Cathy Xuying CaoDepartment of Finance, Seattle University, Seattle, Washington, USA, andChongyang ChenSchool of Business, Pacific Lutheran University, Tacoma, Washington, USA
Abstract
Purpose - This paper examines the relation between political sentiment and future stock price crash risk. Design/methodology/approach - This study employs firm-level political sentiment from earnings conference calls. The empirical analysis applies panel regressions on 40,254 US firm-year observations between 2002 and 2020, controlling for various firm-specific determinants of crash risk and firm-, industry- as well as time-fixed effects. Findings - The study identifies a negative association between both the level and the change of political sentiment and stock crash risk. Further analysis shows that the predictive power of political sentiment is independent of either non-political sentiment or political risk and remains consistently strong during periods of either high or low economic policy uncertainty. Moreover, the predictive effect of political sentiment is more pronounced for firms with high litigation risk. Research limitations/implications - The evidence highlights the important role of political sentiment in predicting stock crash risk. The results are consistent with the signaling hypothesis that managers tend to use their tone in conference calls to convey informative messages on firm outlooks. Practical implications - The study provides a recommendation on risk management: soft information such as political and non-political sentiment in earnings conference calls is useful in managing stock crash risk. The study findings also call for careful consideration of social costs, such as stock crash risk, associated with political policies. Ill-conceived policies may lead to market crashes, which can potentially outweigh the upsides of well-meaning political reforms.
Originality/value - To the authors best knowledge, this is the first study to identify the effect of time-varying firm-level political sentiment conveyed in conference calls on stock price crash.
Keywords Political sentiment, Earnings conference calls, Stock price crash risk, Signaling,
Misleading sentiment
Paper type Research paper
1. Introduction
American business leaders increasingly have been pulled into-or in some cases have dived willingly into - the hot cauldron of today's political system.
Wall Street Journal July 23, 2021.
Companies are getting more political-make statements clarifying their position on political events [1]. According to FactSet, there are a significantly higher number of S&P 500 companies discussing the presidential election during their earnings conference calls in Q3 2020 relative to Q3 2016. Market participants perceive these calls as an increasingly important source of information about the firms (e.g. Frankel et al., 1999). In addition to "hard" data presented in these calls, sentiment ("soft information") reflected in the tone of conference calls conveys incremental information that is associated with contemporaneous and future stock returns (Price et al., 2012). Given the importance of firm-level political sentiment, can it predict firm stock crash risk?
In this paper, we focus on the political sentiment reflected in earnings conference calls and examine how it is related to future stock price crash risk. According to the misleading
The authors would like to thank the Editor and two anonymous referees for valuble comments.
Political sentiment and stock crash risk
139
Received 21 November 2021
Revised 20 December 2021
Accepted 23 December 2021
The Journal of Risk Finance
Vol. 23 No. 2, 2022
pp.
(C) Emerald Publishing Limited
sentiment hypothesis, managers, concerned about their career and personal financial benefits, either inflate or deflate sentiment to affect stock prices. First, they have incentives to downplay negative political impact on firm performance (Verrecchia, 2001; Kothari et al., 2009). The overly optimistic tone in earnings conference calls misleads investors, inducing inflated stock prices (Merkel-Davies and Brennan, 2007). However, the short-term accumulation of hidden bad news exposes firms to sudden and significant drops in stock price or stock crashes (Jin and Myers, 2006). Second, managers can display overly pessimistic sentiment while expecting less negative market reaction in the future. If more negative information is revealed later, the investors are already prepared by the negative sentiment conceded by managers earlier. Therefore, stock crash risk is reduced. Alternatively, if future information is less negative, investors, realizing that the actual situation is better than expected, would react positively, resulting in lower stock crash risk. Hence, this hypothesis predicts a positive association between political sentiment in calls and future stock crash risk.
In comparison, the signaling hypothesis suggests that managers can use their tone in earnings conference calls to signal their private information (Loughran and McDonald, 2016), which affects stock crash risk. Given the interactive and impromptu nature of earnings conference calls, it is easier to detect managers' intention to mislead through the calls than through written documents or pre-rehearsed announcements (Hobson et al., 2012; Larcker and Zakolyukina, 2012). In addition, to reduce potential litigation costs associated with earnings calls, managers have incentives to reveal information about political impact on firm's business, including intensified market competition, costs of doing business and restrictions on business opportunities (Skinner, 1997; Druz et al., 2020). Embedded in the calls, the information concerning the underlying asymmetries in future cash flow shocks drives the skewness in stock returns, which measures stock crash risk (Easterwood et al, 2021). Therefore, lower sentiment in calls is related to high stock price crash risk.
Using a sample of 40,254 US firm-year observations between 2002 and 2020, we investigate the relation between firm-level political sentiment and future stock price crash risk. We employ the political sentiment data constructed by Hassan et al. (2019) that quantify sentiment using textual analysis of earnings call transcripts. The sentiment captures the percentage of positive and negative words in the conversation. Following the literature on crash risk, we measure stock crash risk using two variables: the negative skewness of firmspecific daily returns over a fiscal year (NCSKEW) and the down-to-up volatility (DUVOL).
We document a strong and negative association between the level of political sentiment and future stock price crash risk. A one standard-deviation increase in political sentiment is associated with a decrease in crash risk measured by NCSKEW (DUVOL). The results remain strong after we control for various firm-specific determinants of crash risk, and firm-, industry- as well as time-fixed effects. In addition, we find that the predictive power reflected in political sentiment is independent of either non-political sentiment or political risk. Our analysis shows that political sentiment plays a significant predictive role during periods with both high and low market-wide economic policy uncertainty.
Furthermore, we observe that change in political sentiment is also informative: a positive shift in political sentiment is associated with a decrease in future crash risk. The finding is consistent with the signaling hypothesis, but not with the misleading sentiment hypothesis. The evidence suggests that managers can use their tone to signal future stock performance.
Finally, we conduct tests on how litigation risk influences the relation between political sentiment and crash risk. By interacting political sentiment with an indicator of high litigation risk, we document that higher litigation risk is associated with a greater predictive effect of political sentiment on crash risk. The result is consistent with the argument that litigation concerns incentivize managers to timely reveal their view on a firm's outlook.
We make three contributions to the literature. First, our study belongs to growing literature that examines the link between politics and the stock market (e.g. Santa-Clara and Valkanov,
2003; Addoum and Kumar, 2016). We extend the existing literature that focuses mainly on the impact of political environment. We examine the informational role of political sentiment in predicting stock crashes. Our paper is related to the study by Addoum and Kumar (2016) that examines the effect of market-wide political sentiment, proxied by political sensitivity of industries, on stock returns through the investor-demand channel. However, the measure of aggregate political sentiment mutes the rich variation in political sentiment across firms over time and thus provides limited understanding of the relation between political sentiment and firm-specific stock performance. In addition, we focus on stock crash risk rather than firm cash flows or asset prices in prior studies. To the best of our knowledge, we are the first to focus on time-varying firm-level political sentiment and identify its predictable power in stock crashes.
Second, this paper adds to the stock crash literature by examining the predictive role of political sentiment. Prior studies on stock crash risk mostly focus on market-wide attributes (Chen et al., 2001; Huang and Wang, 2009) and firm characteristics (Jin and Myers, 2006; Hutton et al., 2009). We add to this literature by highlighting the valuable qualitative information in earnings conference calls that predict crash risk.
Third, we contribute to growing literature on market participants' sentiment and its economic consequences. Existing studies focus primarily on the sentiment from investors' perspective, either market-wide (e.g. Baker and Wurgler, 2006; Stambaugh et al., 2012) or firm-specific investor sentiment (Aboody et al, 2018; Fu et al, 2021). Fu et al. (2021) investigate the impact of firm-specific investor sentiment, a composite index constructed from principal analysis of three sentiment proxies: price earnings ratio, average turnover rate and buy-sell imbalance. Our study is different from Fu et al. (2021) in two major ways. First, our measure of sentiment is based on the tone in earnings conference calls and mainly reflects the sentiment of firm managers and analysts rather than that of retail investors. Our results show that qualitative contents or emotive tone can provide predictable crash-relevant information. Second, we restrict our attention to sentiment associated with political topics. We find a strong link between political sentiment and stock crash risk, controlling for non-political sentiment. Our results suggest that political sentiment delivers firm-specific information and affects market perception of the firm and its stock performance. Our findings call for careful consideration of social costs, such as stock crash risk, associated with political policies. Ill-conceived policies may lead to market crashes, which can potentially outweigh the upsides of well-meaning political reforms.
The rest of the paper is organized as follows. Section 2 reviews literature and develops our hypotheses. Section 3 discusses data sources. Section 4 reports the empirical analysis. Finally, section 5 concludes.
2. Literature review and hypotheses development
Political policy and its uncertainty can affect financial and product market as well as individual firms (Pastor and Veronesi, 2013; Brogaard and Detzel, 2015). A proposal of new political policy brings a great deal uncertainty to the market, mainly regarding the effective date of the policy, the magnitude of policy change, and the extent and direction of the influence. Managers' beliefs about the potential political impact can lead to corresponding changes in firm financial policy, investment policy, payout policy and firm risk-taking behavior [2]. As a result, investors try to identify the "winners" and "losers" and revise their portfolios accordingly. One important source of information that investors rely on, through firm announcement and disclosure, is managers' view on the challenges and opportunities from the new political environments.
Compared with polished written financial documents, conference calls are more impromptu and interactive. A burgeoning literature suggests that both verbal (e.g. the content of communication and sentiment) and nonverbal (e.g. facial gestures, vocal expressions and body movements) components of conference calls contain useful information about firm fundamentals and security pricing [3]. For example, Mayew and Venkatachalam (2012) analyze one important type of nonverbal communication-the managerial vocal cues-in conference calls. They find
Political
sentiment and
stock crash
risk
Political sentiment and stock crash risk
that the positive (negative) dimension of a manager's affective or emotional state conveys good (bad) news about firm future performance. Other studies examine the verbal component in conference calls. For example, Frankel et al (1999) find that the verbal contents in conference calls convey significant information that are associated with unusual large stock return volatility, large trade size and high trading volume. In addition, Matsumoto et al (2011) examine the conference call transcripts. They document that the Q&A sections of conference calls are more informative than the introductory parts of the calls and that managers provide more disclosure when firms have poor performance.
In this study, we focus on the tone in the verbal component of conference calls using textual analysis. As one of the important soft signals in verbal contents, the tone in conference calls provides incremental information and therefore is predictive of asset returns. For example, Price et al. (2012) find that conference call linguistic tone explains future abnormal returns and trading volume, especially for non-dividend paying firms. By focusing on managers' tone, Jiang et al. (2019) show that managers' sentiment in conference calls negatively predicts stock returns both at the market level and in the cross-section. Druz et al. (2020) find that increase in the negativity of managerial word choice (managerial tone) in conference calls is a strong indicator of firms' low future earnings and great uncertainty.
Managers' sentiment, defined broadly, is their subjective opinions and beliefs about future cash flows and risks that are not justified by currently available public information. We argue that political sentiment in conference calls is associated with future stock crash risk. The crash risk, a third moment of stock returns, captures the negative skewness of return distribution or the extreme negative returns. Existing theoretical work suggests an important role of return skewness in asset prices (Xu, 2007; Theodossiou and Savva, 2016). There is large literature investigating the potential determinants of stock price crash risk, including firm size, profitability, firm-specific volatility, trading turnover and corporate disclosure quality (e.g. Chen et al, 2001; Jin and Myers, 2006; Hutton et al, 2009). However, we are not aware of any prior work that studies the link between political sentiment and crash risk. Previous literature offers several channels through which political sentiment can affect stock crash risk.
2.1 Misleading sentiment hypothesis
Managers have private information regarding a firm's future cash flows and risk. However, they can engage in strategic tone management to mislead investors about firm future fundamentals due to agency incentives (Huang et al, 2014). First, firm stakeholders react negatively to potentially inferior firm prospects, questioning the managers' capabilities, which can threaten managers' career on issues such as reputation, promotion, employment opportunities, potential termination and postretirement benefits such as directorships (Verrecchia, 2001; Kothari et al, 2009). In addition, negative information can cause stock price to decline, causing monetary loss to managers since executive compensation is linked to stock and firm performance. Finally, the impact of information can be short-lived. Therefore, managers can affect stock price through their sentiment with the hope that the firm's situation changes before they are required to release hard information (Graham et al, 2005). Previous literature shows that managers manipulate the tone in disclosure to mislead and inflate investors' perception of the firm (Merkel-Davies and Brennan, 2007). In addition, Rogers et al. (2011) document that firms involved in shareholder litigation use substantially more optimistic language in their earnings announcements than do non-sued firms.
There are two ways that misleading sentiment affects stock price. First, managers can inflate sentiment to downplay the impact of negative private information concerning firm prospects. Although the misleading positive sentiment in conference calls can inflate stock prices, managers have to reveal the accumulated bad news at a certain point when it becomes too costly or difficult for managers to bury the information (Kothari et al, 2009). When a stockpile of negative messages burst into the market all at once, stock prices drop abruptly
and significantly, causing stock crashes (Jin and Myers, 2006). As a result, strategic high sentiment is associated with high crash risk.
Second, managers can display overly negative sentiment toward potentially detrimental information. There are at least two reasons why managers would utilize such a strategy. They can use the sentiment as a preemptive measure against future negative information. If more negative information is revealed, the investors are already prepared by the negative sentiment conceded by managers earlier. Therefore, the market reaction tends to be less severe. Alternatively, if future information is less negative, investors, realizing that the previous sentiment is too negative, would react positively. Consequently, negative sentiment is associated with less crash risk. This leads to our first hypothesis:
H1. There is a positive relation between political sentiment in conference calls and future firm crash risk.
2.2 Signaling hypothesis
Prior literature suggests that managers have incentives to signal their private information to the market. Doing so helps firms reduce information asymmetry, resulting in lower cost of capital (Glosten and Milgrom, 1985; Diamond and Verrecchia, 1991) and litigation risk (Skinner, 1997; Druz et al, 2020).
Managers can use tone to signal their private information concerning firm future performance and risk (Loughran and McDonald, 2016). Several empirical studies on linguistic tone and vocal cues show that sentiment during earnings conference calls induces stock market reactions such as unusual large return volatility and high trading volume (e.g. Tetlock et al, 2008; Mayew and Venkatachalam, 2012; Price et al, 2012). These studies generally interpret their findings as consistent with the notion that managers use language to signal their expectations about firm future performance. In addition, Davis et al. (2012) and Demers and Vega (2014) examine tone in earnings press releases and find that tone is related to current and future performance. Similarly, Davis et al. (2015) study manager-specific optimism during earnings conference calls and find that manager-specific tone is positively linked with future operating performance. In a recent study, Druz et al. (2020) analyze the negativity of managerial word choice (managerial tone) on conference calls and show that increase in the negativity of tone strongly predicts lower future earnings and greater uncertainty.
Furthermore, the pessimistic tone in conference calls is indicative of negative future cash flow shocks and large downside return risk. Easterwood et al. (2021) examine the information content of stock return skewness. They show that stock return skewness can be driven by information concerning the underlying asymmetries in future cash flow shocks. Therefore, information embedded in low sentiment is related to a greater degree of left skewed return distribution (i.e. high price crash risk). Taken together, we have the following hypothesis:
H2. There is a negative association between political sentiment in conference calls and future stock crash risk.
It is worth noting that managers' tone is jointly determined by economic fundamentals and managerial incentives (Huang et al., 2014). In other words, both signaling and misleading sentiment exist. We examine which plays a dominant role: the signaling or the misleading sentiment.
3. Data
3.1 Sample selection
Our sample includes publicly listed US firms for years 2002-2020. We collect financial data from Compustat and stock information from Center for Research in Security Prices (CRSP).
Political sentiment and stock crash risk
We exclude financial firms with Standard Industrial Classification (SIC) code between 6000 and 6999 because their financial statements tend to be influenced by factors unique to the financial industry [4]. The final sample contains 40,254 firm-year observations.
3.2 Variable construction
3.2.1 Stock price crash risk. We employ two measures of firm-specific stock price crash risk following Chen et al. (2001) and Callen and Fang (2013). We first estimate firm-specific daily returns for every year following the expanded market model:
where is the return on stock in day is the return on the CRSP value-weighted market index in day , and is the return on the value-weighted industry portfolio in day . The industry is defined based on the two-digit SIC code. The firm-specific daily return for stock in day is calculated as the natural logarithm of one plus the residual return from Equation 1.
Our first measure of crash risk is the negative skewness of firm-specific daily returns over the fiscal year, NCSKEW, calculated as in Equation 2:
where is firm-specific daily return, and is the number of daily returns during fiscal year . According to the definition, a higher value of NCSKEW indicates higher crash risk.
Our second measure of crash risk is the down-to-up volatility measure ( . For each firm over a fiscal-year period , we separate firm-specific daily returns into two groups: "down" trading days (i.e. days with returns below the annual mean) and "up" trading days, (i.e. days with returns above the annual mean). We calculate in the following way:
where and are the number of up and down trading days in year , respectively. A larger value of indicates greater crash risk.
3.2.2 Political sentiment. Our firm-level political sentiment measure is from Hassan et al's (2019) database, which is based on textual analysis of the transcripts of earnings calls [5]. The measure is defined as the number of political bigrams conditional on proximity to positive and negative words divided by the total number of bigrams. The treatment of positive or negative sentiment follows the sentiment dictionary of Loughran and McDonald (2011). Intuitively, frequently used positive words include "good", "strong" and "great", representing high sentiment; in contrast, the use of negative words, such as "loss", "decline", and "difficult", reflects low sentiment (Hassan et al, 2019). To have a better comparison among the sentiment across different firms, following Gad et al. (2021), we winsorize the sentiment measures at the tails and then normalize the values.
3.2.3 Control variables. We control for a list of variables that have been identified in the literature to explain stock crash risk. Following the literature on crash risk (e.g. Chen et al, 2001; Jin and Myers, 2006; Hutton et al, 2009; Callen and Fang, 2013), we include (1) stock price crash
risk over the previous fiscal year (NCSKEW ), (2) the kurtosis and the standard deviation (IDV) of firm-specific daily returns over the fiscal year, (3) firm size (AT), defined as the natural logarithm of total book assets, (4) the market-to-book ratio (MB), (5) firm leverage (LEV), calculated as total long-term debts divided by total book assets, (6) firm profitability -return on assets (ROA), (7) an indicator for industry-level litigation risk, LITIG, which is set to 1 for firms in the biotechnology, computer, electronics, and retail industries, and zero otherwise, (8) detrended stock trading volume (DTURN) and (9) opaqueness of financial reports (OPAQUE).
One concern is that the measure of political sentiment is correlated with non-political sentiment and political risk, which may confound our findings. In the robustness analyses, we control for non-political sentiment and political risk, both from Hassan et al. (2019). Nonpolitical sentiment is based on the use of positive and negative words associated with nonpolitical topics. Political risk is constructed by counting the number of occurrences of bigrams indicating discussions of a given political topic within a set of 10 words surrounding a synonym for "risk" or "uncertainty", scaled by the total number of bigrams in the transcript of earnings calls. Detailed variable definitions are in Appendix.
3.3 Summary statistics
Table 1 reports summary statistics for the variables that we use in the analysis. The average NCSKEW is positive ( 0.06 ) and DUVOL is negative ( -0.03 ), indicating that the sample firms' returns are negatively skewed on average. The median values for both NCSKEW and are lower than the means, suggesting that some firms have extremely low returns.
As expected, the normalized sentiment measures have means close to zero and standard deviations of 1 . However, the median of political sentiment is -0.06 , which indicates that, of firm-year observations exhibit a political sentiment level lower than the average sentiment, about 0.06 standard deviations below it [6]. In contrast, the median of non-political sentiment is close to zero, suggesting that a firm's sentiment toward non-political topics is rather symmetric around its long-time mean.
The summary statistics of all control variables are within reasonable ranges and in line with the statistics reported in the literature (Chen et al., 2001; Hoberg et al., 2014; Callen and Fang, 2013).
Variable
Mean
5th Pctl.
Median
95th Pctl.
SD
N
NCSKE
0.061
-2.160
-0.093
3.058
1.668
42,713
DUVOL
-0.030
-0.556
-0.049
0.576
0.341
42,713
-0.005
-1.519
-0.060
1.732
0.995
47,414
NPSentiment
0.006
-1.619
-0.014
1.671
0.994
47,414
8.624
0.754
4.805
35.451
9.375
47,042
IDV
0.025
0.009
0.022
0.054
0.015
47,042
-0.092
-0.300
-0.056
-0.009
0.105
47,042
DTURN
0.004
-0.110
0.001
0.128
0.096
45,630
LOG(MB)
0.671
-1.060
0.691
2.342
1.106
47,287
0.225
0.000
0.204
0.587
0.197
47,412
ROA
0.005
-0.369
0.036
0.191
0.139
47,319
7.032
3.920
6.963
10.429
1.952
47,414
OPAQUE
0.162
0.030
0.117
0.449
0.152
45,809
LITIG
0.278
0.000
0.000
1.000
0.448
47,414
Note(s): This table presents descriptive statistics for the sample of firms included in our study. Our sample consists of firm-year observations for non-financial firms in matched Compustat and CRSP datasets from 2003 to 2020. Following Gad et al. (2021), PSentiment and NPSentiment are first winsorized at the