Economic Policy Uncertainty and Short-term Reversals

Andy C.W. Chui

School of Accounting and Finance

Faculty of Business
The Hong Kong Polytechnic University
Email: [email protected]

This draft: August 17, 2021

Abstract
This study finds that short-term reversals become more profound in the current month
when economic policy uncertainty is larger in the prior month. There is evidence that the
economic policy uncertainty influences return reversals through the liquidity channel.
Short-term reversal profits are also positively related to the VIX index, the Baker-Wurgler
(2007) investor sentiment index, and the Aruoba-Diebold-Scotti (2009) business
conditions index in the prior month. Though, the predictability of the latter two indexes
is less robust. However, adding these indexes and other variables does not weaken the
relationship between economic policy uncertainty and return reversals.

Keywords:
Economic policy uncertainty, Return reversals, Liquidity provision, Investor sentiment

JEL classification:
G11; G12;G14

I appreciate Ken French, Jeffrey Wurgler, the Chicago Board Options Exchange (Cboe)
and the Federal Reserve Bank of Philadelphia for making publicly available the five
Fama-French factors, the Baker-Wurgler (2007) investor sentiment index, the volatility
index (VIX) and the Aruoba-Diebold-Scotti (2009) business conditions index used in this
study. I would like to thank Man Hong Chan for his excellent research assistance. I
acknowledge financial support from the Research Grants Council of the Hong Kong
Special Administrative Region, China (RGC Project No. 15507920).

Electronic copy available at: https://ssrn.com/abstract=3924771

1. Introduction

The objective of this study is to investigate the relationship between economic

policy uncertainty and short-term reversals. Short-term reversal refers to the tendency for

stocks that performed the worst (best) in the prior month or week to perform well (poor)

in the current month or week. The reversal strategy considered in this study is to buy the

losers (worse-performing stocks) and short the winners (best-performing stocks) in the

prior month. The short-term reversal effect pertains to the returns of this reversal strategy.

There is extensive evidence for this short-term reversal effect in both the US and

international stock markets (Jegadeesh, 1990, Griffin, Kelly, and Nardari, 2010, Cheng,

Hameed, Subrahmanyam, and Titman, 2017).

There are both liquidity and behavioural explanations for the short-term reversal

effect. The liquidity explanation is based on the idea that holding the inventory of a stock

is risky. Hence, risk-averse liquidity providers demand compensations in order to absorb

order imbalances, and stock prices return to their intrinsic value when these inventory

imbalances vanish (Jegadeesh and Titman, 1995, Avramov, Chordia, and Goyal, 2006;

Nagel, 2012, Cheng et al., 2017, Chui, Subrahmanyam, and Titman, 2021). 1 The

behavioural explanation posits that investors overreact to information because of

cognitive bias, such as overconfidence. Hence, stock prices may momentarily deviate

from their fundamental value, and these pricing errors are corrected in the near future

(Cooper, 1999; Mase, 1999; Subrahmanyam, 2005, Da, Liu, and Schaumburg, 2014).

Using order flow data and quote mid-point returns for NYSE stocks,

Subrahmanyam (2005) documents that investors' overreaction is the main cause of

1
As in Nagel (2012), liquidity providers in this study refer to investors who can provide liquidity and do
not restricted to designated market makers.

Electronic copy available at: https://ssrn.com/abstract=3924771

monthly return reversals. Da et al. (2014) show that short-term return reversal is related

to both explanations. They find that the reversal effect on the short side is related to

overreaction, while the reversal effect on the long side is attributable to the liquidity

effect. Using the Cboe volatility index (VIX) as a measure of liquidity provision, Nagel

(2012) documents that VIX positively predicts short-term reversal. This result supports

the liquidity explanation of reversal. Cheng et al. (2017) show that short-term reversal

profitability decreases with past 3-month stock returns. This finding indicates that the

liquidity provision of institutional investors is a major source of short-term reversals in

the US market. Recently, Chui et al. (2021) explore short-term reversal in the Chinese A-

shares and B-shares markets and find evidence consistent with the liquidity explanation.

Prior literature has not examined how the macroeconomic environment influences short-

term reversal. This study fills this gap and hypothesizes that economic policy uncertainty

(EPU) predicts the short-term reversal effect.

This study focuses on EPU because economic uncertainty is known to influence

investors’ behaviour. Williams (2015) documents that macroeconomic environments

shape how investors respond to firm-specific news. Bali, Brown, and Tang (2017)

examine how economic uncertainty affects cross-sectional stock returns and find that

stocks with more negative uncertainty beta are commensurate with higher returns.

Recently, Nagar, Schoenfeld, and Wellman (2019) show that increased economic

uncertainty raises bid-ask spreads and reduces stock price reactions to earnings news.

This study uses EPU to measure economic uncertainty because it is unlikely to be

affected by investors and is a crucial aspect of the corporate information environment

(Nagar et al., 2019).

Electronic copy available at: https://ssrn.com/abstract=3924771

 Economic policy uncertainty (EPU) influences short-term reversal through both

liquidity and behavioural channels. So and Wang (2014) find that the compensation for

providing liquidity goes up substantially before earnings announcements when the

uncertainty associated with these announcements increases. Wang, Xu, and Zhong (2019)

document that increases in economic policy uncertainty decreases liquidity provision in

the credit default swap (CDS) market. Furthermore, Nagar et al. (2019) show that a rise in

EPU increases trading costs as measured by bid-ask spreads. These findings indicate that

liquidity provision is more costly when EPU is high and suggest that EPU can predict the

short-term reversal effect. On the other hand, behavioural biases tend to have a more

significant impact on investors’ decisions when market uncertainty is large (Hirshleifer,

2001 and Kumar, 2009). If investors’ cognitive bias causes short-term reversal, high EPU

is expected to be associated with a larger short-term reversal effect.

This study employs data from the US to test the hypothesis that there is a positive

relationship between EPU in the prior month and short-term reversals in the current

month. This positive relationship is robust even after controlling for the Bake-Wurgler

(2007) sentiment index (BWSENT), the Cboe volatility index (VIX), the Aruoba-Diebold-

Scotti (2009) business conditions index (ADS), and the five Fama-French (2015) risk

factors. Further analysis suggests that EPU effect on short-term reversals is not limited to

small stocks.

I divide the whole sample period from January 1985 to June 2019 into two sub-

periods based on the median of the EPU index, high-uncertainty (top 50%) and low-

uncertainty (bottom 50%). I find that the spreads between short-term reversal returns in

the high-uncertainty period and the low-uncertainty period range from 1.4% per month

Electronic copy available at: https://ssrn.com/abstract=3924771

(microcap stocks) to 1% per month (big stocks). These spreads are statistically

significant at the 5% level. Furthermore, the finding from time-series regression indicates

that a one standard deviation increase in the EPU index leads to a 0.6% increase in the

monthly returns on short-term reversal strategy when the Fama-French five factors are

evaluated at their means. Thus, our results are both statistically and economically

significant.

This study contributes to the literature in three aspects. First, this study documents

that economic policy uncertainty forecasts short-term reversal returns. This finding helps

us understand how the short-term reversal effect varies across time. Second, Nagel (2012)

argues that VIX influences liquidity provision and finds that VIX predicts short-term

reversal effect. This study documents that EPU and VIX are important determinants of

short-term reversal, suggesting that liquidity provision in the stock market is lower when

EPU is higher. Finally, after controlling for investor sentiment, this study still finds a

significant positive relationship between lagged EPU and the returns of short-term

reversal. This result provides support to the liquidity explanation of the short-term

reversal effect.

The paper is organized as follows. First, section 2 establishes the hypothesis. Then,

section 3 describes the data. Next, section 4 presents the empirical findings. Finally,

Section 5 concludes the paper.

2. Hypothesis

Since holding inventory of a stock is risky, liquidity providers demand

compensation for providing liquidity (Jegadeesh and Titman, 1995, Avramov, Chordia,

Electronic copy available at: https://ssrn.com/abstract=3924771

and Goyal, 2006; Nagel, 2012, Cheng et al., 2017, Chui et al. 2021). When there is a

positive excess inventory, liquidity providers incline to quote prices below the

fundamental value to discourage sell orders and elicit buy orders. For the same reason,

liquidity providers set prices above the fundamental value when there is negative excess

inventory. Stock prices go back to their fair values when these imbalances in order flow

disappear. This adjustment of order imbalance gives rise to short-term reversal. When

economic uncertainty increases, it is expected that inventory risk will rise, and the supply

of liquidity will decline. Therefore, the compensation for absorbing order imbalance

increases. Looking at earnings announcements, So and Wang (2014) find that the reward

for providing liquidity goes up substantially before earnings announcements when the

uncertainty associated with these announcements increases. Hence, the liquidity

explanation of short-term reversal predicts that economic uncertainty increases the short-

term reversal effect.

On the other hand, the behavioral explanation posits that investors overreact to

information because of cognitive bias, such as overconfidence (Cooper, 1999; Mase,

1999; Subrahmanyam, 2005, Da, Liu, and Schaumburg, 2014). Overconfidence makes

investors overestimate the quality of the information that they received. As a result, when

good (bad) news arrives, investors bid the stock prices up (down) above (below) the

fundamental value. These pricing errors are corrected shortly in the future that generates

short-term reversal. Evidence in psychology suggests that cognitive bias exerts a greater

influence on judgment when decisions are made in more difficult situations (Lichtenstein

and Fischhoff, 1977, and Griffin and Tversky, 1992). Hirshleifer (2001) argues that

investors’ cognitive biases become stronger when investors are trading in informationally

Electronic copy available at: https://ssrn.com/abstract=3924771

opaque environments. Kumar (2009) finds evidence showing that investors exhibit

stronger cognitive biases when market uncertainty is higher. In particular, Kumar (2009)

documents that investors become more overconfident when market uncertainty increases.

Hence, the behavioural explanation indicates that economic uncertainty boosts investors’

overconfidence, and investors overreact more to news, leading to more significant pricing

errors that, in turn, results in a stronger short-term reversal.

In summary, both the liquidity and behaviour explanations suggest that an increase

in economic uncertainty leads to a more significant short-term reversal effect. In this

study, we use economic policy uncertainty as a measure of economic uncertainty. Hence,

we have the following hypothesis.

Ho: When economic policy uncertainty is more prominent in the prior month, the

short-term reversal effect is more substantial in the current month.

3. Data

3.1 Stock sample

Our stock sample consists of all common stocks traded on the NYSE, AMEX, and

NASDAQ exchanges from January 1985 through June 2019. Our sample starts from

January 1985 because the EPU index is not available before 1985. Our tests are mainly

based on monthly stock returns because estimation bias due to the bid-ask spread and thin

trading on estimating short-term return reversals are likely to be small when monthly

returns are employed (Jegadeesh, 1990). Data on stock returns, share prices, number of

shares traded, and shares outstanding are collected from the CRSP. Annual data on book

equity of each firm in our stock sample are collected from Compustat North America.

Electronic copy available at: https://ssrn.com/abstract=3924771

Firm characteristics, such as firm size (SZ), return volatility (Rtnvol), the mean adjusted

average turnover ratio (TNMA), the mean adjusted Amihud (2002) illiquidity ratio

(ILLIQMA), the book-to-market ratio (BM), past 3-month returns (Rt-2, t-4), and past 6-

month returns (Rt-2, t-7) are computed from these data.2 Stocks with closing prices less

than $5 are removed from the sample.3 The appendix provides detailed definitions of

these variables as well as other variables used in this study.

To calculate risk-adjusted returns, we use the Fama and French (2015) five-factor

model. These factors are a market factor based on excess market return (value-weighted

market return minus risk-free rate, EMKTR), a size factor based on firm size (small minus

big, SMB), a book-to-market factor based on the book-to-market ratio (high minus low,

HML), a profitability factor based on operating profitability (robust minus weak, RMW),

and an investment factor based on total assets growth (conservative minus aggressive,

CMA). The Brennan, Chordia, and Subrahmanyam (1998) approach is used to estimate

risk-adjusted returns. Specifically, I estimate the following regression for each firm in

each month:

𝑅𝑗𝑡 = 𝛼𝑗 + 𝛽𝑀𝑘𝑡,𝑗 𝐸𝑀𝐾𝑇𝑅𝑡 + 𝛽𝑆𝑀𝐵,𝑗 𝑆𝑀𝐵𝑗𝑡 + 𝛽𝐻𝑀𝐿,𝑗 𝐻𝑀𝐿𝑗𝑡 + 𝛽𝑅𝑀𝑊,𝑗 𝑅𝑀𝑊𝑗𝑡 +

𝛽𝐶𝑀𝐴,𝑗 𝐶𝑀𝐴𝑗𝑡 + 𝑒𝑗𝑡 , (1)

where Rjt is the stock return of firm j in month t and ejt is the error term. The estimation

period is sixty months prior to each month in the testing period, and each stock is

required to have at least twenty-four observations in the estimation period. Risk-adjusted

2
To make the number of shares traded for the NASDAQ stocks comparable to that of the stocks listed on
NYSE and AMEX, I use a method similar to that in Cheng et al. (2017) to adjust the number of shares traded
for NASDAQ stocks. Specifically, before 2001, NASDAQ number of shares traded is divided by 2. For
2001, NASDAQ number of shares traded is divided by 1.8. From 2002 to 2003, NASDAQ number of shares
traded is divided by 1.6. No adjustment is made after 2003.
3
Specific data requirements are applied to the measures of the mean adjusted Amihud illiquidity ratio, the
mean adjusted turnover ratio, and return volatility. These requirements help make these measures more
reliable. Please refer to the appendix for these requirements.

Electronic copy available at: https://ssrn.com/abstract=3924771

return of firm j in month t (AdjRjt) in the testing period is computed from the following

equation:

𝐴𝑑𝑗𝑅𝑗𝑡 = 𝑅𝑗𝑡 − 𝛽̂𝑀𝑘𝑡,𝑗,𝑡 𝐸𝑀𝐾𝑇𝑅𝑡 − 𝛽̂𝑆𝑀𝐵,𝑗,𝑡 𝑆𝑀𝐵𝑡 − 𝛽̂𝐻𝑀𝐿,𝑗,𝑡 𝐻𝑀𝐿𝑡 −

𝛽̂𝑅𝑀𝑊,𝑗,𝑡 𝑅𝑀𝑊𝑡 − 𝛽̂𝐶𝑀𝐴,𝑗,𝑡 𝐶𝑀𝐴𝑡 , (2)

where the betas, 𝛽̂𝑀𝑘𝑘,𝑗,𝑡 , 𝛽̂𝑆𝑀𝐵,𝑗,𝑡 , 𝛽̂𝐻𝑀𝐿,𝑗,𝑡 , 𝛽̂𝑅𝑀𝑊,𝑗,𝑡 , 𝛽̂𝐶𝑀𝐴,𝑗,𝑡 are estimated from Equation

(1).

3.2 Economic policy uncertainty (EPU) index and other indexes

The economic policy uncertainty index (EPU) is constructed by Baker, Bloom, and

Davis (2006) and is available from the Economic Policy Uncertainty website

(policyuncertainty.com). This index is a weighted average of the number of words

relating to uncertainty or government policies that appeared in major newspapers. The

higher this index is, the more economic policy uncertainty is. The monthly score of this

index is available starting from January 1985.4 This index is widely used in the business

literature (Nagar et al., 2019). The natural logarithm of this EPU index (LnEPU) is used

in the time-series and cross-sectional regressions.

Monthly Baker and Wurgler (2007) investor sentiment index (BWSENT) is

collected from the website of Jeffrey Wurgler. This index is based on the first principal

component of five sentiment proxies. These proxies are value-weighted dividend

premium, first-day returns on IPOs, IPO volume, close-end fund discount, and equity

share in new issues. The construction of this index makes it more relevant to retail

investor sentiment. The higher is BWSENT, the stronger is retail investor sentiment. This

index is available from July 1965 to December 2018.

4
The raw economic uncertainty index is scaled by the mean calculated from a normalization period and
this period for the US is from 1995 to 2009.

Electronic copy available at: https://ssrn.com/abstract=3924771

 Daily Cboe implied volatility index is collected from the Chicago Board Options

Exchange website (cboe.com) and is available starting from January 1986.5 This index is

used as a measure of liquidity provision in Nagel (2012) and is also associated with

investor fear sentiment (Da, Engelberg, and Gao, 2015). Since index options are mainly

traded by institutional investors, this index is more related to the sentiment of institutional

investors. The higher is this index, the greater is fear sentiment, and the weaker is

liquidity provision of institutional investors. The Cboe volatility index (VIX) in a given

month is calculated as the mean of the daily volatility index in that month.

Daily Aruoba, Diebold, and Scotti (2009) business conditions index is from the

Federal Reserve Bank of Philadelphia and is available from 1960. This index is

constructed from several seasonally adjusted macroeconomic variables of mixed

frequencies, such as weekly initial jobless claims, monthly payroll employment, and

quarterly GDP. A more positive index value indicates progressively better-than-average

conditions, while a more negative index value indicates progressively worse-than-average

conditions. Similar to VIX, the ADS business conditions index (ADS) in a given month is

calculated as the mean of the daily Aruoba, Diebold, and Scotti (2009) business

conditions index in that month.

4. Empirical findings

While several tiny firms could drive returns on equal-weighted portfolios, a few

large firms may dominate returns on value-weighted portfolios (Fama and French, 2008).

To make the returns on the reversal portfolios better reflect the profitability of the

5
The Cboe volatility index used in this study is based on the old methodology and Cboe refers this index as
VXO.

Electronic copy available at: https://ssrn.com/abstract=3924771

reversal strategy, reversal portfolios are formed for the whole sample and sub-samples

sorted by firm size. The whole sample is named the Market sample in this study.

Following Cheng et al. (2017), stocks in each month t are divided into three groups based

on their firm size: Microcap (stocks with their firm size in month t-1 that are less than

the 20th NYSE size percentile), Small (stocks with their firm size in month t-1 that are

between the 20th and 50th NYSE size percentiles), and Big (stocks with their firm size in

month t-1 that are larger than the 50th NYSE size percentile).

In each month t, stocks in each sample (Market, Micro, Small, or Big) are allocated

into five portfolios based on their returns in month t-1 from the loser (bottom 20%) to the

winner (top 20%). These portfolios are equal-weighted. A reversal strategy refers to

taking a long position in the loser portfolio and a short position in the winner portfolio.

Accordingly, the return on a reversal portfolio in month t is the difference in returns

between the loser and winner portfolios in that month. Return on the reversal portfolio is

also known as the short-term reversal effect.

4.1 Returns to short-term reversal strategy

Table 1 reports the average returns on the portfolios sorted by prior month returns

in each size-sorted sample and the Market sample. For the Market sample, the average

short-term reversal effect is 0.67% per month, and it is significant at the 1% level. A

closer look at reversal effect in each size-sorted sample reveals that the reversal effect in

the Market sample mainly comes from firms in the Microcap sample. The short-term

reversal effect is decreasing with firm size. While the short-term reversal effect for the

Microcap sample is 0.91% per month with a t-statistics of 5.06, these effects for the

Small and Big firms are positive but insignificant. These findings are consistent with the

Electronic copy available at: https://ssrn.com/abstract=3924771

improvement of liquidity provision in the US market in recent decades (Cheng et al.,

2017).

[Insert Table 1 here]

Returns on reversal portfolios are regressed on the five Fama-French factors, and

the estimated intercepts in these time-series regressions are the risk-adjusted short-term

reversal effects. The Newey-West (1987) heteroscedasticity and autocorrelation

consistent estimates of standard errors are used to compute the t-statistics on the

estimated coefficients from the time-series regressions and the cross-sectional regressions

to be discussed later.6 Similar to the results using raw returns, the risk-adjusted reversal

effects are significantly positive only in the Market and Microcap samples. The risk-

adjusted reversal effects for Small and Big firms are positive but insignificant. Table 4

displays these findings.

4.2 Economic uncertainty and market returns

Bali et al. (2017) find that about 50% of the stocks traded in the NYSE, AMEX, and

NASDAQ have negative betas associated with economic uncertainty. Therefore, if an

increase in EPU reduces liquidity provision and raises investor overreaction, the pricing

errors should be larger in the current month. In other words, investors likely sell (buy)

stocks with negative (positive) uncertainty betas at prices that are deemed to be too low

(high) when EPU is high in the current month, and these pricing errors could be corrected

in the coming month. Therefore, EPU can predict market returns. To examine the

relationship between EPU and market return, I estimate the following regression model:

6
Following Cheng et al. (2017), Newey and West (1987) method with 3 lags is used.

Electronic copy available at: https://ssrn.com/abstract=3924771

 𝑃𝑅𝑗𝑡 = 𝛼𝑗 + 𝛽1𝑗 𝑃𝑅𝑡−1 + 𝛽2𝑗 𝐿𝑛𝐸𝑃𝑈𝑡−1 + 𝛽3𝑗 𝐵𝑊𝑆𝐸𝑁𝑇𝑡−1 + 𝛽4𝑗 𝑉𝐼𝑋𝑡−1

+ 𝛽5𝑗 𝐴𝐷𝑆𝑡−1 + 𝑒𝑗𝑡 , (3)

where PRjt is the equal-weighted return on portfolio j in month t, and j equals Market,

Microcap, Small, or Big sample. The explanatory variables are the lagged portfolio

return (PRjt-1), the natural logarithm of EPU, the Bake-Wurgler (2007) sentiment index

(BWSENT), the Cboe volatility index (VIX), and the Aruoba-Diebold-Scotti (2009)

business conditions index (ADS) in month t-1. Because of the data availability of

BWSENT and VIX, the estimation period is from February 1986 to January 2019. The

lagged portfolio return is used to control for possible lead-lag effect in market returns.

The index BWSENT is included to control for retail investor sentiment, and VIX is used to

control for both liquidity provision and fear sentiment of institutional investors. The ADS

index is in the regression to control for general business conditions.

Panel A of Table 2 shows that the contemporaneous relationships between returns

on the Market and size-sorted portfolios and EPU are significantly negative, suggesting

that stock returns decline when economic policy uncertainty rises. It is noticed that the

contemporaneous relationships between VIX and returns on all portfolios are also

negative and significant. However, the contemporaneous relationships between the other

two indexes and returns on all portfolios, but the portfolio of big firms are weak. While

the contemporaneous relationship between BWSENT and return on the portfolio of big

firms is negative, this relationship between ADS and return on the portfolio of big firms is

positive. These two relationships are statistically significant at the 10% level. Panel B of

Table 2 displays that LnEPU is positively associated with VIX and negatively related to

Electronic copy available at: https://ssrn.com/abstract=3924771

BWSENT and ADS. These correlation coefficients are statistically significant at the 1%

or 5% levels.

Panel C of Table 2 reports the findings from the estimations of Equation (3). The

estimated coefficients on lagged portfolio return are positive and significant for the

Market, Microcap, and Small samples. This coefficient is positive but insignificant for

Big sample. These findings suggest that the lead-lag effect in returns is strong only

among tiny and small firms. The estimated coefficients on BWSENTt-1, VIXt-1, and ADSt-1

are statistically insignificant. On the contrary, the estimated coefficients on LnEPUt-1 are

always positive and statistically significant at the 5% level. This result indicates that

LnEPU predicts market returns. Although increases in EPU correspond with low market

returns in the current month, they predict high market returns (reversal) in the coming

month. This phenomenon is consistent with the conjecture that rises in economic policy

uncertainty decreases liquidity provision and increases investor overconfidence in the

current month that leads to return reversal in the coming month.

[Insert Table 2 here]

4.3 Short-term reversals conditional on economic policy uncertainty

Let us look at how the short-term reversal effect behaves in high- and low-

uncertainty states. The whole sample period from January 1985 to June 2019 is divided

into two sub-periods based on the median EPU index score. A given month t is classified

as low-uncertainty when its EPU index score is less than or equal to the median in month

t-1. Accordingly, a given month t is classified as high-uncertainty when its EPU index

score is higher than the median in month t-1. Average returns on the portfolios

constructed in Section 4.1 are calculated separately for the low- and high-uncertainty

Electronic copy available at: https://ssrn.com/abstract=3924771

periods. Table 3 reports these mean returns. I find that short-term reversal is

significantly positive only in the high-uncertainty period (i.e. months following months

with higher-than-median EPU). Interestingly, big firms also exhibit short-term reversals

(0.80% per month with a t-statistic of 2.46) in months following months with higher-

than-median EPU. The average monthly spreads in short-term reversal effect between

high- and low-uncertainty periods for the Market, Microcap, Small, and Big samples are

1.35%, 1.42%, 1.57%, and 0.99%, respectively. All these spreads are statistically

significant at the 5% or 1% levels. These findings support the hypothesis and suggest

that short-term reversals are concentrated on months following high economic policy

uncertainty even after controlling for firm size.

[Insert Table 3 here]

4.4 Short-term reversals and economic policy uncertainty: Time-series regressions

To explore if the relationship between EPU and short-term reversal documented in

previous sessions is affected by risks and factors closely related to EPU, I estimate the

following time-series regression:

𝐿𝑊𝑗𝑡 = 𝛼𝑗 + 𝛽1𝑗 𝐸𝑀𝐾𝑇𝑅𝑡 + 𝛽2𝑗 𝑆𝑀𝐵𝑗𝑡 + 𝛽3𝑗 𝐻𝑀𝐿𝑗𝑡 + 𝛽4𝑗 𝑅𝑀𝑊𝑗𝑡 + 𝛽5𝑗 𝐶𝑀𝐴𝑗𝑡 +

𝛽6𝑗 𝐿𝑛𝐸𝑃𝑈𝑡−1 + 𝛽7𝑗 𝐸𝑀𝐾𝑇𝑅𝑡−1 + 𝛽8𝑗 𝐵𝑊𝑆𝐸𝑁𝑇𝑡−1 + 𝛽9𝑗 𝑉𝐼𝑋𝑡−1 +

𝛽10𝑗 𝐴𝐷𝑆𝑡−1 + 𝑒𝑗𝑡 , (4)

where LWjt is the short-term reversal return on portfolio j in month t, and j equals Market,

Microcap, Small, or Big sample. The five Fama-French risk factors in month t are

EMKTRt, SMBt, HMLt, RMWt, and CMAt. To investigate whether reversals are driven by

strong liquidity demand in the down market, excess market return in the prior month

(EMKTRt-1) is included in the regression. As in Equation (3), BWSENTt-1, VIXt-1, and

Electronic copy available at: https://ssrn.com/abstract=3924771

ADSt-1 are used to control retail investor sentiment, fear sentiment and liquidity provision

of institutional investors, and general business conditions in the prior month. The

coefficient of interest is β6 which captures the effect of economic policy uncertainty in

the prior month on return reversals.

When the regression model includes only the five Fama-French risk factors and

economic policy uncertainty (LnEPU), the effect of LnEPU on reversals is prominent.

The estimated coefficient on LnEPU of the market portfolio is 1.723, with a t-statistic of

2.94. The estimated coefficients on LnEPU of the Microcap, Small, and Big portfolios

are 1.970 (t-statistic = 3.44), 1.996 (t-statistic = 2.58), and 1.278 (t-statistic = 2.05),

respectively.

When the other three indexes, BWSENTt-1, VIXt-1, and ADSt-1, are included in the

regression model, the estimated coefficients on LnEPU become weaker. Yet, economic

policy uncertainty still has a substantial impact on return reversals. The estimated

coefficients on LnEPU range from 1.220 (Big) to 1.911 (Small), and all of them but that

of Big portfolio are statistically significant at the 5% level. The estimated coefficient of

LnEPU of the Big portfolio is statistically significant at the 10% level.

Consistent with the findings in Nagel (2012), VIXt-1 exerts a strong positive effect

on return reversals. However, the estimated coefficients on EMKTRt-1, BWSENTt-1, and

ADSt-1 are statistically insignificant. I also find that return reversals have significant

exposure to the market, value, and investment factors. While return reversals have

positive loadings on the market and value factors, they have negative loadings on the

investment factor.7

7
All these loadings but one are statistically significant at 1% or 5% levels. Return reversal of the small
portfolio has a positive but insignificant loading on the value factor.

Electronic copy available at: https://ssrn.com/abstract=3924771

 Overall, the findings suggest that economic policy uncertainty in the prior month

has a strong positive effect on return reversals in the current month even after controlling

for the five Fama-French risk factors, retail investor sentiment, fear sentiment and

liquidity provision of institutional investors, and general business conditions. Since both

retail investor sentiment and fear sentiment of institutional investors are taking care of,

the finding indicates that EPU influences return reversals through the liquidity provision

channel. As VIX is also a proxy of liquidity provision, the result also suggests that EPU

affects liquidity provision that is independent of the liquidity effect of VIX. The evidence

is consistent with the situation that the demand for liquidity is greater and the supply of

liquidity is smaller when EPU increases. Apart from VIX, EPU can also serve as a proxy

for liquidity provision.

[Insert Table 4 here]

4.5 Short-term reversal, economic policy uncertainty, and volatility

The liquidity provision and arbitrage are riskier for stocks with larger return

volatilities (Cheng et al. 2017). If economic policy uncertainty affects return reversals

through the liquidity channel, one would expect it would exert a more considerable

impact on return reversals among stocks with greater return volatilities. Return volatility

of a stock in month t (Rtnvolt) is calculated as the standard deviation of daily returns on

this stock in the same month. Each stock must have data on daily return for at least 60%

of the trading days in each month. In each month t, stocks in each size-sorted portfolio

(Market, Microcap, Small, or Big) are divided into two portfolios based on their return

volatilities in month t-1: low (bottom 50%) and high (top 50%). Stocks in each size-

volatility-sorted portfolio are further allocated into quintiles based on their returns in

Electronic copy available at: https://ssrn.com/abstract=3924771

month t-1. Return reversal in month t (LWt) is the difference in returns on the losers

(bottom 20%) and winners (top 20%) in each size-volatility-sorted portfolio. Using these

return reversals conditional on return volatility, I estimate Equation (4) and Panel A of

Table 5 reports the findings.

[Insert Table 5 here]

Consistent with my expectation, the estimated coefficients on economic policy

uncertainty (LnEPU) are positive and significant only among stocks with high return

volatilities. Moreover, the estimated coefficient on LnEPU conditional on high volatility

is significantly larger than that conditional on low volatility in each portfolio. This

finding indicates that economic policy uncertainty affects return reversal through the

liquidity channel. Panel B of Table 5 displays the differences of estimated coefficients

between high- and low-volatility stocks.

Except for the Microcap sample, return reversals’ loadings on the value factor for

stocks with high return volatilities are significantly larger than stocks with low return

volatilities. On the other hand, for each size-sorted portfolio, return reversals’ loadings

on the investment factor for stocks with high return volatilities are significantly less than

stocks with low return volatilities. The effects of VIX on return reversals are positive and

significant for stocks with high return volatilities as well as stocks with low return

volatilities. However, the estimated coefficients on VIX conditional on high return

volatility are not significantly larger than that conditional on low return volatility. The

estimated coefficients on lagged excess market return, retail investor sentiment, and

business conditions remain statistically insignificant.

Electronic copy available at: https://ssrn.com/abstract=3924771

4.6 Cross-sectional regressions

So far, the evidence is based on portfolio returns. This section examines the

relationship between economic policy uncertainty and return reversals using data on

individual stocks. The advantage of this analysis is that one can investigate how prior

month returns and other firm characteristics jointly determine stock returns. This

analysis involves two steps. In the first step, I use the Fama-MacBeth (1973) approach to

estimate the following cross-section regression in each month for each sample.

𝑅𝑗𝑡 = 𝛼𝑗 + 𝛽1𝑡 𝑅𝑗𝑡−1 + 𝛽2𝑡 𝐼𝐿𝐿𝐼𝑄𝑀𝐴𝑗𝑡−1 + 𝛽3𝑡 𝑇𝑁𝑀𝐴𝑗𝑡−1 + 𝛽4𝑡 𝑅𝑡𝑛𝑣𝑜𝑙𝑗𝑡−1 +

𝛽5𝑡 𝐿𝑛𝑆𝑍𝑗𝑡−1 + 𝛽6𝑡 𝐿𝑛𝐵𝑀𝑗𝑡−1 + 𝛽7𝑡 𝑅𝑗,𝑡−2,𝑡−4 + 𝛽8𝑡 𝑅𝑗,𝑡−2,𝑡−7 + 𝑒𝑗𝑡 , (5)

where Rjt and Rjt-1 are returns on stock j in month t and month t-1, respectively. Common

determinants of cross-sectional stock returns are included in the regressions. They are the

mean adjusted Amihud illiquidity ratio (ILLIQMAjt-1), the mean adjusted turnover ratio

(TNMAjt-1), return volatility (Rtnvoljt-1), the natural logarithm of firm size (LnSZjt-1), the

natural logarithm of the book-to-market ratio (LnBMjt-1) for stock j in month t-1. The past

six-month return of stock j (Rjt-2,t-7) is a measure for the momentum effect, and it is

computed from months t-2 to t-7. Past three-month return is also included in the

regression because Cheng et al. (2017) find that return reversals are stronger among

losing stocks based on their past three-month returns. The past three-month return for

stock j (Rjt-2,t-4) is calculated from months t-4 and t-2. As a robustness test, I estimate

Equation (5) using risk-adjusted return on stock j in month t (AdjRjt) as the dependent

variable. The estimation period is from February 1986 to January 2019. The definitions

of these variables are in the appendix. Panel A of Table 6 shows the time-series means of

Electronic copy available at: https://ssrn.com/abstract=3924771

the estimated coefficients in both sets of regressions (raw returns and risk-adjusted

returns) in each sample.

[Insert Table 6 here]

Consistent with the results from portfolios, I find strong reversal effects in both sets

of regressions in each sample. All estimated coefficients on the prior month return (Rjt-1)

are negative and statistically significant at the 1% or 5% levels. For the market sample in

both sets of regressions, the estimated coefficients on the mean adjusted illiquidity ratio

are significantly positive while the estimated coefficients on firm size and past three-

month return are significantly negative. The estimated coefficients on other variables are

insignificant.

The estimated coefficients on the mean adjusted illiquidity ratio become

insignificant in each size-sorted sample in both sets of regressions. In addition, in both

sets of regressions in each size-sorted sample, the estimated coefficients on firm size are

significantly negative for the Microcap and Small samples only. Furthermore, when raw

return is the dependent variable, the estimated coefficients on the past three-month return

are negative and significant only for the Microcap and Small samples. Finally, when risk-

adjusted return is the dependent variable, the estimated coefficient on the past three-

month return is negative and significant only for the Small sample. These results suggest

that the size effect and the past three-month return effect on stock returns are

concentrated on tiny firms or small firms.

In the second step, the estimated coefficients on the prior month return (β1t) in both

sets of regressions in each sample in month t are regressed on the natural logarithm of the

economic policy uncertainty index (LnEPUt-1), the excess market return (EMKTRt-1), the

Electronic copy available at: https://ssrn.com/abstract=3924771

Bake-Wurgler (2007) sentiment index (BWSENTt-1), the Cboe volatility index (VIXt-1),

and the Aruoba-Diebold-Scotti (2009) business conditions index (ADSt-1) in month t-1.

Panel B of Table 6 displays the findings.

All the estimated coefficients but one on LnEPUt-1 in both sets of regressions are

negative and statistically significant at the 1% or 5% levels. The estimated coefficient on

LnEPUt-1 of the Microcap sample is negative and is significant at the 10% level for the

regression involving risk-adjusted returns. On the whole, the findings suggest that return

reversals in the current month become stronger when economic policy uncertainty in the

prior month increases.

Consistent with the results reported in previous sections, the Cboe volatility index

exhibits a strong positive impact on return reversals. For both sets of regressions, all the

estimated coefficients on VIXt-1 are negative and statistically significant at the 1% level.

The excess market return in the prior month (EMKTRt-1) continues to have no effect on

return reversals. For the market portfolio, the estimated coefficients on the Bake-Wurgler

(2007) sentiment index (BWSENTt-1) are significantly negative in both sets of regressions.

Nonetheless, this retail investor sentiment effect on return reversals is concentrated on

Microcap stocks. Besides, the estimated coefficients on the Aruoba-Diebold-Scotti (2009)

business conditions index (ADSt-1) are significantly negative in both sets of regressions

for the Small and Big portfolios. This finding suggests that general business conditions

have a significant positive impact on return reversals for small and big firms.

Overall, results from the cross-sectional regressions indicate that return reversals

are related to economic policy uncertainty, retail investor sentiment, fear sentiment and

liquidity provision of institutional investors, and general business conditions in the prior

Electronic copy available at: https://ssrn.com/abstract=3924771

month. Both economic policy uncertainty and fear sentiment and liquidity provision of

institutional investors influence return reversals of firms regardless of their firm size. In

addition, the retail investor sentiment effect on return reversals is focused on tiny firms,

while the business conditions effect on return reversals is concentrated on small and big

firms.

5. Conclusion

This study sheds light on how economic policy uncertainty influences short-term

reversal in the US stock market. Specifically, return reversals are larger in the current

month when economic policy uncertainty is greater in the prior month. The economic

policy uncertainty effect on return reversals is pervasive and is not limited to tiny stocks.

The finding is consistent with both the liquidity and behaviour explanations of short-term

reversals. An increase in economic policy uncertainty reduces liquidity provision and

boosts investor overconfidence in the current month, leading to stronger return reversals

in the coming month. Moreover, further analyses indicate that the economic policy

uncertainty effect on return reversals goes through the liquidity provision channel. This

evidence shows that liquidity providers do demand higher returns when economic policy

uncertainty rises.

The findings from both the portfolio and individual stock analyses show that

economic policy uncertainty index (EPU) and the Cobe volatility index (VIX), a proxy for

fear sentiment and liquidity provision of institutional investors, are important

determinants of short-term reversals. These findings suggest that both EPU and VIX have

a significant association with liquidity provision in the stock market. Since VIX is more

Electronic copy available at: https://ssrn.com/abstract=3924771

related to institutional investors and EPU influences all investors, the former reflects the

liquidity provision of institutional investors, and the latter shows the liquidity provision

of general investors.

Results from individual stock analysis display that return reversals are also related

to the Bake-Wurgler (2007) sentiment index (BWSENTt-1) and the Aruoba-Diebold-Scotti

(2009) business conditions index (ADSt-1) in the prior month. The positive effect of

BWSENTt-1 on return reversals is concentrated on microcap stocks, while the positive

effect of ADSt-1 on return reversals is focused on small and big stocks. Since the Bake-

Wurgler (2007) sentiment index is a proxy for retail investor sentiment, the result

indicates that retail investor sentiment effect on return reversals is more significant

among tiny stocks. However, the business condition effect on return reversals seems

more noticeable among small and big firms.

Cross-country differences in liquidity pricing are influenced by differences in the

legal system and political risk across countries (Bekaert, Harvey, and Lundblad, 2007).

Liang and Wei (2012) find that variations in corporate governance across nations can

explain differences in the liquidity premium across countries. These findings indicate

substantial variations in liquidity provision across countries, which are related to

differences in institutions. Since economic policy uncertainty is closely related to

liquidity provision, its effect on return reversals is likely affected by differences in

institutions. Hence, further research may examine how differences in economic policy

uncertainty influence short-term reversal across countries.

Electronic copy available at: https://ssrn.com/abstract=3924771

Appendix: Variable definitions and sources

Aruoba-Diebold-Scotti (2009) business conditions index (ADSt): The business conditions

index in month t is collected from the Federal Reserve Bank of Philadelphia.

Bake-Wurgler (2007) sentiment index (BWSENTt): The investor sentiment index in

month t is collected from the webpage of Jeffrey Wurgler.

Book-to-market ratio (BM): The BM for a firm from July of year s to June of year s+1 is
the ratio of its book equity (BE) at fiscal year-end in year s-1 to its firm size (SZ) in
December of year s-1. BE is computed following the method outlined in Cohen,
Polk, and Vuolteenaho (2003). The variable LnBM is the natural logarithm of BM.
Stocks with negative BM are eliminated from the sample for the cross-sectional
regression.

Cboe Volatility index (VIXt): The volatility index is the average of the daily Cboe
volatility index in month t. The daily index is collected from the Chicago Board
Options Exchange (Cboe).

Economic policy uncertainty index (EPUt): Economic policy uncertainty index in month t.
This index is constructed by Baker, Bloom, and Davis (2006) and is available from
the Economic Policy Uncertainty website (policyuncertainty.com). The variable
LnEPUt is the natural logarithm of EPU in month t.

Excess market return (EMKTRt): Value-weighted market return minus the risk-free rate in
month t, and it is the market factor of the five Fama-French factors. The five
Fama-French factors are collected from the webpage of Ken French.

Firm size (SZt): Shares outstanding times the end-of-month closing price in month t. The
variable LnSZt is the natural logarithm of SZ in month t. Stocks with missing firm
size are eliminated from the sample.

Investment factor (CMAt): This factor in month t and it is one of the five Fama-French
factors.

Electronic copy available at: https://ssrn.com/abstract=3924771

Mean adjusted Amihud illiquidity ratio (ILLIQMAt): Amihud illiquidity ratio of stock i in
1 𝐷 |𝑅𝑖𝑑𝑠 |
year s is calculated as 𝐼𝑙𝑙𝑖𝑞𝑖𝑠 = 𝑖𝑠
∑𝑑=1 × 106 where |𝑅𝑖𝑑𝑠 | is the absolute
𝐷𝑖𝑠 𝐷𝑉𝑎𝑙𝑖𝑑𝑠
return on stock i on day d in year s, 𝐷𝑉𝑎𝑙𝑖𝑑𝑠 is the dollar trading volume stock i on
day d in year s, and 𝐷𝑖𝑠 is the number of trading days for stock i in year s. Each
stock is required to have more than 200 observations on the daily ratio between
return and trading volume in each year. Stocks with a closing price of less than $5
or missing firm size at the end of December each year are removed from the sample.
Stocks whose Illiqis in year s is higher than the 99th percentile or lower than the 1st
percentile of the NYSE distribution are eliminated from the sample. NASDAQ
trading volume is adjusted using a method similar to that in Cheng et al. (2017).
Specifically, before 2001, NASDAQ trading volume is divided by 2. For 2001,
NASDAQ trading volume is divided by 1.8. From 2002 to 2003, NASDAQ trading
volume is divided by 1.6. No adjustment is made after 2003. The mean adjusted
Amihud illiquidity ratio of stock i in year s (ILLIQMAis) is Illiqis divided by the
average market illiquidity across stocks in year s. The mean adjusted Amihud
illiquidity ratios in year s are used to explain cross-sectional monthly stock returns
in year s+1. The mean adjusted Amihud illiquidity ratio of stock i in month t in
year s+1 (ILLIQMAit) is ILLIQMAis in year s.

Mean adjusted average turnover ratio (TNMAt): Turnover ratio of stock i in a given
month is the number of shares traded divided by total shares outstanding in that
month. Average turnover ratio of stock i in year s is calculated as the mean of the
monthly turnover ratio of stock i in year s. To compute the average turnover ratio,
stocks are required to have at least six observations on the monthly turnover ratio in
year s. Stocks with a closing price of less than $5 at the end of December each year
are removed from the sample. Stocks whose average turnover ratio in year s is
higher than the 99th percentile or lower than the 1st percentile of the NYSE
distribution are eliminated from the sample. NASDAQ number of shares traded is
adjusted using a method similar to that in Cheng et al. (2017). Specifically, before
2001, NASDAQ number of shares traded is divided by 2. For 2001, NASDAQ
number of shares traded is divided by 1.8. From 2002 to 2003, NASDAQ number of
shares traded is divided by 1.6. No adjustment is made after 2003. The mean
adjusted average turnover ratio of stock i in year s (TNMAis) is the average turnover
ratio of stock i divided by the average market turnover ratio across stocks in year s.
The mean adjusted turnover ratios in year s are used to explain cross-sectional
monthly stock returns in year s+1. The mean adjusted average turnover ratio of
stock i in month t in year s+1 (TNMAit) is TNMAis in year s.

Past three-month return (Rt-2,t-4): Cumulative return (%) calculated from month t-2 to
month t-4.

Past six-month return (Rt-2,t-7): Cumulative return (%) calculated from month t-2 to
month t-7.

Electronic copy available at: https://ssrn.com/abstract=3924771

Portfolio return (PRt): Return on the portfolio (%) in month t.

Profitability factor (RMWt): This factor in month t and it is one of the five Fama-French
factors.

Return volatility (Rtnvolt): Stock return volatility is calculated as the standard deviation
of daily stock returns (%) in month t. Each stock is required to have data on returns
for more than 60% of the days in a given month.

Risk-adjusted stock return (AdjRt): Stock return (%) adjusted for the five Fama-French
factors in month t.

Size factor (SMBt): This factor in month t and it is one of the five Fama-French factors.

Stock return (Rt): stock return (%) in month t.

Value factor (HMLt): This factor in month t and it is one of the five Fama-French factors.

Electronic copy available at: https://ssrn.com/abstract=3924771

References

Amihud, Yakov, 2002, Illiquidity and stock returns: cross-section and time-series effects,
Journal of Financial Markets 5, 31-56.

Aruoba, S.B., F.X. Diebold, and C. Scotti, 2009, Real-time measurement of business
conditions, Journal of Business and Economics Statistics 27, 417-427.

Avramov, Doron, Tarun Chordia, and Amit Goyal, 2006, Liquidity and autocorrelation in
individual stock returns, Journal of Finance 61, 2365-2394.

Baker, Malcolm, and Jeffrey Wurgler, 2007, Investor sentiment in the stock market,
Journal of Economic Perspectives 21, 129-151.

Bali, Turan G., Stephen J. Brown, and Yi Tang, 2017, Is economic uncertainty priced in
the cross-section of stock returns? Journal of Financial Economics 126, 471-489.

Bekaert, Geert, Campbell R. Harvey, and Cgristian Lundblad, 2007, Liquidity and
expected returns: lessons from emerging markets, Review of Financial Studies 20,
1783-1831.

Brennan, Michael J., Tarun Chordia, and Avanidhar Subrahmanyam, 1998, Alternative
factor specifications, security characteristics, and the cross-section of expected stock
returns, Journal of Financial Economics 49, 345-373.

Cheng, Si, Allaudeen Hameed, Avanidhar Subrahmanyam, and Sheridan Titman, 2017,
Short-term reversals: the effects of past returns and institutional exits, Journal of
Financial and Quantitative Analysis 52, 143-173.

Chui, Andy C.W., Avanidhar Subrahmanyam, and Sheridan Titman, 2021, Momentum,
reversals, and investor clientele, UCLA working paper.

Cohen, Randolph B., Christopher Polk, and Tuomo Vuolteenaho, 2003, The value spread,
Journal of Finance 58, 609-641.

Cooper, Michael, 1999, Filter rules based on price and volume in individual security
overreaction, Review of Financial Studies 12, 901–935.

Da, Zhi, Qianqiu Liu, and Ernst Schaumburg, 2014, A closer look at the short-term return
reversal, Management Science 60, 658-674.

Da, Zhi, Joseph Engelberg, and Pengjie Gao, 2015, The sum of all fears investor
sentiment and asset prices, Review of Financial Studies 28, 1-32.

Fama, Eugene F., and James MacBeth, 1973, Risk, return, and equilibrium: Empirical
tests, Journal of Political Economy 81, 607-636.

Electronic copy available at: https://ssrn.com/abstract=3924771

Fama, Eugene F., and Kenneth R. French, 2008, Dissecting anomalies, Journal of
Finance 63, 1653-1678.

Fama, Eugene F., and Kenneth R. French, 2015, A five-factor asset pricing model,
Journal of Financial Economics 116, 1-28.

Griffin, John M., Patrick J. Kelly, and Federico Nardari, 2010, Do market efficiency
measures yield correct inferences? A comparison of developed and emerging markets,
Review of Financial Studies 23, 3225-3277.

Griffin, D., and A. Tversky, 1992, The weighting of evidence and the determinants of
confidence, Cognitive Psychology 24, 411-435.

Hirshleifer, D.A., 2001, Investor psychology and asset pricing, Journal of Finance 56,
1533-1597.

Jegadeesh, Narasimhan, 1990, Evidence of predictable behavior of security returns,

Journal of Finance 45, 881-898.

Jegadeesh, Narasimhan and Sheridan Titman, 1995, Short-horizon return reversals and
the bid-ask spread, Journal of Financial Intermediation 4, 116-132.

Kumar, Alok, 2009, Hard-to-value stocks, behavioral biases, and informed trading,
Journal of Financial and Quantitative Analysis 44, 1375-1401.

Liang, Samuel Xin, and John K.C. Wei, 2012, Liquidity risk and stock returns around the
world, Journal of Banking and Finance 36, 3274-3288.

Lichtenstein, S., and B. Fischhoff, 1977, Do those who know more also know more about
how much they know? The calibration of probability judgment, Organization
Behavior and Human Performance 20, 159-183.

Mase, Bryan, 1999. The predictability of short-horizon stock returns, European Finance
Review 3, 161–173.

Nagar, Venky, Jordan Schoenfeld, and Laura Wellman, 2019, The effect of economic
policy uncertainty on investor information asymmetric and management disclosures,
Journal of Accounting and Economics 67, 36-57.

Nagel, S., 2012, Evaporating liquidity, Review of Financial Studies 25, 2005-2039.

Newey,W., and K.West, 1987, A Simple Positive-Definite Heteroskedasticity and

Autocorrelation Consistent Covariance Matrix, Econometrica, 55, 703-708.

Electronic copy available at: https://ssrn.com/abstract=3924771

So, Eric C., and Sean Wang, 2014, News-driven return reversals: liquidity provision
ahead of earnings announcements, Journal of Financial Economics 114, 20-35.

Subrahmanyam, Avanidhar, 2005, Distinguishing between rationales for short-horizon

predictability of stock returns, Financial Review 40, 11-35.

Wang Xinjie, Weike Xu, and Zhaodong (Ken) Zhong, 2019, Economic policy uncertainty,
CDS spreads, and CDS liquidity provision, Journal of Future Markets 39, 461-480.

Williams, Christopher D., 2015, Asymmetric responses to earnings news: A case for
ambiguity, The Accounting Review 90, 785-817.

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 1
Average returns on reversal portfolios

This table reports average returns on reversal portfolios over the period from January
1985 to June 2019. In each month t, stocks are allocated into three samples, Microcap,
Small, and Big, based on their firm size. The Microcap ample consists of stocks with
their firm size less than the NYSE 20th breakpoint, and the Big sample consists of stocks
with their firm size larger than the NYSE 50th breakpoint. The stocks with their firm size
between these two breakpoints belong to the Small sample. The Market sample consists
of stocks in the entire sample. Within each group, stocks are further divided into quintile
based on their returns in month t-1 form loser (bottom 20%) to winner (top 20%). The
return on the reversal portfolio (L ‒ W) in month t is the difference in returns between the
loser and winner portfolios in that month. Returns on these portfolios are equal-weighted.
The t-statistics are in parentheses.

Firm Size
Sample Microcap Small Big Market
Loser (L) 2.289 1.326 0.959 1.750
(7.60) (3.69) (3.06) (5.57)
2 1.448 1.182 1.093 1.312
(6.39) (4.31) (4.64) (5.67)
3 1.362 1.104 0.964 1.193
(6.88) (4.42) (4.51) (5.75)
4 1.288 0.937 0.834 1.060
(6.26) (3.75) (3.88) (5.08)
Winner (W) 1.377 0.894 0.649 1.077
(5.22) (2.83) (2.48) (4.03)
L‒W 0.912 0.432 0.310 0.673
(5.06) (0.432) (1.36) (3.49)

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 2 Economic policy uncertainty index and average market returns

In each month t, stocks are allocated into four samples, Microcap, Small, Big, and Market.
The definitions of Microcap, Small, Big, and Market samples are described in Table 1.
Each sample represents one portfolio, and returns on these portfolios are equal-weighted.
Panel A reported the contemporaneous correlation coefficients between portfolio returns
and the four indexes. Panel B displays the contemporaneous correlation coefficients
among the indexes. The p-values of these correlation coefficients are in parentheses.
Portfolio return in month t (PRt) is regressed on the portfolio return in month t-1 (PRt-1),
the natural logarithm of the Economic uncertainty index in month t-1, the Bake-Wurgler
(2007) sentiment index in month t-1 (BWSENTt-1), the Cboe volatility index in month t-1
(VIXt-1), and the Aruoba-Diebold-Scotti (2009) business conditions index in month t-1
(ADSt-1). Panel C shows the findings from the time-series regressions. The t-statistics are
adjusted using the Newey-West (1987) method with three lags and are in parentheses.

Panel A Correlation coefficients with PRt-1

Sample LnEPUt-1 BWSENTt-1 VIXt-1 ADSt-1
Market −0.120 −0.037 −0.266 0.064
(0.017) (0.459) (0.000) (0.200)
Microcap −0.104 −0.002 −0.251 0.048
(0.038) (0.972) (0.000) (0.342)
Small −0.112 −0.055 −0.247 0.058
(0.027) (0.268) (0.000) (0.247)
Big −0.133 −0.083 −0.288 0.085
(0.008) (0.097) (0.000) (0.090)

Panel B Correlation coefficients with LnEPUt-1

BWSENTt-1 VIXt-1 ADSt-1
LnEPUt-1 −0.109 (0.030) 0.357 (0.000) −0.339 (0.000)

Panel C Estimated coefficients

Sample Microcap Small Big Market
Intercept ‒9.055 ‒10.142 ‒8.040 ‒9.157
(‒2.53) (‒2.30) (‒2.17) (‒2.43)
PRt-1 0.212 0.112 0.092 0.170
(3.90) (2.03) (1.42) (2.99)
LnEPUt-1 2.020 2.194 1.790 2.013
(2.53) (2.24) (2.20) (2.42)
BWSENTt-1 0.025 ‒0.349 ‒0.482 ‒0.178
(0.08) (‒0.85) (‒1.40) (‒0.53)
VIXt-1 0.047 0.051 0.034 0.047
(1.57) (1.41) (1.06) (1.54)
ADSt-1 0.857 1.156 1.236 1.042
(1.36) (1.46) (1.68) (1.53)
Adj. R2 0.05 0.02 0.03 0.04

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 3
Economic policy uncertainty and average returns on reversal portfolios

In each month t, stocks in each sample (Micro, Small, Big, or Market) are allocated into
quintiles based on their returns in month t-1 from loser (bottom 20%) to winner (top
20%). The definitions of samples are described in Table 1. The return on the reversal
portfolio (L ‒ W) in month t is the difference in returns between the loser and winner
portfolios in that month. Returns on these portfolios are equal-weighted. This table shows
the average returns on these portfolios conditional on the economic policy uncertainty
index score in month t-1 over the period from January 1985 to June 2019. The High (Low)
uncertainty period is recorded when the economic policy uncertainty index score in
month t-1 is higher than (lower than or equal to) the median score of this index computed
from the entire sample period. The t-statistics are in parentheses.

Panel A Market
Loser (L) 2 3 4 Winner (W) L‒W
Low 0.533 0.535 0.559 0.538 0.540 ‒0.007
(1.32) (1.80) (2.08) (1.94) (1.39) (‒0.03)
High 2.909 2.042 1.789 1.537 1.566 1.343
(6.24) (5.92) (5.78) (5.03) (4.29) (4.66)
High minus Low 1.350
(3.54)
Panel B Microcap
Loser (L) 2 3 4 Winner (W) L‒W
Low 1.011 0.621 0.733 0.739 0.815 0.196
(2.66) (2.12) (2.82) (2.70) (2.14) (0.85)
High 3.506 2.220 1.953 1.790 1.886 1.620
(7.81) (6.66) (6.70) (5.95) (5.23) (5.98)
High minus Low 1.424
(4.01)
Panel C Small
Loser (L) 2 3 4 Winner (W) L‒W
Low 0.002 0.272 0.343 0.335 0.356 ‒0.354
(0.00) (0.77) (1.09) (1.00) (0.77) (‒1.19)
High 2.603 2.043 1.827 1.492 1.387 1.216
(4.92) (4.99) (4.79) (4.07) (3.22) (3.74)
High minus Low 1.570
(3.56)
Panel D Big
Loser (L) 2 3 4 Winner (W) L‒W
Low ‒0.024 0.493 0.501 0.379 0.168 ‒0.192
(‒0.06) (1.62) (1.77) (1.32) (0.45) (‒0.61)
High 1.895 1.659 1.396 1.251 1.100 0.795
(4.13) (4.67) (4.39) (3.94) (3.00) (2.46)
High minus Low 0.987
(2.18)

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 4
Economic policy uncertainty, sentiment, risks, and reversal

Stocks are allocated into four samples:- Market, Micro, Small, and Big. The definitions of these samples are described in Table 1.
Returns on reversal portfolio in month t in each group are regressed on the five Fama-French (2015) factors in month t, the natural
logarithm of the economic uncertainty index in month t-1 (LnEPUt-1), the excess market return in month t-1 (EMKTRt-1), the Bake-
Wurgler (2006) sentiment index in month t-1 (BWSENTt-1), the Cboe volatility index in month t-1 (VIXt-1), and the Aruoba-Diebold-
Scotti (2009) business conditions index in month t-1 (ADSt-1). The five Fama-French factors are the market factor (EMKTRt), the size
factor (SMBt), the book-to-market factor (HMLt), the profitability factor (RMWt), and the investment factor (CMAt). Except for the
tests related to the BWSENT and VIX indexes, the sample period is from February 1985 to June 2019. For the test related to the
BWSENT and VIX indexes, the sample period is from February 1986 to January 2019. The t-statistics are adjusted using the Newey-
West (1987) method with three lags. These t-statistics are in parentheses.

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 4 (continued)

Sample Market Microcap Small Big

Intercept 0.523 ‒7.499 ‒7.614 0.756 ‒8.415 ‒7.248 0.264 ‒9.027 ‒10.117 0.114 ‒5.838 ‒7.040
(2.10) (‒2.71) (‒2.75) (2.99) (‒3.10) (‒2.60) (0.89) (‒2.47) (‒2.83) (0.46) (‒2.01) (‒2.36)
EMKTRt 0.223 0.213 0.215 0.204 0.193 0.203 0.230 0.219 0.216 0.275 0.268 0.262
(3.35) (3.24) (3.20) (3.63) (3.53) (3.68) (3.03) (2.92) (2.75) (3.30) (3.20) (3.06)
SMBt ‒0.012 ‒0.031 ‒0.058 ‒0.019 ‒0.041 ‒0.049 ‒0.007 ‒0.029 ‒0.069 ‒0.037 ‒0.051 ‒0.090
(‒0.09) (‒0.23) (‒0.41) (‒0.17) (‒0.37) (‒0.43) (‒0.05) (‒0.20) (‒0.44) (‒0.23) (‒0.33) (‒0.55)
HMLt 0.219 0.246 0.280 0.159 0.189 0.236 0.221 0.252 0.266 0.299 0.319 0.352
(1.78) (1.96) (2.08) (1.50) (1.77) (2.06) (1.44) (1.60) (1.56) (2.02) (2.10) (2.18)
RMWt 0.106 0.099 0.065 0.124 0.116 0.081 0.170 0.161 0.134 0.115 0.109 0.078
(0.49) (0.47) (0.29) (0.62) (0.60) (0.38) (0.67) (0.65) (0.52) (0.49) (0.48) (0.33)
CMAt ‒0.371 ‒0.385 ‒0.471 ‒0.270 ‒0.285 ‒0.360 ‒0.401 ‒0.417 ‒0.494 ‒0.415 ‒0.425 ‒0.523
(‒2.38) (‒2.43) (‒2.81) (‒2.25) (‒2.32) (‒2.74) (‒2.00) (‒2.06) (‒2.33) (‒1.89) (‒1.93) (‒2.25)
LnEPUt-1 1.723 1.404 1.970 1.380 1.996 1.911 1.278 1.220
(2.94) (2.35) (3.44) (2.31) (2.58) (2.40) (2.05) (1.85)
EMKTRt-1 0.012 ‒0.061 0.070 0.071
(0.18) (‒1.01) (1.04) (0.97)
BWSENTt-1 0.713 0.609 0.697 0.701
(1.65) (1.63) (1.28) (1.43)
VIXt-1 0.079 0.080 0.072 0.071
(3.12) (3.51) (2.07) (2.39)
ADSt-1 0.291 0.124 0.379 0.454
(0.87) (0.44) (0.79) (1.11)
Adj. R2 0.09 0.11 0.14 0.07 0.10 0.15 0.07 0.09 0.11 0.09 0.10 0.12

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 5
Economic policy uncertainty, volatility, and reversal

In each month t, stocks are allocated into four samples based on their firm size, Microcap,
Small, Big, or Market. The definitions of samples are described in Table 1. Stocks are
divided independently into two groups based on their return volatility, low (bottom 50%)
and high (top 50%). Return volatility of a stock in a given month is computed as the
standard deviation of its daily stock returns in that month. Stocks in each size-volatility
sorted group are further sorted into quintiles based on their returns in month t-1 from the
loser (bottom 20%) to the winner (top 20%). Returns on these 30 size-volatility-return
sorted portfolios are equal-weighted. Return on reversal portfolio in month t in each size-
volatility sorted group are regressed on the five Fama-French (2015) factors in month t,
the natural logarithm of the economic uncertainty index in month t-1 (LnEPUt-1), the
excess market return in month t-1 (EMKTRt-1), the Bake-Wurgler (2006) sentiment index
in month t-1 (BWSENTt-1), the Cboe volatility index in month t-1 (VIXt-1), and the
Aruoba-Diebold-Scotti (2009) business conditions index in month t-1 (ADSt-1). The five
Fama-French factors are the market factor (EMKTRt), the size factor (SMBt), the book-to-
market factor (HMLt), the profitability factor (RMWt), and, the investment factor (CMAt).
The sample period is from February 1986 to January 2019. Panel A of this table reports
the estimated coefficients, and Panel B of this table shows the differences in estimated
coefficients between high-volatility and low-volatility groups. The t-statistics are adjusted
using the Newey-West (1987) method with three lags. These t-statistics are in
parentheses.

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 5 (continued)

Panel A Estimated coefficients

Sample Market Microcap Small Big
Volatility Low High Low High Low High Low High
Intercept ‒1.718 ‒10.535 ‒1.365 ‒9.816 ‒2.295 ‒13.752 ‒2.112 ‒10.484
(‒1.07) (‒3.36) (‒0.80) (‒2.97) (‒1.16) (‒3.30) (‒1.11) (‒3.13)
EMKTRt 0.138 0.221 0.139 0.230 0.172 0.182 0.171 0.299
(4.33) (3.08) (5.55) (3.33) (5.30) (2.05) (3.90) (3.64)
SMBt 0.073 ‒0.071 0.076 ‒0.120 0.060 ‒0.003 0.018 0.016
(1.28) (‒0.46) (2.05) (‒0.76) (0.99) (‒0.02) (0.30) (0.10)
HMLt 0.120 0.369 0.104 0.310 0.028 0.404 0.143 0.478
(1.68) (2.46) (1.84) (2.13) (0.34) (2.01) (1.55) (2.76)
RMWt ‒0.025 0.048 ‒0.064 0.070 0.017 0.161 0.050 0.097
(‒0.38) (0.19) (‒0.17) (0.25) (0.27) (0.56) (0.71) (0.43)
CMAt ‒0.158 ‒0.602 ‒0.157 ‒0.490 ‒0.035 ‒0.650 ‒0.087 ‒0.743
(‒1.76) (‒3.11) (‒2.19) (‒2.94) (‒0.33) (‒2.64) (‒0.77) (‒3.04)
LnEPUt-1 0.226 1.958 0.165 1.845 0.318 2.564 0.254 1.707
(0.63) (2.93) (0.44) (2.64) (0.72) (2.80) (0.59) (2.34)
EMKTRt-1 0.018 ‒0.011 ‒0.040 ‒0.072 0.009 0.060 0.036 0.035
(0.53) (‒0.16) (‒1.19) (‒1.00) (0.24) (0.82) (0.96) (0.44)
BWSENTt-1 0.388 0.873 0.065 0.993 0.635 0.704 0.521 0.540
(1.93) (1.72) (0.38) (2.00) (2.55) (1.09) (2.57) (0.89)
VIXt-1 0.067 0.103 0.062 0.109 0.074 0.091 0.065 0.108
(4.33) (3.39) (3.99) (4.10) (3.94) (2.16) (3.34) (3.28)
ADSt-1 0.282 0.391 0.006 0.070 0.256 0.691 0.560 0.668
(1.54) (1.00) (0.04) (0.20) (1.19) (1.17) (2.15) (1.45)
Adj. R2 0.17 0.18 0.19 0.17 0.16 0.10 0.13 0.12

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 5 (continued)

Panel B Differences in coefficients (High volatility minus Low volatility)

Sample Market Microcap Small Big
Intercept ‒8.817 ‒8.451 ‒11.457 ‒8.372
(‒3.38) (‒2.75) (‒3.01) (‒3.02)
EMKTRt 0.083 0.091 0.010 0.128
(1.60) (1.52) (0.13) (1.98)
SMBt ‒0.144 ‒0.196 ‒0.063 ‒0.002
(‒1.21) (‒1.36) (‒0.46) (‒0.02)
HMLt 0.249 0.206 0.376 0.335
(2.21) (1.51) (2.26) (2.59)
RMWt 0.073 0.134 0.144 0.047
(0.36) (0.52) (0.56) (0.25)
CMAt ‒0.444 ‒0.333 ‒0.615 ‒0.656
(‒3.32) (‒2.43) (‒2.99) (‒3.50)
LnEPUt-1 1.732 1.680 2.246 1.453
(3.04) (2.54) (2.64) (2.43)
EMKTRt-1 ‒0.029 ‒0.032 0.051 ‒0.001
(‒0.62) (‒0.57) (0.82) (‒0.02)
BWSENTt-1 0.485 0.928 0.069 0.019
(1.19) (2.04) (0.13) (0.04)
VIXt-1 0.036 0.047 0.017 0.043
(1.51) (1.75) (0.47) (1.55)
ADSt-1 0.109 0.064 0.435 0.108
(0.33) (0.17) (0.84) (0.28)

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 6
Cross-sectional regression analysis

Panel A reports average estimated coefficients obtained from monthly cross-sectional

regressions. The estimation period is from February 1986 to January 2019. The
dependent variable is the stock return (Rt) or risk-adjusted return (AdjRt) in month t. The
independent variables are stock return in month t-1 (Rt-1), the mean adjusted Amihud
illiquidity measure (ILLIQMAt-1), the mean adjusted average turnover ratio (TNMAt-1), the
return volatility (Rtnvolt-1), the natural logarithm of firm size (LnSZt-1), the natural
logarithm of book-to-market ratio (LnBMt-1), the past 3-month returns (Rt-2,t-4), and past 6-
month return (Rt-2,t-7) in month t-1. The risk-adjusted return of a stock is computed with
respect to the five Fama-French factors. Panel B shows the estimated coefficients from a
time-series regression. The dependent variable is the estimated coefficient of the reversal
variable (Rt-1) from the monthly cross-sectional regressions. The independent variables
are the natural logarithm of the economic uncertainty index in (LnEPUt-1), the excess
market return (EMKTRt-1), the Bake-Wurgler (2006) sentiment index (BWSENTt-1), the
Cboe volatility index (VIXt-1), and the Aruoba-Diebold-Scotti (2009) business conditions
index (ADSt-1) in month t-1. The t-statistics are adjusted using the Newey-West (1987)
method with three lags. These t-statistics are in parentheses.

38

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 6 (continued)

Panel A Estimated coefficients from monthly cross-sectional regressions

Raw returns Risk-adjusted returns
Sample Market Microcap Small Big Market Microcap Small Big
Intercept 1.565 2.336 2.188 1.235 0.811 1.767 1.507 0.401
(4.46) (4.92) (3.81) (3.22) (4.20) (3.50) (2.67) (1.50)
Rt-1 ‒0.029 ‒0.040 ‒0.022 ‒0.013 ‒0.049 ‒0.059 ‒0.041 ‒0.034
(‒6.61) (‒8.44) (‒4.22) (‒2.19) (‒8.70) (‒9.45) (‒6.87) (‒5.13)
ILLIQMAt-1 0.044 0.027 ‒0.069 ‒0.337 0.068 0.031 0.091 ‒0.384
(2.01) (1.25) (‒0.95) (‒0.79) (4.07) (1.48) (1.10) (‒0.87)
TNMAt-1 −0.015 −0.026 −0.031 0.029 0.009 ‒0.009 0.044 0.041
(−0.21) (−0.29) (−0.39) (0.36) (0.17) (‒0.12) (0.68) (0.61)
Rtnvolt-1 ‒0.038 0.003 ‒0.062 ‒0.068 ‒0.034 0.012 ‒0.096 ‒0.086
(‒0.94) (0.08) (‒1.31) (‒1.14) (‒1.26) (0.33) (‒2.69) (‒1.97)
LnSZt-1 ‒0.093 ‒0.242 ‒0.176 ‒0.053 ‒0.106 ‒0.292 ‒0.200 ‒0.044
(‒2.64) (‒2.98) (‒2.21) (‒1.47) (‒4.78) (‒3.19) (‒2.38) (‒1.69)
LnBMt-1 0.023 0.042 0.038 ‒0.045 ‒0.119 ‒0.126 ‒0.109 ‒0.126
(0.33) (0.55) (0.48) (‒0.64) (‒2.16) (‒1.73) (‒1.70) (‒2.31)
Rt-2,t-4 ‒0.006 ‒0.006 ‒0.008 ‒0.006 ‒0.006 ‒0.005 ‒0.008 ‒0.004
(‒2.42) (‒2.35) (‒2.36) (‒1.46) (‒1.84) (‒1.40) (‒2.06) (‒0.92)
Rt-2,t-7 0.004 0.005 0.006 0.004 0.002 0.003 0.004 0.002
(1.62) (1.78) (1.92) (1.40) (0.94) (1.14) (1.28) (0.63)
Adj. R2 0.06 0.04 0.06 0.09 0.03 0.03 0.04 0.06

39

Electronic copy available at: https://ssrn.com/abstract=3924771

Table 6 (continued)

Panel B Time-series regressions (dependent variable: estimated coefficient on Rt-1 in Panel A)

Raw returns Risk-adjusted returns
Group Market Microcap Small Big Market Microcap Small Big
Intercept 0.239 0.229 0.273 0.257 0.166 0.129 0.192 0.226
(3.99) (3.85) (3.73) (3.03) (2.97) (2.00) (2.95) (2.92)
LnEPUt-1 −0.047 −0.046 −0.053 −0.047 ‒0.032 ‒0.025 ‒0.037 ‒0.043
(−3.55) (−3.51) (−3.28) (−2.53) (‒2.60) (‒1.80) (‒2.52) (‒2.50)
EMKTRt-1 −0.001 0.000 −0.002 −0.002 ‒0.001 0.000 ‒0.001 ‒0.002
(−0.72) (0.01) (−1.19) (−1.05) (‒0.92) (0.05) (‒1.17) (‒1.32)
BWSENTt-1 −0.012 −0.016 −0.009 −0.011 ‒0.018 ‒0.021 ‒0.015 ‒0.015
(−2.19) (−3.21) (−1.24) (−1.18) (‒2.96) (‒3.32) (‒2.03) (‒1.59)
VIXt-1 −0.003 −0.003 −0.003 −0.003 ‒0.003 ‒0.003 ‒0.003 ‒0.003
(−4.81) (−5.01) (−3.58) (−3.87) (‒4.81) (‒4.30) (‒4.39) (‒3.81)
ADSt-1 −0.018 −0.006 −0.027 −0.032 ‒0.010 0.003 ‒0.020 ‒0.022
(−2.28) (−0.76) (−3.06) (−3.01) (‒1.09) (0.29) (‒2.25) (‒2.27)
Adj. R2 0.11 0.13 0.08 0.05 0.11 0.11 0.07 0.06

40

Electronic copy available at: https://ssrn.com/abstract=3924771