Why Don't We Agree? Evidence from a Social Network of Investors

• Published date: 01 February 2020
• DOI: http://doi.org/10.1111/jofi.12852
• Author: MARINA NIESSNER,J. ANTHONY COOKSON
• Date: 01 February 2020

THE JOURNAL OF FINANCE •VOL. LXXV, NO. 1 •FEBRUARY 2020

Why Don’t We Agree? Evidence from a Social

Network of Investors

J. ANTHONY COOKSON and MARINA NIESSNER∗

ABSTRACT

We study sources of investor disagreement using sentiment of investors from a so-

cial media investing platform, combined with information on the users’ investment

approaches (e.g., technical, fundamental). We examine how much of overall disagree-

ment is driven by different information sets versus differential interpretation of infor-

mation by studying disagreement within and across investment approaches. Overall

disagreement is evenly split between both sources of disagreement, but within-group

disagreement is more tightly related to trading volume than cross-group disagree-

ment. Although both sources of disagreement are important, our findings suggest

that information differences are more important for trading than differences across

market approaches.

DISAGREEMENT AMONG INVESTORS HAS LONG been thought to be central to trading

in financial markets. Indeed, it is difficult to explain why investors would

trade at all without some source of disagreement (Milgrom and Stokey (1982),

Karpoff (1986)). Motivated in part by this observation, a growing literature

examines the effects of investor disagreement on financial market outcomes

∗J. Anthony Cookson is affiliated with University of Colorado at Boulder - Leeds School of Busi-

ness. Marina Niessner is with AQR Capital Management. We are grateful to Nick Barberis, Joey

Engelberg (discussant), James Choi, Diego Garcia, Harrison Hong (discussant), Rawley Heimer,

Toby Moskowitz, Tyler Muir, Justin Murfin, Matt Spiegel, Johannes Stroebel (discussant), Paul

Tetlock (discussant), Heather Tookes, Martin Weber (discussant), Paul Irvine (discussant), and

Scott Yonker (discussant) for helpful comments. We thank Jason Klusowski and Toomas Laarits

for outstanding research assistance. This draft has also benefited from the comments of confer-

ence and seminar participants at the 2015 European Summer Symposium for Financial Markets

(evening session), the 2015 IDC Summer Conference (early ideas), Universidad de Chile, Univer-

sity of Colorado Consumer Financial Decision Making Group, Yale School of Management, 2016

National Bureau of Economic Research (NBER) Spring Behavioral Finance Meeting, 2016 NBER

Summer Institute Asset Pricing, 2016 IDC Summer Conference, 6th Helsinki Finance Summit on

Investor Behavior,2016 SITE New Models in Financial Markets Session, 2016 CMU Summer Sym-

posium, Michigan State University, University of Washington, MIT (Accounting), Boston College,

American Finance Association 2017, Jackson Hole Finance Conference 2017, European Winter

Finance Conference 2017, AQR Capital Management, Finance Down under Finance Conference,

Front Range Finance Seminar 2017, Harvard Business School (NOM), Harvard Business School

(Finance), The European Summer Symposium in Financial Markets 2017, and the Red Rock Fi-

nance Conference 2017. AQR Capital Management is a global investment management firm, which

may or may not apply similar investment techniques or methods of analysis as described herein.

The views expressed here are those of the authors and not necessarily those of AQR. We have read

The Journal of Finance’s disclosure policy and have no conflicts of interest to disclose.

DOI: 10.1111/jofi.12852

C2019 the American Finance Association

173

174 The Journal of Finance R

(e.g., Varian(1985), Harris and Raviv (1993), Kandel and Pearson (1995), Nagel

(2005), Banerjee and Kremer (2010), Carlin, Longstaff, and Matoba (2014)).

Prior research has linked disagreement to trading volume and stock returns,

and has studied its dynamic effects (Ajinkya, Atiase, and Gift (1991), Diether,

Malloy, and Scherbina (2002), Banerjee and Kremer (2010)).

However, despite abundant evidence on the consequences of investor dis-

agreement, much less is known empirically about the sources of disagreement.

That is, why do investors disagree in the first place? Leading theories identify

two main sources of disagreement—differences in information sets and dif-

ferences in the models that investors use to interpret information (Hong and

Stein (2007)). To examine these questions empirically, we study disagreement

among investors on the social media investing platform StockTwits, where

users regularly express their opinions (e.g., bullish or bearish) about stocks

and where user profile information explicitly conveys the user’s broad invest-

ment approach (e.g., fundamental, technical). Using this setting, we provide

novel insights into the relative importance of different information sets versus

different investment models.1

Separating the roles of different information sets and different models in de-

termining investor disagreement is empirically challenging, given the typical

data limitations. First, disagreement corresponds to differences in investors’

opinions, which are difficult to observe. Even if a researcher had individual-

level trading data, which itself is hard to come by, it is difficult to impute in-

vestors’ opinions from their trades, as investors can trade for reasons unrelated

to their opinions—such as liquidity. Second, as Rothschild and Sethi (2016)

and Baron et al. (2019) point out, to determine whether differences in investor

opinions are due to differences in information sets or differences in investors’

models, researchers would ideally observe investors’ trading strategies—not

just their executed trades.

Our data set enables us to empirically distinguish between information-

driven and model-driven sources of disagreement because, as we will show,

disagreement across investment approaches is more likely to arise due to dif-

fering investment models, whereas disagreement within investment approach

disagreement is more likely due to different information sets. We find that

differences of opinion across the broad investment approaches in our data are

responsible for approximately half of overall disagreement. At the same time,

within-group differences of opinion are much more strongly related to trading

1Specifically, Hong and Stein (1999) posit that gradual information diffusion is an important

source of disagreement that can drive trading decisions. More recently,Chang et al. (2014)provide

evidence that different information sources lead to a divergence of opinion and greater trading vol-

ume. On the other hand, differential interpretation of information is central to the models of Harris

and Raviv (1993) and Kandel and Pearson (1995). Kandel and Pearson (1995) provide evidence

of differential interpretation by financial analysts and argue that this differential interpretation

leads to greater trading volume after public announcements of information (earnings announce-

ments). A central aim of our paper is to use our decomposition of overall disagreement to speak to

the relative weight of these two theories of trading.

Why Don’t We Agree? Evidence from a Social Network of Investors 175

volume than are differences of opinion across groups, suggesting that model

disagreement is less likely to induce trading than different information sets.

Given that these investment philosophies are self-reported, we carefully

check that adherence to an investment philosophy on StockTwits reflects ad-

herence to an investment model in reality. We first analyze the textual content

of tweets by users of different philosophies. We find that users of different

philosophies use language that is consistent with the underlying philosophy

(e.g., fundamental traders discuss earnings, technical traders discuss charts,

and momentum traders discuss trends). Next, speaking to the external valid-

ity of the language used, we find that the language used on the StockTwits

platform closely resembles public writings of prominent investors with par-

ticular investment philosophies. Furthermore, using hand-classified lists of

information words (i.e., referring to news sources or timing) and model words

(i.e., referring to substantive analyses), we find that information words tend

to be used across investment philosophies, while model words tend to focus

on one or two investment philosophies. Beyond language usage, we show that

investor sentiment reactions to earnings news concentrate among fundamen-

tal investors, while sentiment reactions to “technical view” events identified

by the news analytics database RavenPack concentrate among technical in-

vestors.2Taken together, these findings support the view that the differences

across investment philosophies are significant, substantive, and a function of

differential beliefs about investing.

Turning to our main findings, we observe that both within-group and cross-

group disagreement significantly predict abnormal trading volume, but that

within-group disagreement exhibits a much stronger relation to trading vol-

ume. Specifically, we find that a one-standard-deviation increase in within-

group disagreement is associated with four times the increase in abnormal

trading volume as a one-standard-deviation increase in cross-group disagree-

ment. This finding is robust to alternative specifications for the differences

between within-group and cross-group disagreement. Moreover, we continue

to find a similarly large effect on within-group disagreement when we restrict

attention to opinions from before the market opens. We therefore conclude

that both types of disagreement are important determinants of trading, but

that within-group (informational) differences matter more than differences in

investment philosophies. This result suggests that disagreement due to slow

information diffusion is important for trading volume.

We provide two additional pieces of evidence on the slow information diffu-

sion hypothesis using self-reported experience classifications to split our sam-

ple of investors into sophisticated and unsophisticated investors. First, we find

that within-strategy disagreement across sophisticated and unsophisticated in-

vestors predicts trading volume. This result suggests that information diffuses

from sophisticated investors to unsophisticated slowly over time, consistent

2This test is analogous to the work of Jia, Wang, and Xiong (2015) who show that local and

foreign investors react differently to recommendations of local and foreign analysts in the context

of the Chinese stock market.