1 Introduction

In the literature, it has been well documented that revenues disclosed in quarterly financial reports contain information that can predict future earnings growth and explain the post-earnings-announcement drift (PEAD) (Jegadeesh and Livnat 2006a, b; Chen et al. 2014). There is also a growing literature on whether interim information benefits investors. Fu et al. (2012), Gigler et al. (2014), Wagenhofer (2014), and Tsao et al. (2018) show that interim accounting disclosures are beneficial to investors. Using a unique dataset from Taiwan, where listed firms are required to disclose monthly revenues, this paper integrates these two topics (revenue disclosures and interim information) and studies the informativeness of interim revenue disclosures.Footnote 1 Using data with mandatory monthly revenue disclosures gives us two advantages in empirical settings over previous studies. First, we can rule out the potential sample-selection bias that only firms with benefits outweighing costs in disclosing revenue information or firms in some specific sectors are selected into the sample (Van Buskirk 2012; Froot et al. 2017). Second, in contrast to the case of the U.S., where firms’ revenues and earnings are disclosed simultaneously in quarterly financial statements, investors in Taiwan can observe firms’ monthly revenues at least 20 days earlier than the announcements of their quarterly earnings. In this disclosure timeline, the monthly revenue disclosures become a piece of leading information rather than merely a supplement information additional to quarterly earnings. This timeline also helps us better focus on the effects of revenues because monthly revenues are generally the most prominent pieces of fundamental information between earnings announcements. In such a framework, we obtain a clearer picture of whether monthly revenue information itself contributes to stock price changes or analysts’ forecast revisions during the period of two earnings announcements. In sum, this unique empirical setting distinguishes our study from previous ones that test the information effects of revenues based on revenue disclosures from quarterly financial reports (e.g., Ertimur et al. 2003; Jegadeesh and Livnat 2006a, b; Chen et al. 2014). These advantages motivate us to investigate the informativeness of monthly revenues by answering the following questions: (1) Do monthly revenue disclosures have a positive effect on analysts’ earnings forecasts and predict future earnings surprises? (2) How do stock prices react to monthly revenue surprises?

We begin our empirical tests by examining the effect of monthly revenue disclosures on analysts’ following earnings forecast revisions. In stock markets, analysts are an important information source for investors (e.g., Womack 1996; Kothari 2001). If monthly revenue disclosure is relevant to investors, we expect that it will affect analysts’ earnings forecasts. We find that monthly revenue surprises are positively associated with subsequent revisions of analysts’ earnings forecasts. We then turn our focus into quarterly earnings surprises. Following the literature on earnings surprises, our first measure of earnings surprise is standardized unexpected earnings (SUE). Jegadeesh and Livnat (2006b) show that the preceding quarter revenue surprise can help predict future earnings growth, indicating that earnings growth driven by an increase in revenue growth is likely to be more persistent than earnings growth driven by a reduction in expenses. In our paper, investigating the association between monthly revenue surprises and SUE enables us to test whether monthly revenue information provides leading information about the persistence of earnings growth.Footnote 2 We find monthly revenue surprises indeed forecast future earnings growth.

Following Mendenhall (2004) and Jegadeesh and Livnat (2006b), our second measure of earnings surprise is analysts’ earnings forecast errors (hereafter “forecast errors”) since analysts’ forecasts generally capture the most updated market expectations. If analysts fully incorporate the information in monthly revenue surprises into their earnings forecast revisions, we should observe that monthly revenue surprises and analysts’ forecast errors after the revisions are uncorrelated. Otherwise, under-reaction (over-reaction) of analysts would cause a positive (negative) association between revenue surprises and forecast errors. We find that quarterly forecast errors are still positively related to monthly revenue surprises. Together with the results for the analyst revisions, this result suggests that analysts use monthly revenue information to revise earnings expectations, yet the revisions are insufficient and therefore are subject to an underreaction bias. This finding still holds after controlling for past earnings surprises, past analysts’ earnings forecast revisions (hereafter “forecast revision”), and past stock returns, suggesting that the predictive power of monthly revenue surprises on quarterly earnings surprises goes beyond past earnings information. In sum, our finding is consistent with Jegadeesh and Livnat (2006b) that analysts incorporate but underreact to the information conveyed in revenue surprises. In a reporting regime that separates the disclosure timing of earnings and revenue, our results complement and extend theirs by showing that monthly revenue information has its own distinct value that helps stock analysts to revise their earnings forecasts. Shane and Brous (2001) show that analysts use the information other than earnings surprises available between earnings announcements to revise earnings forecasts, and that non-earnings-surprise information helps investors reduce underreaction to past earnings surprises. Our empirical results suggest that interim revenue disclosures are one of the sources of this non-earnings-surprise information.

We then investigate how stock prices react to monthly revenue surprises (Ertimur et al. 2003; Jegadeesh and Livnat 2006a, b; Chen et al. 2014). Using monthly data, we find that stock prices react significantly and positively to monthly revenue surprises. We also find that monthly revenue surprises have a positive post-announcement price drift. These results hold for both small and large stocks, for each of the three months in a quarter, and for both earnings-announcement and non-earnings-announcement months. These findings suggest that interim revenue disclosures do contain information relevant to firm value. When turning our focus into quarterly returns, consistent with Shane and Brous (2001), we find that prices keep drifting with monthly revenue surprises during the quarter prior to an earnings announcement. Unlike Jegadeesh and Livnat (2006b), however, we find that stock prices are no longer driven by revenue surprises after new earnings information is revealed. Overall, our results imply that investors increase the speed of adjustments to fundamental news when interim revenue information is available.

This paper contributes to several strands of literature. First, for the literature on information value of revenue releases, we show that monthly revenue disclosure by itself provides leading information for investors to assess firms’ fundamentals and better predict future earnings surprises. This paper uses the data of firms that make mandatory interim revenue disclosures prior to quarterly earnings announcements. As mentioned, this unique empirical setting mitigates the sample-selection bias concern. It also helps us better focus on the distinct informativeness of monthly revenue between earnings announcements, which differentiate our study from previous ones that use revenues and earnings in the quarterly financial reports (e.g. Ertimur et al. 2003; Jegadeesh and Livnat 2006a, b; Chen et al. 2014).

Second, this paper contributes to the literature discussing the trade-offs between costs and benefits of frequent financial reporting (Fu et al. 2012; Gigler et al. 2014; Wagenhofer 2014; Tsao et al. 2018). It shows that investors and analysts can benefit from the revenue disclosures in a regime in which revenues are disclosed more frequently and separately from earnings disclosures. In this literature, our paper is closely related to Tsao et al. (2018). With the data of Taiwanese listed firms, their paper shows that switching to a higher reporting frequency increases the speed at which information in accruals is incorporated into stock prices and reduces mispricing of accruals. Unlike Tsao et al. (2018), we focus on how financial analysts react to interim revenue disclosures and whether such information helps investors mitigate underreaction to earnings surprises (Ghosh et al. 2005; Jegadeesh and Livnat 2006b; Keung 2010). Finally, this paper provides direct evidence to support the notion of Shane and Brous (2001) that investors can use interim non-earnings-surprise information to correct for their mispricing or delayed reactions to past earnings surprises. We show that monthly revenue reports are a non-earnings-surprise information source that plays this corrective role.

The rest of the paper is organized as follows. Section 2 provides a literature review. Section 3 describes the institutional background of our empirical setting as well as the data used in our empirical tests. Section 4 presents results on whether and how analysts incorporate monthly revenue information into earnings forecasts. Section 5 presents results on how stock prices react to monthly revenue disclosures. Section 6 concludes the paper.

2 Market and institutional background, literature review, and empirical hypotheses

2.1 Market and institutional background

Our empirical tests are based on a sample from the Taiwan stock market. Compared to diversely owned corporations in Western countries, Taiwanese companies are usually owned by families who might have fewer incentives to disclose private information to minority shareholders (Yeh et al. 2001; Claessens et al. 2002; Fan and Wong 2002; Ali et al. 2007).Footnote 3 In addition, most participants in the Taiwan stock market are less sophisticated individual investors whose interpretation of companies’ fundamental information is more likely to be under- or over-reactive to accounting disclosures (Barber et al. 2009; Lin and Swanson 2010).Footnote 4 Hence, to some extent, the Taiwanese market relies on regulatory requirements to disclose more information (for example, the monthly interim revenues investigated in this study) to the public.

The rule of monthly revenue disclosure in Taiwan can be traced back to 1988. Under the listing rules, listed firms in Taiwan must report their monthly operating status, including their preceding month’s unaudited net operating revenues, on the Taiwan Stock Exchange website. According to a legislative document published by the Legislative Yuan of Taiwan, the purpose of mandatory monthly revenue disclosure was to improve the transparency of disclosures and to provide timely information related to public firms’ financial status and operating performance, aiming to protect the rights and interests of investors.Footnote 5

In addition to monthly reporting, as in most western markets, all Taiwan-listed companies also must issue quarterly reports duly reviewed by a certified public accountant. Before 2013, Taiwan public firms must follow the Generally Accepted Accounting Principles in Taiwan (TW GAAP).Footnote 6 Since 2013, all Taiwan-listed companies have been required to adopt the International Financial Reporting Standards (IFRS).Footnote 7Footnote 8 Note that, both before and after the enforcement of the IFRS, the monthly revenues are disclosed earlier than the quarterly reports by at least 20 days.

Figure 1 shows the event timeline of information available to investors in a typical year, including twelve monthly revenue disclosures, twelve analysts’ earnings forecasts, and four quarterly earnings announcements. In the timeline, EAM indicates the earnings announcement months we defined to separate the effect of revenue surprises from earnings announcements.

Fig. 1
figure 1

Timeline of information available to investors in a typical year in Taiwan. In the figure, there are twelve monthly revenue disclosures for monthly revenue surprises (SURGEM), twelve monthly consensus earnings forecasts for the forecast revision (FRM), and four quarterly earnings announcements for the earnings surprises (SURPRISE). These pieces of information are gathered in the following schedule: before the tenth day of a month for the monthly revenue, at the end of each month for the earnings forecast, and within a quarter for the preceding quarter’s earnings. SURGEM is defined as unexpected revenue growth divided by the standard deviation of unexpected revenue growth. FRM is the mean analysts’ forecast of annual EPS for the current year minus the preceding month’s mean annual forecast EPS, divided by the stock price at the end of the preceding month. Earnings surprises (SURPRISE) could be standardized unexpected earnings (SUE) or the forecast error (FE). SUE is the unexpected earnings growth (actual EPS minus expected EPS) divided by the standard deviation of unexpected earnings growth. Forecast errors are defined as actual quarterly EPS minus the mean analysts’ forecast EPS, divided by the stock price at the end of the preceding quarter. The suffix of SURGEM, FRM, and SURPRISE indicate the period during which the information is generated. The SURGEM1:Q, SURGEM2:Q, and SURGEM3:Q (FRM1:Q, FRM2:Q, and FRM3:Q) represent the sequential first, second, and third month revenue surprises (analysts’ earnings forecast revisions) in a quarter Q, where Q could be Q1, Q2, Q3, Q4. The SURPRISE:Q,Y represents the earnings surprise for the quarter Q of year Y, where the pair (Q, Y) could be (Q4,Y−1), (Q1,Y), (Q2,Y), and (Q3,Y). SURPRISE:Q,Y followed with * indicates the earnings announcement dates before the implementation of IFRS beginning in 2013. The EAM areas are the earnings announcement months we defined

Full size image

This feature of monthly revenue disclosures on the Taiwan stock market therefore provides an ideal empirical setting to test whether interim revenue disclosures contain relevant and timely information beyond formal quarterly financial reports in predicting firms’ performance.

2.2 Literature review

2.2.1 Market reactions to revenue information

Beginning with Beaver (1968) and Ball and Brown (1968), various researchers have studied the information value of earnings and its components. Among the accounting components, the information content of revenues has been widely discussed in the context of pricing effects and value relevance (Ertimur et al. 2003; Ghosh et al. 2005; Jegadeesh and Livnat 2006a, b; and Chen et al. 2014), analysts’ revenue forecasts (Keung 2010; Lorenz and Homburg 2018; Bilinski and Eames 2019), managers’ revenue management (Edmonds et al. 2013; Zhao 2017), credit risk assessment (Melgarejo 2018), and real-time private information (Froot et al. 2017; Agarwal et al. 2021).

Our study is related to the literature discussing the pricing effects and value relevance of revenue information. In this stream of literature, previous papers show evidence that revenue surprises contain incremental information useful for predicting future earnings growth and for explaining the announcement return and post-earnings-announcement drift (PEAD). For example, Ertimur et al. (2003) find that the market responds more to revenue surprises than to expense surprises, especially for firms with higher growth potential. Ghosh et al. (2005) show that firms whose revenue growth accompanies earnings growth have higher long-term analyst earnings forecasts. Jegadeesh and Livnat (2006b) argue that earnings surprises are more persistent if driven by revenue surprises, and show that revenue surprises predict future earnings and that the market underreacts to this information. Kim et al. (2009) divide earnings into sales-induced earnings and non-sales-induced earnings, and find that stock prices respond more sensitively to sale-induced earnings. Chen et al. (2014) investigate the information contained in earnings surprises, revenue surprises, and price momentum, and find that revenue surprises contain exclusive information content that drives the return drift over and above earnings surprises and price momentum. Researchers also find that the effect of revenue surprises can be stronger than earnings surprises under certain conditions. For example, Chandra and Ro (2008) show that revenues convey more information than earnings in extreme earnings situations and in the technology industry. Kama (2009) show that the influence of revenue surprises on stock returns is greater than that of earnings surprises for R&D intensive firms and firms in oligopolistic competition industries. In sum, this literature suggests that revenues do contain incremental information beyond earnings, and this information is value relevant.

However, this literature can only study the additional information value that revenues provide beyond concurrent earnings because quarterly revenues and earnings are released simultaneously in the reports.Footnote 9 In our empirical setting, monthly revenues are disclosed earlier than quarterly earnings. This disclosure timeline enables us to study the distinct influence of revenue information because monthly revenues are generally the most prominent and timely information, rather than merely a piece of supplemental information to earnings between two earnings announcements. Therefore, monthly revenue could influence analysts’ forecasts and stock prices. Further, we can test whether interim revenue information helps analysts better predict earnings and correct their underreaction to past earnings surprises (Shane and Brous 2001; Jegadeesh and Livnat 2006b).

2.2.2 Effects of frequent or interim financial reporting

Our study is also related to an important strand of accounting literature that studies the effects of frequent or interim financial reporting. Previous studies focus on the benefits and costs of higher financial reporting frequency or interim disclosures in the context of semi-annual or quarterly earnings announcements. This literature suggests that more frequent reporting is associated with increased stock price revisions (Beaver et al. 2018), higher analyst forecast accuracy and more revision activities (Stickel 1989; Ivković and Jegadeesh 2004; Tsao et al. 2016), lower information asymmetry (Cuijpers and Peek 2010; Fu et al. 2012),Footnote 10 lower mispricing of accruals (Tsao et al. 2018), improved earnings timeliness (Butler et al. 2007), lower surprise around annual earnings announcements (McNichols and Manegold 1983), and increased monitoring on managers (Downar et al. 2018). Studies also show that investors and analysts use other sources of information during non-reporting periods to help them make decisions and forecasts. For example, Shane and Brous (2001) show that analysts use non-earnings information to revise earnings forecasts. Arif and De George (2020) find that, during non-reporting periods, investors rely on the earnings news of peer-firms that report earnings quarterly to assess values of those reporting earnings semi-annually. These results suggest that investors use other information sources during the absence of own-firm earnings to offset the information loss due to lower reporting frequency.

By contrast, scholars have also found that increased reporting frequency could result in greater short-termism (Gigler et al. 2014; Ernstberger et al. 2017; Kraft et al. 2018) and may increase stock price volatility and the cost of equity (Botosan and Plumlee 2002; Mensah and Werner 2008). For example, Kraft et al. (2018) show that increased reporting frequency is associated with a decline in investments, consistent with the notion that frequent financial disclosures induce myopic managerial behaviors. However, based on a sample from Singapore, Kajüter et al. (2019) cannot find evidence of informational benefits or myopic investment for firms changing disclosure frequency. In sum, the debate over the costs and benefits of frequent financial reporting remains inconclusive. Our study adds to this literature by providing results regarding mandatory interim revenue disclosures. It shows that a higher disclosure frequency of revenue information benefits investors for its value relevance and can help investors make investment decisions.

2.3 Research questions and empirical hypotheses

As previously mentioned, this research bridges the void of two strands of literature: revenue disclosures and interim information. Our empirical setting allows us to explore the information value of monthly revenue disclosures between two earnings announcements. Based on this setting, we discuss our research questions and testable predictions as follows.

2.3.1 Do monthly revenue disclosures have a positive effect on analysts’ earnings forecasts and predict future earnings surprises?

Jegadeesh and Livnat (2006b) show that revenue surprises contain information about future earnings growth. Since revenues are less manipulated than expenses, they conclude that earnings surprises driven by revenue surprises are more persistent. In the same paper, they also show analysts revise their forecasts in response to this information. In our paper, the information value of monthly revenue surprises is twofold. First, similar to Jegadeesh and Livnat (2006b), monthly revenue surprises should also convey information about the persistence of earnings growth, and analysts could revise their forecasts according to this information. Thus, monthly revenue surprises should be positively associated with analysts’ forecast revisions and quarterly earnings surprises measured by SUE. Second, unlike Jegadeesh and Livnat (2006b) who use quarterly revenues, the monthly revenues we use allow investors to obtain timely information about a firm’s operational status before quarterly earnings are announced. As suggested in the literature, interim earnings disclosures contain information that helps analysts revise their forecasts for annual earnings (Stickel 1989; Ivković and Jegadeesh 2004). We expect that interim revenue disclosures would have similar effects. Since the disclosure of monthly revenue and quarterly earnings are at different dates, we can test whether monthly revenue disclosures, similar to interim earnings, contain value relevant information that helps analysts update their forecasts between earnings announcements and help investors better predict future SUE. These discussions lead to the following empirical hypothesis.

Hypothesis 1.1

Analysts positively revise their earnings forecasts in response to monthly revenue surprises, such that analysts’ revisions are positively associated with monthly revenue surprises.

Hypothesis 1.2

Monthly revenue surprises positively predict future SUE.

Since analysts usually have the most updated expectation on earnings, our second measure of earnings surprises is analysts’ forecast errors. The forecast error addresses the question of whether analysts could fully incorporate revenue information without under- or over-reaction. Previous studies show that analysts do not fully incorporate the information in the most recent earnings surprises into their forecasts, which causes a significant association between analysts’ forecast errors and past earnings surprises (Abarbanell and Bernard 1992; Elgers and Lo 1994; Easterwood and Nutt 1999).Footnote 11 Jegadeesh and Livnat (2006b) extend this literature to revenue surprises, and find that analysts also under-react to past revenue information. Their results suggest that analysts do revise their earnings forecasts according to revenue surprises, but the magnitude of adjustments is insufficient.Footnote 12 It is interesting to investigate whether the results documented in the literature regarding quarterly revenue information still hold when monthly revenue information is available to analysts. Our empirical predictions on this issue can be summarized as follows.

Hypothesis 1.3

The analysts’ earnings forecast revisions in response to monthly revenue surprises are insufficient, such that the forecast errors after revision are positively associated with monthly revenue surprises.

2.3.2 How do stock prices react to monthly revenue surprises?

In the literature, the papers studying revenue surprises usually conclude that revenue surprises can help predict future stock returns. However, because firms’ earnings and revenues are always released simultaneously in these studies, an unanswered question is whether revenue surprises alone can influence stock prices. In our paper, since monthly revenues and quarterly earnings are disclosed at different dates, we can study the independent effect of revenue disclosures on stock returns. Consistent with Jegadeesh and Livnat (2006a, b), Chen et al. (2014), and Beaver et al. (2018), we predict that stock prices react positively to monthly revenue surprises during the revenue announcement window and continue to drift in the same direction with revenue surprises. Note that this prediction implies that investors under-react, so the information in monthly revenue surprises is gradually incorporated into stock prices. In addition, because the revenue information is released far before the release of earnings, we can examine how fast the information in monthly revenue surprises is fully incorporated into stock prices. Van Buskirk (2012) shows that monthly sales disclosures reduce stock price and trading volume reactions around earnings announcements, suggesting that stock price has reflected the information about upcoming earnings conveyed in monthly sales.Footnote 13 Similarly, we conjecture that the information in monthly revenue surprises should have been largely incorporated into stock prices when quarterly earnings, which is an important piece of information, become publicly available. Therefore, when quarterly earnings are announced, monthly revenue surprises should have no incremental positive effects on stock returns after earnings surprises have been considered. Our predictions regarding stock price reactions to monthly revenue surprises can be summarized as follows.

Hypothesis 2.1

Stock prices positively react to monthly revenue surprises and continue to drift in the same direction with revenue surprises.

Hypothesis 2.2

Monthly revenue surprises have no incremental positive effects on stock returns when quarterly earnings are announced.

3 Sample and research design

3.1 Sample and data

We obtain data on monthly revenues, stock returns, and financial statements from a local data vendor, the Taiwan Economic Journal (TEJ) database, which compiles the data from the Market Observation Post System (MOPS) website maintained by the Taiwan Stock Exchange (TWSE). Furthermore, to test whether analysts fully incorporate monthly revenue information into their earnings forecasts, we also employ analysts’ earnings forecasts compiled by the CMoney database, which has collected earnings forecasts of listed companies from brokers since Dec. 2002.Footnote 14 Because almost all the observations of earnings forecasts in 2002 are missing, we start our sample period from 2003. The sample period ends with 2015 when we initiate this research. Because the estimations are based on a rolling window, we also collect extra revenue and earnings observations between 2000 and 2002 to calculate the values of revenue and earnings surprises in 2003. All variables are winsorized at 1% and 99% to prevent the effects of outliers.

Our initial sample comprises all the publicly held companies whose common stocks are listed on the Taiwan Stock Exchange (TWSE) and the Gre Tai Securities Market (GTSM) from 2000 through 2015. As of 31 December 2015, the TWSE, the second largest emerging stock market in the Morgan Stanley Capital International (MSCI) Emerging Market index,Footnote 15 has 874 listed companies with a combined market capitalization of US$795 billion. We exclude Depository Receipts, Beneficiary Certificates, and Exchange-Traded Funds (ETFs) from our sample. We then exclude stocks with a nominal price of less than NT$10 since these stocks are usually in distress and extremely illiquid.Footnote 16 We also exclude firm-month observations from which firms have experienced M&A. In addition, we also exclude stocks in the financial and construction sectors because their accounting treatments on revenues are different from those of industrial firms.Footnote 17Footnote 18 Then, we also delete firms covered by less than two analysts and firm-month observations with missing values for calculating stock returns and control variables.Footnote 19 Panel A of Table 1 shows the sample selection criteria and the corresponding numbers of observations (or those being deleted). Our final sample comprises 50,413 firm-months (11,630 firm-quarters) from 1049 (817) firms. Since some firms may not have data available for entire sample years, the final sample comprises 462 firms on average per year. Table 1 Panel B shows the distribution of observations across earnings announcement months. The sample size of earnings announcement months (EAM) is slightly larger than non-EAM because of our setting. If the earnings-announcement dates are at the end of a month, we classified both the concurrent month and subsequent month as our EAM to avoid confounding with earnings information. Panel C shows industries and their corresponding Taiwan Standard Industrial Classification (TW SIC) codes.Footnote 20 The final sample comprises mostly manufacturing firms, followed by the trading and service industry.Footnote 21 This industry distribution generally represents the market structure of the Taiwan stock market and mitigates selection bias concerns.

Table 1 Sample selection criteria and the distribution of observations
Full size table

3.2 Measures for revenue surprises, earnings surprises, and earnings forecast revisions

We measure revenue surprises by a standardized unexpected revenue growth estimator (i.e., SURGE) as in Jegadeesh and Livnat (2006a, b), and revise it to fit a monthly basis. The calculation of monthly SURGE, denoted as SURGEM, is

$$SURGEM_{{i,m}} = \frac{{REV_{{i,m - 1}} - {\text{E}}\left( {REV_{{i,m - 1}} } \right)}}{{\xi _{{i,m - 1}} }}$$
(1)

where \(REV_{{i,m - 1}}\) and \({\text{E}}\left( {REV_{{i,m - 1}} } \right)\) are firm i’s actual and expected monthly revenue per share of month m−1 disclosed in month m, and their difference is unexpected monthly revenue growth. \(\xi _{{i,m - 1}}\) is the standard deviation of the unexpected monthly revenue growth in month m-1. We estimate Eq. (1) based on the rolling window and assume that monthly revenues follow a seasonal random walk with a drift term,

$$\begin{aligned} {\text{E}}\left( {REV_{{i,m}} } \right) & = S_{{i,m - 12}} + v_{{i,m}} , \hfill \\ v_{{i,m}} & = \frac{{\mathop \sum \nolimits_{{j = 1}}^{{24}} \left( {S_{{i,m - j}} - S_{{i,m - j - 12}} } \right)}}{{24}}, \hfill \\ \end{aligned}$$

where \(S_{{i,m}}\) is monthly revenue per share and \(v_{{i,m}}\) is a drift term.Footnote 22 Because the estimation is based on a rolling window, we collect extra three year revenue observations before 2003 (2000 to 2002) to calculate the SURGEMs in 2003.Footnote 23 To investigate how analysts respond to revenue surprises, we measure the monthly forecast revision \(FRM_{{i,m}}\), as in Jegadeesh and Livnat (2006b) as

$$FRM_{{i,m}} = \frac{{FA_{{i,m}} - FA_{{i,m - 1}} }}{{P_{{i,m - 1}} }} \times 100,$$
(2)

where \(FA_{{i,m}}\) is firm i’s mean analysts’ forecast of annual EPS for the current year at the end of month m, and \(P_{{i,m - 1}}\) is the stock price at the end of month m−1.

To investigate if monthly revenue surprises contain leading information about the persistence of future earnings growth in predicting quarterly earnings surprises announced subsequently in quarterly financial reports, and test if analysts fully incorporate the interim monthly revenue information into their earnings forecast, we calculate quarterly standardized unexpected earnings (SUE) and the forecast error, respectively. The definition of SUE is similar to that of Jegadeesh and Livnat (2006a, b) based on the rolling window, which is:

$$SUE_{{i,t}} = \frac{{Q_{{i,t}} - {\text{E}}\left( {Q_{{i,t}} } \right)}}{{\sigma _{{i,t}} }},$$
(3)

where \(Q_{{i,t}}\) and \({\text{E}}\left( {Q_{{i,t}} } \right)\) are firm i’s actual and expected quarterly earnings per share (EPS) in quarter t, and the difference is unexpected quarterly earnings growth. SUE is unexpected quarterly earnings growth divided by its standard deviation \(\sigma _{{i,t}}\). We assume that EPS follows a seasonal random walk with a drift term,

$$\begin{aligned} {\text{E}}\left( {Q_{{i,t}} } \right) & = Q_{{i,t - 4}} + \varrho _{{i,t}} , \\ \varrho _{{i,t}} & = \frac{{\mathop \sum \nolimits_{{j = 1}}^{8} \left( {Q_{{i,t - j}} - Q_{{i,t - j - 4}} } \right)}}{8}, \\ \end{aligned}$$

where \(Q_{{i,t}}\) is quarterly earnings per share for firm i in quarter t, and \(\varrho _{{i,t}}\) is a drift term.Footnote 24

Following Mendenhall (2004) and Jegadeesh and Livnat (2006b), we also measure earnings surprises by analyst forecast errors since analysts’ forecasts capture the most updated market expectations. Since analysts’ forecasts are provided on an annual rather than on a quarterly basis, we calculate the forecast error \(FE_{{i,t}}\) as follows:

$$ FE_{{i,t}} = \frac{{Q_{{i,t}} - {\text{Achievement\_Rate}}_{{i,t}} \times FA_{{i,t}} }}{{P_{{i,t - 1}} }}\times 100,$$
(4)

where \(Q_{{i,t}}\) is firm i’s actual quarterly EPS in quarter t, Achievement_Rate is the percentage of quarterly earnings over total annual earnings averaged over the past three years, and \(~FA_{{i,t}}\) is the mean analysts’ forecast of annual earnings per share for the current year at the end of quarter t. We scale the difference between the actual and expected EPS by the stock price at the preceding quarter-end to obtain the forecast error \(FE_{{i,t}}\).

Since the dependent variable is quarterly data, we also aggregate monthly independent variables into quarterly ones. Quarterly SURGE (denoted as SURGEQ) is the sum of the three SURGEMs in a given quarter: \(SURGEQ_{{i,t}} = SURGEM1_{{i,t}} + SURGEM2_{{i,t}} + SURGEM3_{{i,t}}\), where \(SURGEM1_{{i,t}} ,\) \(SURGEM2_{{i,t}}\) and \(SURGEM3_{{i,t}} ~\) are revenue surprises for firm i in the first, second, and third months of quarter t, respectively.Footnote 25 The quarterly forecast revision \(FRQ_{{i,t}}\) is defined as

$$FRQ_{{i,t}} = \frac{{FA_{{i,t}} - FA_{{i,t - 1}} }}{{P_{{i,t - 1}} }} \times 100,$$
(5)

where \(FA_{{i,t}}\) is the mean forecast of annual EPS at the end of quarter t, and \(P_{{i,t - 1}}\) is the stock price at the end of quarter t−1.

3.3 Summary statistics

Panel A of Table 2 presents the summary statistics for our sample, including year-by-year sample size and the mean and median number of analysts covering a firm. During our sample period, 2003 through 2015, the number of sample firms gradually increases from 324 to 620 firms per year. The average number of sample firms per year is 468. We calculate the mean and median number of analysts covering a firm per firm-month in each year and then average them across years. The mean and median number of analysts covering a firm per firm-month averaged across years are 5.86 and 4.62, respectively.

Table 2 Descriptive statistics
Full size table

Panel B of Table 2 describes the characteristics of the overall monthly firm variables. The summary statistics we report here represent the simple average over all firm-months. The average SURGEM of the overall sample is 4.0%, indicating that, on average, our sample firms demonstrate positive monthly revenue surprises. In addition, the averages are NT$34.107 billion in firm size (SIZE), 2.231 in the price-to-book ratio (PB), −12.427% in forecast revision (FRM), 1.118 in market beta (BETA), and 19.253% in share turnover (TURN).Footnote 26 Panel C of Table 2 describes the characteristics of each quintile sorted by SURGEM and earnings-announcement month (EAM). We find that stocks with higher monthly revenue surprises tend to be large, growth, and actively traded, and have higher forecast revisions. We therefore control for these firm characteristics in our following tests. Panel D of Table 2 describes the quarterly firm characteristics. The monthly SURGEMs (FRMs) are aggregated into quarterly SURGEQ (FRQ). The simple average numbers are 4.6% in quarterly revenue surprise (SURGEQ), −0.096 in standardized unexpected earnings (SUE), −0.312% in forecast error (FE), −0.388% in forecast revision (FRQ), and −0.745% in abnormal stock return (RET). The negative forecast error (FE) indicates that on average analysts are prone to overestimate earnings. Panel E of Table 2 describes the quarterly variables sorted by SURGEQ. Examining average firm characteristics from the smallest to the largest SURGEQ quintile, we find that stocks with higher SURGEQ tend to have higher SUE as well, consistent with Jegadeesh and Livnat (2006a, b). Panel E also shows that the forecast error (FE) is monotonically increased with the SURGEQ, indicating that analysts tend to underestimate earnings for those firms with positive revenue surprises and overestimate earnings, by contrast, for those firms with negative revenue surprises. The under- and over-estimation of earnings also increase with more extreme revenue surprises. It also shows that analysts’ forecast errors are about two-thirds of the forecast revision for the highest SURGEQ firms, indicating that analysts generally make insufficient revisions for the highest SURGEQ firms.

4 Revenue surprises, earnings surprises, and analysts’ earnings forecast revisions

In this section, we examine the information effect of monthly revenues on analysts’ earnings forecasts to determine whether this information helps investors better forecast earnings growth and correct their underreaction to prior earnings surprises (Shane and Brous 2001; Ghosh et al. 2005; Jegadeesh and Livnat 2006b; Keung 2010). First, we test the premise that earnings and revenue surprises are persistent in our sample in Sect. 4.1. Second, in our empirical setting, we collect monthly revenue information and calculate the mean of analysts’ updated earnings forecasts at the end of each month. We can therefore determine whether analysts incorporate revenue information into their earnings forecasts in Sect. 4.2. Third, the monthly revenue information is publicly available before the announcement of quarterly earnings. We can therefore test whether monthly revenue surprises are useful in predicting SUE, compared to the information provided by analysts’ earnings forecasts, in Sect. 4.3. Finally, Sect. 4.4 shows the result of whether analysts underreact to the information provided by revenue surprises.

4.1 Persistence of revenue surprises and earnings surprises

The literature usually interprets the post-announcement price drift regarding earnings or revenue surprises as investors’ underreaction to the persistence of such fundamental information. It is therefore important to test the premise that earnings and revenue surprises are persistent in our sample before we investigate the information content of the interim monthly revenue disclosures.

Following Abarbanell and Bernard (1992), we estimate the time series autocorrelations of revenue and earnings surprises. We first calculate each firm’s autocorrelation coefficient with a constant and then test whether the average coefficient across firms is equal to zero. Panel A of Table 3 shows the results for monthly revenue surprises. The first row indicates that monthly revenue surprises based on the seasonal random walk with drift, SURGEM, are persistent and significant for up to eight months. In particular, the average first-order autocorrelation in our sample is significantly positive at 0.52. The autocorrelations then become significantly negative after the tenth lagged month. Similarly, the second row shows that revenue surprises based on the seasonal random walk without drift, RevYoY, also demonstrate significantly positive autocorrelations. Panel B of Table 3 presents the results for the quarterly variables. It shows that the first-order autocorrelations for SUE and SURGEQ are significantly positive at 0.26 and 0.51, respectively. SUE and SURGEQ both persist up to two quarters and then become negatively autocorrelated after the fourth lagged quarter.

Table 3 Firm-specific autocorrelation estimates of SURGEM, RevYoY, SUE, and SURGEQ
Full size table

In sum, we document that revenue surprises (both monthly and quarterly) and earnings surprises (quarterly) persist up to at least six months (or two quarters). The duration of this persistence is very similar to that documented in previous studies. Abarbanell and Bernard (1992), for example, show in their Table 1 that forecast errors (i.e., earnings surprises) persist up to three quarters. The evidence of persistency motivates us to investigate whether market participants have exploited this information.

4.2 Monthly revenue surprises and analysts’ earnings forecast revisions

Lang and Lundholm (1996) and Shane and Brous (2001) suggest that analysts may use the information from sources other than earnings announcements, such as conference calls and interim sales numbers, to revise their earnings forecasts. We present direct evidence here that interim monthly revenue is important information that analysts employ to revise their earnings forecasts. To investigate the effect of SURGEM on the monthly forecast revision (FRM), we estimate the following regression and control for firm and time fixed effects:

$$\begin{aligned} FRM_{{i,m}} & = {{\alpha }} + \mathop \sum \limits_{{k = 0}}^{2} \beta _{k} \cdot SURGEM_{{i,m - k}} + \mathop \sum \limits_{{k = 1}}^{2}\gamma _{k} \cdot FRM_{{i,m - k}} \\ & \quad + \phi _{1} \cdot RET\left( { - 4, - 1} \right)_{{i,m}} + \phi _{2} \cdot PB_{{i,m - 1}} \\ & \quad + \phi _{3} \cdot SIZE_{{i,m - 1}} + \phi _{4} \cdot SUE_{{i,m}} + v_{i} + \delta _{m} + \varepsilon _{{i,m}} , \\ \end{aligned}$$
(6)

where \(FRM_{{i,m}}\) and \(SURGEM_{{i,m}}\) are the monthly forecast revision and revenue surprise for firm i in month m. The significances of lag terms allow us to examine whether there is a delayed reaction. We control for firm i’s past return (\(RET\left( { - 4, - 1} \right)_{{i,m}}\)), preceding month-end price-to-book ratio (\(PB_{{i,m - 1}}\)) and market capitalization (\(SIZE_{{i,m - 1}}\)), the latest quarterly earnings surprise (\(SUE_{{i,m}}\)), and firm and time fixed effects \(v_{i}\) and \(\delta _{m}\).

To further exclude the possibility that forecast revisions are due to the announcement of earnings, we include an earnings-announcement-month dummy (EAM) to interact with SURGEM and remove all lag terms from Eq. (6).Footnote 27 We estimate the following regression model:

$$\begin{aligned} FRM_{{i,m}} & = {{\alpha }} + \beta _{1} \cdot SURGEM_{{i,m}} + \beta _{2} \cdot SURGEM_{{i,m}} \times EAM_{{i,m}} \\ & \quad + \gamma \cdot FRM_{{i,m - 1}} + \phi _{1} \cdot RET\left( { - 4, - 1} \right)_{{i,m}} + \phi _{2} \cdot PB_{{i,m - 1}} \\ & \quad + \phi _{3} \cdot SIZE_{{i,m - 1}} + \phi _{4} \cdot SUE_{{i,m}} + v_{i} + \delta _{m} + \varepsilon _{{i,m}} , \\ \end{aligned}$$
(7)

where \(EAM_{{i,m}}\) is an earnings-announcement-month dummy that takes the value of one if (1) firms announce their quarterly earnings within the month m, or (2) firms announce their quarterly earnings at the end of the preceding month m−1, and zero otherwise.Footnote 28 In this model, the coefficient \(\beta _{1}\) represents the effect of SURGEM during non-earnings-announcement months and \(\beta _{2}\) represents the difference in this effect between earnings-announcement and non-earnings-announcement months. Control variables are the same as in Eq. (6).

Table 4 shows the regression results. From univariate regression Models (1) to (3), we find that the regression coefficients on \(SURGEM_{{i,m}}\), \(SURGEM_{{i,m - 1}}\), and \(~SURGEM_{{i,m - 2}}\) are all significantly positive. Further, when we include \(SURGEM_{{i,m}}\), \(SURGEM_{{i,m - 1}}\), and \(~SURGEM_{{i,m - 2}}\) all in Model (4), the regression coefficients all remain significantly positive, suggesting that the current and two lag monthly revenue surprises all provide incremental information in determining analyst forecast revisions. This finding also implies that analysts delay the full incorporation of the information in monthly revenue surprises into their forecasts by at least two months. As shown in Models (5) and (6), even when we include all control variables, SURGEMs are still significantly positive. As the accounting standard changing from TW GAAP to IFRS may affect the accuracy of analysts’ earnings forecasts, we further divide our sample into two sub-periods by 2013.Footnote 29 Models (7) and (8) report the results for the sub-period before and after the adoption of IFRS in 2013, respectively. In Models (7) and (8), the concurrent term \(SURGEM_{{i,m}}\) is significant and qualitatively similar to Model (6), indicating that our findings are robust to the accounting standard change.Footnote 30 Model (9) shows the regression results when including an earnings-announcement-month dummy (EAM) in the regression. The coefficient on \(SURGEM_{{i,m}}\) is significant at 4.157 (p-value < 1%), and the coefficient on \(SURGEM_{{i,m}} \times EAM_{{i,m}}\) is also significant at 1.324 (p-value < 5%), indicating that our results are robust to both earnings-announcement and non-earnings-announcement months. This also excludes the possibility that forecast revisions are merely due to the announcement of earnings.Footnote 31Footnote 32 The impacts of SURGEMs are even more prominent during earnings announcement months. A possible explanation is that earnings announcements generally attract more investor attention and thus intensify the effects of SURGEMs.

Table 4 Effects of monthly revenue surprises on analysts’ earnings forecast revisions
Full size table

Overall, the results indicate that interim monthly revenue surprises have significant effects on analysts’ earnings forecasts. Therefore, our Hypothesis 1.1 is supported. Analysts employ the information contained in monthly revenue reports to revise their earnings expectations. As suggested by Lang and Lundholm (1996) and Shane and Brous (2001), analysts may use information from sources other than earnings announcements, such as conference calls and interim sales numbers, to revise their earnings forecasts. Our results provide direct evidence that monthly revenue disclosures are important information employed by analysts to form their earnings expectations.

4.3 Predicting earnings surprises and analysts’ earnings forecast errors

Next, we discuss the information content of monthly revenue surprises in predicting quarterly earnings surprises. Based on quarterly financial reports, Jegadeesh and Livnat (2006b) find that earnings surprises are positively and significantly associated with contemporaneous and past revenue surprises, suggesting that preceding revenue surprises can help predict future earnings growth. Here we test whether this relationship also holds under our monthly revenue reporting regime. Following Jegadeesh and Livnat (2006b), we measure the earnings surprise using both SUE and analyst forecast errors.

To show that monthly revenue surprises contain information in predicting quarterly earnings surprises (either SUE or the forecast error), we estimate the following regression and control for firm and time fixed effects:

$$\begin{aligned} Y_{{i,t}} & = \alpha + \beta _{1} \cdot SURGE_{{i,t}} + \beta _{2} \cdot FR_{{i,t}} + \beta _{3} \cdot RET\left( { - 4, - 1} \right)_{{i,t}} + \beta _{4} \cdot SIZE_{{i,t - 1}} \\ & \quad + \beta _{5} \cdot PB_{{i,t - 1}} + \beta _{6} \cdot SUE_{{i,t - 1}} + v_{i} + \delta _{t} + \varepsilon _{{i,t}} ,~ \\ \end{aligned}$$
(8)

where \(Y_{{i,t}}\) is either \(SUE_{{i,t}}\) or the forecast error \(FE_{{i,t}}\) for firm i in quarter t. \(SURGE_{{i,t}}\) and \(FR_{{i,t}}\) represent the revenue surprise and forecast revision for firm i in quarter t. The \(SURGE_{{i,t}}\) (\(FR_{{i,t}}\)) is either \(SURGEM1_{{i,t}}\), \(SURGEM2_{{i,t}}\), \(SURGEM3_{{i,t}}\), or \(SURGEQ_{{i,t}}\) (\(FRM1_{{i,t}}\), \(FRM2_{{i,t}}\), \(FRM3_{{i,t}}\), or \(FRQ_{{it}}\)), representing firm i’s sequential first, second, and third month SURGEMs (FRMs) and quarterly aggregated SURGEQ (FRQ) in quarter t, respectively. We include the forecast revision \(FR_{{i,t}}\) in the regression to examine whether the predictive power of revenue surprises for earnings growth goes beyond the information delivered by analysts’ earnings forecasts. Control variables include firm i’s past return (\(RET\left( { - 4, - 1} \right)_{{i,t}}\)), preceding quarter-end market capitalization (\(SIZE_{{i,t - 1}}\)) and price-to-book ratio (\(PB_{{i,t - 1}}\)), past earnings surprises (\(SUE_{{i,t - 1}}\)), and firm and time fixed effects \(v_{i}\) and \(\delta _{t}\).

In Panel A of Table 5, we report the results of regressing SUE on different revenue surprises, including the sequential first, second, and third month SURGEMs in a quarter and quarterly aggregated SURGEQ, to investigate the predictability of monthly revenues surprises for quarterly earnings surprises. When the dependent variable is SUE, a positive slope coefficient on SURGE implies that revenue information conveys the information about the persistence of earnings growth. Otherwise, it is not significant. In Models (1), (3), and (5), we run univariate regressions and take SURGEM1, SURGEM2, and SURGEM3 respectively as the sole independent variable. We find that the coefficients are all significantly positive, suggesting that monthly revenue surprises carry the information about future earnings growth, which supports our Hypothesis 1.2. As leading information for investors, monthly revenue information can be used to predict future earnings growth. In Models (2), (4), and (6), we include other control variables. We find that the coefficients on SURGEMs remain significantly positive, indicating that the predictive power goes beyond analysts’ forecast revisions and past earnings surprises. Furthermore, in Models (7) and (8), we aggregate SURGEMs and FRMs of a quarter into SURGEQ and FRQ and rerun the regressions. We find that the SURGEQ coefficient is also positive and significant. The overall results suggest that monthly revenue surprises provide timely and leading information in predicting future earnings growth. Since the change of accounting standard could affect the relationship between revenue surprises and SUE, in Panels B and C we show the results for two sub-periods divided by the adoption of IFRS in 2013. The results for the two sub-periods are qualitatively similar to those shown in Panel A, indicating that the change of accounting standard does not change our findings.

Table 5 Predictions of monthly revenue surprises for standardized unexpected earnings (SUE)
Full size table

Recall that the estimations of SUE are based on the naïve seasonal random walk model and may not reveal the true expected earnings after market participants incorporate interim information such as monthly revenue surprises into their earnings expectations. Hence, we also use analysts’ earnings forecasts as expectations and measure the earnings surprise by the forecast error, which is the difference between actual earnings and the mean of analysts’ earnings forecasts at the end of the earnings quarter. When the dependent variable is the forecast error, a positive (negative) slope coefficient on SURGE implies that analysts under- (over-) react to SURGE. Conversely, if the slope coefficient of SURGE is not significant, it implies that analysts have fully incorporated the monthly revenue information into their earnings forecasts. We interpret the slope coefficients on past SUE in the same manner. Table 6 Panel A shows the results by taking the forecast error \(FE_{{i,t}}\) as the dependent variable. Similar to the results in Table 5, all the coefficients on SURGEM are significantly positive whether we include control variables. This suggests that analysts should increase the magnitude of revisions to fully reflect the information in monthly revenue surprises when they make earnings forecasts. It also indicates that Hypothesis 1.3 is supported. Further, we find that most of the coefficients on forecast revisions \(FR_{{i,t}}\) are also significantly positive. This indicates that the larger the revisions made by analysts, the greater the forecast errors after revisions. In addition, consistent with the findings in the literature that analysts under-react to past earnings surprises, most coefficients on past SUE are significantly positive. Furthermore, we report in Panels B and C the results for two before and after subsamples divided by 2013 because of the adoption of IFRS. The results are similar to those in Panel A, indicating that our findings are robust to the change of accounting standards.Footnote 33Footnote 34 The overall results suggest that monthly revenue surprises significantly predict quarterly earnings surprises and that this predictive power goes beyond the information provided by analysts’ earnings forecasts and other firm characteristics.

Table 6 Effects of monthly revenue surprises on analysts’ quarterly earnings forecast errors
Full size table

5 Market reactions to monthly revenue releases

In this section, we investigate how stock prices react to monthly revenue surprises (Ertimur et al. 2003; Jegadeesh and Livnat 2006a, b; Chen et al. 2014). Based on U.S. stock market data, the literature has shown that stock prices react positively to quarterly revenue surprises on earnings-announcement dates, and continue to drift in the direction of revenue surprises when controlling for earnings surprises (see, for example, Jegadeesh and Livnat 2006b). Unlike the literature based on quarterly accounting data, our empirical analyses are based on interim monthly revenue disclosures, which can separate market reactions to revenue surprises from market reactions to earnings surprises. Thus, this empirical setting enables us to investigate the pure information content of revenue surprises. We investigate market reactions to monthly revenue surprises in different timings. Section 5.1 studies market reactions when monthly revenues are disclosed. Section 5.2 shows the results of post-revenue-announcement drifts. Section 5.3 examines how interim revenue surprises and earnings surprises affect quarterly stock returns. Finally, Sect. 5.4 studies whether monthly revenue disclosures affect stock returns during and after quarterly earnings are disclosed.

5.1 Revenue-announcement window returns

In this subsection, we investigate how stock prices react to revenue surprises around the short-term period of monthly revenue announcements. Around the earnings-announcement dates, Jegadeesh and Livnat (2006b) show that stock prices react positively to quarterly revenue surprises. Following Jegadeesh and Livnat (2006b), we test whether stock prices positively react to monthly revenue surprises.

We define revenue-announcement-window return (RCAR) as the six-day cumulative abnormal return within a revenue-announcement window. A revenue-announcement window is defined as trading day τ− 3 through trading day τ + 2, where τ is the revenue-announcement date. Following Jegadeesh and Livnat (2006b), we calculate the abnormal return as the raw return minus the return on the value-weighted size and price-to-book ratio matched portfolio benchmark. We form six benchmark portfolios using the intersection of two size and three price-to-book groups at the end of month m−1. Among these six portfolios, the portfolio to which a stock belongs is the benchmark for that stock.

Table 7 shows the equally weighted average cumulative abnormal return (RCAR) of each SURGEM-sorted quintile. Panel A presents results for all firms, and Panels B and C show results for two size groups (market capitalization below/above TWSE median), small firms and large firms, respectively. The first column shows results for whole firm-months. We find that the average RCAR monotonically increases from low SURGEM quintile (R1) to high SURGEM quintile (R5). The difference of RCAR between R1 and R5 (R5–R1) is significantly positive at 1.94%, with a p-value of less than 1%. As shown in Panels B and C, this monotonically and significantly positive relationship between announcement-window returns and monthly revenue surprises is robust to both large and small firms. For example, the difference of RCAR between R1 and R5 (R5–R1) for small (large) firms is significant at 2.19% (1.56%).

Table 7 Revenue-announcement window returns (RCAR) by SURGEM-sorted quintiles
Full size table

We then investigate whether the information value of monthly revenue surprises demonstrates certain patterns of seasonality. For example, Ivković and Jegadeesh (2004) find that forecast revisions are more informative in the period before an earnings announcement than after an earnings announcement. Kama (2009) find that revenue surprises contain more information than earnings surprises during the fourth quarter because firm managers are more likely to manage earnings in the fourth quarter. Specifically, we test whether the price impact is more significant as the end of a quarter approaches or in the months of an earnings announcement. In the middle columns of Table 7, we find that the return differences between R5 and R1 (R5–R1) remain significant for all sequential months in a quarter. In addition, no major difference in magnitude exists across the three sequential months. Next, from the last two columns of Table 7, we find that the return differences between R5 and R1 (R5–R1) remain significantly positive regardless of whether the month overlaps with earnings announcements or not.

Overall, the results of the short-window price impact test are consistent with those of Jegadeesh and Livnat (2006b), who document a positive price reaction to quarterly revenue surprises around preliminary earnings-announcement dates. Our results suggest that the interim monthly revenues contain relevant fundamental information that influences stock prices.

5.2 Post-revenue-announcement drift

We then investigate how stock prices drift in association with revenue surprises after the revenue disclosure. We test whether stock prices continue to drift in the direction of the revenue surprises as in Jegadeesh and Livnat (2006a, b). We define the post-revenue-announcement drift (PRAD) as the abnormal return within the 18-day post-revenue-announcement window. We set the post-revenue-announcement window as trading day τ + 3 through τ + 20, where τ is the revenue-announcement date.Footnote 35 We calculate the cumulative abnormal return in the same manner as RCAR.Footnote 36

Table 8 shows the equally weighted average PRAD of each SURGEM-sorted quintile. In Panel A, we show the results for all firms. The first column shows that the average PRAD monotonically increases from low SURGEM quintile (R1) to high SURGEM quintile (R5) for entire sample months. The difference in the average PRAD between R5 and R1 (R5–R1) is significantly positive at 1.33%, with p-value less than 1%. This result shows that stock prices continue to drift in the same direction with revenue surprises in the post-revenue-announcement period, suggesting that the market slowly incorporates this information into stock prices. Panels B and C show the average PRAD for small and large firms (market capitalization below/above the TWSE median), respectively. In general, we find that the results of the PRAD are robust to both large and small firms. The differences in average PRAD between R5 and R1 (R5–R1) are significant and positive at 1.43% for small firms and 1.21% for large firms.

Table 8 Post-revenue-announcement drift (PRAD) by SURGEM-sorted quintiles
Full size table

Additional tests show the robustness of our results regarding the post-revenue-announcement drift. In the middle columns of Table 8, we find that the return differences between R5 and R1 (R5–R1) are all significant and are similar in magnitude for the three sequential months in a quarter. In the last two columns of Table 8, we find that the return differences between R5 and R1 (R5–R1) remain significant regardless of whether it covers earnings-announcement months or not. This result implies that the effects of revenue information on stock price drift are robust to whether the month covers an earnings announcement date. This result also confirms Chen et al.’s (2014) finding that revenue surprises contain exclusive information content that drives the return drift over and above earnings momentum (price drift driven by earnings surprises) and price momentum.

In addition to the portfolio approach, we also estimate the following Fama–Macbeth regression to examine the effect of revenue surprises on RCAR and PRAD during non-earnings-announcement monthsFootnote 37:

$$\begin{aligned} AR_{{i,m}} & = \alpha + \beta _{1} \cdot SURGEM_{{i,m}} + \beta _{2} \cdot FRM_{{i,m-1}} + \beta _{3} \cdot RET\left( { - 4, - 1} \right)_{{i,m}} + \beta _{4} \cdot PB_{{i,m - 1}} \\ & \quad + \beta _{5} \cdot SIZE_{{i,m - 1}} + \beta _{6} \cdot SUE_{{i,m - 1}} + \varepsilon _{{i,m}} ,~ \\ \end{aligned}$$
(9)

where \(AR_{{i,m}}\) is either RCAR or PRAD for firm i in month m. \(SURGEM_{{i,m}}\) and \(FRM_{{i,m}}\) are the monthly revenue surprise and forecast revision. We control for firm i’s past return (\(RET\left( { - 4, - 1} \right)_{{i,m}}\)), preceding month-end price-to-book ratio (\(PB_{{i,m - 1}}\)) and market capitalization (\(SIZE_{{i,m - 1}}\)), and the latest quarterly earnings surprise (\(SUE_{{i,m - 1}}\)). We focus on whether \(\beta _{1}\) is significantly positive as it represents the effect of monthly revenue surprises on stock returns during non-earnings-announcement months.

Table 9 shows the regression results. Models (1) to (3) present the results of RCAR. Model (1) is the result based on our full sample. Models (2) and (3) present the results for small and large firms, respectively. We see that SURGEM coefficients from Models (1) to (3) are all significantly positive at the 1% level even when we control for RET(−4, −1), PB, SIZE, and past SUE. This confirms the results of the portfolio approach in Table 7 that revenue surprises contain information about stock returns. Similarly, Models (4) to (6) present the results for PRAD. Model (4) shows the result based on our full sample, and Models (5) and (6) present the results for small and large firms, respectively. From Models (4) to (6), all SURGEM coefficients are significantly positive at the 1% level even when we control for RET(−4, −1), PB, SIZE, and past SUE. The results are consistent with the results of the portfolio approach in Table 8 that the market slowly incorporates the monthly revenue information into stock prices. Overall, these results indicate that revenue surprises based on interim monthly revenue information contribute positively to both the abnormal returns during revenue announcements and post-revenue-announcement drifts.

Table 9 Results of regressing revenue-announcement window return (RCAR) and post-revenue-announcement drift (PRAD) on monthly revenue surprises using the sample of non-earnings-announcement months
Full size table

5.3 Explaining quarterly returns by interim revenue surprises and earnings surprises

In this subsection, we run multivariate regressions to further support our hypothesis that the information contained in revenue surprises goes beyond that contained in earnings surprises, in analysts’ earnings forecasts, or in past stock returns. Shane and Brous (2001) suggest that analysts and investors use non-earnings-surprise information available between earnings announcements to correct for their underreaction to earnings surprises. They proxy this in-between information by interim analysts’ earnings forecast revisions and conjecture that weekly sales could be a type of information collected by analysts. Here we test whether the information contained in monthly revenue surprises is a type of in-between information that can be used by investors to correct their underreaction to earnings surprises.

Following Shane and Brous (2001), we estimate the following regression using the Fama–Macbeth method:

$$\begin{aligned} RET_{{i,t}} ~ & = \alpha + \phi _{1} \cdot SURGEM1_{{i,t}} + \phi _{2} \cdot SURGEM2_{{i,t}}^{ \bot } + \phi _{3} \cdot SURGEM3_{{i,t}}^{ \bot } \\ & \quad+ \beta _{1} \cdot FRM1_{{i,t}} + \beta _{2} \cdot FRM2_{{i,t}} + \beta_{3} \cdot FRM3_{{i,t}} + \beta _{4} \cdot SUE_{{i,t - 1}} \\ & \quad + \beta_{5} \cdot RET_{{i,t - 1}} + \beta _{6} \cdot SIZE_{{i,t - 1}} + \beta_{7} \cdot PB_{{i,t - 1}} + \varepsilon _{{i,t}} , \\ \end{aligned}$$
(10)

where \(RET_{{i,t}}\) is the quarterly abnormal return for firm i in quarter t that covers three sequential months. All pieces of interim information available during the quarter, including three sequential revenue surprises (SURGEMs), three sequential forecast revisions (FRMs), and the earnings surprise disclosed in quarter t (SUE) are used to explain the quarterly stock return. We also control for firm i’s past return (\(RET_{{i,t - 1}}\)), preceding quarter-end market capitalization (\(SIZE_{{i,t - 1}}\)) and price-to-book ratio (\(PB_{{i,t - 1}}\)). To isolate the marginal effects of revenue surprises, we use the residual of regressing SURGEM2 on SURGEM1, denoted as \(SURGEM2_{{i,t}}^{ \bot }\), and the residual of regressing SURGEM3 on SURGEM2 and SURGEM1, denoted as \({SURGEM3}_{i,t}^{\bot}\), as the explanatory variables. The rationale for doing so is that SURGEM is persistent and we intend to capture the marginal effect of each monthly revenue surprise on stock returns.

Table 10 reports the regression results. Model (1) shows the baseline result that we regress quarterly abnormal returns \((RET_{t} )\) on SURGEMs. We find that all regression coefficients on SURGEMs are significantly positive at the 1% level. This result indicates that interim revenue surprises during the quarter positively influence the quarterly returns. In Model (2), we test whether forecast revisions (FRMs) during the quarter influence the quarterly returns. The result shows that forecast revisions also have marginal price impacts on stock returns, consistent with Shane and Brous (2001). The regression coefficients on FRMs are all significantly positive at the 1% level. In Model (3), we include both monthly revenue surprises and forecast revisions into the regression. We find that all coefficients on monthly revenue surprises and on forecast revisions remain positive and significant at the 1% level. This result suggests that the information contained in monthly revenue surprises and that in analysts’ earnings forecasts both have their marginal effects on stock prices. In Model (4), we control for earnings surprises (\(SUE_{{i,t - 1}}\)), past return (\(RET_{{i,t - 1}}\)), market capitalization (\(SIZE_{{i,t - 1}}\)), and price-to-book ratio (\(PB_{{i,t - 1}}\)). We find that coefficients on monthly revenue surprises and on forecast revisions all remain significantly positive at the 1% level. This indicates that the marginal effect of revenue surprises and forecast revisions on quarterly abnormal returns are robust to changes in model settings.

Table 10 Effects of monthly revenue surprises on the contemporaneous quarterly abnormal returns
Full size table

Overall, the regression results from Table 10 complement those of Shane and Brous (2001), who document that analysts and investors use non-earnings-surprise information available between earnings announcements to correct for their underreaction to earnings information. Our results suggest that monthly revenue disclosures contribute to such interim non-earnings-surprise fundamental information that helps investors confirm the persistence of prior earnings surprises and thereby correct their underreaction to the persistence of earnings growth. The results from Sects. 5.1 to 5.3 suggest that stock prices positively react to monthly revenue surprises and continue to drift in the same direction as revenue surprises around monthly revenue announcements and during the quarter prior to earnings announcements. Therefore, Hypothesis 2.1 is supported.

5.4 Market reactions to revenue surprises around earnings announcements

We have shown in previous subsections that monthly revenue surprises have significant price impacts when they are announced. The question remains, however, whether such revenue information still has incremental effects on CAR and PEAD when firms finally announce earnings. Jegadeesh and Livnat (2006b) show that abnormal returns around earnings-announcement dates (CAR) and the post-earnings-announcement drift (PEAD) are positively related to both revenue surprises and earnings surprises. In our empirical setting, revenue information is available to investors in the sequential periods that precede the announcement of quarterly earnings. Therefore, the results may be different.

To examine the incremental effects of revenue information on CAR and PEAD around earnings announcements, we follow Jegadeesh and Livnat (2006b) to estimate the following Fama–Macbeth regressions:

$$R_{{i,t}} = \alpha + \beta _{0} \cdot SURGE_{{i,t}} + \beta _{1} \cdot FRQ_{{i,t}} + \beta _{2} \cdot SUE_{{i,t}} + \beta _{3} \cdot RET_{{i,t}} + \beta _{4} \cdot SIZE_{{i,t - 1}} + \beta _{5} \cdot PB_{{i,t - 1}} + \varepsilon _{{i,t}} ,$$
(11)

where \(R_{{it}}\) is either \(CAR_{{i,t}}\) or \(PEAD_{{i,t}}\). \(CAR_{{i,t}}\) (\(PEAD_{{i,t}}\)) is the abnormal return adjusted by the value-weighted size and price-to-book matched portfolio benchmark within the earnings (post-earnings) announcement window. We set the earnings (post-earnings) announcement window as day s−2 through day s+1 (day s+2 through day s+21), where day s is the earnings-announcement date. SURGE is either quarterly aggregated \(SURGEQ_{{i,t}}\), or three sequential monthly \(SURGEM1_{{i,t}}\), \(SURGEM2_{{i,t}}^{ \bot }\), and \(SURGEM3_{{i,t}}^{ \bot }\) in quarter t.Footnote 38 FRQ is the quarterly forecast revision.Footnote 39 Since quarter t’s earnings are disclosed at least one month after that quarter, variables including \(CAR_{{i,t}}\), \(PEAD_{{i,t}}\), and \(SUE_{{i,~t}}\) are collected at least one month after the end of quarter t. We control for firm i’s stock returns in quarter t (\(RET_{{i,t}}\)), market capitalization in quarter t−1 (\(SIZE_{{i,t - 1}}\)), and price-to-book ratio in quarter t−1 (\(PB_{{i,t - 1}}\)).

Jegadeesh and Livnat (2006b) find that both coefficients on SURGE and SUE are significantly positive, suggesting that both SURGE and SUE contain incremental information on firms’ fundamental prospects. Furthermore, in the same paper, they test whether there is a delayed reaction to past surprises by replacing contemporaneous revenue surprises (SURGE) and earnings surprises (SUE) with those in preceding quarters. If the market prices do indeed react efficiently, future returns should be unrelated to past surprises. They find a significant positive coefficient on the immediately preceding quarter SURGE, suggesting that the market does not fully react to SURGE on the earnings announcement date. In our empirical setting, if the market fully reflects the information content of revenue surprises in predicting earnings surprises, we expect that the coefficient on \(SURGE_{t}\) is not significant when we replace contemporaneous SUE with that of the preceding quarter (\(SUE_{{t - 1}}\)). Otherwise, if investors underreact (overreact) to the content of revenue surprises in predicting earnings surprises, we expect that the coefficient on SURGE is significantly positive (negative).

Table 11 presents the results. In Models (1), (3), and (5), we investigate the effects of quarterly aggregated SURGEQ. In Models (2), (4), and (6), we investigate the separate effects of each of the three monthly SURGEMs on CAR and PEAD by replacing \(SURGEQ_{{i,t}}\) with \(SURGEM1_{{i,t}}\), \(SURGEM2_{{i,t}}^{ \bot }\), and \(SURGEM3_{{i,t}}^{ \bot }\) in regression. Models (1) to (4) show the regression coefficients for CAR as the dependent variable. In Model (1), we find that the coefficient on SUE is significantly positive, a result that is consistent with the literature in which stock prices react positively to earnings surprises around earnings announcement windows. Different from the situation in Jegadeesh and Livnat (2006b), however, the coefficient on SURGEQ becomes significantly negative. In Model (2), we investigate the separate effects of each of the three monthly SURGEMs on CAR. We still find that the coefficients on two of the three monthly SURGEMs are significantly negative. The results suggest that the revenue information released in earlier periods has largely been incorporated into stock prices and has a negative effect on earnings-announcement-window returns (CAR) once the quarterly earnings information hits the market.

Table 11 Effects of revenue surprises on earnings-announcement window return (CAR) and post-earnings-announcement drift (PEAD)
Full size table

In Models (3) and (4), we test whether there is a delayed reaction of past surprises on CAR by replacing \(SUE_{{i,t}}\) with the immediately preceding quarter \(SUE_{{i,t - 1}}\) and remain \(SURGE_{{i,t}}\) in Eq. (11) since the latter has become past information during quarter t’s earnings announcement. The coefficient of \(SUE_{{i, t - 1}}\) is not significant, which is different from previous evidence and suggests that there is no delayed reaction to past earnings surprises. Moreover, unlike the finding in Jegadeesh and Livnat (2006b, Table 5, p. 157) that the coefficient of the past revenue surprise is significantly positive, our findings show that none of the coefficients of revenue surprises are statistically significant. This result suggests that monthly revenue surprises have been fully incorporated into the stock prices and there are no delayed reactions when the quarterly earnings information is finally made public to the market.

In Models (5) and (6) in Table 11, we turn our focus to the PEAD, 20-days abnormal returns within the post-earnings announcement window. The coefficients on SURGEQ (SURGEMs) are no longer significant when both SUE and SURGEQ (SURGEMs) are included in the regression. This result indicates that the revenue information released in earlier periods does not contribute to post-earnings-announcement drift. In contrast to the findings of Jegadeesh and Livnat (2006b, Table 7, p. 162) that both SUE and SURGE are statistically significant in predicting PEAD, our results suggest that the market no longer reacts to the information contained in the monthly revenue surprises.

Overall, the results presented in Table 11 suggest that interim monthly revenue disclosures mitigate delayed reactions to past surprises. In our empirical setting, firms publicly release three pieces of monthly revenue information in periods prior to earnings announcements. The information content of revenue surprises in predicting future earnings growth is gradually incorporated into stock prices prior to earnings announcements. During or following earnings announcements, there is no further incremental positive effect of revenue surprises on stock prices after earnings surprises have been considered. Therefore, our Hypothesis 2.2 is supported.

6 Conclusions

This paper studies the information role of interim monthly accounting disclosures based on a unique sample from the Taiwan stock market where firms are required to report revenues monthly, which is more frequent than the quarterly revenue disclosures employed in previous studies. We find that interim monthly revenue surprises have significant effects on analysts’ earnings forecasts, suggesting that analysts employ the information contained in monthly revenue reports to revise their earnings expectations. We also find that interim monthly revenue surprises significantly predict quarterly earnings surprises (SUE and forecast error), and that this predictive power goes beyond the information provided by analysts’ earnings forecasts and other firm characteristics, suggesting that monthly revenue surprises provide leading information about future earnings growth but analysts do not fully reflect this information. Consistent with Shane and Brous (2001), the overall results suggest that interim monthly revenue is an important piece of non-earnings-surprise information employed by analysts to revise their earnings forecasts between earnings announcements.

In determining the information effect on stock prices, we find significant and positive price reactions to monthly revenue surprises around and post the time of monthly revenue releases. In addition, using quarterly stock returns, we find further price drifts in the same direction as monthly revenue surprises during the quarter. However, stock prices no longer show incremental drifts with revenue surprises on the quarterly earnings-announcement date when controlling for earnings surprises, suggesting that the information content of monthly revenues has been fully incorporated into stock prices by the time of the earnings announcements.

Our results provide insight into a long-debated issue among researchers and practitioners as to whether more frequent financial reporting is useful for investors to assess companies’ fundamental value or whether it just bothers companies and investors with short-term noise. Based on monthly revenues reports, our results suggest that investors can benefit from more frequent accounting disclosures, which hasten the incorporation of fundamental information into stock prices.

Our results also imply that investors who own and can better process the real-time operational information could gain an information advantage over others. In that sense, future empirical research can investigate whether analysts with greater capability to gain more real-time operational information can provide better earnings forecasts and recommendations.