: 10.1111/jofi.12754

985

986 The Journal of FinanceR

investor sentiment (e.g., closed-end fund discounts, mutual fund flows, odd-

lot transactions) have been selected precisely because they are designed to

capture the behavior of individual investors.

Contrary to this widely held assumption, in this paper, we demonstrate that

sentiment metrics capture the demand shocks of institutional, rather than in-

dividual, investors. It follows that if these metrics capture investor sentiment,

then the traders driving the sentiment-induced mispricing are institutional,

rather than individual, investors (in aggregate). We further show that our

results make sense intuitively and economically, as previously documented

common institutional investment styles contribute to the relations between

investor sentiment metrics, institutional demand shocks, and returns. Con-

tributing factors include institutions’ risk management, reputational concerns,

momentum trading, herding, bubble riding, and, to a lesser extent, underlying

investor flows. Beyond these explanations, however, a substantial portion of the

relation between institutional demand and sentiment metrics remains unex-

plained, leading to the possibility that omitted variables play a role in driving

the relations between institutional demand shocks, investor sentiment, and

equity returns.

Prior to Baker and Wurgler (henceforth BW) (2006,2007), empirical support

for the investor sentiment hypothesis was, at best, mixed (see the Internet

Appendix2for a review of this literature). BW’s innovative approach to exam-

ining the relation between sentiment and asset prices led to a renewed focus on

investor sentiment and the widespread adoption of their sentiment metric—

the two BW studies have been cited more than 5,800 times. Specifically, BW

propose that sentiment traders shift from safe to speculative securities when

sentiment increases, and from speculative to safe securities when sentiment

declines, and that these sentiment-induced demand shocks drive mispricing

in financial markets. Consistent with their hypothesis, BW find positive senti-

ment betas for speculative stocks and negative sentiment betas for safe stocks,

and that speculative stocks average lower returns than safe stocks following

periods of high sentiment levels. The authors conclude that these return pat-

terns serve as “ ...a powerfulconfirmation of the sentiment-driven mispricing

view” (2007, p. 135).

The sentiment hypothesis requires that sentiment-induced demand shocks

(i.e., net buying by sentiment traders) impact prices. However, because every

buyer requires a seller, sentiment traders’ net demand shocks must be offset

by supply from traders who are less susceptible to changes in sentiment. By

recognizing this market-clearing condition, we can identify whether the traders

captured by sentiment metrics are in aggregate individual or institutional

investors, as changes in sentiment will be positively related to changes in

sentiment traders’ demand (i.e., demand shocks) for speculative stocks and

inversely related to sentiment traders’ demand shocks for safe stocks. Thus,

in our initial test we examine correlation, not causation, as sentiment metrics

2The Internet Appendix may be found in the online version of this article.

Sentiment Metrics and Investor Demand 987

capturing individual investors’ demand shocks is a necessary but insufficient

condition of the traditional interpretation of the investor sentiment hypothesis.

Inconsistent with this traditional interpretation, we find that commonly used

measures of investor sentiment capture the demand shocks of institutional,

rather than individual, investors. In particular, an increase in sentiment is

associated with both an increase in institutional investors’ demand for risky

stocks and, by definition, an associated decrease in individual investors’ de-

mand for speculative stocks. Further supporting the hypothesis that the BW

sentiment metric captures institutional, rather than individual, investors’ de-

mand, we find that the level of institutional investors’ speculative stock hold-

ings, relative to the level of their holdings of safe stocks, is higher when the

level of sentiment is higher. Moreover, none of the 17 investor sentiment met-

rics we examine—the BW metric, the six individual components of the BW

metric, three measures of mutual fund flows, two consumer sentiment mea-

sures, one survey-based measure of individual investors’ sentiment, two mea-

sures of venture capital flows as proxies for sophisticated investors’ sentiment,

and two measures of aggregate economic activity or stress—capture individual

investors’ demand shocks. In contrast, 10 of these 17 measures are meaning-

fully related to institutional investors’ demand shocks. Further, the ability of

sentiment metrics to predict cross-sectional return patterns is limited only to

those metrics that capture institutional investors’ demand shocks. In short,

the relations between sentiment metrics, returns, and institutional demand

shocks are pervasive.

The balance of our study focuses on understanding why sentiment metrics

capture institutional, rather than individual, investors’ behavior. (For ease of

exposition, we henceforth refer to the relation between institutions and sen-

timent as “institutional sentiment trading”). We first consider the possibility

that retail investors actually drive our findings either because the complement

of 13(f) demand shocks does not capture retail investors’ demand shocks or

because retail investor flows drive aggregate institutional investor demand.

Inconsistent with the first possibility, individual investors’ demand shocks—as

captured by the Odean (1998) retail broker data—are strongly related to the in-

verse of 13(f)-inferred institutional demand shocks. With respect to the second

possibility, that underlying retail investor flows drive institutional investors’

trades, we conduct four separate tests. First, we decompose 13(f) institutional

demand shocks into flow-induced or managers’ decisions. While we find that

underlying investors’ flows affect which securities 13(f) institutions buy and

sell, there is no evidence that these flows drive the relation between aggregate

13(f) institutional demand shocks and changes in sentiment. Similarly, in a

second test we examine mutual fund holdings data, which allows us to mea-

sure flows within a fund family. In this test we find some evidence that flows

between funds in the same family contribute to institutional sentiment trad-

ing. Nonetheless, managers’ decisions (whether at the 13(f) level or the mutual

fund level) play a dominant role in driving the relation between institutional

trading and the change in sentiment metrics. Third, following Edelen, Ince,

and Kadlec (2016), we examine the relation between changes in sentiment and