Let me sleep on it: sleep and investor reactions to earnings surprises

Abstract

We explore if sleep deprivation affects how investors react to relevant news. Using the transition to Daylight Saving Time (DST) in spring as a disruption to sleeping patterns, we show that investors underreact to a firm’s earnings surprise in the days after the transition to DST. Further, an earnings surprise in the days after the transition to DST is associated with a positive drift in the post-announcement period. Our findings are consistent with sleep-deprived investors mispricing and subsequently revisiting relevant information. Overall, our results highlight the importance of investors’ cognitive ability for efficient market pricing.

KEYWORDS:

• sleep deprivation
• market reactions
• daylight saving time
• earnings surprise

JEL codes:

• G10
• G14
• G41

1. Introduction

A key assumption underlying the efficient market hypothesis is that investors have infinite cognitive processing capacity and price value-relevant signals instantaneously and accurately. One testable implication of this assumption is that if the cognitive ability of investors was somehow impaired, the efficiency of market pricing will decline.

In this study, we use sleep deprivation to measure the effects of cognitive impairment on equity returns. The sleep literature reports that the loss of as little as one hour of sleep leads to a significant deterioration in a person's cognitive abilities that remains detectable for days (see Harrison 2013; Stojanoski et al. 2019; Gibbings et al. 2021). We exploit the transition to daylight saving time (DST) as a proxy for sleep deprivation and examine the impact of DST on the stock price reactions to earnings surprises. We find that when investors are sleep-deprived, a degree of return predictability emerges. Sleep-deprived investors underreact to earnings surprises. Further, subsequent returns display a positive drift as investors revisit their initial reaction to the news when their sleeping patterns are restored.

We are not the first to investigate the effects of sleep deprivation on financial markets using DST (e.g. Siganos 2019; Mugerman, Yidov, and Wiener 2020; Kamstra, Kramer, and Levi 2000). However, our study is the first to offer a direct test of the transition to DST on the information processing capabilities of investors.¹ We do so by linking the start of DST to the pricing patterns generated by news about individual stocks. Examining the effects that a disruption to sleeping patterns has on the pricing of individual stocks allows us to identify the role of information processing by investors. That is because we can link the news on individual firms (primarily, the magnitude of the earnings surprise) to the observed pricing patterns, including post-announcement drift. By contrast, studying the effects of DST on market-wide volatility does not allow for this granular analysis and leaves open the possibility that other explanations, which simultaneously affect many stocks around the transition to DST, explain the pricing patterns.

We study a large sample of earnings announcements between 1993 and 2018 which we contrast against the consensus earnings forecasts available at the time. We compute the announcement returns experienced by firms that announce earnings surprises (the difference between announced earnings and forecast earnings). We refer to firms that announce earnings surprises in the days following the transition to DST as treatment firms. To assemble a suitable counterfactual to the returns experienced by treatment firms, we compare the returns of treatment firms to a control group that announce earnings exactly one week earlier and are propensity-matched based on company characteristics. We also assemble a second control group using synthetic matching (see Acemoglu et al. 2016) where the returns of control firms mirror those of the treatment firms before DST. Under both approaches, our identifying assumption is that, absent DST, the same type of earnings surprise would have received the same price reaction by firms in the treatment and control group.

Our setting to study how sleep deprivation affects investors’ response to earnings announcements has several advantages. First, before firms making announcements, the capital market generates expectations and speculations regarding these announcements. Earnings surprises offer an unpredictable event, with any investor expectations being different before the announcements, and it is thus more likely to capture reactions from sleepy investors who may not have anticipated the outcomes. Investor reactions to earnings surprises, which deviate from market expectations, are crucial in assessing market efficiency as per the efficient market hypothesis. Second, earnings surprises are a setting where investors perform similar investment-related tasks, while experiencing randomly assigned sleep duration (one-hour less vs. normal sleep) in our research design. Third, all firms also make relevant announcements and so firm characteristics on the relation between sleep and market reaction could be explored. Fourth, we can quantify the sensitivity in investors’ reactions to earnings surprises due to sleep loss, and their progressive reassessing of their decisions in the subsequent periods following DST change.

We find that following the transition to DST, the stock prices of treated firms underreact to earnings surprises by 35.87% relative to control firms. Further, there is a sizable post-announcements drift experienced by treated firms. In the weeks following earnings surprises, treated firms experience returns that are about twice the level of those experienced by firms in the control group. The latter is consistent with investors revisiting (and reversing) their initial underreaction to earnings surprises. Since the reversal is evident as early as 10 days post-announcement, this gives further support to our interpretation that the return patterns are due to investors revisiting the original earnings surprise as their sleep recovers (rather than the arrival of new information since the earnings announcements).

Additional analysis offers further support for our interpretation that the cognitive ability of investors affects returns. First, we do not find changes in return patterns when DST ends in the autumn. While the ‘clocks going back’ in the autumn exogenously affects mood (via reduction in the hours of daylight), it is widely documented that people do not sleep longer as a result (Barnes and Wagner 2009; Shambroom and Fabregas 2010; Harrison 2013; Sexton and Beatty 2014). This strengthens our interpretation that changes in cognitive abilities (and not mood) drive the main result.

Second, we do not observe asymmetry in the effect of sleep deprivation by a firm’s information environment (measured by the size of a firm and the number of analysts which follow it). We argue that if investors’ information processing power was indeed compromised, investors would underreact to earnings news regardless of the amount of information available on a firm.

Third, our findings align with existing literature that refers to the relevance of salient information (e.g. Mugerman, Steinberg, and Wiener 2022; Jarvenpaa 1989). We found that investors’ responses to earnings surprises depends on the salience of these surprises. In particular, the difference in reactions between sleep-deprived and non-sleep deprived investors is more pronounced when the news is non-salient compared to salient news. This suggests that when cognitive resources are limited, investors tend to focus their attention on the most prominent events. Moreover, our findings reveal that the disparities in reactions are less marked when it comes to negative news as opposed to positive news. This further strengthens our argument in favour of cognitive impairment, rather than mood, serving as the underlying mechanism behind our results.

Finally, if sleep deprivation was driving our results, we would expect its effect to be muted for stocks with a higher level of algorithmic trading, where human investors (and their sleep patterns) play less of a role. We do not find an effect of sleep deprivation on these stocks.

This study contributes to the literature in a number of ways. A growing body of behavioural finance research addresses information processing constraints experienced by investors. This includes attention (e.g. DellaVigna and Pollet 2009; Hirshleifer, Lim, and Hong Teoh 2009; Pantzalis and Ucar 2014; Giglio and Shue 2014; Fedyk 2018; Israeli, Kasznik, and Sridharan 2020), mood swings (e.g. Hirshleifer and Shumway 2003; Kaplanski and Levy 2010; Goetzmann et al. 2015; Birru 2018; Siganos 2019), and extraneous factors (e.g. the timing, prominence, or quantity of information). Our study is the first to look at the effects of the choices that investors make (sleep is mostly a choice) on their ability to process relevant financial information.

Second, our investigation provides new insights into the DST anomaly. There is conflicting evidence on whether or not DST affects market pricing. Kamstra, Kramer, and Levi (2000) and Mugerman, Yidov, and Wiener (2020) report that DST transitions are associated with lower mean returns in the days that follow. However, Pinegar (2002), Worthington (2003), Lamb, Zuber, and Gandar (2004) and Gregory-Allen, Jacobsen, and Marquering (2010) reject the existence of a DST effect. We contribute to this research by utilising an empirical setting around earnings surprises that links the information content of announcements made by individual firms to pricing patters around DST. We offer a different explanation to that in existing research which is that the effect of DST will vary by stock depending on the information released by firms when investors are sleep-deprived. Therefore, the effects of DST will not necessarily be reflected in the mean returns across entire markets that have been studied in previous work.

Third, our study contributes to the literature on the economic costs of sleep deprivation. Inadequate sleep is a substantial public health problem that affects more than one in three adults (Hillman et al. 2018). Gibson and Shrader (2018) find that a one-hour increase in the long-term average sleep time is associated with 16% wage increase. In a similar study, Costa-Font and Flèche (2017) find that one hour less sleep significantly reduces labour market performance and household income.

The remainder of the paper is organised as follows: In Section 2, we discuss why sleep deprivation can affect market reactions to earnings news. Section 3 describes our empirical strategy. Section 4 presents our OLS baseline results. Section 5 explores alternative explanations (i.e. mood and information availability) and presents additional results. Section 6 presents alternative methodologies and Section 7 concludes.

2. Institutional background: how DST affects information processing

Each spring, millions experience a disruption to their sleeping patterns as the clocks are moved forward. Janszky et al. (2012) describe the transition to DST as a large-scale natural experiment to study the effects of mild sleep deprivation and circadian rhythm disturbances. Even though the timing of DST may be anticipated, Janszky et al. (2012) argue that it can reasonably be considered a source of exogenous variation in sleep duration.

Using sleep diaries, Lahti et al. (2006a) document that the transition to DST results in a one-hour reduction in subjects’ average sleep duration and a 10% reduction in average sleep efficiency. Harrison (2013) and Sexton and Beatty (2014) find that increased sleep fragmentation and sleep latency caused by DST persist beyond a single night. Similarly, Barnes and Wagner (2009) report that on the Monday after the transition to DST, workers sleep on average 40 min less than usual. Lahti et al. (2006a) and Lahti et al. (2008) document that sleeping patterns are disrupted over several nights and only fully restored after five nights.

There is accumulating evidence that sleep deprivation has detrimental effects on cognitive ability, decision-making, and physical well-being. The neurology literature shows that a reduction in sleep duration causes symptoms similar to alcohol intoxication (Dawson and Reid 1997; Williamson and Feyer 2000) and jet lag (Lahti et al. 2006b). The reduction in average sleep duration dampens the cognitive accuracy and speed (Dawson and Reid 1997; Williamson and Feyer 2000; Whitney, Hinson, and Nusbaum 2019). People who are sleep-deprived find it hard to cope with challenging tasks, maintain concentration levels, and adjust to new information (e.g. Jennings, Monk, and van der Molen 2003; Kreutzmann et al. 2015; Whitney, Hinson, and Nusbaum 2019). Consistent with this, evidence shows that the transition to DST is associated with more workplace injuries (Barnes and Wagner 2009) and higher traffic accidents (Lahti et al. 2010; Smith 2016; Robb and Barnes 2018).

Importantly, as little as a one-hour reduction in sleep has a significant impact on peoples’ ability to process information. Recent studies find that mild, small amounts of sleep loss (e.g. only a couple of hours, for only a single night) reduce processing capacity for decision-making (Gibbings et al. 2021), behavioural preparedness and responses (Stojanoski et al. 2019).

In the finance context, the process of incorporating earnings news about a firm, digesting its implications on future profits, and make trading decisions is likely to be highly cognitive intensive. Sleep deprivation might restrict the allocation of limited cognitive resources. Kamstra, Kramer, and Levi (2000) find the average stock market return on Monday after the time change is economically and statistically significantly lower than the typical Monday return and suggest that the desynchronosis may result in the difficulty of reaching rational decisions during the first trading session following the DST change. Investors may prefer safer investments and avoid risk in trading on days when their sleep patterns are disturbed. Kamstra, Kramer, and Levi (2002) show that the daylight-saving effect remains intact in the international financial markets. Siganos (2019), on the other hand, documents that investors of target firms tend to overreact to the announcement of M&A deals, using the event of transition to DST as exogenous disruptions to investors’ sleep pattern.

In our setting, investors suffering from sleep deprivation may choose to avoid making difficult investment decisions such as trading on earnings announcements. This might amount to an ‘limited attention' effect that has been shown to lead to announcement underreaction to earnings news and post-announcement drifts (e.g. Hirshleifer, Lim, and Hong Teoh 2009; DellaVigna and Pollet 2009).

3. Empirical strategy

3.1 Data

We use earnings announcements data and consensus earnings forecasts data from I/B/E/S and return data from CRSP between 1993 and 2018. Earnings, forecasts, and stock prices are all split-adjusted. We follow Hirshleifer, Lim, and Hong Teoh (2009) in terms of the covariates that are associated with investor reactions to earnings announcements, including size, book-to-market, number of analysts following, reporting lag, earnings persistence, earnings volatility, and share turnover.

Size is the market value of equity at the end of June, ranked into deciles. B/M ratios are calculated as the book value of equity for the last fiscal year-end in the previous calendar year divided by the market value of equity for December of the previous calendar year, ranked into deciles and scaled to range between −0.5 and 0.5 (Drake, Gee, and Thornock 2016). no._analysts is the log (1 + number of analysts following the firm during the most recent fiscal year). r_lag is the log of the number of days from the quarter-end until the earnings announcement date. e_vol is the standard deviation during the preceding 4 years of the deviations of quarterly earnings from 1-year-ago earnings, ranked into deciles and scaled to range between −0.5 and 0.5 (split-adjusted; minimum four observations required) (Cao and Narayanamoorthy 2011). s_turn is defined as the average monthly share trading volume divided by the average number of shares outstanding during a 1-year period ending at the end of the corresponding fiscal quarter. e_pers is the first-order autocorrelation coefficient of quarterly earnings per share during the past 4 years (split-adjusted; minimum four observations required). We include industry dummies using Fama-French 10 industry classification, and report White (1980) standard errors in the main regression.

Our treatment group consists of earnings announcements from Monday to Thursday after the transition to DST between 1993 and 2018 (excluding 1994 when there are no observations in the sample). We include Monday to Thursday because the sleep literature indicates that sleeping patterns are disrupted over several nights and only fully restored after five nights (Lahti et al. 2006 and Lahti et al. 2008). Therefore, the announcement period event window [0,1] that we study lies within the range that DST influences investors. This yields a treatment group contains 807 earnings surprises. Table 1 presents the number of earnings surprises by year.

Table 1. Number of earnings surprises by year. This table presents the annual breakdown of earnings surprises. The surprises are built by contrasting earnings announcements against the consensus earnings forecasts from I/B/E/S for the period from 1993 to 2018 (excluding 1994 when there are no observations).

Year	#Surprises	Year	#Surprises	Year	#Surprises
1993	6	2003	19	2012	136
1995	13	2004	17	2013	218
1996	8	2005	13	2014	197
1997	16	2006	25	2015	171
1998	11	2007	74	2016	169
1999	7	2008	96	2017	270
2000	12	2009	181	2018	313
2001	10	2010	107
2002	9	2011	134	Total	2,232

To build a group containing counterfactuals, that captures returns in response to earnings surprises outside the transition to DST, we match treatment and control earnings announcements using the weekday of the announcements. Since all the stocks are simultaneously affected by the transition to DST, we use the earnings announcements on the same day one week before and match using propensity scores based on size, B/M, no._analysts, e_pers, e_vol, and s_turn. We trim the sample to common support based on propensity scores. The control group contains 1,425 earnings surprises. Table 2 presents the summary statistics of our sample. There are 515 distinct firms in our sample.

Table 2. Summary statistics. This table presents the summary statistics of our sample. Treatment firms are firms that announce earnings surprises in the days following the transition to DST. Control firms are firms that announce earnings on the same days in the week before the transition to DST week. Please refer to the Appendix 1 for definitions of variables.

	Mean	Median	S.D.	Min	Max
	Panel A: Treated firm (N = 807)
CAR[0,1]	−0.0138	−0.0118	0.0994	−0.527	0.487
CAR[2,5]	0.0076	0.0034	0.0937	−0.282	0.84
CAR[2,10]	0.0097	0.0069	0.126	−0.569	0.958
CAR[2,61]	0.0233	0.0251	0.382	−1.8	1.716
CAR[2,90]	0.0245	0.0209	0.51	−2.385	2.442
ES (unranked)	−0.272	0.0002	4.192	−80.41	4.1
size (unranked)	1939	246.7	9267	0.259	106413
B/M (unranked)	4.179	0.161	52.61	−10.37	993.8
no._analysts (raw)	5.242	4	5.785	0	40
r_lag (raw)	66.38	72	19.29	2	77
e_pers	0.215	0.136	0.442	−0.706	3.866
e_vol (unranked)	1797	0.272	28074	0.0053	551638
s_turn	0.248	0.116	1.397	0	27.29
	Panel B: Control firm (N = 1425)
CAR[0,1]	0.0007	−0.0021	0.0987	−0.452	0.598
CAR[2,5]	0.0072	0.0031	0.0902	−0.378	1.212
CAR[2,10]	0.0079	0.001	0.106	−0.638	0.964
CAR[2,61]	0.0499	0.0103	0.443	−1.583	7.628
CAR[2,90]	0.0675	0.0215	0.519	−2.514	7.628
ES (unranked)	0.0213	0.0003	0.767	−1.815	24.7
size (unranked)	2656	745.6	7285	2.638	87882
B/M (unranked)	1.009	0.111	20.58	−49.04	543.3
no._analysts (raw)	7.726	6	6.78	0	49
r_lag (raw)	60.13	59	6.284	54	88
e_pers	0.211	0.151	0.584	−5.807	9.005
e_vol (unranked)	38.16	0.26	565.8	0.0036	10058
s_turn	0.202	0.138	0.299	0	6.309
	Panel C: Full sample (N = 2232)
DST	0.3616	0	0.4806	0	1
CAR[0,1]	−0.0046	−0.0052	0.0992	−0.5273	0.5976
CAR[2,5]	0.0073	0.0032	0.0915	−0.378	1.2122
CAR[2,10]	0.0085	0.0018	0.1134	−0.6384	0.9636
CAR[2,61]	0.0403	0.0122	0.422	−1.8002	7.6276
CAR[2,90]	0.052	0.0213	0.5163	−2.514	7.6276
ES (unranked)	−0.0848	0.0003	2.5968	−80.4116	24.6975
size (unranked)	2396.7663	512.439	8063.4148	0.25871	6412.99
B/M (unranked)	2.1551	0.1252	35.6728	−49.0373	993.7962
no._analysts (raw)	6.828	5	6.5465	0	49
r_lag (raw)	62.3875	60	12.9856	2	88
e_pers	0.2124	0.1462	0.5371	−5.807	9.0055
e_vol (unranked)	674.0112	0.2652	16901.1487	0.00365	51638.3814
s_turn	0.2189	0.1319	0.8735	0	27.2852

Table 3 presents a comparison of the characteristics of treatment and control firms. We note that there are three covariates with significant mean differences. We take additional steps to rule out differences between treatment and control groups that cannot be captured by our OLS analysis as an explanation for our findings. We apply the synthetic control method (SCM) in Section 6 and find that our results hold.

Table 3. Characteristics of treatment and control firms. This table compares the characteristics of treatment firms and control firms. Treatment firms are firms that announce earnings surprises in the days following the transition to DST. Control firms are firms that announce earnings on the same days in the week before the transition to DST week. The mean differences and the t-statistic are listed.

	Mean Treated firms	Mean Control firms	Difference in Means	t-statistic
size	4.21	6.23	−2.02***	−16.919
B/M	0.07	−0.04	0.11***	7.687
no._analysts	5.25	7.73	−2.49***	−8.760
r_lag	4.08	4.11	−0.02	−1.355
e_vol	0.01	−0.01	0.02	1.381
s_turn	0.25	0.20	0.05	1.194
e_pers	0.21	0.21	0.00	0.186
#firms	807	1,425

3.2 Hypotheses and methods

If asset pricing patterns are shaped by sleep deprivation following the transition to DST, we expect to find two trends in our data. First, sleep-deprived investors may underreact to earnings surprises as they struggle to assess earnings news in the period immediately following DST. Second, as investors recover, eventually, they will become aware of the information they neglected and trade accordingly. There might be stronger delayed response (larger drift) that reverses the initial underreaction. Therefore, the presence of sleep deprivation is likely to give rise to a weaker immediate reaction to the earnings surprise and a stronger post-earnings announcement drift.

Thus, our hypotheses are:

Hypothesis 1: Announcement price reactions to earnings surprises following the transition to DST are lower than those of matched earnings surprises.

Hypothesis 2: The post-announcement price reactions to earnings surprises following the transition to DST are higher than those of matched earnings surprises.

Following Hirshleifer et al. (2009), we measure earnings surprises using forecast error (FE), i.e. the difference between announced earnings (e_iq) and the consensus earnings forecast (Fiq) which is the median of recent forecasts by analysts (both from I/B/E/S). The difference between the consensus forecast and the announced earnings is normalised by the stock price at the end of the quarter when there is an earnings announcement (P_iq): (1) $F{E}_{\mathrm{iq}}=\frac{e_{\mathrm{iq}}-F_{\mathrm{iq}}}{P_{\mathrm{iq}}}$(1) The announcement abnormal returns (CAR[0,1]) capture the immediate response of investors to earnings announcements made within 5 days of the transition to DST. The post-earnings announcement abnormal returns (CAR[2,X]) capture the post-earnings announcement drift. Since we expect the underreaction to correct gradually as investors not only receive more information but also restore their sleeping deficiencies, X takes the value of 5, 10, 61, and 90.

Abnormal returns are calculated using the market model with an estimation window of 100 trading days ending 30 days prior to the event day 0 (the day when earnings are announced). Following Hirshleifer et al. (2009), we run regressions of CAR on the earnings surprise decile rank (ES).² To test the effect of sleep deprivation on market reactions to earnings surprises, we run the following model: (2) $\mathrm{CAR}={\alpha }_0+{\alpha }_1\mathrm{ES}+{\alpha }_2\mathrm{DST}+{\alpha }_3\mathrm{ES}\ast \mathrm{DST}+\sum _{i=1}^n{b}_i{X}_i+ϵ$(2) DST is a dummy variable that is equal to 1 if the earnings announcement belongs to the treatment group, otherwise 0. X_i is a set of firm-level covariates for firm i. According to our hypothesis, we expect α₃ to be negative when using CAR[0,1] as the dependent variable and α₃ to be positive when using post-earnings announcement abnormal returns CAR[2,X] as the dependent variable.

4. Baseline results

Table 4 presents OLS estimates of the effect of the transition to DST on cumulative abnormal returns (CAR) surrounding the earnings announcements. Abnormal returns are calculated using the market model with an estimation window of 100 trading days ending 30 days prior to event day 0, which is the earnings announcement date in the treatment group.

In column 1 of Table 4, for the announcement return (CAR[0,1]), the coefficient on the interaction term (ES x DST) is negative (−0.0033) and statistically significant at 5% level. This estimate indicates that there is a reduction of 33 basis points in CAR[0,1] when earnings are announced immediately after the transition to DST. The market reactions are significantly less sensitive to earnings surprises by 35.87% on the days following the transition to DST, compared to the control group. (The sensitivity is 0.0092–0.0033 = 0.0059 for treatment group, while for the control group the market reaction to earnings surprises is 0.0092.) The coefficient on ES (92 basis points) is positive and significant at the 1% level, showing that at the time of the announcement, there is a reaction to the earnings surprise, which is consistent with the evidence reported in Hirshleifer, Lim, and Hong Teoh (2009).

Table 4. Market reaction to earnings surprises (Spring). This table reports the relation between transition to daylight saving time and the CAR to earnings surprises over different event windows relative to the earnings announcement date. CAR is the cumulative abnormal return. ES is the decile of the difference between the announced earnings and the consensus forecast normalised by the stock price at the end of the corresponding quarter. DST is a dummy variable equal to 1 if the earnings was released from Monday to Thursday following the transition to daylight saving time, otherwise 0. size is the market value of equity at the end of June, ranked into deciles. B/M ratios are calculated as the book value of equity for the last fiscal year-end in the previous calendar year divided by the market value of equity for December of the previous calendar year, ranked into deciles and scaled to range between −0.5 and 0.5. no._analysts is the log (1 + number of analysts following the firm during the most recent fiscal year). r_lag denotes the log of reporting lag that is the number of days from the quarter-end until the earnings announcement date. e_vol denotes earnings volatility that is the standard deviation during the preceding 4 years of the deviations of quarterly earnings from 1-year-ago earnings, ranked into deciles and scaled to range between −0.5 and 0.5 (split-adjusted; minimum four observations required). s_turn denotes share turnover that is defined as the average monthly share trading volume divided by the average number of shares outstanding during a 1-year period ending at the end of the corresponding fiscal quarter. e_pers denotes earnings persistence that is the first-order autocorrelation coefficient of quarterly earnings per share during the past 4 years (split-adjusted; minimum four observations required). We include industry dummies using Fama-French 10 industry classification. Standard errors are in parentheses. *, **, and *** denote statistical significance at 10%, 5%, and 1% levels, respectively.

	Announcement return	Post-earnings announcement return
	1	2	3	4	5
	CAR [0,1]	CAR [2,5]	CAR [2,10]	CAR [2,61]	CAR [2,90]
ES	0.0092***	−0.0004	−0.0018	−0.0195***	−0.0252***
	(0.0011)	(0.0012)	(0.0013)	(0.0065)	(0.0073)
DST	0.0091	−0.0093	−0.0184	−0.1499***	−0.2273***
	(0.0096)	(0.0105)	(0.0130)	(0.0540)	(0.0645)
ES x DST	−0.0033**	0.0013	0.0033*	0.0185**	0.0266***
	(0.0016)	(0.0017)	(0.0020)	(0.0083)	(0.0100)
size	0.0012	−0.0004	−0.0007	−0.0198***	−0.0279***
	(0.0010)	(0.0014)	(0.0014)	(0.0044)	(0.0055)
B/M	−0.0026	0.0021	0.0051	0.0477	0.0703
	(0.0077)	(0.0089)	(0.0106)	(0.0370)	(0.0473)
no._analysts	0.0051	0.0054	0.0089	0.0725**	0.0901**
	(0.0077)	(0.0077)	(0.0085)	(0.0291)	(0.0350)
r_lag	−0.0038	−0.0063	−0.0011	−0.0528***	−0.0756***
	(0.0048)	(0.0058)	(0.0066)	(0.0160)	(0.0203)
e_vol	−0.0025***	0.0012*	0.0023**	−0.0008	0.0007
	(0.0008)	(0.0007)	(0.0009)	(0.0041)	(0.0049)
s_turn	−0.0014	−0.0077***	−0.0073*	0.0212***	0.0387***
	(0.0012)	(0.0017)	(0.0041)	(0.0047)	(0.0082)
e_pers	−0.0024	0.0106	0.0122*	0.0120	0.0031
	(0.0042)	(0.0075)	(0.0064)	(0.0166)	(0.0187)
Industry	Yes	Yes	Yes	Yes	Yes
Intercept	−0.0153	0.0293	0.0161	0.4556***	0.6203***
	(0.0245)	(0.0283)	(0.0327)	(0.0943)	(0.1153)
Obs	2232	2232	2232	2232	2232
R-sq	0.0636	0.0160	0.0081	0.0255	0.0332

Table 5. Market reaction to earnings surprises (Autumn). This table reports the relation between the end of daylight saving time and the CAR to earnings surprises over different event windows relative to the earnings announcement date. CAR is the cumulative abnormal return. ES is the decile of the difference between the announced earnings and the consensus forecast normalised by the stock price at the end of the corresponding quarter. EDST is a dummy variable that is equal to 1 if the earnings announcement is released from Monday to Thursday following the end of daylight saving time, otherwise 0. ES x EDST is the interaction between ES and EDST. size is the market value of equity at the end of June, ranked into deciles. B/M ratios are calculated as the book value of equity for the last fiscal year-end in the previous calendar year divided by the market value of equity for December of the previous calendar year, ranked into deciles and scaled to range between −0.5 and 0.5. no._analysts is the log (1 + number of analysts following the firm during the most recent fiscal year). r_lag denotes the log of reporting lag that is the number of days from the quarter-end until the earnings announcement date. e_vol denotes earnings volatility that is the standard deviation during the preceding 4 years of the deviations of quarterly earnings from 1-year-ago earnings, ranked into deciles and scaled to range between −0.5 and 0.5 (split-adjusted; minimum four observations required). s_turn denotes share turnover that is defined as the average monthly share trading volume divided by the average number of shares outstanding during a 1-year period ending at the end of the corresponding fiscal quarter. e_pers denotes earnings persistence that is the first-order autocorrelation coefficient of quarterly earnings per share during the past 4 years (split-adjusted; minimum four observations required). We include industry dummies using Fama-French 10 industry classification. Standard errors are in parentheses. *, **, and *** denote statistical significance at 10%, 5%, and 1% levels, respectively.

	Announcement return	Post-earnings announcement return
	1	2	3	4	5
	CAR [0,1]	CAR [2,5]	CAR [2,10]	CAR [2,61]	CAR [2,90]
ES	0.0091***	0.0009*	0.0005	−0.0026	−0.0060**
	(0.0006)	(0.0005)	(0.0007)	(0.002)	(0.0027)
EDST	0.0164**	−0.0073	−0.0171**	−0.0092	−0.0394
	(0.0071)	(0.0055)	(0.0080)	(0.0223)	(0.0299)
ES x EDST	−0.0013	0.0008	0.0022**	0.0032	0.0050
	(0.0008)	(0.0006)	(0.0009)	(0.0026)	(0.0035)
Controls	Yes	Yes	Yes	Yes	Yes
Industry	Yes	Yes	Yes	Yes	Yes
Intercept	−0.0392***	−0.0226**	−0.0186	0.1080***	0.125**
	(0.0122)	(0.0095)	(0.0138)	(0.0384)	(0.0516)
Obs	4980	4980	4980	4980	4980
R-sq	0.0818	0.0091	0.0114	0.0054	0.0079

Post-announcement returns are summarised in columns 2, 3, 4 and 5. We concentrate on narrow post-earnings announcement windows, i.e. CAR[2,5] and CAR[2,10], because we want to make sure that post-earnings announcement returns are related to sleep recovery rather than to other factors such as the emergence of other news.³ We still report, however, results for wider windows for completeness. We find that in CAR[2,5], the coefficients on ES x DST are positive and insignificant. Further, in column 3, the coefficient on ES x DST is 0.0033 and significant at the 10% level. This indicates that the 9-trading day post-earnings announcement CAR is 33 basis points higher compared to the control group. Our interpretation is that sleep deprived investors make corrections as they recover and therefore, we observe a higher post-earnings announcement CAR compared to the control group. This is consistent with our sleep deprivation hypothesis.

The coefficients on our control variables are largely consistent with prior literature. e_vol is positively associated with market reactions to earnings announcements in the post-announcement period, which is consistent with Cao and Narayanamoorthy (2011). In line with Atiase (1985), size is negatively associated with market reactions to earnings announcements, since market's reaction to an earnings announcement is inversely associated with the amount of pre-disclosure information that is priced before the earnings signal is released (Holthausen and Verrecchia 1988; and Schroeder 1995). Consistent with Kross and Schroeder (1984), r_lag is negatively associated with firm abnormal returns, as delayed earnings announcements more often convey bad news (i.e. lower than the expected earnings).

The fact that investors reverse their initial reaction before a significant amount of new information is likely to have been released, e.g. CAR[2,10], gives further support to our interpretation that after investors regain their sleep patterns, they revisit the information contained in earnings announcements. This finding is consistent with DellaVigna and Pollet (2009) that investors affected by Fridays reassess their decisions in subsequent periods, the information should eventually be incorporated in stock prices. This also adds to the findings by Hirshleifer, Lim, and Hong Teoh (2009) that the market has a weaker immediate reaction to the earnings surprise due to more competing news and a stronger post-earnings announcement drift in the subsequent periods.

Our empirical findings demonstrate the robustness of the results to cluster standard errors by firm or by firm’s announce time (see Appendix 3 and Appendix 4), and we further verify our main results by re-estimating them using the full sample of firms without utilising any matching technique (see Appendix 5) and using only Monday earnings surprises (Appendix 6).

Overall, consistent with our hypothesis, the transition to DST has the opposite effects on the sensitivity of the announcement period reaction to earnings surprises (the effect is negative and statistically significant) versus the sensitivity of the post-earnings announcement reaction to earnings surprises (the effect is positive and statistically significant).

5. Alternative explanations

5.1 Mood

One could argue that, because sleep deprivation leads to both impaired cognitive ability and potential changes in mood, our findings may result from swings in investor mood rather than changes in cognitive abilities.

The end of DST in the autumn presents us with a test to disentangle cognitive and mood explanations. Sleep research shows that the ending of DST does not cause individuals to sleep significantly longer that night (Barnes and Wagner 2009; Shambroom and Fabregas 2010; Harrison 2013; Sexton and Beatty 2014). However, the end of DST is an exogenous shock to the hours of daylight on the following days. Evidence suggests that the circadian disruption resulting from shifts in light alter the functions of brain regions involved in emotion and mood regulation (see, for example, Bedrosian and Nelson 2017). Since changes in ambient light are more likely to affect mood than cognitive ability, we use the end of DST in the autumn to disentangle one effect from the other. If swings in mood rather than compromised cognitive abilities were driving our results, we should find similar results around the end of DST and the transition to DST.

Table 5 reports the results using a control group that consists of earnings announcements from Monday to Thursday in the week right before the end of DST week. We create a dummy variable, EDST, which takes the value of 1 if the earnings announcement is released from Monday to Thursday following the end of daylight saving time, otherwise 0. In column 1, the coefficient on ES x EDST is statistically insignificant. In columns 4 and 5, the coefficients on ES x EDST are also statistically insignificant. In other words, we do not find a significant effect following the end of daylight saving time on abnormal returns. This gives further support to our interpretation that the lower announcement day reaction and the higher post-earnings announcement reaction in Table 4 are driven by sleep deprivation, instead of mood.

5.2 Information availability

If our findings were due to impairment of cognitive ability, the results should be the same irrespective of the information environment. This is because the availability of information will not affect the inability of investors to process information when sleep deprived. Hirshleifer, Lim, and Hong Teoh (2009) show that extraneous events lead to an underreaction to earning news in line with an inattention hypothesis. Size and analysts following proxy for the amount of information available about a firm (e.g. Grant 1980; Atiase 1985; Collins, Kothari, and Rayburn 1987; Freeman 1987; Bhushan 1989; Frankel and Li 2004) as more information is available on large firms and firms with more analyst coverage. In the event that investors seek to reassess their decisions following earnings surprises, it is notable that information tends to be more accessible for larger companies and those with comprehensive analyst coverage. Nevertheless, a sleep-deprived investor may find it challenging to harness all readily available information, even if it is easily accessible.

We split the sample into two sub-groups according to the size of the firm (firm size above the sample median and below the sample median) and re-estimate our baseline regression. The coefficients on ESxDST in column 1 (−0.0042, large firm) and column 2 (−0.0057, small firm) of Table 6 are both negative and significant, showing that sleep-deprived investors underreact to earnings news irrespective of the size of the firm.⁴

Table 6. Additional analysis – market reaction across different firm size and analysts following. This table reports the relation between transition to daylight saving time and the announcement period CAR to earnings surprises in large firm and small firm subsamples (firm size above sample median or below sample median), and in more analysts and less analysts following subsamples (number of analysts following above sample median or below sample median). The dependent variable is CAR[0,1]. ES denotes the deciles of earnings surprise. DST is a dummy variable equal to 1 if the earnings was released from Monday to Thursday following the transition to daylight saving time, otherwise 0. We include industry dummies using Fama-French 10 industry classification. We include size, B/M, no._analysts, r_lag, e_vol, s_turn and e_pers as control variables. Standard errors clustered by DST are in parentheses. *, **, and *** denote statistical significance at 10%, 5%, and 1% levels, respectively.

	Large firm	Small firm	More analyst following	Less analyst following
	1	2	3	4
	CAR [0,1]	CAR [0,1]	CAR [0,1]	CAR [0,1]
ES	0.0070***	0.0129***	0.0086***	0.0081***
	(0.0003)	(0.0003)	(0.0003)	(0.0004)
DST	0.0091***	0.0199***	−0.0036*	0.0086***
	(0.0009)	(0.0017)	(0.0019)	(0.0005)
ESxDST	−0.0042***	−0.0057***	−0.0024***	−0.0021***
	(0.0005)	(0.0010)	(0.0008)	(0.0006)
size	0.0002	−0.0014***	0.0032***	−0.0001
	(0.0011)	(0.0004)	(0.0004)	(0.0016)
B/M	0.0203**	−0.0177**	0.0078***	−0.0132*
	(0.0101)	(0.0098)	(0.0012)	(0.0069)
no._analysts	0.0139	−0.0045	−0.0504***	0.0362
	(0.0092)	(0.0035)	(0.0046)	(0.0213)
r_lag	−0.0098***	0.0016	−0.0050***	−0.0062***
	(0.0003)	(0.0029)	(0.0009)	(0.0005)
e_vol	−0.0034***	−0.0016	−0.0005	−0.0052***
	(0.0012)	(0.0013)	(0.0008)	(0.0002)
s_turn	−0.0562***	−0.0005	−0.0022	0.0159
	(0.0172)	(0.0023)	(0.0028)	(0.0106)
e_pers	−0.0022	−0.0040**	−0.0126***	0.0013***
	(0.0110)	(0.0013)	(0.0001)	(0.0003)
Industry	Yes	Yes	Yes	Yes
Intercept	0.0342***	−0.0065	0.0313***	−0.0229
	(0.0057)	(0.0825)	(0.0005)	(0.0325)
Obs	1117	1115	1228	1004
R-sq	0.0620	0.0719	0.0600	0.0937

Standard errors below coefficients. Significance stars denote: * 0.1 ** 0.05 *** 0.01.

We also split the sample into two sub-groups according to the number of analysts following the firm (number of analysts above the sample median and below the sample median). Then we re-estimate the baseline regression using the two sub-samples. The coefficients on ESxDST in column 3 (−0.0024, more analysts) and column 4 (−0.0021, less analysts) of Table 6 are both negative and significant, implying that sleep-deprived investors underreact to earnings news irrespective of the number of analysts following the firm.⁵

Put together, our findings imply that sleep-deprived investors do not seek/assimilate all supplementary information following earnings surprises, aligning with the hypothesis that their limited cognitive resources hinder their ability to mitigate mispricing during the earnings announcement.

5.3 Information salience

In today’s information-saturated era, characterised by an immense data influx, the concept of salience takes on growing significance. Studies have shown that investors have a propensity to prioritise salient information (e.g. Mugerman, Steinberg, and Wiener 2022; Jarvenpaa 1989). It is conceivable that even though sleep-deprived investors may not process the entirety of available information to immediately adjust to earnings surprises, as demonstrated in Section 5.2, they still possess the capability to discern news items that hold greater prominence and focus their attention on those specific aspects. Consequently, we investigate the impact of salience by examining whether investors react differently to more salient earnings surprises compared to less prominent surprises.

Since positive earnings surprises may be more/less salient than negative surprises, we divide the sample into 4 groups. The first group consists of positive earnings surprises falling within the top quartile, while the second group encompasses the positive earnings surprises in the remaining quartiles. Analogously, the third group consists of negative earnings surprises falling within the top quartile, while the fourth group encompasses the negative surprises in the remaining quartiles. Subsequently, we re-estimate our baseline regression. ES x DST is statistically significant across specifications in columns 1–4 in Table 7. Although the coefficients for ES x DST are mainly positive, which does not support Hypothesis 1, we observe that these differences in reactions tend to be less pronounced when the news is highly noticeable or salient. The coefficient on ES x DST (0.0563) in Table 7, column 1 is significantly smaller than the coefficient on ES x DST (0.4497) in column 2; and the coefficient on ES x DST (−0.0143) in Table 7, column 3, is significantly smaller than that in column 4 (0.0959).⁶ This observation aligns with the idea that the impact of sleep deprivation is mitigated when investors allocate their limited cognitive resources to respond to the most attention-grabbing news.

Table 7. Additional analysis – market reaction across different levels of news salience. This table reports the relation between transition to daylight saving time and the announcement period CAR to earnings surprises in positive salient (first quartile of positive surprises), positive non-salient (the remaining quartiles of positive surprises), negative salient (first quartile of negative surprises), and negative non-salient (the remaining quartiles of negative surprises) earnings surprise subsamples. The dependent variable is CAR[0,1]. We include size, B/M, no._analysts, r_lag, e_vol, s_turn and e_pers as control variables. We include industry dummies using Fama-French 10 industry classification. Standard errors clustered by DST are in parentheses. *, **, and *** denote statistical significance at 10%, 5%, and 1% levels, respectively.

	Positive salient	Positive non-salient	Negative salient	Negative non-salient
	1	2	3	4
	CAR [0,1]	CAR [0,1]	CAR [0,1]	CAR [0,1]
ES	−0.0335***	0.1402**	0.0175***	0.0685***
	(0.0018)	(0.0553)	(0.0029)	(0.0055)
DST	−0.0846***	−0.0243***	−0.0290***	−0.0002
	(0.0069)	(0.0034)	(0.0013)	(0.0004)
ES x DST	0.0563***	0.4497***	−0.0143***	0.0959*
	(0.0010)	(0.0352)	(0.0002)	(0.0545)
Controls	Yes	Yes	Yes	Yes
Industry	Yes	Yes	Yes	Yes
Intercept	0.1221***	0.0526***	−0.3305***	0.0303***
	(0.0282)	(0.0059)	(0.0151)	(0.0052)
Obs	302	904	257	769
R-sq	0.1515	0.0253	0.1945	0.0106

Additionally, the nature of the news, whether positive or negative, also plays a significant role in influencing investor behaviour. Notably, we found more substantial differences in reactions to positive news than negative news. Based on the results in Table 7, we find that the sleep-deprived investors’ difference in reactions to positive surprises (positive salient – non-salient = −0.5672***) is significantly larger than that to negative surprises (negative salient – non-salient = −0.1612***).⁷ This suggests that the effects we observed are not primarily driven by a bad mood resulting from sleep loss, as previous literature has shown a greater sensitivity to bad news when mood is negatively affected (see, Isen et al. 1978 and Forgas and Bower 1988 for psychological evidence and Hirshleifer and Shumway 2003 for a financial application).

5.4 Algorithmic trading

Due to the proliferation of algorithmic trading, where human processing plays a diminished role, the underreaction to earnings news after the transition to DST should be muted for stocks with higher levels of algorithmic trading. Following Rosu et al. (2020), we use QT ratio as a measure of algorithmic trading. QT is the monthly ratio of the number of quotes to the number of trades. QT is associated with the level of algorithmic trading (e.g. Hendershott, Jones, and Menkveld 2011; Boehmer, Fong, and Wu 2020) and high-frequency trading (e.g. Conrad, Wahal, and Xiang 2015; Brogaard, Hendershott, and Riordan 2017). The dramatic increase in the QT ratio has been widely attributed to the emergence of algorithmic trading and HFT (e.g. Hendershott, Jones, and Menkveld 2011).

We obtain the trades and quotes reported in the NYSE Trade and Quote (TAQ) database. We retain stocks listed on the NYSE, AMEX, and NASDAQ for which information is available in TAQ. There are 1,310 surprises in the final sample.

We therefore split the sample into two sub-groups according to the level of algorithmic trading of the stock (level of QT above the sample median and below the sample median) and re-estimate our baseline regression. Column 1 of Table 8 reports the high algorithmic trading sub-sample regression results. The coefficient on ESxDST is not statistically significant, showing that the effect of earnings surprises on CAR does not differ depending on whether investors are sleep-deprived or not when the trading of the stock involves high level of algorithm. Column 2 of Table 8 reports the low algorithmic trading sub-sample regression results. The coefficient on ESxDST (−0.0064) is negative and significant at 5% level, showing a manifestation of investors’ significant underreaction to the earnings surprises following the time change. This finding lends further support to our interpretation that the effect of transition to DST on stock returns is due to sleep deprivation.

Table 8. Additional analysis – market reaction across different levels of algorithmic trading. This table reports the relation between transition to daylight saving time and the announcement period CAR to earnings surprises in high algorithmic trading and low algorithmic trading subsamples (level of QT above 75th percentile or below 25th percentile of the sample). The dependent variable is CAR[0,1]. ES denotes the deciles of earnings surprise. DST is a dummy variable equal to 1 if the earnings was released from Monday to Thursday following the transition to daylight saving time, otherwise 0. We include size, B/M, no._analysts, r_lag, e_vol, s_turn and e_pers as control variables. We include industry dummies using Fama-French 10 industry classification. Standard errors clustered by week are in parentheses. *, **, and *** denote statistical significance at 10%, 5%, and 1% levels, respectively.

	High algorithmic trading	Low algorithmic trading
	1	2
	CAR [0,1]	CAR [0,1]
ES	0.0102***	0.0150***
	(0.0006)	(0.0015)
DST	0.0435*	0.0275*
	(0.0229)	(0.0143)
ESxDST	−0.0043	−0.0064**
	(0.0029)	(0.0031)
size	−0.0006	0.0030
	(0.0027)	(0.0044)
B/M	−0.0027	0.0139
	(0.0259)	(0.0208)
no._analysts	0.0202	−0.0092
	(0.0198)	(0.0217)
r_lag	0.0064	−0.0009
	(0.0069)	(0.0103)
e_vol	−0.0001	−0.0056
	(0.0033)	(0.0067)
s_turn	−0.1639*	0.1209***
	(0.0827)	(0.0424)
e_pers	−0.0091	0.0088
	(0.0122)	(0.0181)
Industry	Yes	Yes
Intercept	0.0245	−0.0640
	(0.0329)	(0.0734)
Obs	326	328
R-sq	0.1546	0.1406

Standard errors below coefficients. Significance stars denote: * 0.1 ** 0.05 *** 0.01.

6. Synthetic matching

Since the exact timing of earnings announcements is endogenously determined by firms, it is important to find a suitable counterfactual of firms that made earnings surprises outside the DST transition period. In the classic framework for permutation inference, the distribution of a test statistic is computed under random permutations of the sample units’ assignment to the treatment and non-treatment groups. In Table 3, we find three covariates (size, B/M, and no._analysts) with significant mean differences between our treatment and control groups, which indicates that unobservable factors might have an impact on the assignments of firms to treatment and control groups. Unconfoundedness requires that the probability of treatment assignment is free of dependence on the potential outcomes, and this motivated us to turn to a matching procedure that does not require this assumption to be satisfied for unbiasedness and that does not rely on the parallel trend assumption either.

We turn to the synthetic control method (SCM), originally proposed by Abadie and Gardeazabal (2003) and Abadie, Diamond, and Hainmueller (2010), and create our own counterfactuals by selecting a weighted average of the outcome variable from a group of units similar to the treated unit. SCM does not rely on parallel pre-implementation trends like difference in difference methods (e.g. Kreif et al. 2015; Roesel 2017, Bouttell et al. 2018), but still mitigates any concern that unobservable characteristics affect the timing of earnings announcements or could determine investor response to the earnings surprises. Full details are provided in Appendix 2.

Following Acemoglu et al (2016), we construct a synthetic match that mirrors the values of the returns of treatment firms before the transition to DST. We estimate the effect as the difference in abnormal returns between treated firms and their synthetic versions in the days following the transition to DST. We then perform a series of placebo studies that confirm that our estimated effects for treated firms are unusually large relative to the distribution of the estimate that we obtain when we apply the same analysis to the firms in control group.

Panel A of Table 9 presents the SCM results for CAR[0,1]. Column 1 reports OLS estimates for comparison. In column 2, the SCM estimate shows that the announcement abnormal return is around 4.8 percentage points lower than what would have been in the absence of transition to DST, which is economically sizable and statistically significant at the 1% level. This indicates that immediate market response is significantly less sensitive to earnings surprises following transition to DST, which confirms our OLS finding that due to impaired cognitive ability, sleep-deprived investors underreact to earnings surprises. Column 2 of Panel B to Panel E reports the SCM estimates for post-announcement periods, which confirm our inference that when investors regain their sleep patterns, e.g. CAR[2,10], their cognitive processing abilities are restored and are able to correct their mispricing before a significant amount of new information is likely to have been released.

Table 9. Market reactions to earnings surprises: SCM analysis (Spring). This table reports synthetic control method estimates of the effect of transition to daylight saving time on CAR surrounding the earnings announcements, where the dependent variables are announcement day CAR and post-announcement CARs. The matching window is the 100 trading days ending 30 days prior to event day 0. Confidence intervals for hypothesis testing of the effect of transition to daylight saving time being equal to zero are computed according to 1,000,000 placebo simulations. OLS results are reported for comparison. *, **, and *** denote statistical significance at 10%, 5%, and 1% levels, respectively.

		1	2
		OLS	SCM
Announcement return	Panel A: Dependent variable is CAR[0,1]
	ES x DST	−0.0033**	−0.0488***
	Confidence interval (2.5%)		−0.0736
	Confidence interval (97.5%)		−0.0241
	Number of firms	2232	2232
	Number in treatment group	807	807
Post-earnings announcement return	Panel B: Dependent variable is CAR[2,5]
	ES x DST	0.0013	−0.0028
	Confidence interval (2.5%)		−0.0096
	Confidence interval (97.5%)		0.0040
	Number of firms		2232
	Number in treatment group		807
	Panel C: Dependent variable is CAR[2,10]
	ES x DST	0.0033*	0.0157***
	Confidence interval (2.5%)		0.0108
	Confidence interval (97.5%)		0.0206
	Number of firms	2232	2232
	Number in treatment group	807	807
	Panel D: Dependent variable is CAR[2,61]
	ES x DST	0.0185**	0.0572***
	Confidence interval (2.5%)		0.0524
	Confidence interval (97.5%)		0.0621
	Number of firms	2232	2232
	Number in treatment group	807	807
	Panel E: Dependent variable is CAR[2,90]
	ES x DST	0.0266***	0.1019***
	Confidence interval (2.5%)		0.0955
	Confidence interval (97.5%)		0.1083
	Number of firms	2232	2232
	Number in treatment group	807	807

Our SCM estimates are larger than the OLS estimates. Abadie, Diamond, and Hainmueller (2010) also find larger estimates using SCM than those reported by Fichtenberg and Glantz (2000) who employ linear regression methods in the study of the effect of 1988 California Proposition 99 which taxed and placed restrictions on the sale of tobacco in California. As in other studies, the smaller OLS estimates compared to the SCM estimates in our study might be due to the assignments of control units, as shown in Table 3, where there are covariate differences across treatment and control groups.

Taken together, Panels A to Panel E confirm the presence of a negative and significant immediate announcement returns, and positive and significant returns over the post-earnings announcement period when using SCM as an alternative to OLS. In other words, we cannot reject our sleep-deprivation hypothesis.

We carry out different robustness checks for SCM by using different specifications and replacement techniques for stale stock returns (returns that are equal to 0). The results are similar to those reported in Table 9. These results are unreported but available upon request.

7. Concluding remarks

For a long time, there has been a debate over DST. From a medical point of view, evidence suggests it should be avoided. From an economic point of view, it was originally believed that DST would reduce using costly energy resources in the early evening hours. The latter argument is not likely to hold in modern times as the use of technology, which is part of our daily lives, requires constant energy usage. In this study, we provide evidence that the economic consequences of DST may actually be negative as the ability of investors to process information on firms is affected. We show that sleep deprivation is associated with a weaker immediate price reaction to earnings surprises and a subsequent post-earnings announcement drift. Our results suggest that sleep disruption affects the pricing of relevant information in individual stocks. Overall, we highlight the importance of cognitive abilities of investors for efficient market pricing by demonstrating how deviations from this assumption give rise to predictable return patterns.

The USA Senate recently passed the Sunshine Protection Act of 2021, where daylight saving time is the new, permanent standard time, effective from November 5, 2023. Other countries like Mexico, experienced in 2023 their first year without DST. Our findings, although based in the USA, have policy implications beyond this country and may inform decision-making processes related to DST changes in other nations that are currently re-evaluating their current practices.

Furthermore, sleep disruption is a growing concern that extends beyond the financial sector. Our study contributes to the interdisciplinary field by providing insights into the effects of sleep deprivation on the pricing of relevant information in individual stocks. This may have implications for various fields, including healthcare and public policy.

Acknowledgements

We would like to thank Jens Hagendorff, Maria Boutchkova, Ben Sila, Khaladdin Rzayev, and seminar participants at The University of Edinburgh for useful comments and suggestions. Special thanks to Chris Adcock (Editor), an anonymous Associate Editor, and anonymous referees for giving us the chance to improve the manuscript’s quality.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Angelica Gonzalez

Angelica Gonzalez is a Senior Lecturer in Finance at the University of Edinburgh Business School. Her research focuses on corporate governance with a strong emphasis on the gender diversity on boards. Angelica also has an interest on corporate social responsibility and the implications of sleep deprivation on financial decisions.

Xuhao Li

Xuhao Li is a doctoral researcher at The University of Edinburgh. His research areas are behavioural finance, sustainable finance, corporate governance, and asset pricing.

Notes

1 An important exception is Siganos (2019) that studies M&A announcements around the transition to DST. While Siganos documents overreaction to announcement returns around M&A (and interpret the results as due to swings in mood), he does not test for post-announcement drifts. Since the effects of DST on sleep are temporary, it is likely that investors will reassess their initial reaction as their sleeping patterns recover. We argue that a post-announcement drift, as documented in our study, is necessary to test whether DST affects the information processing abilities of investors.

2 Our decision to use deciles follows existing literature (see Hirshleifer, Lim, and Hong Teoh 2009; Bernard and Thomas 1990; Bhushan 1994; Bartov, Radhakrishnan, and Krinsky 2000; Livnat and Mendenhall 2006; Johnson and Zhao 2012). The rationale to use deciles is that this procedure can mitigate the negative effects of outliers and nonlinearities in the earnings-surprise returns relationship. Further, this decile ranking procedure can mitigate the intertemporal effects of macroeconomic events that suddenly shift the earnings surprise distribution to the right or left over time.

3 We found some evidence of macroeconomic news being released at wider windows.

4 We also carry out untabulated test on the difference between the market reaction to earnings surprises in relation to DST change for large firm and small firm subsamples and we find the difference to be insignificant.

5 We also carry out untabulated test on the difference between the market reaction to earnings surprises in relation to DST change for more analysts and less analysts following subsamples and we find the difference to be insignificant.

6 Untabulated tests corroborate that these differences are statistically significant.

7 Untabulated tests corroborate that this difference is statistically significant.