The Reluctant Analyst

• Published date: 01 September 2016
• Author: ZHIJIE XIAO,DAN BERNHARDT,CHI WAN
• DOI: http://doi.org/10.1111/1475-679X.12120
• Date: 01 September 2016

DOI: 10.1111/1475-679X.12120

Journal of Accounting Research

Vol. 54 No. 4 September 2016

Printed in U.S.A.

The Reluctant Analyst

DAN BERNHARDT,

∗CHI WAN,

†AND ZHIJIE XIAO

‡

Received 23 February 2015; accepted 11 March 2016

ABSTRACT

We estimate the dynamics of recommendations by financial analysts, uncover-

ing the determinants of inertia in their recommendations. We provide over-

whelming evidence that analysts revise recommendations reluctantly, intro-

ducing frictions to avoid frequent revisions. More generally, we characterize

the sources underlying the infrequent revisions that analysts make. Publicly

available data matter far less for explaining recommendation dynamics than

do the recommendation frictions and the long-lived information that ana-

lysts acquire but the econometrician does not observe. Estimates suggest that

analysts structure recommendations strategically to generate a profitable or-

der flow from retail traders. We provide extensive evidence that our model

describes how investors believe analysts make recommendations, and that in-

vestors value private information revealed by analysts’ recommendations.

JEL codes: G2; G24

Keywords: financial analyst recommendations; recommendation revi-

sions; recommendation stickiness; asymmetric frictions; duration; MCMC

methods

1. Introduction

One of the most important services that financial analysts provide is to

make recommendations to retail and institutional customers about which

∗University of Illinois and University of Warwick; †University of Massachusetts Boston;

‡Boston College.

Accepted by Haresh Sapra. We thank an anonymous referee, Murillo Campello, Roger

Koenker, Jenny Tucker, and seminar participants at the University of Illinois, Queen’sUniver-

sity, the University of Guelph, and the University of Florida for helpful comments. Formerly

entitled “The Pretty Good Analyst.”

987

Copyright C, University of Chicago on behalf of the Accounting Research Center,2016

988 D.BERNHARDT,C.WAN,AND Z.XIAO

stocks to purchase, and which ones to sell. Brokerage houses want to em-

ploy financial analysts who provide recommendations on which investors

can profit, thereby generating profitable trading activity for the brokerage

house. Many researchers (e.g., Womack [1996], Francis and Soffer [1997],

Barber et al. [2001, 2006], Ivkovi´

c and Jegadeesh [2004], Jegadeesh et al.

[2004], Howe, Unlu, and Yan [2009], Bradley et al. [2014]) have docu-

mented the profitability and informativeness of various measures of recom-

mendations and recommendation changes.

In this line, one can contemplate an “idealized” financial analyst who first

gathers and evaluates information from public and private sources about a

set of companies to form assessments about their values, and then com-

pares her value assessment with the stock’s price, issuing recommendations

to her investor audience on that basis. Thus, an idealized analyst employing

a five-tier rating system would issue “Strong Buy” recommendations for the

most undervalued stocks, whose value–price differentials, V−P

P, exceed a

high critical cutoff, μ5. The analyst would establish progressively lower cut-

offs, μ4,μ3,andμ2, that determine “ Buy,” “Hold,” “Sell,” and “Strong Sell”

recommendations, so that, for example, the analyst would issue Buy recom-

mendations for value–price differentials between μ5and μ4,andstrongly

advise customers to sell stocks with the worst value–price differentials below

μ2.

Analysts do not form recommendations in this way. To understand why,

observe that sometimes a stock’s value–price differential will be close to a

cutoff, in which case slight fluctuations in price relative to value lead to re-

peated recommendation revisions. In practice, analysts infrequently revise

recommendations—customers would question the ability of an analyst who

repeatedly revised recommendations on which they based investments.

We develop and estimate a model of a “reluctant” financial analyst. The

analyst assesses value just like an idealized analyst, and, when initiating cov-

erage, she makes an initial recommendation on the same basis. However,

the analyst only downgrades a recommendation if the value–price differen-

tial falls far enough below the critical cutoff, and only upgrades a recom-

mendation if the value–price differential rises far enough above the cutoff.

Thus, a reluctant analyst downgrades a recommendation from a Buy only if

a stock’s value–price differential falls below μ4−δ4↓instead of μ4,andshe

upgrades from a Hold only if the differential rises above μ4+δ4↑instead

of μ4,whereδ4↓and δ4↑are stickiness parameters that measure an analyst’s

strategic “reluctance” to revise recommendations.1

Our paper is the first to identify the drivers and determi-

nants of stickiness in analyst recommendations. We distinguish the

relative importance of recommendation revision frictions, persistent ana-

lyst information, and public information available to an econometrician

for explaining the dynamics of analyst recommendations. In turn, these

1See section 2.1 for an extensive motivation of these stickiness parameters.

THE RELUCTANT ANALYST 989

drivers provide insights into the strategic considerations and information

of analysts. We uncover how incorporating strategic behavior and analyst

information alters our understanding of how various public information

characteristics of a firm (e.g., size, past performance) enter an analyst’s

assessment of firm value. We show how our model informs about the

returns of firms following recommendation revisions inside and out of

earnings announcement (EA) and earnings guidance (EG) windows.

Finally, we show that our model provides a measure of the “surprise”

associated with a recommendation revision or initiation that explains the

magnitudes of returns.

There are many different sources of stickiness in recommendations, and

the econometric model must account for each of them to avoid biasing es-

timates that lead to mistaken inferences about their relative importance.

One source of stickiness is simply that much of the public information that

analysts receive arrives in lumps. Concretely, EAs arrive quarterly, and EG is

given sparingly. An unsurprising consequence, for example, is that recom-

mendation revisions are more likely inside announcement and guidance

windows, generating “stickiness” outside these windows.

A second source of stickiness is the information that analysts uncover to

which an econometrician is not privy. This information could reflect an

analyst’s assessments based on repeated private interactions with firm per-

sonnel.2Alternatively, the information need not be “private,” just unob-

served by the econometrician, and hence not in her model of valuation

even when the information enters price.3The valuation consequences of

such information persist—if an analyst has favorable information that the

econometrician lacks, some of the information likely remains months later,

positively affecting future recommendations.

A third source of stickiness is the strategic choices by analysts. Analysts

can reduce the likelihood of revising recommendation iby increasing its

bin size, μi+1−μi, or by increasing the recommendation frictions δi+1↑

and δi↓into higher and lower recommendations. Moreover, an analyst can

reduce the frequency of recommendation revisions not only with symmet-

ric frictions, but also with asymmetric ones, where, say, the friction from

Buy to Hold is large, but that from Hold to Buy is small. Analysts have

2According to a survey conducted by Brown et al. [2015, p. 3], analysts’ private commu-

nication with management is the most useful input to their decision process when forming

earnings forecasts and stock recommendations, and “over half of the analysts, we survey, re-

port that they have direct contact with the CEO or CFO of the typical company they follow five

or more times a year.”

3For example, the public information could be the quality of the CEO, which is obviously

persistent; it could be approval of a drug by the FDA, which will enter share price both now and

into the future (but possibly not near-term earnings), and the value of this approval will persist;

it could be the near-term entry or exit of rivals, the value of new patents (or future expiration

of old ones), etc. Quite generally, any information that enters distant future revenues will

typically be quite persistent, known to the analyst and the market, but not fully captured by a

set of control variables.