Cost behaviour and reporting frequency during the COVID-19 outbreak

Abstract

We examine the effect of financial reporting frequency on cost management decisions in crisis situations, with a focus on the COVID-19 outbreak. Using the European setting, we find that quarterly reporters exhibit greater cost elasticity relative to semi-annual reporters, meaning they had larger changes in cost for each change in sales. When allowing for cost asymmetry, we see that our results are driven by firms with decreases in sales and that quarterly reporters reduced their costs more. Additional analyses show that managerial learning and monitoring pressure might be potential channels behind the results and that there is a positive performance effect in the short run.

Keywords:

• cost behaviour
• COVID-19
• quarterly reporting
• reporting frequency

1. Introduction

In this study, we examine the relationship between cost behaviour and financial reporting frequency in crisis situations. We focus on the COVID-19 pandemic because it impacted the stock markets unlike any previous infectious disease (Baker et al. 2020, Demers et al. 2021) and induced a sudden combination of supply and demand shocks to firms in a broad range of industries around the world (Baqaee and Farhi 2022, Guerrieri et al. 2022). In the outbreak phase, firms recognised how critical the ability to quickly adjust costs in response to a lower activity level was for success and survival (Hassan et al. 2023). For instance, in a stock exchange release from May 2020, Topi Manner, the CEO of Finnair, stated: ‘In the post-corona market, those who can adapt their costs to the changed market and the competitive situation are the ones who will succeed’.

The role and importance of financial reporting when firms respond to a sudden commercial activity shock is ex-ante unclear. Following Roychowdhury et al. (2019), Shakespeare (2020), and Simpson and Tamayo (2020), we reason that financial reporting and disclosure have real effects on corporate behaviour. Specifically, we conjecture that publicly listed firms’ financial reporting frequency (i.e. quarterly or semi-annual reporting of financial statements) influenced their cost adjustments in response to the outbreak of COVID-19 in the first half (H1) of 2020.

We see two reasons why quarterly reporting may nurture efficient cost management and consequently a more flexible and elastic cost structure during the COVID-19 outbreak. First, it is possible that managers learn from the preparation of financial statements. Shroff (2017) and Cheng et al. (2018), for instance, document how the reporting process provides managers with not yet incorporated decision-relevant information or force them to acknowledge additional information. This information may not have been recognised earlier since managers, like all other economic agents, have limited attention and power to acquire and process information (Simon 1973, Sims 2003). Therefore, more frequent production of financial statements can be informative to managers and lead to greater action (Ghobadian et al. 2022). Second, more frequent reporting mitigates information asymmetries, which allows for timelier monitoring of managers by shareholders (Kanodia and Lee 1998). This increase in monitoring pressures managers, and thereby increases managerial incentives to align costs with new activity levels. Through these, non-mutually exclusive, channels of managerial learning and monitoring pressure we expect reporting frequency to influence firms’ cost behaviour, especially during the COVID-19 outbreak when substantial amounts of new information had to be processed by managers and the information discrepancy between managers and shareholders was high.

For our main empirical analyses, we use a sample of 3197 publicly listed firms from 17 European countries with fiscal year-end in December 2019. Europe is a suitable research arena to study the effects of interim reporting because there is both within- and between-country variation in reporting frequencies. In our sample, we observe an average sales decline of 10.3% between H1 of 2020 and H1 of 2019. Using a log-linear model to measure cost elasticity (i.e. the average sensitivity of costs to sales changes) following prior studies (e.g. Banker et al. 2014a, Holzhacker et al. 2015a), we find a more elastic cost structure among quarterly reporters than among semi-annual reporters during the COVID-19 outbreak. Our estimates without controls suggest that a 1% change in sales is associated with a 0.51% change in operating costs for quarterly reporters, while the corresponding figure is 0.36% for semi-annual reporters.

Anderson et al. (2003) provide evidence that managers, on average, are more inclined to adjust costs upward when activity rises than they are to adjust costs downward when activity falls. This cost asymmetry with ‘sticky costs’ exists because managers are optimistic about the future, and hence retain unused resources when activity falls in anticipation of rebound (Banker et al. 2018). If managers, on the contrary, are pessimistic about the future, and activity falls, they are likely to cut unused resources which will reduce cost stickiness. Under such circumstances, costs can be termed ‘anti-sticky’, meaning that costs are less adjusted upward when activity rises than they are adjusted downward when activity falls (Weiss 2010, Banker et al. 2014b). Allowing for cost asymmetry, we find that quarterly reporters had anti-sticky costs during the COVID-19 outbreak, because of larger cost reductions.

Our focus on the COVID-19 outbreak is beneficial since the change in the activity levels we observe is exogenous. However, we recognise the possibility that firms seeing increased benefits from more elastic cost structures may self-select to report more frequently. To mitigate this endogeneity concern, we implement a reduced-form instrumental variable approach. As the instrument, we use a variable indicating whether the firm’s country in 2004 required quarterly reporting at the main stock exchange. The instrument captures a country’s legacy of quarterly reporting before major harmonisation attempts regarding reporting frequency took place in the European Union (EU). This legacy is not expected to have a direct impact on cost behaviour during the first half of 2020, which, among other assumptions, is necessary for a valid instrument. Our main results remain unchanged with this approach.

In additional tests, we find that our results are driven by firms operating in industries highly affected by the shock. We do not find statistically significant evidence of cost elasticity differences before the pandemic but note that quarterly reporters adjusted their cost structures more than semi-annual reporters in response to the COVID-19 outbreak. Regarding the reasons behind the results, we find some support of the internal learning and external monitoring channels by examining earnings announcement speeds and statements of managers during conference calls as well as analyst behaviour. In terms of future performance, we show that quarterly reporters had higher cumulative abnormal returns around the 2020 half-year earnings announcement and superior accounting profitability in the short-run after the COVID-19 outbreak, potentially due to greater cost elasticity. We find no evidence suggesting that managers of quarterly reporters were engaging in myopic cost cutting. Taken together, the additional tests as well as a battery of sensitivity analyses (e.g. using the global financial crisis of 2008–2009, alternative variable definitions, and controlling for firm-specific measures of managerial pessimism and operating leverage) corroborate the main findings that firms with a higher reporting frequency are more responsive to a crisis in general and to changes in activity levels in particular.

We make the following contributions. First, we add to the literature on the benefits and costs of corporate disclosure and different reporting frequencies. Gigler et al. (2014) theoretically outline that more frequent reporting may induce short-termism that ultimately impacts investment decisions and prior empirical studies find that quarterly reporting indeed has negative real effects (e.g. Ernstberger et al. 2017, Kraft et al. 2018, Fu et al. 2020). In contrast, we document a positive real effect, with our finding that firms reporting more frequently are better at adjusting their cost levels when exposed to an exogenous activity level shock. Second, we contribute to the cost elasticity literature by demonstrating that reporting frequency is associated with cost and activity co-movement during crises. We differ from previous cost elasticity studies on demand uncertainty using more local or industry-specific demand shocks (e.g. Kallapur and Eldenburg 2005, Banker et al. 2014a, Holzhacker et al. 2015a, Holzhacker et al. 2015b), by examining a setting with demand as well as supply shocks that affected most firms in the economy (Baqaee and Farhi 2022, Guerrieri et al. 2022). The COVID-19 pandemic represents a particularly strong setting to test our hypothesis because the impact on activity levels was more pronounced in the outbreak phase. Our finding of more anti-sticky costs among quarterly reporters during the crisis extends prior research where drivers of cost anti-stickiness have been identified using firm-specific measures of pessimism (e.g. Banker et al. 2014b). Third, we contribute to the substantial and growing literature on the COVID-19 crisis where most studies to date focus on aggregate, industry-level, or stock return analyses. For example, Ding et al. (2021) show that firms with strong finances, less exposure to COVID-19, more corporate social responsibility activities, and less entrenched executives had a smaller crisis-induced drop in stock prices. Fahlenbrach et al. (2021) correspondingly find that high financial flexibility was valuable in the early stages of the crisis. Demers et al. (2021) present evidence that investments in intangible assets immunised stocks. Similarly, but with a financial reporting focus, our study sheds light on the frequency of financial reporting as a potential resilience factor that helps firms cope with changes in demand and supply.

2. Literature and development of hypothesis

2.1. Reporting frequency

There is an ongoing regulatory and academic debate regarding the benefits and costs of higher financial reporting frequency. In 1955, the US SEC (United States Securities and Exchange Commission) started requiring semi-annual reporting instead of annual reporting. In 1970, the mandatory reporting frequency was changed to quarterly. To harmonise the reporting frequency regulation for listed firms within the EU, regulators proposed a shift to mandatory quarterly reporting for all listed firms at the beginning of the twenty-first century. However, the EU parliament rejected the proposal and instead adopted the Transparency Directive 2004/109/EC, requiring firms to publish semi-annual financial reports with quarterly statements known as Interim Management Statements, which could be seen as trading updates (Schleicher and Walker 2015).¹ The Transparency Directive soon received plenty of criticism. For example, in 2010, the EU Commission demonstrated a need for simplified reporting to reduce compliance costs of small and medium-sized firms. Kay (2012) argued that short-term decision making is a consequence of more frequent reporting. As a response, the EU parliament issued the Transparency Directive 2013/50/EU, which stated that Member States were not allowed to require more frequent reporting than on a semi-annual basis without any further justification.

Early academic studies identify benefits of higher reporting frequency in the form of lower earnings announcement stock price variability (McNichols and Manegold 1983), more timely earnings (Butler et al. 2007), lower information asymmetry (Cuijpers and Peek 2010), and lower cost of capital (Fu et al. 2012). More recently, Downar et al. (2018) find that European semi-annual reporters have a lower valuation of cash assets and Haga et al. (2022) document a decrease in stock price synchronicity with a larger fraction of peer firms reporting quarterly. These studies highlight information environment improvements that benefit stakeholders. On the other hand, there is a budding literature on the negative real effects of higher reporting frequency. Theoretically, Gigler et al. (2014) show that increased reporting frequency causes lower investments and managerial short-termism. Several empirical studies consistently find more real earnings management (Ernstberger et al. 2017), lower capital expenditures (Kraft et al. 2018, Hitz and Moritz 2019), and lower innovation (Fu et al. 2020) among quarterly reporters. Meanwhile, Nallareddy et al. (2021) do not find any investment effect using firms in the United Kingdom (UK), but they do observe a substantial increase of firms announcing managerial guidance for the upcoming year’s earnings or sales with higher reporting frequency. With an event study, Kajüter et al. (2019) find a 5% decrease in firm value for Singaporean firms that were required to shift from semi-annual to quarterly reporting. Whether the net effects of an increase in reporting frequency are positive or negative remains an open question, Roychowdhury et al. (2019) conclude in their literature review.

2.2. Cost behaviour and the COVID-19 pandemic

Basic cost behaviour models postulate a mechanistic linear relation between a cost driver, such as sales, and concurrent costs (Garrison et al. 2015). In such models, fixed costs stay constant in the short-run and variable costs change proportionally to the cost driver or level of activity. How much costs move in response to activity changes is termed cost elasticity (Holzhacker et al. 2015a, Banker et al. 2018). Greater cost elasticity means having a larger percentage change in cost for each percentage change in activity. Prior studies document that firms increase their cost elasticity as demand uncertainty and financial risk increase (Kallapur and Eldenburg 2005, Holzhacker et al. 2015a, Holzhacker et al. 2015b). The economic shock caused by COVID-19 was unusual because in addition to decreasing demand and increasing uncertainty about future demand, the pandemic also disrupted the supply of input factors in some industries (Baqaee and Farhi 2022). Even though some industries were unaffected by the negative supply shock, such shocks lead to lower aggregate demand for the economy (Guerrieri et al. 2022). Consequently, both the COVID-19 induced demand and supply shock led to lower aggregated sales for firms. Because earnings are a function of sales and costs, greater cost elasticity enables firms to maintain earnings when there is a decline in sales. As such, the ability to have a more elastic cost structure was especially beneficial during a shock like COVID-19.

Many studies examining cost elasticity also distinguish between upward and downward cost elasticity (e.g. Holzhacker et al. 2015a, Holzhacker et al. 2015b, Hall 2016). Prior research documents cost elasticity to be smaller when there is a decline in activity compared to a positive change in activity (Anderson et al. 2003). This is defined as cost asymmetry or cost stickiness. Theory predicts costs to be sticky when adjustment costs exist and when managers are generally optimistic about the future but encounter a decline in activity (Anderson et al. 2003, Banker et al. 2018). Thus, firms with downward demand shocks are expected to avoid adjusting their costs under managerial optimism. However, the COVID-19 pandemic turned managerial optimism into pessimism. Because of pessimistic outlooks, anti-sticky cost behaviour may occur (Banker et al. 2014b). The definition of anti-sticky cost behaviour is that costs rise less in response to sales increases than they fall when sales decrease by an equivalent amount (Weiss 2010). Banker et al. (2014b) provide evidence of anti-sticky cost behaviour among firms with multiple negative growth years. In the US airline industry, Cannon (2014) finds anti-sticky cost behaviour when managers save more cost by removing aircraft capacity when demand is falling than they save by removing capacity when demand is growing. In summary, the literature suggests that crisis situations with pessimism and dramatic declines in sales, such as COVID-19 (Fahlenbrach et al. 2021), result in anti-sticky cost behaviour.

2.3. Hypothesis development

We see two, non-mutually exclusive, channels through which a firm’s reporting frequency may impact its cost behaviour during the COVID-19 outbreak. The first is managerial learning. We conjecture that more frequent reporting provides managers with not yet incorporated decision-relevant information in a timely manner. While top managers have almost unconstrained access to information at any point in time, managers like other economic agents have limited attention and information acquisition and processing power (Simon 1973, Sims 2003). Hence, decision-relevant information may be left unrecognised. When managers prepare financial reports for the purpose of informing shareholders about historical performance and future outlooks, the managers themselves may gain an improved and updated information set. Shroff (2017) provides empirical support for a managerial learning channel, by showing that managers learn from changes in accounting rules that require them to gather new information and alter their investment behaviour accordingly. Similarly, Cheng et al. (2018) confirm that compliance with a new accounting rule induces managers to acquire new information and therefore improves their information sets. The managerial learning channel is further discussed in literature reviews by Roychowdhury et al. (2019), Shakespeare (2020), and Simpson and Tamayo (2020). Outside the crisis setting, Kim et al. (2022) find costs to be stickier for firms less likely to provide managers with timely and precise information. During the COVID-19 outbreak, managers were forced to absorb and act on information regarding lockdowns, downturns in customer demand, and supply chain disruptions (Ghobadian et al. 2022, Hassan et al. 2023). The information guided managers in the development of survival strategies and business model adjustments. We argue that more frequent reporting, through the managerial learning channel, enabled managers to make faster cost adjustment decisions when struck by the sudden activity level shock.

The second channel is the monitoring channel. Kanodia and Lee (1998) argue that more frequent reporting facilitates the closer monitoring of managers by shareholders and lowers monitoring costs by reducing information asymmetries. A higher frequency of financial reporting also allows shareholders to evaluate managerial decisions on a timelier basis and react to problems at an earlier stage. Timelier monitoring played a crucial role during the COVID-19 outbreak. During a crisis, shareholders may lower their perceived value of the shares in case of unfulfilling managerial decisions. This potentially causes a decrease in share prices which reduces the value of managers’ equity in the firm and increases the likelihood of their dismissal. Given that managers anticipate this monitoring, they are incentivised to make timelier decisions (Kanodia and Lee 1998). Moreover, Bharath et al. (2013) argue that the mere threat of a share price reduction has a disciplining effect on managers. Regarding reporting frequency, Downar et al. (2018) find that cash holdings for quarterly reporters are valued higher than for semi-annual reporters, which could be linked to closer monitoring. During the COVID-19 outbreak, the increased monitoring associated with more frequent reporting might lead to timelier changes to cost structures from managers of quarterly reporters.

If higher reporting frequency improves managerial learning and/or shareholder monitoring during the COVID-19 outbreak, we expect larger cost adjustments in response to changes in activity levels from firms that report more frequently. This would allow quarterly reporters to have a more elastic cost structure and less cost stickiness. As such, we formulate the following hypothesis:

Hypothesis: During the COVID-19 outbreak, quarterly reporters exhibit greater cost elasticity relative to semi-annual reporters.

3. Method

3.1. Empirical models

To study the relationship between reporting frequency and cost elasticity, we begin with a log-log specification that links changes in operating costs to contemporaneous changes in sales. Banker et al. (2014a) and Holzhacker et al. (2015a), among others, advocate for such a model, where operating costs change proportionately with activity levels according to the following: (1) $\mathrm{\Delta }\mathrm{lnXOP}{R}_i={\text{β}}_0+{\text{β}}_1\mathrm{\Delta lnSAL}{E}_i+{ϵ}_i$(1) where $\mathrm{\Delta lnXOP}{R}_i$ is the log-change in operating cost (Compustat Global mnemonic XOPR which is the sum of Cost of Goods Sold (COGS), Other Operating Expense (XOPRO), and Selling, General, and Administrative Expense (XSGA)) between H1 2020 and H1 2019 for firm i. $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$ is the log-change in sales revenue for the same period and firm. This specification with changes in logged levels eliminates potential bias from unobserved heterogeneity and serial correlation of error terms (Wooldridge 2009). The coefficient ${\text{β}}_1$ in Eq. (1) provides an empirical measure of the degree of cost elasticity (Holzhacker et al. 2015a). Following our hypothesis, we expect that costs change more if a firm reports quarterly during the COVID-19 outbreak. To test the hypothesis, we rely on Eq. (1) and introduce an indicator variable $\mathrm{QR}{T}_i$ (taking the value one if firm i reports quarterly or semi-annually with business reviews, and zero if firm i reports semi-annually) as well as firm-level and country-level controls. Following Holzhacker et al. (2015a), we include the control main effects as well as their interactions with $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$ and arrive at the following full estimation model for cost elasticity: (2) $\begin{array}{rl}\mathrm{\Delta }\mathrm{lnXOP}{R}_i& ={\text{β}}_0+{\text{β}}_1\mathrm{QR}{T}_i+{\text{β}}_{2}M{V}_i+{\text{β}}_3\mathrm{AIN}{T}_i+{\text{β}}_4\mathrm{\Delta GD}{P}_c+{\text{β}}_5\mathrm{\Delta }\mathrm{lnSAL}{E}_i\\ & +{\text{β}}_6\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i+{\text{β}}_7\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times M{V}_i+{\text{β}}_8\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{AIN}{T}_i\\ & +{\text{β}}_9\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{\Delta GD}{P}_c+{ϵ}_i\end{array}$(2) We use the ordinary least squares (OLS) estimation technique to obtain our coefficients. In Eq. (2), the coefficient of interest is ${\text{β}}_6$ and it captures the difference in cost elasticity between quarterly and semi-annual reporters. According to our hypothesis, we expect that ${\text{β}}_6>0$, implying that firms reporting more frequently have a more flexible and elastic cost structure.

We control for firm size by including the logarithm of the market value ($M{V}_i$) on December 31, 2019 (CSHO × PRCC_F). Larger firms often increase transparency by having greater disclosure, and have better internal control systems (e.g. Kasznik and Lev 1995, Ge and McVay 2005). As such, omitting the size control may bias ${\text{β}}_6$ if firm size is associated with $\mathrm{QR}{T}_i$ and cost behaviour (Weiss 2010). To control for the cost of adjusting operating expenses, we include asset intensity ($\mathrm{AIN}{T}_i$) as the log-ratio of assets (AT) to sales revenue (SALE) following Anderson et al. (2003).² Furthermore, we control for gross domestic product growth ($\mathrm{\Delta GD}{P}_c$) between H1 2020 and H1 2019 for country c following Banker et al. (2013). While commonly used to control for managerial expectations, $\mathrm{\Delta GD}{P}_c$ is a suitable control for the impact of the COVID-19 pandemic on country activity since it captures both the impact of COVID-19 fatalities and mandatory social distancing imposed by the authorities (König and Winkler 2020).

According to Noreen and Soderstrom (1994), costs respond asymmetrically to activity changes. Anderson et al. (2003) developed a comprehensive empirical framework to capture cost asymmetry, again using changes in logged levels of costs and activity, but also with an indicator variable for decreasing sales ($\mathrm{DE}{C}_i$) interacted with $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$.With $\mathrm{QR}{T}_i$ and control variables in three-way interactions with $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i$, we estimate the following full model for cost asymmetry: (3) $\begin{array}{rl}\mathrm{\Delta }\mathrm{lnXOP}{R}_i& ={\text{β}}_0+{\text{β}}_1\mathrm{QR}{T}_i+{\text{β}}_{2}M{V}_i+{\text{β}}_3\mathrm{AIN}{T}_i+{\text{β}}_4\mathrm{\Delta GD}{P}_c+{\text{β}}_5\mathrm{\Delta }\mathrm{lnSAL}{E}_i\\ & +{\text{β}}_6\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i+{\text{β}}_7\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times M{V}_i+{\text{β}}_8\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{AIN}{T}_i\\ & +{\text{β}}_9\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{\Delta GD}{P}_c+{\text{β}}_{10}\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i+{\text{β}}_{11}\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\\ & \times \mathrm{QR}{T}_i+{\text{β}}_{12}\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times M{V}_i+{\text{β}}_{13}\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{AIN}{T}_i\\ \left(3\right)& & +{\text{β}}_{14}\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{\Delta GD}{P}_c+{ϵ}_i\end{array}$(3) Our coefficient of interest is ${\text{β}}_{11}$, which captures the difference in cost asymmetry between quarterly and semi-annual reporters. A positive ${\text{β}}_{11}$ implies that quarterly reporters have less cost stickiness. Following Anderson et al. (2003), we have omitted the main effect for $\mathrm{DE}{C}_i$ in Eq. (3).³

3.2. Sample selection and descriptive statistics

We select a sample of European publicly listed firms due to the cross- and within-country variation in reporting frequency. We collect all financial statement information from Compustat Global and GDP growth rates from Eurostat. Using Compustat Global identifiers, our initial sample covers firms from 25 European countries with data for fiscal year 2019 (6417 firms). To generate our final sample, we first exclude financial institutions by removing firms with SIC code between 6000 and 6999 (less 1517 firms). To obtain a comparable sample, we exclude all firms with fiscal year-end different from December 2019 (less 844 observations). We collect firm reporting frequency information from Bloomberg (less 77 firms).⁴ We only keep firms with available data for the variables in Eq. (3) where some require both current and lagged information (less 644 firms). Finally, we drop countries with less than 30 observations (less 138 firms). The final sample contains 3197 firms from 17 European countries. Out of these firms, 1977 are from quarterly and business review reporters ($\mathrm{QR}{T}_i=1$), and 1220 from pure semi-annual reporters ($\mathrm{QR}{T}_i=0$).

Panel A of Table 1 provides the number of observations and mean values for our main variables, by country and reporting frequency. We observe clear differences in the proportion of quarterly reporters between the countries. For example, countries with a legacy of mandatory quarterly reporting such as Finland, Norway, and Sweden have a high proportion of quarterly reporters. Meanwhile, countries with a legacy of semi-annual reporting, such as the UK, have more semi-annual reporters. The averages of log-changes in sales suggest that firms in most countries have experienced a decrease in activity during the outbreak of COVID-19. On average, only Swedish firms reported growth. The negative means of log-changes in operating costs suggest that firms adjusted their cost structure based on the reduced activity levels. Switzerland was the only country with a positive change in GDP during the first half of 2020. The last two columns in Panel A of Table 1 report that we cover on average 62.3% of the firms and 65.6% of the market capitalisation in Compustat Global. Our coverage is the lowest for the UK with 27.4% of the firms and 45.2% of the market capitalisation. The main reason for the low coverage is that we follow Alves et al. (2021) and Fahlenbrach et al. (2021) by limiting our sample to firms with December fiscal year-end.⁵

Table 1. Descriptive statistics.

Panel A: Summary statistics by country and reporting frequency
Country	Obs.	Mean						Coverage
		ΔlnXOPRi	ΔlnSALEi	DECi	MVi	AINTi	ΔGDPc	% of Obs.	% of MCap.
Austria	30	−0.100	−0.106	0.700	20.461	0.181	−0.077	46.9%	53.2%
QRTi = 1	22	−0.089	−0.092	0.636	20.975	0.304
QRTi = 0	8	−0.129	−0.144	0.875	19.046	−0.155
Belgium	66	−0.059	−0.092	0.636	19.954	0.506	−0.074	54.1%	65.9%
QRTi = 1	23	−0.059	−0.004	0.652	21.006	0.573
QRTi = 0	43	−0.059	−0.138	0.628	19.391	0.470
Croatia	52	−0.284	−0.569	0.712	17.625	0.866	−0.089	85.2%	69.3%
QRTi = 1	52	−0.284	−0.569	0.712	17.625	0.866
QRTi = 0	-	-	-	-	-	-
Denmark	71	−0.016	−0.097	0.577	19.373	0.635	−0.013	50.0%	80.0%
QRTi = 1	57	0.031	0.083	0.561	20.062	0.444
QRTi = 0	14	−0.205	−0.833	0.643	16.566	1.412
Finland	116	−0.031	−0.010	0.448	19.130	0.024	−0.016	77.3%	75.6%
QRTi = 1	96	−0.022	0.002	0.427	19.594	0.057
QRTi = 0	20	−0.072	−0.063	0.550	16.904	−0.133
France	379	−0.083	−0.104	0.697	19.038	0.563	−0.092	63.4%	79.6%
QRTi = 1	216	−0.106	−0.144	0.727	20.277	0.431
QRTi = 0	163	−0.051	−0.052	0.656	17.396	0.738
Germany	335	−0.044	−0.072	0.630	19.466	0.321	−0.043	56.3%	61.4%
QRTi = 1	225	−0.045	−0.103	0.613	20.340	0.259
QRTi = 0	110	−0.041	−0.008	0.664	17.678	0.449
Greece	117	−0.162	−0.224	0.726	17.142	0.590	−0.103	78.5%	61.4%
QRTi = 1	22	−0.119	−0.181	0.727	19.798	0.340
QRTi = 0	95	−0.172	−0.234	0.726	16.527	0.648
Italy	213	−0.140	−0.207	0.704	18.909	0.499	−0.109	69.8%	59.5%
QRTi = 1	120	−0.116	−0.169	0.717	19.875	0.442
QRTi = 0	93	−0.170	−0.256	0.688	17.662	0.572
Netherlands	84	−0.009	−0.034	0.583	20.587	0.269	−0.027	66.1%	75.4%
QRTi = 1	50	0.007	−0.037	0.540	21.435	0.394
QRTi = 0	34	−0.032	−0.031	0.647	19.341	0.085
Norway	123	−0.043	−0.055	0.463	19.218	0.850	−0.142	61.5%	75.2%
QRTi = 1	111	−0.051	−0.053	0.459	19.347	0.727
QRTi = 0	12	0.031	−0.077	0.500	18.025	1.988
Poland	449	−0.119	−0.155	0.579	16.150	0.237	−0.022	67.7%	53.9%
QRTi = 1	288	−0.129	−0.135	0.576	17.100	0.270
QRTi = 0	161	−0.103	−0.190	0.584	14.451	0.179
Romania	57	−0.160	−0.249	0.632	16.955	0.565	−0.035	74.0%	57.4%
QRTi = 1	54	−0.131	−0.228	0.630	17.030	0.548
QRTi = 0	3	−0.678	−0.639	0.667	15.607	0.862
Spain	88	−0.041	−0.079	0.705	20.301	0.608	−0.119	52.4%	52.7%
QRTi = 1	78	−0.064	−0.098	0.718	20.603	0.601
QRTi = 0	10	0.136	0.074	0.600	17.945	0.663
Sweden	508	0.012	0.035	0.488	17.887	0.651	−0.033	71.8%	67.2%
QRTi = 1	496	0.014	0.043	0.492	17.945	0.627
QRTi = 0	12	−0.048	−0.283	0.333	15.485	1.637
Switzerland	133	−0.139	−0.150	0.759	20.618	0.374	0.011	56.6%	75.9%
QRTi = 1	35	−0.138	−0.106	0.829	22.096	0.445
QRTi = 0	98	−0.139	−0.166	0.735	20.090	0.349
UK	376	−0.051	−0.122	0.686	19.098	0.613	−0.064	27.4%	45.2%
QRTi = 1	32	−0.082	−0.229	0.750	21.319	0.762
QRTi = 0	344	−0.049	−0.112	0.680	18.891	0.599
Panel B: Full sample descriptive statistics
Variable	Mean	Std.Dev.	Min	P25	Median	P75	Max
ΔlnXOPRi	−0.070	0.319	−1.255	−0.195	−0.061	0.069	1.072
ΔlnSALEi	−0.103	0.484	−1.962	−0.262	−0.067	0.080	1.683
MVi	18.554	2.604	12.294	16.669	18.354	20.347	24.999
AINTi	0.487	1.025	−1.254	−0.144	0.251	0.847	4.777
ΔGDPc	−0.056	0.038	−0.142	−0.092	−0.043	−0.027	0.011
QRTi	0.618	0.485	0.000	0.000	1.000	1.000	1.000
DECi	0.617	0.486	0.000	0.000	1.000	1.000	1.000

Note: The table reports summary statistics by country and reporting frequency in Panel A and for the full sample in Panel B. QRT_i takes the value one if a firm i reports quarterly or semi-annually with business reviews, and zero otherwise. ΔlnXOPR_i (ΔlnSALE_i) is the log-change in operating cost (sales revenue) between H1 2020 and H1 2019. DEC_i takes the value one if the sales revenue in H1 2020 is smaller than in H1 2019, and zero otherwise. MV_i is the logarithm of the market value on December 31, 2019. AINT_i is the log-ratio of assets to sales revenue. ΔGDP_c is the change in gross domestic product between H1 2020 and H1 2019 for country c. For the sample coverage relative to the Compustat Global database, we display coverage in terms of observations (% of Obs.) and market capitalisation (% of MCap.).

Panel B of Table 1 reports full sample statistics. We observe a mean cost decline of 7.0% that is associated with a mean sales decline of 10.3% between H1 of 2020 and H1 of 2019. In our sample, 61.7% of firms experience a sales decline.

4. Results

4.1. Main results

Column (1) of Table 2 reports the results based on Eq. (2), excluding control variables. Consistent with our expectation, the positive coefficient on the interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$) indicates that there is a difference in cost elasticity between reporting frequencies. Column (2) of Table 2 provides the full Eq. (2) estimates where the coefficient on the main interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$) remains positive and statistically significant (coef. = 0.1436, t-stat = 2.64). When including fixed effects for industry, Column (3) of Table 2 reports consistent results. These results also suggest that the economic magnitude of the difference is meaningful. In Column (1), the coefficient on $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$suggests that semi-annual reporters on average have a 0.36% change in operating costs per 1% change in sales. The corresponding change in operating costs per 1% change in sales for the average quarterly reporter is 0.51%. These results suggest that quarterly reporters have more elastic cost structures after being exposed to a shock in activity levels, which is supportive of our hypothesis.

Table 2. Reporting frequency and cost elasticity.

	(1)	(2)	(3)
	ΔlnXOPRi	ΔlnXOPRi	ΔlnXOPRi
QRTi	0.0143	−0.0026	−0.0008
	(1.06)	(−0.18)	(−0.05)
MVi		0.0053**	0.0061***
		(2.32)	(2.74)
AINTi		−0.0052	−0.0124
		(−0.58)	(−1.65)
ΔGDPc		0.1353	0.1685
		(0.87)	(1.11)
ΔlnSALEi	0.3553***	0.7883***	0.8014***
	(7.11)	(3.23)	(3.03)
ΔlnSALEi × QRTi	0.1583***	0.1436**	0.1362**
	(4.56)	(2.64)	(2.51)
ΔlnSALEi × MVi		−0.0168	−0.0174
		(−1.06)	(−1.02)
ΔlnSALEi × AINTi		−0.0479**	−0.0474**
		(−2.25)	(−2.26)
ΔlnSALEi×ΔGDPc		0.8140	0.8687
		(1.13)	(1.14)
Fixed effects	No	No	Industry
Adj. R-sq.	0.499	0.546	0.554
Obs.	3197	3197	3197

Note: The table reports the results for regressions analyses that examine the difference in cost elasticity between quarterly and semi-annual reporters. The dependent variable is the log-change in operating cost between H1 2020 and H1 2019 (ΔlnXOPR_i). The coefficient on the two-way interaction between ΔlnSALE_i and QRT_i is used to test the difference in cost elasticity. ΔlnSALE_i is the log-change in sales revenue between H1 2020 and H1 2019. QRT_i takes the value one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise. MV_i is the logarithm of the market value on December 31, 2019. AINT_i is the log-ratio of assets to sales revenue. ΔGDP_c is the change in gross domestic product between H1 2020 and H1 2019. Column (1) reports the OLS estimation of Eq. (2) without control variables, Column (2) reports the OLS estimation of Eq. (2), and Column (3) reports the OLS estimation of Eq. (2) with industry fixed effects. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

In addition to the above, Columns (2) and (3) report that the coefficient on the interaction between $\mathrm{\Delta lnSAL}{E}_i$ and $\mathrm{AIN}{T}_i$ is negative and statistically significant. This is consistent with asset intensity capturing adjustment costs. Neither of the other two interactions are statistically significant. Finally, we use the estimated coefficients in Column (2), where control variables are included, to calculate the cost elasticity. Given the coefficients and mean continuous variables, the results suggest that semi-annual reporters and quarterly reporters on average have a 0.41% and 0.55% change in operating costs per 1% change in sales, respectively.⁶

Column (1) of Table 3 reports the cost asymmetry regression results of Eq. (3), excluding control variables. Here, the insignificant but positive coefficient on the two-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i$) provides no support for overall cost stickiness. This contrasts with Anderson et al. (2003) and highlights the extraordinary circumstances during the COVID-19 outbreak.⁷ The coefficient on the three-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$) is positive and statistically significant (coef. = 0.1933, t-stat = 2.51), indicating more anti-sticky cost behaviour among quarterly reporters. Column (2) of Table 3 provides the estimation results of Eq. (3) and the estimate on the three-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$) continues to be positive and statistically significant (coef. = 0.2810, t-stat = 3.21). Using the coefficients in Column (2), we calculate that semi-annual reporters with decreasing sales on average have a 0.34% change in operating costs per 1% change in sales, while the corresponding figure for quarterly reporters is 0.56%.⁸

Table 3. Reporting frequency and cost asymmetry.

	(1)	(2)	(3)
	ΔlnXOPRi	ΔlnXOPRi	ΔlnXOPRi
QRTi	0.0416***	0.0346**	0.0358**
	(3.30)	(2.30)	(2.40)
MVi		0.0007	0.0014
		(0.32)	(0.58)
AINTi		0.0368***	0.0309**
		(3.07)	(2.59)
ΔGDPc		0.3248	0.3111
		(1.34)	(1.18)
ΔlnSALEi	0.3508***	0.4293	0.4491
	(3.00)	(0.98)	(1.05)
ΔlnSALEi × QRTi	0.0281	−0.0655	−0.0642
	(0.69)	(−1.16)	(−1.15)
ΔlnSALEi × MVi		0.0232	0.0214
		(0.84)	(0.78)
ΔlnSALEi × AINTi		−0.1906***	−0.1837***
		(−9.94)	(−10.07)
ΔlnSALEi×ΔGDPc		0.6518	0.7799
		(0.73)	(0.92)
ΔlnSALEi × DECi	0.0073	0.4127	0.4091
	(0.06)	(0.89)	(0.92)
ΔlnSALEi × DECi × QRTi	0.1933**	0.2810***	0.2716***
	(2.51)	(3.21)	(3.05)
ΔlnSALEi × DECi × MVi		−0.0466	−0.0457
		(−1.56)	(−1.60)
ΔlnSALEi × DECi × AINTi		0.1646***	0.1576***
		(5.79)	(5.66)
ΔlnSALEi × DECi×ΔGDPc		0.3039	0.1841
		(0.21)	(0.13)
Fixed effects	No	No	Industry
Adj. R-sq.	0.508	0.587	0.591
Obs.	3197	3197	3197

Note: The table reports the results for regressions analyses that examine the difference in cost asymmetry between quarterly and semi-annual reporters. The dependent variable is the log-change in operating cost between H1 2020 and H1 2019 (ΔlnXOPR_i). The coefficient on the three-way interaction between ΔlnSALE_i, DEC_i, and QRT_i is used to test the difference in cost asymmetry. ΔlnSALE_i is the log-change in sales revenue between H1 2020 and H1 2019. DEC_i takes the value one if the sales revenue in H1 2020 is smaller than in H1 2019, and zero otherwise. QRT_i takes the value one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise. MV_i is the logarithm of the market value on December 31, 2019. AINT_i is the log-ratio of assets to sales revenue. ΔGDP_c is the change in gross domestic product between H1 2020 and H1 2019. Column (1) reports the OLS estimation of Eq. (3) without control variables, Column (2) reports the OLS estimation of Eq. (3), and Column (3) reports the OLS estimation of Eq. (3) with industry fixed effects. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

In Column (3) of Table 3, the results are quantitatively similar including industry fixed effects. Therefore, a higher reporting frequency is associated with a greater degree of anti-sticky costs (i.e. a more positive stickiness coefficient). In Table 3, the coefficients on the two-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$) are insignificant in contrast to our results in Table 2. This suggests that our cost elasticity results are mainly driven by firms with decreases in sales. We reason that the many two- and three-way interactions containing $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$ generate insignificant coefficients on the standalone independent variables in Columns (2) and (3). Finally, we find that the coefficient on the three-way interaction between $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$, $\mathrm{DE}{C}_i$, and $\mathrm{AIN}{T}_i$ is positive and statistically significant at the 1% level. Taken together, the results in Tables 2 and 3 supports our hypothesis that quarterly reporters have a more elastic cost structure and less cost stickiness.

4.2. Reduced-form instrumental variable approach

The results of our main tests indicate that quarterly reporters have more efficient cost management during the COVID-19 outbreak. However, the reporting frequency of firms is endogenously chosen, which limits our ability to draw strong conclusions regarding a causal relationship. For example, firms may select to report more frequently because they are required to have better cost control. Alternatively, firms with more growth opportunities may have a different cost behaviour and a preferred reporting frequency. We aim to mitigate these potential endogeneity concerns by implementing a reduced-form instrumental variable approach.^9,10 As an instrument for firms’ reporting frequency in year 2020, we create $\mathrm{RE}{G}_c$ which takes the value one if the firm’s home country c in 2004 required quarterly reporting at the main stock exchange.¹¹ It is suitable to use 2004 because this was when the Transparency Directive 2004/109/EC was issued, which ultimately stipulated the future reporting frequency in the European countries belonging to the EU. While all countries in our sample were not affected by the Transparency Directive, country-specific legislation or the individual stock exchanges in Europe stipulated the previous reporting frequency. To gather the data for our instrument, we collect information about the EU-15 countries from Ernstberger et al. (2017) and by contacting the main stock exchanges for the remaining countries. We expect path-dependency of disclosure regulation to be one way in which the situation in 2004 affects firms’ reporting frequency choice in 2020. Due to path-dependency, the effect from regulation remains after the regulation is revoked, because firms continue to be indirectly affected by the regulation through institutional expectations and procedures shaped according to the prior regulation. Moreover, the instrument proxies for countries’ cultural attitude towards more frequent reporting. Link (2012) exemplifies how Denmark, Netherlands, and the UK, where quarterly reporting was voluntary, opposed the suggested introduction of mandatory quarterly reporting during the discussions leading up to the Transparency Directive 2004/109/EC.

To support our claim that there is a causal effect of reporting frequency on cost behaviour, the instrument must satisfy the relevance condition and exclusion restriction. Column (1) of Table 4 shows that our instrument is correlated with firms’ reporting frequency in 2020 after controlling for the standalone control variables in Eq. (2). As such, the relevance condition is fulfilled. The exclusion restriction requires the instrument to be correlated with cost behaviour only through its effect on firms’ reporting frequency in 2020. While we cannot explicitly test that restriction, we find it unlikely that the countries’ reporting frequency requirement in 2004 had any systematic and direct impact on firms’ cost behaviour in 2020. Our country-level instrument may yield biased estimates if cultural attitudes towards quarterly reporting (external reporting) relate to attitudes towards internal reporting, which in turn affects cost management practices. We are not aware of any empirical evidence suggesting such a relation, however, we are unable to formally rule it out. Link (2012) empirically shows that the choice to voluntarily report quarterly is associated with the level of investor protection in the country. Stronger investor protection could potentially be associated with cultures with a preference towards better external and internal reporting. Following Link (2012), we use the anti-self-dealing-index by (Djankov et al. 2008) as proxy for investor protection, and with a univariate regression (untabulated) we find that investor protection is negatively correlated with our instrument. This suggests that the instrument is not an indirect proxy for stronger investor protection.

Table 4. Reduced-form instrumental variable approach.

	(1)	(2)	(3)
	QTRi	ΔlnXOPRi	ΔlnXOPRi
REGc	0.4579***	0.0234**	0.0495***
	(28.08)	(2.15)	(4.41)
MVi	0.0684***	0.0067***	0.0037
	(22.27)	(4.13)	(1.54)
AINTi	−0.0046	−0.0113	0.0311**
	(−0.63)	(−1.46)	(2.64)
ΔGDPc	1.1319***	0.1419	0.3869
	(5.59)	(0.86)	(1.55)
ΔlnSALEi		0.5315**	0.3480
		(2.08)	(0.83)
ΔlnSALEi × REGc		0.1900***	0.0392
		(4.35)	(0.71)
ΔlnSALEi × MVi		−0.0047	0.0211
		(−0.31)	(0.88)
ΔlnSALEi × AINTi		−0.0462**	−0.1769***
		(−2.10)	(−9.95)
ΔlnSALEi×ΔGDPc		0.9323	0.2665
		(1.17)	(0.29)
ΔlnSALEi × DECi			0.2089
			(0.49)
ΔlnSALEi × DECi × REGc			0.2184**
			(2.51)
ΔlnSALEi × DECi × MVi			−0.0305
			(−1.22)
ΔlnSALEi × DECi × AINTi			0.1511***
			(5.31)
ΔlnSALEi × DECi×ΔGDPc			0.9830
			(0.66)
Fixed effects	Industry	Industry	Industry
Adj. R-sq.	0.277	0.559	0.593
Obs.	3197	3197	3197

Note: The table reports the results for reduced-form instrumental variable analyses that examine the difference in cost elasticity and cost asymmetry between quarterly and semi-annual reporters. In Column (1), the dependent variable is one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise (QRT_i). REG_c takes the value one if the firm’s country in 2004 required quarterly reporting at the main stock exchange, and zero otherwise. MV_i is the logarithm of the market value on December 31, 2019. AINT_i is the log-ratio of assets to sales revenue. ΔGDP_c is the change in gross domestic product between H1 2020 and H1 2019. In Columns (2) and (3), the dependent variable is the log-change in operating cost between H1 2020 and H1 2019 (ΔlnXOPR_i). In Column (2), the coefficient on the two-way interaction between ΔlnSALE_i and REG_c is used to test the difference in cost elasticity. In Column (3), the coefficient on the three-way interaction between ΔlnSALE_i, DEC_i, and REG_c is used to test the difference in cost asymmetry. ΔlnSALE_i is log-change in sales revenue between H1 2020 and H1 2019. DEC_i takes the value one if the sales revenue in H1 2020 is smaller than in H1 2019, and zero otherwise. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Column (2) of Table 4 reports the result for the reduced-form instrumental variable approach for cost elasticity. When we replace the indicator for quarterly reporting with $\mathrm{RE}{G}_c$ in Eq. (2), the two-way interaction of interest is positive and statistically significant (coef. = 0.1900, t-stat = 4.35), which supports the results in Table 2. Similarly, when we replace the indicator for quarterly reporting with $\mathrm{RE}{G}_c$ in Eq. (3), the three-way interaction including $\mathrm{RE}{G}_c$ in Column (3) of Table 4 is positive and statistically significant (coef. = 0.2184, t-stat = 2.51) with a similar magnitude as in Table 3. These results corroborate our main findings.

4.3. COVID-19 exposure

The COVID-19 outbreak affected firms differently depending on their producing and selling organisations. We expect our observed relationship to be stronger among firms with greater exposure to the pandemic. Following Koren and Peto (2020), we use the aggregate social distancing exposure of different industries and classify firms belonging to industries in the top quartile of the distribution as highly affected. As such, the classification directly relates to the shock in supply of labour, but we recognise that such supply shocks also cause shortfalls in demand (Guerrieri et al. 2022). For subsamples of highly affected and less affected firms, we estimate regression coefficients for Eq. (2) and (3) augmented with industry fixed effects. Following Holzhacker et al. (2015a) and Hall (2016), we estimate separate regressions for ease of interpretation and subsequently conduct pairwise comparisons of the coefficients of interest using a z-test (Clogg et al. 1995).

Table 5 reports the results. In Panel A, Columns (1) and (2) display the results for cost elasticity for highly and less affected firms, respectively. The coefficient on the main interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$) is statistically significant only with the highly affected subsample (coef. = 0.1619, t-stat = 2.83). Allowing for cost asymmetry in Columns (3) and (4), we note a statistically significant coefficient on the main three-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$) in highly affected firms (coef. = 0.3379, t-stat = 2.89) but not in less affected firms. For the subsample with highly affected firms the coefficient for $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$ is negative and statistically significant (coef. = −0.7098, t-stat = −1.91). When calculating the average operating cost changes, we find that semi-annual reporters without a decline in sales have a 0.86% change in operating costs per 1% change in sales while the corresponding figure for quarterly reporters is 0.78%.¹² In Panel B of Table 5, we test the differences in coefficients between the subsamples. The z-tests indicate that the coefficients of interest are significantly larger in the highly affected subsample compared with the less affected subsample. This suggests that our main results are driven by firms with a greater exposure to the pandemic.

Table 5. Reporting frequency and cost behaviour among differentially affected firms.

Panel A: Cost elasticity and cost asymmetry in highly and less affected firms
	(1)	(2)	(3)	(4)
	ΔlnXOPRi	ΔlnXOPRi	ΔlnXOPRi	ΔlnXOPRi
QRTi	−0.0102	−0.0017	0.0436**	0.0128
	(−0.81)	(−0.09)	(2.56)	(0.59)
MVi	0.0099***	0.0048**	−0.0004	0.0031
	(5.89)	(2.15)	(−0.10)	(0.89)
AINTi	−0.0025	−0.0171	0.0236	0.0239*
	(−0.18)	(−1.62)	(1.29)	(1.75)
ΔGDPc	−0.1859	0.3552**	0.3114	0.2441
	(−0.75)	(2.60)	(0.88)	(0.85)
ΔlnSALEi	0.4476***	0.8588**	−0.7098*	0.7751**
	(3.88)	(2.52)	(−1.91)	(2.32)
ΔlnSALEi × QRTi	0.1619***	0.0439	−0.0830	0.0172
	(2.83)	(1.02)	(−0.82)	(0.41)
ΔlnSALEi × MVi	0.0061	−0.0178	0.0815***	0.0006
	(0.68)	(−0.91)	(3.19)	(0.03)
ΔlnSALEi × AINTi	−0.0362**	−0.0899***	−0.1303***	−0.1977***
	(−2.06)	(−3.80)	(−5.58)	(−5.44)
ΔlnSALEi×ΔGDPc	0.9533	0.5462	−2.1731	1.0841
	(0.52)	(0.84)	(−0.68)	(1.18)
ΔlnSALEi × DECi			1.4159***	0.0208
			(3.71)	(0.03)
ΔlnSALEi × DECi × QRTi			0.3379***	0.0846
			(2.89)	(0.77)
ΔlnSALEi × DECi × MVi			−0.0945***	−0.0248
			(−3.77)	(−0.65)
ΔlnSALEi × DECi × AINTi			0.1022***	0.1633***
			(3.49)	(3.19)
ΔlnSALEi × DECi×ΔGDPc			3.9566	−1.4137
			(1.14)	(−0.85)
Fixed effects	Industry	Industry	Industry	Industry
Adj. R-sq.	0.668	0.446	0.697	0.488
Obs.	1337	1860	1337	1860
Panel B: Tests of differences in coefficients between highly and less affected firms
			Result	p-value
β6ΔlnSALEi × QRTi in Col. (1) > Col. (2)			0.1181**	(0.0493)
β11ΔlnSALEi × DECi × QRTi in Col. (3) > Col. (4)			0.2533*	(0.0577)

Note: In Panel A, the table reports the results for regressions analyses that examine the difference in cost behaviour between quarterly and semi-annual reporters, for highly affected firms in Columns (1) and (3), and less affected firms in Columns (2) and (4) based on Koren and Peto (2020). The dependent variable is the log-change in operating cost between H1 2020 and H1 2019 (ΔlnXOPR_i). In Columns (1) to (2), the coefficient on the two-way interaction between ΔlnSALE_i and QRT_i is used to test the difference in cost elasticity. In Columns (3) to (4), the coefficient on the three-way interaction between ΔlnSALE_i, DEC_i, and QRT_i is used to test the difference in cost asymmetry. ΔlnSALE_i is the log-change in sales revenue between H1 2020 and H1 2019. QRT_i takes the value one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise. DEC_i takes the value one if the sales revenue in H1 2020 is smaller than in H1 2019, and zero otherwise. MV_i is the logarithm of the market value on December 31, 2019. AINT_i is the log-ratio of assets to sales revenue. ΔGDP_c is the change in gross domestic product between H1 2020 and H1 2019. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. In Panel B, the table reports tests of differences in coefficients between the highly and less affected industries in Panel A. Following Clogg et al. (1995), the z-statistic is calculated as follows: $z={\displaystyle \frac{{\widehat{\text{β}}}_1-{\widehat{\text{β}}}_2}{\sqrt{SE{\left({\widehat{\text{β}}}_1\right)}^2+SE{\left({\widehat{\text{β}}}_2\right)}^2}}}$.

4.4. Pre-period analyses

We also expand the sample one year before the COVID-19 outbreak to examine whether the differences in cost behaviour between reporting frequencies existed already before pandemic and whether the differences became more pronounced in the outbreak phase. For this purpose, we collect reporting frequency data from Bloomberg for the first half of 2019 and estimate regression coefficients of Eq. (2) and (3) with industry fixed effects based on the log-change in operating cost and sales between H1 2019 and H1 2018. We compare the separate coefficients obtained with our pre-period sample with the coefficients from Table 2 and 3 using z-tests following Clogg et al. (1995), Holzhacker et al. (2015a), and Hall (2016).

Column (1) in Panel A of Table 6 reports the results for cost elasticity where the coefficient on the two-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$) is positive but insignificant (coef. = 0.0324, t-stat = 1.13). Column (2) reports an insignificant coefficient on the three-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$), showing no evidence of reporting frequency related differences in asymmetric cost behaviour in the pre-period.¹³ In Column (2), the coefficients suggest that semi-annual reporters with a sales decline on average have a 0.12% lower change in operating costs per 1% change in sales, where the equivalent difference is 0.13% for quarterly reporters.¹⁴ Panel B of Table 6 presents the tests of differences in coefficients between the main results and the coefficients in Panel A of Table 6. We note that the difference in coefficients on $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$ between Column (3) of Table 2 and Column (1) in Panel A of Table 6 is 0.1038 and that the difference in three-way interaction coefficients ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$) between Column (3) of Table 3 and Column (2) in Panel A of Table 6 is 0.2659. Both z-tests indicate statistically significant differences at conventional levels. Taken together, the results in Table 6 show that there was no statistically significant difference between quarterly reporters and semi-annual reporters before the pandemic but that quarterly reporters adjusted their cost structures more than semi-annual reporters in response to the COVID-19 outbreak.

Table 6. Reporting frequency and cost behaviour before the COVID-19 outbreak.

Panel A: Cost elasticity and cost asymmetry regression results
	(1)	(2)
	ΔlnXOPRi	ΔlnXOPRi
QRTi	−0.0041	−0.0018
	(−0.38)	(−0.20)
MVi	−0.0007	−0.0022
	(−0.45)	(−1.05)
AINTi	0.0308***	0.0255***
	(2.86)	(2.88)
ΔGDPc	0.0383	0.0472
	(0.09)	(0.10)
ΔlnSALEi	−0.0028	−0.1399
	(−0.01)	(−0.69)
ΔlnSALEi × QRTi	0.0324	0.0238
	(1.13)	(0.35)
ΔlnSALEi × MVi	0.0306**	0.0414***
	(2.62)	(3.52)
ΔlnSALEi × AINTi	−0.1179***	−0.1213***
	(−7.64)	(−10.56)
ΔlnSALEi×ΔGDPc	3.7216	3.6000
	(1.54)	(1.47)
ΔlnSALEi × DECi		0.4151
		(0.96)
ΔlnSALEi × DECi × QRTi		0.0057
		(0.05)
ΔlnSALEi × DECi × MVi		−0.0304
		(−1.32)
ΔlnSALEi × DECi × AINTi		0.0096
		(0.29)
ΔlnSALEi × DECi×ΔGDPc		−0.3105
		(−0.06)
Fixed effects	Industry	Industry
Adj. R-sq.	0.565	0.571
Obs.	3048	3048
Panel B: Tests of differences in coefficients before and during COVID-19
	Result	p-value
Β6ΔlnSALEi × QRTi in Table 2 Col. (3) > Table 6 Col. (1)	0.1038**	(0.0452)
β11ΔlnSALEi × DECi × QRTi in Table 3 Col. (3) > Table 6 Col. (2)	0.2659**	(0.0417)

Note: In Panel A, the table reports the results for regressions analyses that examine the difference in cost behaviour between quarterly and semi-annual reporters. The dependent variable is the log-change in operating cost between H1 2019 and H1 2018 (ΔlnXOPR_i). In Column (1), the coefficient on the two-way interaction between ΔlnSALE_i and QRT_i is used to test the difference in cost elasticity. In Column (2), the coefficient on the three-way interaction between ΔlnSALE_i, DEC_i, and QRT_i is used to test the difference in cost asymmetry. ΔlnSALE_i is the log-change in sales revenue between H1 2019 and H1 2018. QRT_i takes the value one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise. DEC_i takes the value one if the sales revenue in H1 2019 is smaller than in H1 2018, and zero otherwise. MV_i is the logarithm of the market value on December 31, 2018. AINT_i is the log-ratio of assets to sales revenue. ΔGDP_c is the change in gross domestic product between H1 2019 and H1 2018. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. In Panel B, the table reports tests of differences in coefficients between Panel A Column (1) and Column (2) and corresponding coefficients in Column (3) of Table 2 and Column (3) of Table 3. Following Clogg et al. (1995), the z-statistic is calculated as follows: $z={\displaystyle \frac{{\widehat{\text{β}}}_1-{\widehat{\text{β}}}_2}{\sqrt{SE{\left({\widehat{\text{β}}}_1\right)}^2+SE{\left({\widehat{\text{β}}}_2\right)}^2}}}$.

4.5. Managerial learning and external monitoring

Higher reporting frequency could facilitate managerial learning internally and monitoring externally. To support the proposed channels behind our main results, we additionally test how the speed of earnings announcements, information content in conference calls, and analyst behaviour differ between quarterly reporters and semi-annual reporters in 2020, relative to 2019. First, we use the logarithm of the number of calendar days between the fiscal year-end (December 31) and the date of the full-year earnings announcement multiplied by negative one ($\mathrm{FYSPEE}{D}_{\mathrm{it}}$) as the dependent variable, based on data from Bloomberg. The earnings announcement speed serves as a proxy for the internal information environment (Leventis and Weetman 2004, Gallemore and Labro 2015, Cheng et al. 2018). We also use the speed of the half-year earnings announcement ($\mathrm{HYSPEE}{D}_{\mathrm{it}}$). On average, quarterly (semi-annual) reporters announced their full-year earnings after 64.2 (94.4) and 67.1 (104.2) days in 2019 and 2020, respectively.¹⁵ In the regressions, the independent variable of interest is the interaction $\mathrm{COVI}{D}_t\times \mathrm{QR}{T}_{\mathrm{it}}$, whose coefficient captures the difference in announcement speed of quarterly reporters in 2019 and 2020 relative to semi-annual reporters. $\mathrm{COVI}{D}_t$ takes the value one for announcements in 2020, and zero for the pre-period announcements. We add common firm characteristics as control variables and include firm fixed effects to control for potential omitted time-invariant firm characteristics.¹⁶ Column (1) in Panel A of Table 7 reports a positive and statistically significant (coef. = 0.0580, t-stat = 6.95) coefficient on the interaction $\mathrm{COVI}{D}_t\times \mathrm{QR}{T}_{\mathrm{it}}$. This coefficient is also positive and statistically significant (coef. = 0.0287, t-stat = 2.77) in Column (2), with $\mathrm{HYSPEE}{D}_{\mathrm{it}}$ as the dependent variable. These coefficients indicate a relative decrease in the preparation time for quarterly reporters in 2020, potentially because more frequent reporting is associated with faster information collection and dissemination.

Table 7. Managerial learning and external monitoring.

Panel A: Managerial learning
	(1)	(2)	(3)	(4)
	FYSPEEDit	HYSPEEDit	WORDCOUNTit	NUMCOUNTit
COVIDt	−0.0820***	−0.0409***	−0.0176	−0.1755
	(−10.04)	(−4.84)	(−0.24)	(−1.57)
COVIDt × QRTit	0.0580***	0.0287***	0.1555**	0.2599**
	(6.95)	(2.77)	(2.23)	(2.46)
MVit	0.0108	0.0106	0.0683	0.2053
	(1.01)	(0.80)	(0.85)	(1.62)
MBit	0.0016**	0.0016	0.0044	−0.0023
	(2.37)	(1.39)	(0.91)	(−0.34)
LEVit	−0.0096	−0.1408**	−0.6520**	0.2186
	(−0.22)	(−2.51)	(−2.16)	(0.50)
TANGit	0.0411	0.1227	0.9072**	0.0018
	(0.63)	(1.43)	(2.03)	(0.00)
CASHit	−0.0078	−0.0057	0.5371	1.0553*
	(−0.16)	(−0.11)	(1.38)	(1.79)
ROAit	0.0956***	−0.0173	0.1984	−0.0630
	(3.69)	(−0.31)	(0.38)	(−0.06)
Fixed effects	Firm	Firm	Firm	Firm
Adj. R-sq.	0.887	0.877	0.666	0.568
Obs.	5466	5296	1214	1214
Panel B: External monitoring
	(1)	(2)	(3)	(4)
	UPDATEit	NEWit	CONFIRMit	ANALYSTSit
COVIDt	1.5355***	0.9655***	0.5700***	−0.8047***
	(5.55)	(4.87)	(5.93)	(−3.54)
COVIDt × QRTit	0.9201***	0.6239***	0.2962***	0.7645***
	(3.51)	(3.22)	(3.18)	(4.18)
MVit	0.5717*	0.4800**	0.0917	0.3920**
	(1.73)	(2.12)	(0.72)	(2.17)
MBit	0.0668	0.0207	0.0461	0.0082
	(0.96)	(0.53)	(1.42)	(0.22)
LEVit	−1.7687	−0.9794	−0.7893	−2.3959**
	(−1.28)	(−1.22)	(−1.11)	(−2.33)
TANGit	1.5693	0.4979	1.0714	4.9529**
	(1.05)	(0.44)	(1.47)	(2.59)
CASHit	−1.4636	−1.4689	0.0053	1.3610
	(−1.18)	(−1.66)	(0.01)	(1.14)
ROAit	−0.4311	−0.4038	−0.0272	3.4796**
	(−0.33)	(−0.45)	(−0.05)	(2.15)
Fixed effects	Firm	Firm	Firm	Firm
Adj. R-sq.	0.781	0.771	0.692	0.804
Obs.	2558	2558	2558	1278

Note: In Panel A, the table reports the results for regressions analyses that examine the difference in earnings announcement speeds and conference call content between quarterly and semi-annual reporters. In Column (1), the dependent variable is the logarithm of the full-year earnings announcement speed in days, multiplied by −1 (FYSPEED_it). In Column (2), the dependent variable is the logarithm of the half-year earnings announcement speed in days, multiplied by −1 (HYSPEED_it). In Column (3), the dependent variable is the logarithm of the word count in the statements of managers in the discussion period of the half-year conference call (WORDCOUNT_it). In Column (4), the dependent variable is the logarithm of the number count (excluding years) in the statements of managers in the discussion period of the half-year conference call (NUMCOUNT_it). In Panel B, the table reports the results for regressions analyses that examine the difference in analyst behaviour between quarterly and semi-annual reporters. In Column (1), the dependent variable is the number of analysts that updated their sales forecast for the firm (NEW_it) plus the number of analysts that confirmed their sales forecast (CONFIRM_it) during the first 90 days of the calendar year (UPDATE_it). In Columns (2) and (3), the dependent variable is NEW_it and CONFIRM_it, respectively. In Column (4), the dependent variable is the number of analysts attending the half-year conference call (ANALYSTS_it). The coefficient on the two-way interaction between COVID_t and QRT_it is used to test the difference between quarterly reporters in 2019 and 2020 relative to semi-annual reporters. COVID_t takes the value one if the fiscal year is 2020, and zero otherwise. QRT_it takes the value one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise. MV_it is the logarithm of the market value on December 31 each year. MB_it is the ratio of market value to equity book value. LEV_it is the ratio of total debt to assets. TANG_it is the ratio of net property, plant, and equipment to assets. CASH_it is the ratio of cash and short-term investments to assets. ROA_it is the ratio of earnings before interest and taxes to assets. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Second, we focus on statements of managers during the discussion period (or Q&A) of conference calls for the half-year financial results with a restricted sample based on available transcripts in Refinitiv. Following Matsumoto et al. (2011), we use the logarithm of answer length in words ($\mathrm{WORDCOUN}{T}_{\mathrm{it}}$) as a measure of information content. We expect longer answers for quarterly reporters if more frequent reporting improves the information sets of managers. On average, the number of words spoken by managers of quarterly (semi-annual) reporters is 2694.6 (2793.9) and 3060.1 (2936.7) in 2019 and 2020, respectively. With a formal test, Column (3) in Panel A of Table 7 show a positive and statistically significant (coef. = 0.1555, t-stat = 2.23) coefficient on the interaction which indicates more information coming from quarterly reporters during the COVID-19 outbreak. We also expect more concrete answers for quarterly reporters. Following Elliott et al. (2015), we associate specific numbers or digits with concrete language and use the logarithm of the number count (excluding years) during conference calls ($\mathrm{NUMCOUN}{T}_{\mathrm{it}}$) as the dependent variable in Column (4) in Panel A of Table 7. The coefficient on the interaction is positive and statistically significant (coef. = 0.2599, t-stat = 2.46) suggesting that more quantitative information is communicated by quarterly reporting firms during the COVID-19 outbreak.

Next, we turn to external monitoring, and examine how sales forecasts of sell-side equity analysts change for quarterly reporters and semi-annual reporters in the first three months of 2020 (Q1), relative to Q1 2019 with a sample of available analyst data from Refinitiv I/B/E/S Estimates. Column (1) in Panel B of Table 7 show regression results for the indicator dependent variable $\mathrm{UPDAT}{E}_{\mathrm{it}}$ (new or confirmed sales forecast for firm i in period t). We note a positive and statistically significant (coef. = 0.9201, t-stat = 3.51) coefficient on the interaction. When we use the components separately ($\mathrm{NE}{W}_{\mathrm{it}}$ and $\mathrm{CON}{F}_{\mathrm{it}}$) in Columns (2) and (3), we continue to see positive and statistically significant coefficients on the interaction, suggesting that analysts more frequently updated their sales forecasts of quarterly reporters relative to semi-annual reporters.¹⁷

Finally, Column (4) in Panel B of Table 7 reports regression results where the number of analysts attending the half-year conference call is the dependent variable based on transcripts from Refinitiv. On average, the number of analysts attending conference calls of quarterly (semi-annual) reporters are 5.48 (5.54) and 5.56 (4.96) in 2019 and 2020, respectively. In the regression, we note a positive and statistically significant (coef. = 0.7645, t-stat = 4.18) coefficient on the interaction. Taken together, Panel B of Table 7 provide suggestive results of increased and timelier external monitoring among firms with higher reporting frequency.

4.6. Market reactions

Based on the benefits of a more flexible and elastic cost structure (Holzhacker et al. 2015a), the cost behaviour we observe in Table 2 and Table 3 suggests that investors may prefer quarterly reporters. If investors perceive the cost behaviour of quarterly reporters as beneficial, we expect higher abnormal returns for them compared to semi-annual reporters. To examine investor perceptions, we conduct an event study on 2020 half-year earnings announcements using a restricted sample due to additional data requirements. These earnings announcements contained the information needed to assess the cost behaviour of firms during the outbreak of COVID-19. Using the Fama and French (2015) European five-factor model and the estimation period [−270, −21], we estimate cumulative abnormal returns for [−3, 3] and [−5, 5]. When regressing the return windows on $\mathrm{QR}{T}_i$ and $\mathrm{\Delta ln}\mathrm{SAL}{E}_i$ to control for the content of the earnings announcement, Columns (1) and (2) of Table 8 report significantly more positive abnormal returns for quarterly than for semi-annual reporters. The estimated economic magnitude in Column (2) suggests that firms reporting quarterly had 1.91 percentage points higher abnormal returns to the half-year earnings announcement. This finding indicates that investors perceived the cost behaviour of quarterly reporters as beneficial. In addition, the results suggest that quarterly reporting was an important indicator of share price resilience during the COVID-19 outbreak.

Table 8. Cumulative abnormal returns around H1 2020.

Dependent variable:	(1)	(2)
	CARi[−3, 3]	CARi[−5, 5]
QRTi	0.0151*	0.0191**
	(1.89)	(2.34)
ΔlnSALEi	0.0035	0.0024
	(0.73)	(0.47)
Fixed effects	Industry, Country	Industry, Country
Adj. R-sq.	0.014	0.022
Obs.	2653	2653

Note: The table reports the results for regressions analyses that examine the difference in cumulative abnormal returns surrounding the 2020 half-year earnings announcement between quarterly and semi-annual reporters. Cumulative abnormal returns (CAR_i) for two windows were calculated using the Fama and French (2015) European five-factor model and the estimation period of [−270, −21]. The dependent variables in Columns (1) and (2) are CAR[−3, 3] and CAR[−5, 5], respectively. The coefficient on QRT_i is used to test the difference. QRT_i takes the value one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise. ΔlnSALE_i is the log-change in sales revenue between H1 2020 and H1 2019. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

4.7. Firm performance

While our results indicate that quarterly reporters have more efficient cost structures relative to semi-annual reporters during the COVID-19 outbreak, the performance implications are unclear. Hence, we next compare accounting profitability during the crisis years for quarterly reporters relative to semi-annual reporters after controlling for normal circumstances. For the test, we use a sample covering fiscal years 2019–2021 and $\mathrm{RO}{A}_{\mathrm{it}}$ and $\mathrm{RO}{E}_{\mathrm{it}}$ as dependent variables.¹⁸ The independent variables of interest are interactions between the separate crisis years and reporting frequency ($Y{2020}_t\times \mathrm{QR}{T}_{\mathrm{it}}$ and $Y{2021}_t\times \mathrm{QR}{T}_{\mathrm{it}}$). As control variables, we include fiscal year dummies and firm characteristics together with firm fixed effects.

In Column (1) of Table 9, where $\mathrm{RO}{A}_{\mathrm{it}}$ is the dependent variable, we note a positive and statistically significant coefficient on $Y{2020}_t\times \mathrm{QR}{T}_{\mathrm{it}}$ (coef. = 0.0129, t-stat = 2.69). This coefficient suggests that quarterly reporters performed better than semi-annual reporters in fiscal year 2020, after considering their performance difference in the pre-period. The coefficient on $Y{2021}_t\times \mathrm{QR}{T}_{\mathrm{it}}$ is positive, however, statistically insignificant (coef. = 0.0046, t-stat = 0.78). With $\mathrm{RO}{E}_{\mathrm{it}}$ as the dependent variable, Column (2) reports a positive and statistically significant coefficient on $Y{2020}_t\times \mathrm{QR}{T}_{\mathrm{it}}$ (coef. = 0.0520, t-stat = 1.93) and an insignificant coefficient on $Y{2021}_t\times \mathrm{QR}{T}_{\mathrm{it}}$ (coef. = 0.0333, t-stat = 1.12). Taken together, the results in Table 9 indicate that quarterly reporters performed better after the immediate COVID-19 outbreak but that this effect did not persist in 2021. Potentially, the cost elasticity of quarterly reporters enabled the superior performance in the short run.

Table 9. Firm performance.

Dependent variable	(1)	(2)
	ROAit	ROEit
Y2020t	−0.0043	−0.0760***
	(−0.82)	(−3.06)
Y2021t	0.0242***	0.0149
	(3.82)	(0.63)
Y2020t × QRTit	0.0129***	0.0520*
	(2.69)	(1.93)
Y2021t × QRTit	0.0046	0.0333
	(0.78)	(1.12)
MVit	−0.0467***	−0.1932***
	(−4.65)	(−5.40)
MBit	0.0000	0.0214***
	(0.02)	(4.22)
LEVit	0.0421	0.3313*
	(1.53)	(1.74)
TANGit	−0.0390	−0.1400
	(−1.20)	(−0.81)
CASHit	−0.0372	0.0829
	(−1.37)	(0.78)
Fixed effects	Firm	Firm
Adj. R-sq.	0.779	0.435
Obs.	7995	7455

Note: The table reports the results for regressions analyses that examine the difference in firm performance between quarterly and semi-annual reporters. In Column (1), the dependent variable is the ratio of earnings before interest and taxes to assets (ROA_it). In Column (2), the dependent variable is the ratio of income before extraordinary items to equity (ROE_it). The coefficient on the two-way interaction between Y2020_t (Y2021_t) and QRT_it is used to test the difference in firm performance of quarterly reporters between 2019 and 2020 (2021) relative to semi-annual reporters. Y2020_t (Y2021_t) takes the value one if the fiscal year is 2020 (2021), and zero otherwise. QRT_it takes the value one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise. MV_it is the logarithm of the market value on December 31 each year. MB_it is the ratio of market value to equity book value. LEV_it is the ratio of total debt to assets. TANG_it is the ratio of net property, plant, and equipment to assets. CASH_it is the ratio of cash and short-term investments to assets. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

4.8. Insider trading and seasoned equity offerings

To examine whether higher cost elasticity is indicative of myopic actions among quarterly reporters, we study insider trading and equity offerings. We expect to see more insiders selling and a greater likelihood of firms issuing equity if managers and board members view the cost adjustments as value-destroying (Bhojraj et al. 2009). First, we use the sum of the proportion of shares outstanding held by non-employee directors plus the proportion held by executives at the end of the year ($\mathrm{INSIDE}{R}_{\mathrm{it}}$) from Bloomberg as the dependent variable. We then estimate regressions for a restricted sample comparing quarterly and semi-annual reporters in 2020, relative to the figures in 2019.¹⁹ In Column (1) of Table 10, the coefficient on $\mathrm{COVI}{D}_t\times \mathrm{QR}{T}_{\mathrm{it}}$ is positive and statistically significant (coef. = 0.0092, t-stat = 2.10). This suggests that in 2020, insiders of quarterly reporters were increasing their ownership more than insiders of semi-annual reporters.

Table 10. Insider holdings and seasoned equity offerings.

Dependent variable	(1)	(2)
	INSIDERit	SEOit
COVIDt	−0.0056	0.0309***
	(−1.32)	(3.02)
COVIDt × QRTit	0.0092**	−0.0162
	(2.10)	(−1.38)
MVit	0.0005	−0.1166***
	(0.10)	(−3.69)
MBit	0.0001	−0.0003
	(0.29)	(−0.24)
LEVit	0.0096	0.2477***
	(0.29)	(3.43)
TANGit	−0.0460	−0.1429
	(−1.19)	(−1.48)
CASHit	−0.0125	−0.2456***
	(−0.51)	(−2.98)
ROAit	0.0018	−0.0987
	(0.07)	(−1.06)
Fixed effects	Firm	Firm
Adj. R-sq.	0.860	0.217
Obs.	1370	5636

Note: The table reports the results for regressions analyses that examine the difference in insider holdings and seasoned equity offerings between quarterly and semi-annual reporters. In Column (1), the dependent variable is the proportion of insider holdings (INSIDER_it). In Column (2), the dependent variable is an indicator variable taking the value one if the company issued equity, and zero otherwise (SEO_it). The coefficient on the two-way interaction between COVID_t and QRT_it is used to test the difference in analyst reactions to quarterly reporters in 2019 and 2020 relative to semi-annual reporters. COVID_t takes the value one if the fiscal year is 2020, and zero otherwise. QRT_it takes the value one if the firm reports quarterly or semi-annually with business reviews, and zero otherwise. MV_it is the logarithm of the market value on December 31 each year. MB_it is the ratio of market value to equity book value. LEV_it is the ratio of total debt to assets. TANG_it is the ratio of net property, plant, and equipment to assets. CASH_it is the ratio of cash and short-term investments to assets. ROA_it is the ratio of earnings before interest and taxes to assets. A constant term is included in all estimations, but not reported. t-statistics, reported in the parentheses, are based on standard errors clustered at the industry level (2-digit SIC). ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Second, we use data from the Securities Data Corporation and create an indicator variable taking the value one if the firm issued seasoned equity during the second half of the fiscal year, and zero otherwise ($\mathrm{SE}{O}_{\mathrm{it}}$). In Column (2) of Table 10, we examine $\mathrm{SE}{O}_{\mathrm{it}}$ using a linear probability model. The coefficient on the interaction is negative, however, not statistically significant (coef. = −0.0162, t-stat = −1.38). Taken together, the results in Table 10 do not suggest that managers of quarterly reporters were engaging in value-destroying cost cutting. The results instead point to the opposite, especially because insiders of quarterly reporters increased their holdings in 2020.

4.9. Additional sensitivity analyses

We conduct several untabulated additional tests to examine how sensitive our results are to an alternative crisis setting, alternative variable definitions for cost and reporting frequency, extreme observations, and firm- and country-specific factors that we do not control for in the main specifications. First, we check whether our results are robust to an alternative crisis setting by using the global financial crisis of 2008–2009 as the shock. Based on available data on reporting frequency in Bloomberg and firm financials in Compustat Global, we obtain a sample of 2834 firms. We conduct the analyses based on the log-change in operating cost and sales between H1 2009 and H1 2008.^20,21 We find that the coefficient on the two-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$) is positive and statistically significant (coef. = 0.1821, t-stat = 2.82) also during the global financial crisis. When we examine cost asymmetry, the coefficient on the three-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$) is positive but insignificant. We reason that differences in crisis characteristics explain this result, such as large differences regarding sectorial effects and the chain of events. For example, the number of European non-financial firms experiencing decreases in sales just after the collapse of Lehman was relatively small, indicating a slower spread than in the case of COVID-19.

Second, we examine whether the behaviour of other cost categories vary between quarterly and semi-annual reporters in the same way as aggregated operating costs. For this purpose, we use the log-change in operating costs including depreciation, the log-change in selling, general and administrative costs, and the log-change in staff costs as dependent variables following Anderson et al. (2003), Dierynck et al. (2012), and Banker et al. (2013). We also examine cash outflows more generally by using the log-change in capital expenditures. The analyses reveal that the results are consistent with the main results when examining operating costs including depreciation and with selling, general, and administrative costs. The results are insignificant when examining staff costs, possibly due to sample attrition. With capital expenditures, we note a significant effect regarding cost elasticity and an insignificant effect regarding cost asymmetry. These analyses indicate that the most discretionary of the cost types drive our main results while quarterly reporters did not necessarily cut long-run expenditures more than semi-annual reporters.

Third, in the main tests we code firms reporting semi-annually with business reviews as quarterly reporters. Since these are not pure quarterly reporters, we exclude them from the regressions and run the analyses based on the remaining 2831 observations. The inferences from our main results remain unchanged with this approach. We also find consistent results when we exclude firms that required manual data collection. Because the financial reporting frequency choice could coincide with the pandemic, we additionally confirm that the results hold when we use a lagged $\mathrm{QR}{T}_i$ variable.

Fourth, we examine whether extreme observations affect our results. We follow the suggestion of Anderson and Lanen (2007) and exclude observations where costs move in the opposite direction of sales (e.g. cost increases following sales declines). We also follow Cannon (2014) and remove observations with absolute studentised residuals greater than 3. In addition, we follow Holzhacker et al. (2015a) and winsorise the upper and lower 0.5 percentile tails of the distribution for log-changes in cost and sales to reduce the influence of outliers. Based on this sensitivity check, we conclude that extreme observations influence our cost elasticity results primarily when winsorising, since the coefficient on $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$ is positive but no longer statistically significant. Meanwhile, the cost asymmetry results remain intact with all three approaches.²² To mitigate concerns over outliers, we also conduct manual checks of observations with absolute studentised residuals greater than 2 using quarterly and semi-annual reports from corporate websites. These manual checks confirm that the extreme values in our data set are accurate and consistent with reality.

Fifth, cost structure and cost behaviour may also vary across firms for reasons unrelated to our control variables. Therefore, we additionally control for managerial pessimism with an indicator for prior sales decreases following Banker et al. (2014b). In our estimation, prior sales decreases are associated with anti-sticky costs, however, our main results remain qualitatively and quantitatively similar. Recent COVID-19 studies (Fahlenbrach et al. 2021, Demers et al. 2021, Ding et al. 2021) highlight financial flexibility as a resilience factor. Since large cash holdings can be used to avoid cost cuts, we control for cash to assets. The results suggest that more cash holdings are associated with lower cost elasticity, but our main results remain intact. Further, we ensure that our results are not an artefact of different operating leverage levels (Banker et al. 2013, Lev and Thiagarajan 1993, Weiss 2010), by using the log-ratio of net PP&E to sales and gross margin as additional explanatory variables. Our results remain unaffected by the addition of operating leverage as control variables. Following Chen et al. (2012), Gigler (2014), and Wagenhofer (2014) we also control for corporate governance and shareholder base. We retrieve data on the structure of the board and free-floating shares from Refinitiv Eikon and Datastream, respectively. Then we use board size (logarithm of board members), board independence (percentage of independent board members), and institutional holdings (percentage of free-floating shares) as additional controls. The additional corporate governance controls reduce the sample size to 891 observations. The takeaways regarding cost elasticity remain unchanged, however, the difference in cost asymmetry between quarterly and semi-annual reporters is no longer statistically significant with the small sample. Meanwhile, our results remain unaffected to the inclusion of the investor base control.

Finally, we acknowledge that country factors other than GDP growth explain cross-country variation in cost behaviour. Following Banker et al. (2013), we control for employment protection legislation. We also include a common-law indicator for Ireland and the UK since Djankov et al. (2007) note that the legal origin of a country is a driver of several factors (e.g. corporate governance, access to financing, and business regulation) that are likely to affect firm-level cost behaviour. Neither inclusion of these additional country variables nor inclusion of small countries with less than 30 observations (Czech Republic, Estonia, Ireland, Lithuania, Luxembourg, Malta, Portugal, and Slovenia) nor inclusion of country fixed effects alter our takeaways.

5. Conclusion

Our study examines real effects of reporting frequency. We find that firms reporting quarterly were better able to cushion the blow and adapt to the sudden shock of COVID-19. Specifically, our results show that quarterly reporters had greater operating cost elasticity than semi-annual reporters during the COVID-19 outbreak in the first half of 2020. A more elastic cost structure allows for more flexibility in uncertain times. When allowing for asymmetrical cost behaviour, we find that the difference in cost elasticity originates from firms with sales decreases. We reason that higher reporting frequency could facilitate managerial learning internally and monitoring externally and these could be potential channels behind our results. To give more context to our findings, we investigate market reactions to the 2020 half-year earnings announcements when the cost behaviour of all sample firms became public. We find that firms reporting quarterly had 1.91 percentage points higher abnormal returns than semi-annual reporters. Such a significant difference is also visible in terms of short-run accounting profitability, potentially due to greater cost elasticity.

How generalisable are our results? We acknowledge that our sample does not include all publicly listed firms in Europe. However, we cover most of the total market capitalisation and the pandemic resulted in a shock to activity levels around the world. Hence, our results are at least generalisable to other listed firms, countries, or worldwide large variations in activity levels. We also replicate the cost elasticity findings using the global financial crisis of 2008–2009, which indicates that our results should also be generalisable to previous and forthcoming crises. Because we do not examine unlisted firms, we are hesitant to generalise our findings to firms without external monitoring from capital markets.

Our study is subject to limitations. Due to its archival nature, we are unable to avoid endogeneity concerns. For example, variations in internal information quality (Kim et al. 2022) or firm transparency (Nallareddy et al. 2021) associated with both reporting frequency and cost behaviour may bias our estimates. Moreover, our sample is short and future research could investigate the long-run effects of COVID-19 or other similar settings. An interesting avenue for future research would be to examine whether reporting frequency also facilitates cost elasticity for expansions, in periods after crises.

The most pronounced contribution of our study is to the reporting frequency literature. Recent theoretical (Gigler et al. 2014) and empirical (Ernstberger et al. 2017, Kraft et al. 2018, Fu et al. 2020) studies argue that a higher reporting frequency may hurt shareholder value, because it induces managerial short-termism which might lead to suboptimal investments decisions. However, we document that higher reporting frequency may protect shareholder value in a situation of sudden and unexpected shortfall in sales. The empirical results in Downar et al. (2018) similarly highlight that cash holdings are more valuable in firms with a higher reporting frequency, potentially because these firms use their cash more efficiently. It remains an open question whether higher reporting frequency is value enhancing or not and our study provides additional evidence to that discussion.

Acknowledgements

We appreciate helpful comments from Per Olsson (the associate editor), two anonymous reviewers, Mansoor Afzali, Patricia Bromley, Ellie Chapple, Sander De Groote, Peter Frii, Henrik Höglund, Henry Jarva, Bjørn Jørgensen, Juha-Pekka Kallunki, Jukka Kettunen, Eva Labro, Zeping Pan, Zihang Peng, Carsten Rohde, Petri Sahlström, Amir Sasson, David Schröder, Stefan Sundgren, Stephen Taylor, Aljoša Valentinčič, Sami Vähämaa and participants in the 2021 Nordic Accounting Conference, XVI International Accounting Research Symposium and workshop participants at Copenhagen Business School, Stanford University, Queensland University of Technology, Umeå University, University of Oulu, University of Vaasa, University of Technology Sydney, UNSW Sydney, and Victoria University of Wellington. We gratefully acknowledge financial support from the Foundation for Economic Education, Hanken Support Foundation, OP Group Research Foundation, the Finnish Foundation for Share Promotion, the Foundation for the Advancement of Finnish Securities Markets, and research assistance from Jiangnan Zhang.

Disclosure statement

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

Notes

1 The Transparency Directive 2004/109/EC did not require Interim Management Statements to include quantitative sales and earnings numbers. For their European sample, Ernstberger et al. (2017) state that 51.4% of all Interim Management Statements contain quantitative earnings figures.

2 Following Banker et al. (2013) and Prabowo et al. (2018), we do not control for employee intensity because the measure requires total employees which is often not reported in Compustat Global. Thus, including it could lead to sample attrition and distorted results in our cross-country sample.

3 We acknowledge that this forces the intercept to be equal for observations with increases and decreases in sales. To ensure the robustness of our results, we also re-estimate Eq. (3) with the main effect for $\mathrm{DE}{C}_i$. This analysis (untabulated) indicates that our main results are robust to the alternative model specification, and we note that the coefficient on $\mathrm{DE}{C}_i$ is negative and statistically significant.

4 Appendix 1 describes the classification process for reporting frequency in detail.

5 In Compustat Global, the three countries with most observations in our initial sample are the UK (22.9%), Sweden (11.7%), and Poland (10.5%). Out of the observations we drop due to the fiscal year-end requirement, 65.6% are from the UK where it is common with fiscal years ending in March or June (only 5.0% and 2.5% are from Sweden and Poland, respectively).

6 We calculate the cost elasticity by using the following estimated formula: $0.7883\times \mathrm{\Delta ln}\mathrm{SAL}{E}_i+0.1436\times \mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i-0.0168\times \mathrm{\Delta }\mathrm{lnSAL}{E}_i\times M{V}_i-0.0479\times \mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{AIN}{T}_i+0.8140\times \mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{\Delta GD}{P}_i$. In the calculations, we use sample means for $M{V}_i$, $\mathrm{AIN}{T}_i$, and $\mathrm{\Delta GD}{P}_i$, which are 18.5544, 0.4875, and −0.0564, respectively. When calculating the cost elasticity for $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$ equal to 1%, with these input values, we find the operating cost change to be 0.41% for semi-annual reporters and 0.55% for quarterly reporters.

7 When estimating the baseline Anderson et al. (2003) equation in untabulated tests, we observe a significantly positive coefficient (coef. = 0.2006, t-stat = 2.02) on the two-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i$) for quarterly reporters and an insignificant coefficient (coef. = 0.0073, t-stat = 0.06) for semi-annual reporters.

8 We calculate the cost elasticity for observations with declining sales by using the coefficients for $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times M{V}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{AIN}{T}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{\Delta GD}{P}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times M{V}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{AIN}{T}_i$, and $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$. In the calculations, we use sample means of $M{V}_i$, $\mathrm{AIN}{T}_i$, and $\mathrm{\Delta GD}{P}_i$, and $\mathrm{DE}{C}_i$ equal to one.

9 Prior reporting frequency studies have utilised potentially exogenous variation due to regulatory changes in the US and the EU (Butler et al. 2007, Ernstberger et al. 2017). We are not implementing the European setting proposed by Ernstberger et al. (2017), mainly because the regulatory changes do not overlap with our more recent time period. An alternative approach would be to compare cost elasticity for firms under a mandatory quarterly reporting regime with the semi-annual reporters. However, the semi-annual reporters would still have endogenously chosen their reporting frequency with such an approach.

10 In the reduced-form approach the instrument is directly regressed on the outcome. Chernozhukov and Hansen (2008) highlight that the approach yields valid test and confidence intervals with weak instruments (and strong instruments). We are not concerned about a weak instrument, however, the multiple interactions and the fact that the instrument is on the country-level make us prefer the reduced-form approach.

11 In 2004, the country legislation stipulated quarterly reporting for firms listed in Finland, Greece, Italy, Norway, Poland, and Spain. At the same time, quarterly reporting was required by the stock exchange for firms on the main markets in Austria, Croatia, Germany, Romania, and Sweden.

12 We calculate the cost elasticity for observations without declining sales using the coefficients for the following variables $\mathrm{\Delta }\mathrm{lnSAL}{E}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times M{V}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{AIN}{T}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{\Delta GD}{P}_i$ and sample means of $M{V}_i$, $\mathrm{AIN}{T}_i$, and $\mathrm{\Delta GD}{P}_i$.

13 Consistent with cost stickiness under normal circumstances, we observe a significantly negative coefficient (coef. = −0.1847, t-stat = −2.71) on the two-way interaction ($\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i$) when estimating the baseline Anderson et al. (2003) equation for the pre-period in an untabulated test.

14 We calculate the cost asymmetry using the coefficients for $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{QR}{T}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times M{V}_i$, $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{AIN}{T}_i$, and $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{DE}{C}_i\times \mathrm{\Delta GD}{P}_i$. In the calculations, we use sample means of $M{V}_i$, $\mathrm{AIN}{T}_i$, and $\mathrm{\Delta GD}{P}_i$, and $\mathrm{DE}{C}_i$ equal to one.

15 On average, quarterly (semi-annual) reporters announced their half-year earnings after 46.2 (66.9) and 47.0 (69.0) days in 2019 and 2020, respectively.

16 The sample excludes firms with incomplete information for year 2019 and 2020. Further, we also exclude firms that changed their reporting frequency during the sample period.

17 The results are robust to using forecasts of earnings per share (untabulated).

18 We only include firms with observations for all years in the sample. Further, we exclude firms that changed their reporting frequency during the sample period. For the regressions with $\mathrm{RO}{E}_{\mathrm{it}}$ as the dependent variable, we exclude observations with negative $\mathrm{RO}{E}_{\mathrm{it}}$.

19 We restrict the sample to firms with observations for all years of the sample period and without changes in reporting frequency.

20 During the global financial crisis, the half-year with the largest drop in activity level (i.e., sales) in our sample was the first half of 2009.

21 At the end of H1 2009, the Transparency Directive 2004/109/EC was effective and demanded at minimum an Interim Management Statement after Q1 and Q3 for firms listed on the EU-regulated market in Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Portugal, Spain, Sweden, and the UK. The Transparency Directive 2004/109/EC did not require quantitative sales and earnings numbers, hence the variation in reporting frequencies for our alternative crisis sample remained.

22 We note that the coefficients on $\mathrm{\Delta }\mathrm{lnSAL}{E}_i\times \mathrm{QR}{T}_i$ are negative in the cost asymmetry regressions and significantly so with the Cannon (2014) approach. We reason that this is a result of managerial pessimism, so even though sales increase, managers are not prepared to adjust costs upwards by the same proportion.

Appendix 1

This appendix presents details on how we obtain the financial reporting frequency of the firms in our sample. We use data from Bloomberg Finance L.P. to classify part of the firms according to defined rules described in detail below. We begin the process by looking up earnings announcement dates and periods for the firms in our sample using the ISINs as provided by Compustat Global. We classify a firm as a quarterly reporter if the following variables have non-missing information for Q3 2019 (September 30, 2019) and Q1 2020 (March 31, 2020):

• Announcement date (field: EARN_ANN_DT_TIME_HIST_WITH_EPS, item: Announcement Date)

• Announcement period (field: EARN_ANN_DT_TIME_HIST_WITH_EPS, item: Announcement Time)

• Sales revenue (field: SALES_REV_TURN)

• Net income (field: NET_INCOME)

• Accounting standard (field: ACCOUNTING_STANDARD)

As an additional check for stale data, we require that the values for sales revenue and net income have changed since the previous quarter (December 31, 2019 (Q4) or June 30, 2019 (Q2)).

We classify a firm as a semi-annual reporter if all of the aforementioned variables are missing and the primary periodicity (field: PRIMARY_PERIODICITY) is not empty and not set to quarterly. Using this procedure, we classify the reporting frequency of 3300 firms (1670 quarterly reporters and 1630 semi-annual reporters). The remaining firms are manually classified as quarterly, semi-annual, or business review reporters based on corporate filings in Bloomberg for H1 2020. We classify a firm as a business review reporter if the firm reports semi-annually with additional financial reports not containing a complete income statement and balance sheet for Q1 2020. The manual collection yielded 195 quarterly reporters, 403 business review reporters, and 81 semi-annual reporters. Thus, before dropping firms with unavailable data for the variables in Eq. (3) we have 3979 firms. After dropping firms with unavailable financial data for the variables in Eq. (3) we have 3335 firms (1688 quarterly reporters, 372 business review reporters, and 1275 semi-annual reporters). After dropping firms in countries with less than 30 observations (Czech Republic, Estonia, Ireland, Lithuania, Luxembourg, Malta, Portugal, and Slovenia) we have 3197 firms (1611 quarterly reporters, 366 business review reporters, and 1220 semi-annual reporters).

For the additional test with the pre-period in 2019, we use the same procedure and retrieve a usable sample of 3048 firms (1492 quarterly reporters, 319 business review reporters, and 1237 semi-annual reporters). We also use the same procedure for the additional test with the global financial crisis of 2008–2009 and retrieve a usable sample of 2834 firms (1575 quarterly reporters, 170 business review reporters, and 1089 semi-annual reporters).