Sentiment, order imbalance, and co‐movement: An examination of shocks to retail and institutional trading activity

• DOI: http://doi.org/10.1111/eufm.12146
• Author: Neophytos Lambertides,Patricia Chelley‐Steeley,Christos S. Savva
• Date: 01 January 2019
• Published date: 01 January 2019

DOI: 10.1111/eufm.12146

ORIGINAL ARTICLE

Sentiment, order imbalance, and co-movement:

An examination of shocks to retail and institutional

trading activity

Patricia Chelley-Steeley

1

|

Neophytos Lambertides

2

|

Christos S. Savva

2

1

University of Birmingham Business

School, University of Birmingham,

Edgbaston Park Road, Birmingham B15

2TY, UK

Email: p.l.chelley-steeley@bham.ac.uk

2

Department of Commerce, Finance and

Shipping, Cyprus University of

Technology, 3603 Lemesos, Cyprus

Emails: n.lambertides@cut.ac.cy;

christos.savva@cut.ac.cy

Abstract

Using order flow imbalance as a measure of sentiment we

show that positive and negative shocks to sentiment lead

to lower co-movement between portfolio and market returns

in the post-shock period. Furthermore, an asymmetry is

present as positive shocks to sentiment have less impact on

co-movement changes than negative shocks. Moreover,

shocks to retail sentiment and the sentiment of two types of

institutional investors lead to a reduction in co-movement.

Positive shocks to institutional order flow imbalance lead to

smaller reductions in co-movement than associated with

retail shocks. These effects exist even after controlling for

firm-specific and market-wide news.

KEYWORDS

co-movement, order flow shock and sentiment, smooth transition model

JEL CLASSIFICATION

G12, G14

We would like to thank the editor and an anonymous referee for the generosity of their time and comments, which have

allowed us to improve the paper considerably.

116

|

© 2017 John Wiley & Sons, Ltd. wileyonlinelibrary.com/journal/eufm Eur Financ Manag. 2019;25:116–159.

1

|

INTRODUCTION

Compelling evidence shows that the mood or sentiment of the stock market can play an important role

in influencing the extent to which individual stock returns co-move with the rest of the market.

A pioneering paper by Kumar and Lee (2006) suggests that order flow imbalance (buying activity

relative to selling activity) is a useful measure of sentiment. This connection exists because an

optimistic mood encourages more buying activity and less selling activity, while a pessimistic mood

encourages more selling and less buying activity. Kumar and Lee (2006) use a measure of order

flow imbalance adjusted for common factors to capture sentiment and show that sentiment increases

co-movement.

A premise of Kumar and Lee (2006) is that market sentiment is captured by the sentiment of retail

investors because these investors do not have access to the information resources that institutions have,

so they make more irrational investment decisions. However, there is growing evidence to suggest that

institutional investors also act irrationally and are influenced by sentiment (see, e.g., Bagnoli, Clement,

Crawley, & Watts, 2014; Barberis, Shleifer, & Wurgler, 2005; Brown & Cliff, 2005; DeVault, Sias, &

Starks, 2016). Institutional investors become sentiment traders because they are influenced by

‘reputational trading,’which encourages institutions to trade in the direction of sentiment to avoid their

performance standing out from the average, while the consistent short-term predictability of sentiment

strategies and the impediments to corrective low-cost arbitrage cause institutions to take advantage of

short-term predictability driven by sentiment because it is profitable. As a result, the order flow of

institutional investors reflects sentiment. Moreover, recent analysis of the order flow of retail and

institutional investors by DeVault et al. (2016) shows that institutions are more driven by sentiment

than retail investors. We are therefore motivated to extend the analysis of Kumar and Lee (2006) and

consider whether the sentiment of retail and institutional investors can influence co-movement.

One of the key results presented in Kumar and Lee (2006) shows that the portfolio returns of small

firms is positively influenced by the order flow imbalance or sentiment of retail investors. They do not

find that there is a relationship between the portfolio returns of larger firms and retail sentiment. Large

institutional investors invest more heavily in large firms, while retail investors are more concentrated in

small firms. This suggests that for large firms to reflect investor sentiment the measure of sentiment

may need to be broadened to include institutional sentiment, as these investors are more likely to have

an influence over large firms. Moreover, the results of Kumar and Lee (2006) do not provide direct and

conclusive evidence of the relationship between sentiment and co-movement, only that sentiment

influences returns and therefore indirectly must influence co-movement.

Using an adaptation of the Lee and Radhakrishna (2000) algorithm we identify the trades

associated with three types of traders. Small trades are classified as retail trades; medium trades are

classified as institutional informed trades or stealth trades; and the largest trades are classified as large

institutional trades. For each of these three investor groups we use the traders’daily order flow

imbalance to capture their sentiment. For each type of investor this is measured as a ratio of their dollar

value of buyer-initiated trades to seller-initiated trades adjusted for common factors. We also measure

the total sentiment of the market using all trades. Over the period examined we find that sentiment to

the different investor groups on average across all stocks is optimistic, but varies over time.

Empirically, we find that firm characteristics are also important in determining sentiment, as some

firm characteristics are associated with average order imbalances greater than unity, which suggests

greater optimistic sentiment. These characteristics are firms that are S&P500 index constituents, larger,

more liquid, have low book-to-market ratios, have higher prices, are older firms, have higher levels of

institutional ownership, have lower earnings-to-price ratios and higher earnings growth, because on

average over time buying pressure for these stocks outweighs selling pressure. Large institutional

CHELLEY-STEELEY ET AL.

|

117

traders on average display positive sentiment across all the stock characteristics we examine, but

have the greatest positive sentiment for smaller firms, stocks with low illiquidity, and firms

with high prices and earnings growth. Retail investors have pessimistic sentimentforsmallstocks,

low priced stocks, and young stocks, but the most positive sentiment for high priced stocks, firms

with low earnings per share (EPS) ratios, and firms with high earnings growth. Stealth traders

generally have positive sentiment for each firm characteristic which is especially strong for large

firms, and stocks with high prices, high institutional ownership, low EPS, and high earnings

growth.

We calculate the average pairwise correlation between changes in the order flow imbalance of one

stock and another to identify whether changes to order flow are systematically correlated, an indication

that trading decisions are coordinated and therefore influenced by sentiment. Our results suggest that

changes to the overall order flow imbalances across all stocks are not highly correlated, but changes

amongst retail investors appear to be correlated. We also find that firm characteristics influence the

average pairwise correlations and contribute to higher levels of coordination amongst firms with some

shared characteristics. Multivariate analysis identifies which characteristics have an independent

influence over correlations.

We next examine whether the sentiment level of the investor group is related to portfolio excess

returns. This is similar to some previous analysis undertaken by Kumar and Lee (2006). We extend

their analysis by regressing portfolio returns in excess of the risk-free rate against a set of market factors

and the sentiment of our distinctive types of investors. We find that total sentiment, retail, and large

institutional sentiment are correlated with portfolio returns, even after controlling for market risk

factors. This confirms that the sentiment of retail investors influences co-movement, but also shows

that the sentiment of institutional investors matters.

To our knowledge, we are the first paper to comprehensively examine the impact that shocks to

sentiment, measured as order flow imbalance shocks, have on co-movement. The study of shocks

rather than levels offers a number of advantages. First, shocks capture unanticipated changes in order

flow imbalance so reflect the element of order flow imbalance or sentiment that represents an

innovation or change in behavior, so may have a different impact on returns and co-movement to order

flow imbalance levels. We also show that order flow imbalance levels are non-stationary, but that

changes to order flow imbalance are stationary, which provides an additional motivation to focus on

shocks. Within the paper we consider two types of shocks. The first is the change in the order flow

imbalance level between consecutive periods. This is a useful measure as it adequately captures the

concept of a shock and can be easily calculated.

Using this concept of shock, we estimate the average pairwise correlation between the market

excess return and changes in sentiment for each of the investor groups. This analysis is similar to

some that Kumar and Lee (2006) undertook, but we extend their analysis to also include the

sentiment of stealth and institutional traders. We find that the changes in total order flow

imbalance and retail order flow imbalance are negatively correlated with changes in the market

excess return. This suggests that changes in order flow imbalance are associated with reduced

co-movement and motivates us to examine the impact of order flow shocks on stock market

correlation further.

A weakness of the analysis undertaken by Kumar and Lee (2006) is that a direct link between

co-movement and order flow imbalance is not established. This motivates us to examine a second

type of shock to order flow imbalance, which is derived from the Silvennoinen and Teräsvirta

(2005) and Berben and Jansen (2005) Smooth Transition Conditional Correlation Generalized

Autoregressive Conditional Heteroskedasticity model (STCC GARCH model). Estimation of this

model allows us to capture the conditional return correlation between individual portfolio returns

118

|

CHELLEY-STEELEY ET AL.