# Do firms use early guidance to disclose the effect of conservatism on future earnings?

## Abstract

### Purpose

By deferring profits and anticipating losses, conservatism makes earnings increases more persistent and earnings declines more likely to revert. Therefore, the level of conservatism in current earnings has implications for future earnings expectations. Past research shows that outsiders can fail to understand these implications. This paper aims to investigate whether firms help outsiders by voluntarily disclosing their expectations about how conservatism will affect future earnings trends.

### Design/methodology/approach

The authors examine the likelihood and content of “early” earnings guidance – i.e. guidance about future earnings that is released around or before the announcement of current earnings. The sample is made of 8,820 annual earnings announcements, 62 per cent of which are combined with early guidance.

### Findings

The authors find that the more conservative current earnings, the higher: the likelihood that the firm releases early guidance; the likelihood that the firm predicts a positive change in earnings; and the difference between the forecasted earnings and current earnings. The authors also find such guidance to be relevant to analysts, who use it to update their forecasts.

### Practical implications

By showing that firms use early guidance to disclose the effect of conservatism on future earnings, the study is interesting to users and preparers because it shows that analysts need and use such disclosure; and regulators because it alleviates concerns about the information consequences of conservatism.

### Originality/value

The findings show that firms do not refrain from committing to positive early guidance to disclose the earnings effects of conservatism. This is interesting in light of the difficulty of predicting such effects, the manager incentives to keep expectations low and the cost of committing to positive guidance instead of less risky qualitative disclosure alternatives. In this way, the authors contribute to the literature on the interrelation between voluntary disclosure and conservatism in financial reports.

## Keywords

## Citation

D’Augusta, C. and Redigolo, G. (2019), "Do firms use early guidance to disclose the effect of conservatism on future earnings?", *Review of Accounting and Finance*, Vol. 18 No. 3, pp. 432-455. https://doi.org/10.1108/RAF-09-2018-0203

## Publisher

:Emerald Publishing Limited

Copyright © 2019, Emerald Publishing Limited

## 1. Introduction

We investigate whether firms use early earnings guidance to voluntarily disclose their expectations about how conservative accounting will affect next year’s earnings. We define guidance to be “early” if it is released on or before the current year’s earnings announcement day. Following Weil *et al.* (2012), we define conservative accounting as the implementation of accounting methods that, in the absence of certainty, tend to anticipate expenses and losses but to defer revenues and gains. Because of the properties of the double-entry system, conservatism has two main effects. On the balance sheet, it results in the understatement of *accumulated* net assets because of the higher (lower) valuation of liabilities (assets): this type of effect has been referred to as “balance sheet conservatism” (Sunder *et al.*, 2018). On the income statement, conservatism results in lower *current* earnings because of the faster (slower) recognition of economic expenses (revenues). In this paper, we focus on the income statement effects of conservatism. We define current earnings to be “conservatively reported” if they are lower due to the application of conservative accounting methods; in a given year, we define a firm to be “more conservative” than another firm if it reports current earnings more conservatively[1]. As noted by Weil *et al.* (2012, p.754), the effects of conservatism are temporary: “over long-enough time spans, income is cash-in less cash-out. If a (conservative) reporting method shows less income in early periods, it must show higher income in some later period”. In other words, higher conservatism on the current year’s income statement will reduce *current* earnings because of the faster recognition of expenses and losses and/or the deferral of uncertain profits. However, it will also increase *future* earnings when the recognition of the deferred profits eventually begins and the recognition of expenses and losses stops or declines. Therefore, conservatism in the current income statement affects the earnings time-series in a way that makes current increases (decreases) in earnings more (less) persistent (Basu, 1997). Thus, the level of conservatism in current earnings has implications for future earnings expectations: if in a given year two firms report the same *current* change in earnings but, in that year, one firm has been more conservative than the other, the *future* change in earnings will likely be higher (i.e. more positive or less negative) for the more conservative firm relative to the less-conservative one.

The literature shows that outsiders may fail to fully understand these implications, which can have negative repercussions on the usefulness of income statement numbers (Chen *et al.*, 2014; Penman and Zhang, 2002; Mensah *et al.*, 2004; Louis *et al.*, 2014; Barth *et al.*, 2017). This has led regulators to question the desirability of conservatism in financial statements (Watts, 2003; FASB, 2010). Compared to outsiders, managers likely have superior information about the effects of conservatism on future earnings. For instance, they could be able to predict how much of the deferred revenues and gains will meet recognition requirements the next year or how much of the current expenses and losses will not persist. In this paper, we investigate whether managers share this information with outsiders. More specifically, we study whether managers, when current earnings have been reported conservatively, voluntarily disclose their expectations of higher future earnings.

We focus on earnings guidance because it is arguably the most direct way to convey information about future earnings. In addition, the nature of our research question requires that we do not consider all the earnings guidance released by a firm but that instead we discriminate based on its timing. Absent any guidance, outsiders may have to wait until future earnings are announced to learn that such earnings are more favorable than what past earnings suggested. Releasing guidance later during the year, perhaps just before future earnings are announced, would be of little benefit, as outsiders would be left for most of the period with an incomplete information set. On the other hand, releasing guidance together with (or before) the current earnings announcement would ensure that outsiders can combine the information about the current earnings with managers’ predictions. Our research question is, therefore, whether firms release early guidance that incorporates the higher (lower) expected persistence of positive (negative) earnings components so as to predict a future earnings trend that is more positive – or less negative – than what outsiders would infer based on current (conservative) earnings numbers.

As earnings guidance is voluntary, the answer to this question is not clear *ex ante*. One reason to disclose the earnings effects of conservatism is represented by managers’ asymmetric payoff function. As their compensation and tenure are tied to earnings performance, managers benefit (suffer) from revealing positive (negative) earnings trends (Burgstahler and Dichev, 1997; Graham *et al.*, 2005). It is, therefore, in managers’ interest to inform outsiders of the higher (lower) persistence of profits (losses) whose recognition has been slowed down (accelerated) due to conservatism. Based on this argument, we make three hypotheses: the more conservative current earnings, the more likely the firm to release early guidance; the more conservative current earnings, the more likely the firm to predict a positive change in earnings; and regardless of the sign of the predicted change in earnings, such a change will be more favorable (i.e. more positive or less negative) when current earnings are more conservative.

Several considerations add tension to these hypotheses. An early commitment to higher future earnings is risky for firms, due to litigation and reputation concerns and audit costs (Francis *et al.*, 1994; Krishnan *et al.*, 2012; Lee *et al.*, 2012). They may prefer to wait in cautious silence until conservatism’s effects on earnings are known with more certainty. Also, firms could simply be unable to predict such effects so long in advance. Finally, they may prefer to use less risky qualitative disclosure venues or simply prefer to keep expectations down to a level that is easier to beat. It is, therefore, an empirical question of whether firms use early guidance to disclose their expectations about the earnings effects of conservatism.

We build our sample using annual earnings announcements made by industrial companies listed on the NYSE, AMEX or NASDAQ and managers’ annual earnings forecast data relative to the fiscal years 1996 to 2011. The propensity to release guidance varies across firms, many of which do not release any forecast at all. Prior literature suggests that two reasons why firms issue guidance are to reduce information asymmetry and to mitigate litigation risk (Coller and Yohn, 1997; Skinner, 1994). Because both can be attenuated by conservatism, conservative firms may be less inclined to release forecasts (Hui *et al.*, 2009). Our goal is not to examine how conservatism affects firms’ motivations to become forecasters or to release frequent guidance throughout multiple years, which is examined by other studies (Hui *et al.*, 2009; Jaggi and Xin, 2014; Li, 2008). Rather, our question is whether a forecaster is willing to use early guidance to reveal the effects of conservatism on future earnings. Therefore, we only include in the sample firms that have released one or more forecast of the current year’s earnings (i.e. we exclude non-forecasters) and control for the number of those forecasts. The final sample is made of 8,820 earnings announcements. We label “early guidance” the management forecast estimate about next year’s earnings that is outstanding when current earnings are announced.

Thus, the early guidance estimate will be either the most recent forecast released before the earnings announcement; or a forecast bundled with the earnings announcement. If early guidance is outstanding at the time of the announcement, we define the announcement to be “combined with early guidance”. Approximately 62 per cent of the earnings announcements in our sample are combined with early guidance.

Because our focus is on the income statement, we use conservatism measures that reflect the downward effect of conservative reporting on current earnings: the annual build-up of conservative reserves (Penman and Zhang, 2002), the recording of negative non-operating accruals, the recording of quarterly earnings that are negatively skewed relative to cash flows (Givoly and Hayn, 2000) and the extent that current earnings incorporate economic shocks to performance (Callen *et al.*, 2010). Therefore, the more current earnings have been lowered by conservatism, the higher the values of the conservatism measures.

The results support our hypotheses. Using a binomial logit model we find that an earnings announcement is more likely to be combined with early guidance when current earnings have been reported conservatively. In a multinomial logit setting, we find that this effect is driven by conservatism increasing the probability of *positive* guidance. In an Ordinary Least Squares (OLS) regression, we find the predicted change in earnings to be significantly positively associated with the level of conservatism of current earnings.

In additional analyses, we examine earnings guidance that is released after the announcement of current earnings (i.e. “later guidance”). We find that conservatism is not significantly associated with the likelihood of later guidance, suggesting that, when firms want to disclose the earnings effects of conservatism, they do it in a timely fashion. We also examine the usefulness of early guidance to analysts. Our results suggest that analysts *need* early guidance to understand the effect of conservatism on future earnings; and *use* such guidance to revise their outstanding forecasts upward.

In the remainder of the paper, we review the literature and develop the hypotheses in Section 2, we describe the research design in Section 3, we discuss the results in Sections 4 and 5 and we present concluding remarks in Section 6.

## 2. Literature review and hypothesis development

### 2.1 Valuation implications of the earnings effects of conservatism

The literature shows that information users may fail to fully understand how conservatism affects current and future earnings, which, in turn, can have negative repercussions on the usefulness of earnings information. For instance, Penman and Zhang (2002, p. 238) show that, when investments are growing, conservative accounting makes current earnings a low-quality indicator of future earnings by temporarily reducing the former in favor of the latter. Their results suggest that investors fail to appreciate this effect as they “fixate” on the reported earnings number “unaware that conservative accounting may lead to reported earnings of doubtful quality”. Mensah *et al.* (2004, p. 161) show that conservatism, by affecting the timing of earnings “with bad news reflected in earnings earlier than good news”, increases earnings volatility, and thus, makes it harder for analysts to forecast future earnings. Narayanamoorthy (2006) suggests that investors struggle to fully appreciate the implications of conservatism on future earnings changes, which causes price drifts. Separately analyzing conditional and unconditional conservatism, Chen *et al.* (2014, p. 240) show that both forms lead to lower earnings persistence and smaller valuation multiples, suggesting that uncertainty about the timing and magnitude of the effects of conservatism on earnings “may cause investors to misestimate (and misprice) the impact of conservative earnings”. In line with Chen *et al.*’s results, Louis *et al.* (2014) find that only sophisticated analysts are able to incorporate the earnings effects of conditional conservatism into their long-horizon forecasts and Barth *et al.* (2017) find that conditional conservatism can make investors take a longer time to understand the valuation implications of current earnings. In line with the aforementioned concerns, standard setters have taken a stance against conservatism and in favor of neutral reporting (Watts, 2003; FASB, 2010)[2].

### 2.2 Conservatism and guidance

The literature has explored some aspects of the relationship between conservatism and earnings guidance. For instance, Hui *et al.* (2009) hypothesize that conservatism mitigates litigation risk and information asymmetry, which, in turn, reduces the need for frequent guidance. Accordingly, they find that firms that were more conservative between 1992 and 1996 were less likely to release earnings or sales forecasts between 1997 and 2002. By showing that *past* conservatism decreases firms’ likelihood to release forecasts over multiple *future* years, Hui *et al.* suggest that conservatism and guidance are linked by an indirect negative association mediated by lower litigation risk and information asymmetry. We investigate the existence of a different, additional link. Our question is whether, after controlling for differences in firms’ forecasting behavior and other firm-specific confounds, the fact that *current* earnings have been reported conservatively makes firms more likely to release *positive* guidance *when or before* current earnings are announced[3].

Different from Hui *et al.* (2009) and Li (2008) analyzes a later sample period and finds evidence that conservatism makes firms *more* likely to issue downward guidance. She argues that, after the passing of Regulation Fair Disclosure, analysts may be unable to anticipate whether earnings that are soon to be announced will be depressed by conservative charges. This prompts managers to release *negative*, *short-horizon* guidance as the earnings announcement approaches, to walk analysts’ expectations of *current* earnings down to levels that are easier to meet. Different from Li (2008a), we do not aim to re-examine Hui *et al.* (2009) argument on the association between conservatism and forecast frequency, which we rather control for. Also, different from Li (2008), we investigate whether firms issue *upward*, *long-horizon* guidance to predict how conservatism, having lowered current earnings, will result in a positive *future* earnings trend. Li (2008) arguments add tension to our hypotheses because they suggest that conservative managers have incentives to keep analyst expectations down. Therefore, managers could be reluctant to reveal the positive effects of conservatism on future earnings.

Re-examining the association between conservatism and forecast frequency on a period different from those analyzed by Hui *et al.* (2009) and Li (2008a), Jaggi and Xin (2014) find it to be negative but weakly significant. They also document that the substitution effect exists for informative forecasts and especially for bad news ones, but not for opportunistic and good news forecasts. Therefore, prior literature has examined the relationship between conservatism and firms’ incentives to release frequent (if any) forecasts, finding mixed results. As mentioned earlier, our paper is different from the papers discussed above in that we do not examine whether (or under what conditions) conservatism is associated with forecast frequency.

### 2.3 Hypothesis development

When earnings are reported conservatively, revenues and gains are more persistent because their recognition is deferred and spread over time, while expenses and losses are less persistent because their recognition is anticipated. As a result, conservatism makes earnings increases more persistent and earnings declines more likely to revert. If two firms report the same change in current earnings (e.g. *Δ*E_{t} = E_{t} − E_{t-1} = +$100), a more-conservative firm (hereafter MC) is more likely to expect the positive earnings trend to continue than a less conservative firm (hereafter LC). That is, the future change in earnings forecasted by the firm (hereafter F*Δ*E_{t}), which is defined as the difference between the expected future earnings and current earnings (E_{t+1} − E_{t}), will be more favorable for MC than for LC (i.e. F*Δ*E_{t}^{MC} > F*Δ*E_{t}^{LC}).

Similarly, if both firms report a current earnings decline (e.g. *Δ*E_{t} = E_{t} − E_{t-1} = −$100), MC is more likely to expect that the current trend will attenuate or revert, whereas LC is more likely to expect the negative trend in reported earnings to continue. As a consequence, the forecasted earnings trend will be more favorable (i.e. less negative or more positive) for MC than for LC (i.e. again, F*Δ*E_{t}^{MC} > F*Δ*E_{t}^{LC}).

Research shows that outsiders may fail to understand the effects of conservatism on current and future earnings trends – i.e. fail to understand the likelihood that F*Δ*E_{t}^{MC} > F*Δ*E_{t}^{LC} even though *Δ*E_{t}^{MC} = *Δ*E_{t}^{LC}. However, managers are arguably better informed about the earnings effects of conservatism. Managers are also likely to benefit from communicating positive earnings trends: if a company’s prospects look better than its competitors’, its shareholders are less likely to sell the stock or fire the managers; similarly, its creditors are less likely to cut the credit supply or charge higher interest rates. Therefore, it is in MC’s managers’ interest to help outsiders distinguish MC from LC.

Based on this argument, we make three empirical predictions. First, conservatism will be associated with the likelihood of early guidance. Releasing early guidance is costly: if they fail to meet their own predictions, managers face a significant risk of litigation (Francis *et al.*, 1994; Kellog (1984), termination of employment (Lee *et al.*, 2012) and adverse market reaction (Skinner and Sloan, 2002; Mindak *et al.*, 2016). Even when not missed, positive guidance is associated with higher audit costs (Krishnan *et al.*, 2012), as managerial pressure to meet higher targets can increase audit risk (Desai, 2016). As managers benefit from explaining the effects of current conservatism on future earnings, the more conservative current earnings, the more likely the benefits of releasing guidance to exceed the costs, all else equal. As a consequence, conservatism will increase the likelihood of a firm releasing early guidance rather than remaining silent. Therefore, we express our first hypothesis as follows:

*H1.*

The more conservative current earnings, the more likely the current earnings announcement to be combined with early guidance.

Second, conservatism will be associated with the sign of early guidance. We argue that MC uses early guidance to disclose their expectations of *favorable* earnings trends (i.e. trends that are more positive or less negative than those of LC, all else equal). It follows that conservatism’s effect on the probability of early guidance, predicted by *H1*, and should be driven by conservatism increasing the probability of *positive* guidance (i.e. F*Δ*E_{t} > 0) over the probability of no guidance. Therefore, we express our second hypothesis as follows:

*H2.*

The more conservative current earnings, the more likely the current earnings announcement to be combined with early guidance that predicts a positive change in earnings.

Finally, if economic circumstances make both MC and LC predict a positive (negative) change in earnings, then MC will predict a *more positive (less negative)* change in earnings than LC because MC expects the future earnings increase (decline) to be augmented (attenuated) by the effect of current conservatism on future earnings. As a result, conservatism will be positively associated with the difference between the forecasted earnings and the current earnings (i.e. F*Δ*E_{t}^{MC} > F*Δ*E_{t}^{LC}). Therefore, we express our third hypothesis as follows:

*H3.*

The more conservative current earnings, the higher the difference between the earnings forecasted by early guidance and current earnings.

## 3. Research design

### 3.1 Measurement of variables

Our measure of conservatism (*CONS*) is a composite measure that is equal to the average of the decile ranks of four different proxies[4]. The first one is the earnings skewness (*SKEW*) proxy proposed by Givoly and Hayn (2000), which is calculated as the difference between the skewness in operating cash flows and earnings. The second proxy, also based on Givoly and Hayn (2000), is equal to −1 multiplied by the magnitude of non-operating accruals (*NOACCR*). The third proxy is the year-over-year change in the balance sheet reserves created by conservative accounting, estimated following Penman and Zhang (2002) (Δ*RES*). These three measures are all accounting-based: i.e. they measure conservatism based on its downward effect on current earnings. We consider using accounting-based proxies to be an appropriate decision in the context of our hypotheses. Because of the way they are constructed, accounting-based proxies are likely to capture how conservatism has decreased current earnings in a way that makes expected earnings trends more favorable. This is what, we hypothesize, motivates firms to release positive early guidance in a specific year. Different from the first three proxies, the fourth one is based on both accounting earnings and market returns (i.e. it is both accounting- and market-based). It is the conservatism ratio (*CRATIO*) developed by Callen *et al.* (2010), which exploits the Vuolteenaho (2002) market-returns decomposition model to estimate the percentage of a shock to current and future cash flows that is incorporated in current earnings. By capturing how more-conservative firms will recognize a larger (smaller) portion of a negative (positive) shock, *CRATIO*, as the other three proxies, likely captures how conservatism depresses current earnings in favor of future ones.

As noted by Callen *et al.* (2010, p. 146), *CRATIO* is “in the spirit” of the model developed by Basu (1997), which is based on the intuition that conservatism results in an asymmetric association between earnings and stock returns. We acknowledge that the coefficient estimate of the asymmetry in the Basu (1997) model is arguably one of the most common market-based measures of conservatism adopted in the literature. However, we believe that it is not appropriate for our specific research design. Because it is based on estimating the asymmetry in the earnings-return relationship across many firm-year observations, the Basu (1997) model does not yield a firm-year specific proxy. For this reason, the model can be used to test hypotheses where conservatism is the dependent construct (and therefore, the asymmetric timeliness coefficient is the dependent variable). However, when conservatism is the independent variable, a firm-year specific proxy is necessary to use it as a variable in a regression model. While one could use a rolling window of previous earnings and returns to estimate a firm-year coefficient of asymmetric timeliness, this is not a feasible solution in the context of our hypotheses: the resulting measure would, in fact, only be available for firms with a long availability of time-series data, and would end up measuring the firm’s conservatism in its (sometimes distant) past. As explained in the previous section, our focus is on analyzing the link between early guidance and *current* conservatism in earnings[5].

While *CRATIO* and *SKEW* have been used in the literature to measure the conditional form of conservatism (Garcìa Lara *et al.*, 2011; Kim and Pevzner, 2010), *NOACCR* and Δ*RES* have often been used to capture the unconditional form (Chen *et al.*, 2014; Kim *et al.*, 2018). As our hypotheses do not differentiate between the two forms, throughout the paper, we mainly present and discuss the results of the composite measure *CONS* and report the results of the other proxies in Section 5.1.

We define early guidance as to the management forecast estimate about the earnings of the year t + 1 that is outstanding (i.e. most recently released) when the earnings of the year *t* are announced. Therefore, the early guidance estimate will be either a forecast bundled with the earnings announcement; or an earlier forecast, if no bundled forecast is released. We measure the change in earnings predicted by the early guidance (*FΔE)* as the difference between the guidance estimate (or midpoint, if it is a range forecast) and the current earnings per share, scaled by price[6].

Several control variables are added to all test models. We control for the current earnings change (*ΔE*) and also for a binary variable equal to one if *ΔE* is negative (*NEGΔE*). We add the logarithm of the firm’s market value of equity (*MVAL*), financial leverage (*LEV*) and the market to book ratio (*MB*) to control for the firm’s size, external scrutiny and growth opportunities. To control for past earnings levels (*PAST_E*), we include the average earnings per share in the years *t-1*, *t-2* and *t-3*, scaled by stock price and a binary variable equal to one if *PAST_E* is negative (*NEGPAST_E*). To control for firm-specific uncertainty, we include the standard deviation of daily returns (*SDRET*) and the standard deviation of quarterly earnings’ changes (*SDEARN*). Because a firm’s choice to release guidance can depend on the quality of the information provided by analysts, we add several variables related to analysts’ forecasts: the number of analysts following the firm (*FOLLOW*), the forecast dispersion (*DISP*) and the absolute magnitude of the analyst forecast error (*AFE*). We also include *MISSEXP*, a dummy variable equal to one when current earnings fall below analysts’ expectations and 0 otherwise. As firms that frequently issue forecasts are less likely to remain silent, we add the total number of forecasts issued by the firm relative to the earnings of year *t* as a control for forecast frequency (*FREQ*). Finally, to capture the firm’s ability in forecasting the earnings of year *t*, we include the average management forecast error (*F_BIAS*) and its absolute value (*F_INACCUR*). All our models also include industry and year fixed effects[7].

### 3.2 Sample construction

The sample is composed of annual earnings announcements made by industrial firms listed on the NYSE, AMEX or NASDAQ, obtained from the CRSP/Compustat data set[8]. We merge this sample with managers’ forecast data relative to the fiscal years 1996 to 2011, obtained from First Call. The initial data set comprises 68,670 announcements. We delete 1,338 observations not denominated in US$ and 2,961 observations whose share price is lower than $1 (illiquid stocks). Forecast data are not available for 52,158 firm-year observations. It could be that these firms never issue forecasts, whether early or not. We cannot include them in our sample because key variables related to forecasting characteristics (e.g. *FREQ*, *F_BIAS*, *F_INACCUR*) cannot be computed[9]. We also delete observations lacking data to compute *CONS* or any of the control variables that are included in all models.

After the adjustments described above, the sample consists of 8,820 earnings announcements, 5,502 of which are combined with early guidance (62 per cent of observations). For around 88 per cent of these 5,502 observations, the early guidance estimate is a bundled forecast, while for the other observations it is an earlier, stand-alone forecast that was on average released 110 days before the announcement. Of the 3,318 observations with no early guidance, 1,174 (i.e. around 35 per cent) released at least one forecast at a later point in time. On average, the first forecast released by these 1,174 firms was issued 142[10] days after the earnings announcement. Table I reports the descriptive statistics (all continuous variables are winsorized at the 1st and 99th percentiles). The values of the four conservatism proxies are in line with those reported by other studies (Kim and Pevzner, 2010; Kim *et al.*, 2013; Biddle *et al.*, 2016; Aier *et al.*, 2014; Mensah *et al.*, 2004; Qiang, 2007).

Table II shows the pairwise correlations. In line with prior research (Hui *et al.*, 2009; Khan and Watts, 2009; LaFond and Watts, 2008), *CONS* is positively correlated with variables related to information asymmetry and uncertainty (e.g. *LEV*, *SDRET*, *SDEARN*, *DISP*) and negatively with variables related to firm size (*MVAL* and *FOLLOW*). Conservatism is also negatively correlated with *FREQ*, which is consistent with Hui *et al.*’s (2009) argument that more-conservative firms do not need to release guidance as frequently as less-conservative ones because of lower information asymmetry and litigation risk concerns.

## 4. Main results

### 4.1 Conservatism and the probability of early guidance (hypothesis 1)

To test *H1*, we create a binary variable, *EARLYGUID*, which identifies earnings announcements that are combined with early guidance. We then run the binomial logit model specified by the following equation:

*EARLYGUID _{it} = β_{0} + β_{1}CONS_{it} + β_{k}CONTROLS_{kit} + ɛ* (1)

The results are reported in Panel A of Table III. Consistent with *H1*, the coefficient *β _{1}* is significantly positive, suggesting that reporting conservatively makes firms more likely to release early guidance to explain the effect of conservatism on future earnings. The results are in line.

[…] with expectations. For instance, the coefficients of FREQ, MVAL, SDRET, SDEARN, AFE and FOLLOW suggest that firms are reluctant to release early guidance when they are smaller, when they do not frequently forecast or when they operate in a turbulent, unpredictable environment

(Ajinkya *et al.*, 2005; Chen *et al.*, 2011; Dye, 1986; Houston *et al.*, 2010). Also, the coefficients of *NEGΔE* and *MISSEXP* suggest that reporting negative current performance makes firms unlikely to venture into making early predictions.

### 4.2 Conservatism and the probability that FΔE > 0 (hypothesis 2)

Different from *H1*, *H2* discriminates based on the sign of the guidance. To test this hypothesis, we run a multinomial logit regression specified by the following equation:

where the dependent variable is a categorical variable that can take three alternative values: zero, if the announcement is not combined with guidance; one if the announcement is combined with positive guidance (i.e. if *FΔE > 0*); and minus one if the announcement is combined with negative guidance. Compared to the binomial model used to test *H1*, which did not distinguish between positive and negative guidance, the multinomial model allows testing whether reporting conservatively increases the probability of *positive* guidance relative to the probability of no guidance. The results are reported in Panel B of Table III. Consistent with *H2*, *CONS* is significantly positive, suggesting that conservative firms are more likely to release positive early guidance rather than no guidance[11]. The coefficients of the other variables are also in line with expectations. For instance, the coefficients of *FREQ, MVAL, SDRET, SDEARN, DISP* and *FOLLOW* are consistent with those of Panel A. Similarly, the coefficients of *NEGΔE* and *MISSEXP* indicate that firms prefer cautious silence over predicting earnings increases when current earnings are disappointing, probably because of litigation risk.

### 4.3 The association of conservatism with FΔE when early guidance is released (hypothesis 3)

*H3* predicts that, if early guidance is released, conservatism is positively associated with the difference between forecasted earnings and current earnings. To test *H3* we use an OLS model and regress *FΔE* on *CONS* and the other controls. The results are presented in Panel C of Table III. Supporting *H3*, the coefficient of *CONS* is significantly positive, suggesting that the early guidance released by firms that have reported more conservatively is more favorable (i.e. more positive or less negative) than the early guidance of other firms[12]. Different from Panel A, the coefficients of *F_BIAS* is positive, which suggests that conditional on releasing early guidance, firms with a reputation for optimistic forecasting tend to predict positive earnings changes. Also, the coefficients of *FREQ*, *SDRET* and *SDEARN* are not significant, suggesting that guidance frequency and information uncertainty may affect a firm’s likelihood to remain silent, but not the forecast estimates if a firm chooses to release guidance. Similar to Panel B, firms reporting disappointing earnings seem to incorporate less good news in early guidance, perhaps waiting to assess the situation with better precision before committing to risky predictions.

## 5. Additional analyses

### 5.1 Alternative measures

In this section, we discuss the results obtained using alternative proxies for the two main constructs of our models, i.e. the forecasted earnings trend (*FΔE*) and the conservatism in current earnings (*CONS*).

#### 5.1.1 The forecasted earnings trend (FΔE).

In the results reported so far, *FΔE* is measured using the actual earnings number from the First Call database. This number often excludes items that may be considered non-operating or non-recurring, and therefore, is more likely to reflect a measurement of the “core” earnings of the firm rather than the reported earnings. This exclusion, however, could also be removing part of the earnings effects of conservatism from *FΔE*: Conservatism reduces current earnings by deferring revenues and gains and by anticipating expenses and losses, which can be accomplished by disseminating negative charges throughout the income statement. Some of these charges could be “special” items[13] – e.g. asset write-downs, goodwill impairment charges or restructuring costs. Special items have various implications for future earnings trends. Being less likely to recur than other items, they can result in a more favorable (i.e. more positive or less negative) future earnings trend. Also, current write-downs can decrease future core expenses such as depreciation or cost of goods sold. Moreover, special items such as restructuring costs can reflect managerial efforts to cut unprofitable investments and focus on the core business, which will have a favorable impact on future revenues and operating profits. Nevertheless, there is evidence that the capital markets fail to understand special items’ implications for future earnings[14], which could prompt managers to release early guidance to disclose such effects. Therefore, it is important to repeat the tests of our hypotheses without excluding from the measurements the effects of conservatism on current earnings that occur through special items. We do so by calculating *FΔE* using the current earnings number from Compustat, which includes special items.

The results (not tabulated for brevity) are consistent with those reported in Table III. We note that the coefficients’ magnitude grows materially (0.077 vs 0.055 in Panel B, 0.049 vs 0.018 in Panel C). This growth is consistent with *FΔE* now capturing also the earnings effects of conservatism that occur through special items and managers’ desire to disclose such effects using early guidance.

#### 5.1.2 The effects of conservatism on current earnings (CONS).

In the literature, *CRATIO* and *SKEW* have often been used to measure the conditional application of conservative accounting (Garcìa Lara *et al.*, 2011; Kim and Pevzner, 2010), while *NOACCR* and Δ*RES* have often been used to capture the unconditional form (Chen *et al.*, 2014; Kim *et al.*, 2018). To test whether either form is the main driver of our results, we analyze the two forms separately by calculating a conditional (unconditional) proxy, labeled *CCONS* (*UCONS*), as the average of the decile ranks of *CRATIO* and *SKEW* (of the decile ranks of *NOACCR* and *ΔRES).* We then further split each of the two proxies into its individual components (i.e. the decile rank of each proxy).

When testing *H1*, the results (not tabulated for brevity) appear to be somewhat stronger if we use *UCONS* (0.044, *t*-stat 3.37) compared to *CCONS* (0.025, *t*-stat 1.87). This effect seems to be mostly driven by *NOACCR*. On the other hand, the significance of *CRATIO* and *SKEW* appears to be weak when isolating the two proxies, which support *H1* only when combined. *H2* and *H3* are supported whether we use *CCONS* or *UCONS*. The main drivers of these results appear to be *CRATIO* and *NOACCR* (for both *H2* and *H3*), and *ΔRES* (for *H3*), while *SKEW* is not significant for either hypothesis. However, *SKEW* significantly supports both *H2* and *H3* when the measurement of *FΔE* includes special items. Overall, the main results appear to be driven by both conditional and unconditional conservatism proxies, though using *SKEW* yields results that are sensitive to the measurement of the dependent variable.

### 5.2 Later guidance

In this section, we look at post-announcement guidance patterns. We create a dummy variable (*LATERGUIDE*) equal to one if the firm has released a forecast of the earnings of the year t + 1 after the three-day window centered on the day of the announcement of the year *t* earnings, and 0 otherwise. We then run the binomial logit regression of equation (1) using *LATERGUIDE* as a dependent variable, first on a subsample made of firms that did not release early guidance, then on the subsample made of firms that did. In neither case we find *CONS* to be statistically significant (the t-statistics are 1.41 and 0.48, respectively). This suggests that when firms *do not* use early guidance to disclose the effects of conservatism on future earnings, they are unlikely to decide to do it later; and when firms *do* use early guidance to disclose the effects of conservatism on future earnings, such explanation is unlikely to require later updates[15]. Overall, these additional findings reinforce the role played by early guidance (as opposed to any guidance) to disclose how conservatism affects earnings.

### 5.3 The relevance of the early guidance that explains the earnings effects of conservatism

The results of the main analysis show that firms use early guidance to disclose how conservatism is expected to affect future earnings. A different question is whether such disclosure conveys information that is relevant to analysts (i.e. whether analysts need it and use it). We address this question in two ways. First, we analyze whether such information is new to analysts. If this is the case, then conservatism will increase the likelihood that and the extent to which early guidance beats analyst expectations. To test it, we repeat the analysis of *H2* and *H3* using the analyst pre-guidance consensus estimate as a benchmark, to isolate the portion of the forecast that is new to analysts. We label the variable so constructed *FNEWS* (as opposed to *FΔE*). If analysts are able to predict all earnings effects of conservatism, conservative firms’ guidance will, on average, contain no more news than any other firms’, and *FNEWS* will be unassociated with *CONS*. The results reported in Table IV suggest the opposite, indicating that analysts need early guidance to predict the effect of conservatism on future earnings.

Second, we analyze whether analysts *use* the information about the earnings effects of conservatism contained in early guidance, i.e. whether they decide to revise their expectations upward. Specifically, we analyze the change in the analyst consensus estimate (i.e. the change in the mean value of analyst forecasts relative to the earnings of year *t + 1*) that occurs after the release of positive early guidance (Δ*CONSENSUS*). If analysts use (ignore) such information, positive forecast news will be associated (unassociated) with upward revisions. Therefore, we test whether the association between the magnitude of positive *FNEWS* and Δ*CONSENSUS* becomes significantly weaker among more-conservative firms relative to other firms, which would suggest that conservative firms’ guidance contains information that analysts do not use. The results reported in Table V suggest the opposite as the coefficient of *FNEWS* is not significant in the low-conservatism subsample but becomes stronger and significant when conservatism is high. Overall, we conclude that early guidance released by conservative firms is relevant to analysts.

## 6. Concluding remarks

In this paper, we show that firms resort to voluntary disclosure to reveal their expectations about how conservatism affects future earnings trends. In particular, we show that they do so as early as the announcement of current earnings, which allows outsiders to combine managers’ predictions with the currently reported numbers. Our findings also suggest that analysts *need* such disclosure to correctly predict conservatism’s effects on future earnings, and *use* it by revising their expectations upward. Our findings have practical implications for regulators, preparers and users. Prior literature suggests that outsiders can struggle to understand the earnings effects of conservatism (Chen *et al.*, 2014; Penman and Zhang, 2002; Mensah *et al.*, 2004; Louis *et al.*, 2014; Barth *et al.*, 2017). This has led regulators to question the desirability of conservatism in financial statements (Watts, 2003; FASB, 2010). By showing that firms use early guidance to mitigate the potential negative impact of conservatism, our findings can alleviate regulators’ concerns. Moreover, our findings are interesting for preparers and users of accounting information. When considering the release of positive guidance to disclose the effect of conservatism on future earnings, managers weigh the risks and costs of such strategy against its potential benefits: if the former outweigh the latter, they will not do it. By finding that analysts need and use the information about conservatism that is contained in early guidance, we show evidence of the benefits of such disclosures, which could encourage managers to pursue it.

This paper contributes to growing academic research on the relationship between conservative financial reports and voluntary disclosure practices (Guay and Verrecchia, 2018; Hui *et al.*, 2009; Jaggi and Xin, 2014; D’Augusta and DeAngelis, 2017; Li, 2008; D’Augusta, 2018). Moreover, this study contributes to the literature on the consequences of conservatism. Recent studies have focused on its effects on information *users*, such as investors in the stock market (D’Augusta *et al.*, 2016; Kim and Zhang, 2016; Kim *et al.*, 2013; Barth *et al.*, 2017; Chen *et al.*, 2014; Louis *et al.*, 2014) or the debt market (Balakrishnan *et al.*, 2016; Sunder *et al.*, 2018). By contrast, the main focus of our paper is on the actions of *preparers* – i.e. their decision to release positive early guidance. A thorough analysis of the effect of conservatism on users’ reaction to earnings guidance is beyond the scope of this paper and we leave it to future research. Similarly, while our study focuses on US companies, future research could expand the analysis to the international setting, as conservatism is a phenomenon that has been empirically observed and studied in various countries (Ding and Stolowy, 2006; Noravesh *et al.*, 2007; Dimitropoulos and Asteriou, 2008; Ball *et al.*, 2008).

Descriptive statistics

Variable |
N | Mean | SD | 25th percentile | 50th percentile | 75th percentile |
---|---|---|---|---|---|---|

SKEW |
8,567 | 0.591 | 1.557 | −0.353 | 0.423 | 1.500 |

NOACCR |
3,999 | 0.025 | 0.056 | −0.001 | 0.020 | 0.045 |

CRATIO |
8,111 | −0.514 | 2.002 | −0.509 | −0.213 | 0.056 |

RES |
1,188 | 0.064 | 0.192 | 0.002 | 0.018 | 0.077 |

ΔRES |
960 | −0.007 | 0.110 | −0.006 | 0.000 | 0.007 |

FΔE |
5,502 | 0.024 | 0.052 | 0.003 | 0.008 | 0.022 |

ΔE |
8,820 | −0.015 | 0.123 | −0.013 | 0.005 | 0.017 |

NEGΔE |
8,820 | 0.380 | 0.485 | 0.000 | 0.000 | 1.000 |

FREQ |
8,820 | 3.634 | 2.318 | 2.000 | 4.000 | 5.000 |

F_BIAS |
8,820 | 0.029 | 0.070 | −0.002 | 0.003 | 0.032 |

F_INACCUR |
8,820 | 0.035 | 0.072 | 0.002 | 0.008 | 0.034 |

MVAL |
8,820 | 21.187 | 1.622 | 20.039 | 21.048 | 22.242 |

MB |
8,820 | 3.199 | 2.954 | 1.518 | 2.332 | 3.746 |

LEV |
8,820 | 0.796 | 1.001 | 0.208 | 0.460 | 0.981 |

SDRET |
8,820 | 0.027 | 0.013 | 0.018 | 0.025 | 0.034 |

SDEARN |
8,820 | 0.023 | 0.030 | 0.007 | 0.012 | 0.025 |

AFE |
8,820 | 0.004 | 0.011 | 0.000 | 0.001 | 0.003 |

DISP |
8,820 | 0.003 | 0.007 | 0.000 | 0.001 | 0.002 |

FOLLOW |
8,820 | 9.372 | 6.577 | 4.000 | 8.000 | 13.000 |

MISSEXP |
8,820 | 0.320 | 0.466 | 0.000 | 0.000 | 1.000 |

PAST_E |
8,820 | 0.038 | 0.069 | 0.027 | 0.048 | 0.065 |

NEGPAST_E |
8,820 | 0.123 | 0.328 | 0.000 | 0.000 | 0.000 |

SKEW and NOACCR, CRATIO, RES and ΔRES, are conservatism proxies based on, respectively, Givoly and Hayn (2000), Callen *et al.* (2010) and Penman and Zhang (2002). FΔE is the difference between the early guidance estimate and current earnings. ΔE is the difference between the current and previous year’s earnings. NEGΔE is a binary variable equal to one if ΔE is negative and 0 otherwise. FREQ is the total number of forecasts issued by the firm relative to current earnings. F_BIAS is the average error of the forecasts issued by the firm relative to current earnings. F_INACCUR is the absolute value of F_BIAS. MVAL is the logarithm of the total market value of equity. MB is the market-to-book ratio. LEV is the financial leverage. SDRET is the standard deviation of daily stock returns during the year. SDEARN is the standard deviation of quarterly earnings’ changes, computed over a rolling window of up to 16 quarters leading up to the last quarter of the current year. AFE is the absolute magnitude of analyst forecast error relative to the current earnings before the announcement. DISP is the analyst forecast dispersion before the announcement. FOLLOW is the number of analysts following the firm before the announcement. MISSEXP is a binary variable equal one if current earnings miss analyst consensus, 0 otherwise. PAST_E is the average value of the earnings of the past three years. NEGPAST_E is a binary variable equal to one if PAST_E is negative, 0 otherwise. Detailed definitions of all variables are provided in the Appendix

Pearson (lower diagonal) and spearman correlations (upper diagonal)

Variable |
1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

CONS |
1 | 0.097*** | −0.230* | 0.169* | −0.046* | 0.078* | 0.088* | −0.061* | 0.014 | −0.001 | 0.116* | 0.144* | 0.070* | 0.067* | −0.036* | 0.040* | −0.172* | 0.096* |

FΔE |
0.043** | 1.000 | 0.080*** | −0.059*** | −0.108*** | 0.524*** | 0.473*** | −0.106*** | 0.036** | −0.030* | 0.045*** | 0.050*** | −0.038** | −0.084*** | 0.006 | 0.025 | ||

ΔE |
−0.197* | −0.052*** | 1 | −0.716* | 0.067* | −0.248* | −0.127* | 0.071* | 0.157* | −0.117* | −0.060* | 0.031* | −0.131* | −0.179* | 0.002 | −0.145* | −0.168* | 0.102* |

NEGΔE |
0.216* | −0.031* | −0.449* | 1 | −0.065* | 0.186* | 0.136* | −0.096* | −0.209* | 0.183* | 0.102* | 0.035* | 0.183* | 0.243* | −0.042* | 0.115* | 0.113* | −0.036* |

FREQ |
−0.047* | −0.097*** | 0.073* | −0.069* | 1 | −0.142* | −0.124* | 0.236* | 0.040* | 0.053* | −0.224* | −0.124* | −0.095* | −0.096* | 0.208* | −0.069* | 0.078* | −0.059* |

F_BIAS |
0.067* | 0.824*** | −0.371* | 0.133* | −0.137* | 1 | 0.663* | −0.109* | −0.109* | 0.098* | 0.100* | 0.011 | 0.036* | 0.085* | −0.034* | 0.171* | 0.107* | −0.036* |

F_INACCUR |
0.056* | 0.751*** | −0.375* | 0.124* | −0.139* | 0.945* | 1 | −0.202* | −0.192* | 0.168* | 0.211* | 0.070* | 0.270* | 0.216* | −0.108* | 0.071* | 0.097* | 0.022* |

MVAL |
−0.065* | −0.119*** | 0.172* | −0.142* | 0.254* | −0.176* | −0.193* | 1 | 0.404* | −0.029* | −0.503* | −0.223* | −0.339* | −0.243* | 0.682* | −0.053* | 0.057* | −0.114* |

MB |
0.006 | −0.038** | 0.121* | −0.163* | 0.01 | −0.096* | −0.111* | 0.316* | 1 | −0.600* | −0.173* | 0.110* | −0.390* | −0.420* | 0.239* | −0.095* | −0.254* | 0.005 |

LEV |
0.023* | 0.069*** | −0.342* | 0.185* | 0.005 | 0.241* | 0.289* | −0.130* | −0.298* | 1 | −0.078* | −0.235* | 0.298* | 0.379* | −0.050* | 0.087* | 0.360* | −0.028* |

SDRET |
0.134* | 0.052*** | −0.257* | 0.151* | −0.215* | 0.181* | 0.205* | −0.468* | −0.081* | 0.123* | 1 | 0.318* | 0.273* | 0.256* | −0.196* | 0.029* | −0.250* | 0.148* |

SDEARN |
0.146* | 0.013 | −0.018 | 0.025* | −0.095* | 0.022* | 0.037* | −0.170* | 0.110* | −0.130* | 0.310* | 1 | 0.151* | 0.147* | −0.095* | 0.001 | −0.365* | 0.219* |

AFE |
0.097* | 0.135*** | −0.460* | 0.198* | −0.098* | 0.398* | 0.438* | −0.279* | −0.144* | 0.359* | 0.317* | 0.101* | 1 | 0.578* | −0.232* | 0 | 0.003 | 0.089* |

DISP |
0.110* | 0.065*** | −0.437* | 0.232* | −0.107* | 0.323* | 0.353* | −0.260* | −0.161* | 0.383* | 0.326* | 0.116* | 0.675* | 1 | −0.053* | 0.118* | 0.01 | 0.081* |

FOLLOW |
−0.036* | −0.064*** | 0.037* | −0.045* | 0.186* | −0.063* | −0.076* | 0.682* | 0.183* | −0.086* | −0.150* | −0.041* | −0.145* | −0.097* | 1 | −0.045* | −0.029* | −0.062* |

MISSEXP |
0.040* | 0.038** | −0.128* | 0.165* | −0.083* | 0.116* | 0.099* | −0.091* | −0.075* | 0.108* | 0.061* | 0.006 | 0.155* | 0.162* | −0.063* | 1 | −0.001 | 0.017 |

PAST_E |
−0.168* | 0.077*** | −0.153* | 0.101* | 0.087* | 0.078* | 0.064* | 0.123* | −0.142* | 0.133* | −0.282* | −0.508* | −0.065* | −0.070* | 0.038* | 0.01 | 1 | −0.315* |

NEGPAST_E |
0.163* | −0.014 | 0.068* | −0.052* | −0.101* | −0.016 | 0.011 | −0.209* | 0.048* | 0.005 | 0.309* | 0.469* | 0.110* | 0.121* | −0.099* | 0.01 | −0.719* | 1 |

CONS is the composite firm-year measure of conservatism. FΔE is the difference between the early guidance estimate and current earnings. ΔE is the difference between the current and previous year's earnings. NEGΔE is a binary variable equal to one if ΔE is negative and 0 otherwise. FREQ is the total number of forecasts issued by the firm relative to current earnings. F_BIAS is the average error of the forecasts issued by the firm relative to current earnings. F_INACCUR is the absolute value of F_BIAS. MVAL is the logarithm of the total market value of equity. MB is the market-to-book ratio. LEV is the financial leverage. SDRET is the standard deviation of daily stock returns during the year. SDEARN is the standard deviation of quarterly earnings' changes, computed over a rolling window of up to 16 quarters leading up to the last quarter of the current year. AFE is the absolute magnitude of analyst forecast error relative to the current earnings before the announcement. DISP is the analyst forecast dispersion before the announcement. FOLLOW is the number of analysts following the firm before the announcement. MISSEXP is a binary variable equal one if current earnings miss analyst consensus, 0 otherwise. PAST_E is the average value of the earnings of the past three years. NEGPAST_E is a binary variable equal to one if PAST_E is negative, 0 otherwise. * indicates significance at less than the 10% level. Detailed definitions of all variables are provided in the Appendix

Test of the hypotheses – Do firms use early guidance to predict the effect of conservatism on future earnings?

Panel A: Binomial logit model (H1) |
Panel B: Multinomial logit model (H2) |
Panel C: OLS model (H3) |
|||||||||
---|---|---|---|---|---|---|---|---|---|---|---|

Effect of conservatism (CONS) on the probability combined with an early positive change in earnings (CAT_FΔE) = 1) relative to guidance (FΔE) | Effect of conservatism (CONS) on the probability that an announcement is combined with early guidance predicting the magnitude of the change in the probability of no early guidance | Association between conservatism (CONS) and the that an announcement is predicted by early guidance relative to the probability of no early guidance (FΔE) | |||||||||

DEPVAR: EARLYGUID |
DEPVAR: CAT_FΔE |
DEPVAR: FΔE |
|||||||||

Exp. sign |
Coeff |
t-stat |
Exp. sign |
Coeff |
t-stat |
Exp. sign |
Coeff |
t-stat |
|||

CONS |
+ |
0.034** |
(2.39) |
CONS |
+ |
0.055*** |
(3.77) |
CONS |
+ |
0.018*** |
(3.85) |

ΔE |
+ | 0.007** | (2.39) | ΔE |
+ | 0.010*** | (2.89) | ΔE |
+ | 0.0148** | (5.52) |

NEGΔE |
− | −0.199*** | (−3.25) | NEGΔE |
− | −0.333*** | (−5.14) | NEGΔE |
− | −0.087*** | (−4.90) |

FREQ |
+ | 0.934*** | (22.53) | FREQ |
+ | 0.944*** | (22.61) | FREQ |
+/− | −0.001 | (−0.17) |

F_BIAS |
+/− | −0.015 | (−1.05) | F_BIAS |
+ | 0.108*** | (4.15) | F_BIAS |
+ | 0.158*** | (16.87) |

F_INACCUR |
+/− | 0.010 | (0.70) | F_INACCUR |
+/− | −0.100*** | (−3.84) | F_INACCUR |
+/− | −0.002 | (−0.24) |

MVAL |
+ | 0.202*** | (4.15) | MVAL |
+ | 0.213*** | (4.26) | MVAL |
+/− | −0.033** | (−2.48) |

MB |
+/− | −0.000 | (−0.00) | MB |
+/− | 0.016 | (0.49) | MB |
+/− | 0.007 | (0.73) |

LEV |
+/− | −0.007 | (−0.19) | LEV |
+/− | −0.038 | (−0.98) | LEV |
+/− | −0.032* | (−1.68) |

SDRET |
− | −0.091** | (−2.09) | SDRET |
− | −0.113** | (−2.48) | SDRET |
+/− | −0.016 | (−0.98) |

SDEARN |
− | −0.048 | (−1.44) | SDEARN |
− | −0.070** | (−1.97) | SDEARN |
+/− | 0.006 | (0.51) |

AFE |
− | −0.125*** | (−3.25) | AFE |
− | −0.159*** | (−3.49) | AFE |
+/− | 0.010 | (0.24) |

DISP |
− | −0.027 | (−0.70) | DISP |
− | −0.058 | (−1.25) | DISP |
+/− | −0.006 | (−0.22) |

FOLLOW |
− | −0.020*** | (−3.10) | FOLLOW |
− | −0.023*** | (−3.45) | FOLLOW |
+/− | −0.001 | (−0.89) |

MISSEXP |
− | −0.212*** | (−3.69) | MISSEXP |
− | −0.178*** | (−3.03) | MISSEXP |
− | −0.051*** | (−2.89) |

PAST_E |
+/− | 0.010* | (1.72) | PAST_E |
+/− | 0.011* | (1.73) | PAST_E |
+/− | 0.002 | (0.75) |

NEGPAST_E |
+/− | 0.042 | (0.37) | NEGPAST_E |
+/− | 0.043 | (0.35) | NEGPAST_E |
+/− | 0.029 | (0.70) |

Constant |
+/− | −4.704*** | (−7.16) | Constant |
+/− | −5.024*** | (−7.41) | Constant |
+/− | 0.800*** | (3.49) |

Industry/Year FE |
YES | Industry/Year FE |
YES | Industry/Year FE |
YES | ||||||

Observations |
8,820 | Observations |
8,820 | Observations |
5,502 | ||||||

Pseudo R^{2} |
0.209 | Pseudo R^{2} |
0.203 | Adj. R^{2} |
0.718 |

All continuous variables are winsorized at the first and 99th percentiles. The *t*-statistics are reported in parentheses. Standard errors are heteroskedasticity robust (White 1980) and clustered at the firm level. The dependent variable in the binomial logit model (Panel A), EARLYGUID, is a categorical variable that takes the value of one if there is early guidance, and 0 otherwise. The dependent variable in the multinomial logit model (Panel B), CAT_FΔE, is a categorical variable that takes the value of 0 if there is no early guidance; one if the announcement is combined with positive early guidance (FΔE > 0); minus one if the announcement is combined with negative early guidance (FΔE < 0). The dependent variable in the OLS model (Panel C), FΔE, is the difference between the early guidance estimate and current earnings. CONS is the composite firm-year measure of conservatism. *, ** and *** indicate significance at the 10, 5 and 1%levels respectively. Detailed definitions of all variables are provided in the Appendix

Relevance of early guidance – Do analysts need the information about the earnings effects of conservatism contained in early guidance?

Panel A: Multinomial logit model | Panel B: OLS model | ||||||
---|---|---|---|---|---|---|---|

DV: CAT_FNEW | Exp. sign |
Coeff |
t-stat |
DV: FNEWS |
Exp. sign |
Coeff |
t-stat |

Effect of conservatism (CONS) on the probability that an announcement is combined with early guidance exceeding analysts’ consensus (CAT_FNEWS = 1) rather than not combined with guidance (CAT_FNEWS = 0) | Association between conservatism (CONS) and the difference between the early guidance estimate and analysts’ consensus before the announcement (FNEWS) | ||||||

CONS |
+ |
0.058*** |
(3.49) |
CONS |
+ |
0.017*** |
(3.16) |

ΔE |
+ | 0.016*** | (3.92) | ΔE |
+ | 0.021*** | (6.84) |

NEGΔE |
− | −0.327*** | (−4.53) | NEGΔE |
− | −0.063** | (−2.44) |

FREQ |
+ | 0.972*** | (21.68) | FREQ |
+/− | 0.009 | (0.71) |

F_BIAS |
+ | 0.047*** | (2.81) | F_BIAS |
+ | 0.103*** | (5.80) |

F_INACCUR |
+/− | −0.015 | (−0.92) | F_INACCUR |
+/− | 0.042** | (2.30) |

MVAL |
+ | 0.175*** | (3.03) | MVAL |
+/− | −0.037 | (−1.44) |

MB |
+/− | 0.031 | (0.89) | MB |
+/− | 0.005 | (0.32) |

LEV |
+/− | −0.029 | (−0.69) | LEV |
+/− | −0.085*** | (−3.46) |

SDRET |
− | −0.134** | (−2.55) | SDRET |
+/− | −0.076*** | (−3.48) |

SDEARN |
− | −0.044 | (−1.11) | SDEARN |
+/− | −0.024* | (−1.71) |

AFE |
− | −0.086* | (−1.88) | AFE |
+/− | −0.136** | (−2.48) |

DISP |
− | −0.153*** | (−2.74) | DISP |
+/− | −0.033 | (−1.03) |

FOLLOW |
− | −0.019** | (−2.48) | FOLLOW |
+/− | −0.003 | (−1.12) |

MISSEXP |
− | −0.581*** | (−8.13) | MISSEXP |
− | −0.168*** | (−7.71) |

PAST_E |
+/− | 0.022*** | (2.97) | PAST_E |
+/− | 0.007** | (2.28) |

NEGPAST_E |
+/− | −0.030 | (−0.21) | NEGPAST_E |
+/− | −0.054 | (−1.18) |

Constant |
+/− | −5.256*** | (−6.79) | Constant |
+/− | 1.226*** | (3.40) |

Industry/Year FE |
YES | Industry/Year FE |
YES | ||||

Observations |
8,680 | Observations |
5,362 | ||||

Pseudo R^{2} |
0.182 | Adj. R^{2} |
0.585 |

Notes: All continuous variables are winsorized at the first and 99th percentiles. The *t*-statistics are reported in parentheses. Standard errors areheteroskedasticity robust (White 1980) and clustered at the firm level. The dependent variable in the multinomial logit model (Panel A) CAT_FNEWS, is a categorical variable that takes the value of zero if there is no early guidance; one if the announcement is combined with early guidance that is higher than the pre-guidance analyst consensus (FNEWS > 0); minus one if the announcement is combined with early guidance that is not higher than the pre-guidance analyst consensus (FNEWS < 0). The dependent variable in the OLS model (Panel B), FNEWS, is the difference between the early guidance estimate and the pre-guidance analyst consensus. CONS is the composite firm year measure of conservatism. *, ** and *** indicate significance at the 10, 5 and 1% levels respectively. Detailed definitions of all variables are provided in the Appendix

Relevance of early guidance – Do analysts use the new information about the earnings effects of conservatism disclosed through early guidance?

Split: high-conservatism vs. low-conservatism firms | ||||
---|---|---|---|---|

Subsample: LOW CONS | Subsample: HIGH CONS | Difference Test | Exp. Sign | |

FNEWS |
0.005 (0.16) |
0.091*** (2.59) |
0.086** (2.03) |
? |

Controls and FE |
YES | YES | ||

Observations |
1,520 | 1,219 | ||

Adj. R^{2} |
0.139 | 0.153 |

Notes: All control variables are included in the models, but not tabulated for brevity. All continuous variables are winsorized at the 1st and 99th percentiles. The *t*-statistics are reported in parentheses. Standard errors are heteroskedasticity robust (White 1980) and clustered at the firm level. ΔCONSENSUS is the change in analyst consensus relative to the earnings of year t + 1 that occurs after the forecast release. FNEWS is the difference between the early guidance estimate and the pre-guidance analyst consensus. To facilitate the comparison of coefficients across the panels of this table, ΔCONSENSUS and FNEWS are scaled by their sample standard deviation. *, ** and *** indicate significance at the 10, 5 and 1% levels, respectively. Detailed definitions of all variables is provided in the Appendix

In the equation above, the coefficient *β*1 represents the association between the change in the mean analyst forecast (ΔCONSENSUS) and the magnitude of FNEWS when the sample is limited to early guidance that beat the pre-guidance consensus

## Notes

The literature distinguishes between the conditional and the unconditional application of conservative accounting.As per Beaver and Ryan (2005, p. 269), “under unconditional conservatism, the book value of net assets is understated due to predetermined aspects of the accounting process. Under conditional conservatism, book value is written down under sufficiently adverse circumstances, but not up under favorable circumstances.” Although the two forms of conservatism are not merely substitutes of each other as they arise for different reasons and have different implications for contracting efficiency, the literature shows that “both forms of conservatism bias […] earnings downward” (Qiang, 2007, p. 770). Our paper does not focus on the causes of conservatism or its contracting usefulness, but rather on its effects on earnings and their relation to earnings guidance. Therefore, our analysis is not limited to any of the two forms, but rather examines the earnings effects of conservatism whether it is applied conditionally or unconditionally.

The Statement of Financial Accounting Concepts No. 8 rejects “conservatism as an aspect of faithful representation because including [it] would be inconsistent with neutrality (BC 3.27, p. 28)”. Specifically, standard setters’ concerns appear to stem from the undesirability of the earnings effects of conservatism, which would impair the usefulness of earnings information relative to neutral reporting: “An admonition to be [conservative] is likely to lead to a bias. Understating assets or overstating liabilities in one period frequently leads to overstating financial performance in later periods – a result that cannot be described as prudent or neutral. (BC 3.28, p. 28)”

The diversity between our research question and Hui’s *et al.* (2009) is the reason for various differences in the research design of the two papers. For instance, Hui *et al.* (2009) measure conservatism over a six-year window *before* the six-year window used to measure forecast frequency and other forecast attributes. We measure conservatism in the. Year *t* and then measure whether the announcement of year *t* earnings is combined with guidance relative to the year t + 1. In addition, we do not look at the probability or the number of forecasts released at any time in a multiple-year period. We look at the probability of early guidance and whether it predicts the effects of conservatism on future earnings.

To maximize the power of our tests, if a conservatism proxy is unavailable, *CONS* is equal to the average of the other available proxies.

Similarly, we choose not to use the conditional conservatism proxy developed by Khan and Watts (2009) (i.e. *C_Score*). *C_Score* is a firm-year specific estimate of the coefficient of asymmetric timeliness in the Basu (1997) model based on the value predicted by the firm market capitalization, market leverage and market-to-book ratio. Because these three predictors tend to be very stable over time, *C_Score* can be an optimal choice to test hypotheses related to the forces that demand conservatism, to a firm’s reputation for conservatism and to how the markets price such reputation. It would be suboptimal, however, in the context of our hypotheses: as between-firms variations in *C_Score* in a given year would only reflect variations in the three predictors and not in the extent that conservatism has depressed a given firm’s current earnings (i.e. *C_Score* would not capture the difference between MC and LC in the example of section 2.3), there would be no reason to hypothesize that *C_Score* is associated with positive early guidance.

In the main analysis, *FΔE* is computed using current earnings obtained from First Call, which excludes many types of non-operating or non-recurring items. In Section 5.1 we report the results obtained when using current earnings obtained from Compustat, which includes such items.

In untabulated analyses we also include additional controls, such as the number and the value of the company’s shares owned by the CEO, the tenure of the CEO, the size of the board, the percentage of executive directors and independent auditors on the board. We do not use these variables in the main tests because their lack of availability would greatly reduce the sample size. Including the additional controls always leaves the coefficients of interest practically unaffected, and sometimes even stronger in magnitude, suggesting that our main models do not suffer from significant omitted-variable issues.

Financial firms, excluded from the sample, are identified according to Fama and French (1997). Market data are obtained from the CRSP daily stock database and analysts’ forecast data from I/B/E/S’s Detail History. All data are obtained through the Wharton Research Data Services (WRDS) website.

Moreover, our study does not aim to investigate whether conservatism affects the likelihood of being a forecaster but, rather, whether it makes a forecaster release positive guidance together with or before the annual earnings announcement. Anyway, for robustness we repeat the test the hypotheses on a broader sample (26,231 observations) that includes observations with missing forecast data. We find the results to be qualitatively similar, and therefore, robust to alternative sampling criteria.

This suggests that such “late” guidance may be of little help for the purpose of disclosing the effects of current conservatism on future earnings: indeed, Narayanamoorthy (2006) finds that investors’ failure to understand such effects gives rise to anomalous return patterns for approximately 90 days after the current earnings announcements, after which the information content of quarterly announcements begins to resolve the mispricing.

In an untabulated test, we run a binomial logit regression where the dependent variable is equal to 1 for positive guidance and zero for negative guidance. Consistent with expectations, the coefficient of *CONS* is significantly positive. This is also consistent with our test of *H3* where *CONS* is found to increase *FΔE*: if *FΔE* is higher, it is more likely to be positive than negative, all else equal. We thank an anonymous reviewer for suggesting this test.

To ensure that the results are not influenced by observations that have very low levels of conservatism and issue negative guidance to reduce litigation risk, we repeat the test using only observations that are above the fifth decile of *CONS*. The results (not tabulated for brevity) are qualitatively similar.

Chen *et al.* (2014, pp. 236, 244) “find that conservative reporting affects earnings persistence beyond the effects of special items” and note that “special items represent just one possible way to incorporate conservatism in earnings and not all special items result in more conservative reporting”. Consistent with this notion, Frankel and Roychowdhury (2008, p. 6) show that “[more-conservative firms] are no more likely to report significantly large negative special items. Nor are their negative special items any larger in magnitude as a percentage of sales, on average”. However, they “predict and find that negative special items of [less-conservative] firms are […] more persistent with respect to future net income.” Their findings suggest that conservatism’s effects on special items pertains not so much to their frequency and magnitude, but rather to their implications for future earnings.

For instance, Dechow and Ge (2005, pp. 253, 256) find that special items explain future stock returns, which is “consistent with investors misunderstanding the transitory nature of special items” and “overweigh the probability that the firm will be unsuccessful [in turning around the current earnings trend]”. While investors’ misunderstanding of special items could be alleviated by analyst production of information, Louis *et al.* (2014, p.20) note that “there is no consensus among analysts regarding the treatment of special items and, consequently, the earnings forecasted by analysts are not generally fully adjusted for special items”. Investor valuation of special items can be made harder by managerial opportunism: Frankel and Roychowdhury (2008) show that not all special items are “equally special”. Special items reported by more-conservative firms are more predictive of favorable earnings trends because they reflect fuller recognition of losses and managerial actions to turn the firm around. By contrast, less-conservative firms are more likely to classify recurring operating expenses as “special”, which makes their special items more persistent.

These findings should not be interpreted as conservative firms necessarily remaining silent after the release of early guidance: they may well release additional forecasts for various reasons. However, such reasons do not appear to include managers’ desire to disclose how the level of conservatism in current earnings is going to affect future earnings.

## Appendix

*Variables’ definition*:

*CONS*composite firm-year measure of conservatism, equal to the average of the decile ranks of_{it}=*SKEW*,*NOACCR*,*CRATIO*, and*ΔRES.*To maximize the power of our tests, if a conservatism proxy is unavailable,*CONS*is equal to the average of the other available proxies.*CRATIO*annual decile ranking of the conservatism ratio, expressing the proportion of total earnings news that is recognized in current reported earnings. The ratio is calculated according to Callen_{it}=*et al.*(2010), that is, η2_{i,t}/Ne_{i,t}. The numerator and denominator are estimated following Callen and Segal (2010). The ratio is multiplied by −1 if the denominator is positive, to reflect the asymmetric timeliness in the recognition of good and bad news (Biddle*et al.*, 2016). Decile rankings are calculated separately for positive and negative shocks.*SKEW*conservatism measure developed by Givoly and Hayn (2000), equal to the difference between the skewness in cash flows from operations and the skewness in earnings before extraordinary items (both scaled by total assets) estimated over a rolling window of up to 20 quarters (at least nine required) ending in the fourth quarter of year t._{it}=*NOACCR*non-operating accruals, measured as per Givoly and Hayn (2000), multiplied by −1._{it}=*ΔRES*difference between_{it}=*RES*and_{t}*RES*_{t}_{-1}, where*RES*is the conservatism proxy developed by Penman and Zhang (2002) based on inventory, research and development (R&D), and advertising reserves.*EARLYGUID*= binary variable that assumes the value of one if the announcement of year_{it}*t*earnings is combined with early guidance (i.e. if the announcement is bundled with or preceded by a management forecast) predicting year t+1 earnings, and 0 otherwise.*FΔE*_{it}difference between the forecasted earnings per share (relative to the year*=**t*+ 1) and the current earnings per share (year t), scaled by the stock price.*CAT_FΔE*_{it}*=*Categorical transformation of, which can take three different values: 0, if the announcement is not combined with early guidance (i.e. if there is no value of*FΔE**FΔE*); one, if the announcement is combined with early guidance predicting a positive earnings change (i.e. if*FΔE*> 0); minus one, if the announcement is combined with early guidance not predicting a positive earnings change (i.e. if*FΔE*≤ 0).*ΔE*difference between the current earnings per share (year t) and the previous year’s earnings per share (year_{it}=*t*−1), scaled by the stock price.*NEGΔE*binary variable that assumes the value of one if_{it}=is negative and 0 otherwise.*ΔE*_{it}*MVAL*natural logarithm of the total market value of equity at the end of the fiscal year._{it}=*LEV*financial leverage, measured as total liabilities scaled by the total market value of equity at the end of the fiscal year._{it}=*MB*market-to-book ratio, measured as the market value of equity divided by the book value of equity at the end of the fiscal year._{it}=*SDRET*standard deviation of daily stock returns during the year._{it}=*FOLLOW*number of analysts following the shares of firm i on the day before the announcement._{it}=*DISP*standard deviation of analysts’ forecasts of the earnings per share of year_{it}=*t*scaled by stock price. The standard deviation is calculated on the day before the announcement, based on forecasts that have been issued or confirmed within the previous 65 days.*AFE*= magnitude of analyst forecast error relative to the earnings per share of year_{it}*t*, measured as the absolute difference between the actual earnings per share and the analyst consensus estimate measured on the day before the announcement, scaled by stock price.*SDEARN*standard deviation of quarterly earnings’ changes, computed over a rolling window of up to 16 quarters before the end of year t. Earnings changes are scaled by total assets at the beginning of the quarter._{it}=*FREQ*number of forecasts issued by the firm relative to the earnings being currently announced (i.e. the earnings of year t)._{it}=*F_BIAS*average error of the forecasts issued by the firm relative to the earnings being currently announced (i.e. the earnings of year t), scaled by stock price. The error is calculated as forecasted earnings less actual earnings._{it}=*F_INACCUR*absolute value of_{it}=*F_BIAS*._{it}*PAST_E*average value of the earnings per share of the years t-1, t-2 and t-3, scaled by stock price._{it}=*NEGPAST_E*binary variable that assumes the value of one if_{it}=*PAST_E*is negative and 0 otherwise._{it}*MISSEXP*binary variable that assumes the value of one if the earnings per share of firm i in year_{it}=*t*is below the analyst consensus estimate measured on the day before the announcement, and 0 otherwise.*FNEWS*difference between the forecasted earnings per share and the average of analyst estimates relative to next year’s earnings, measured five days before the early guidance release, scaled by stock price._{it}=*CAT_FNEWS*Categorical transformation of_{it}=*FNEWS*, which can take three different values: 0, if the announcement is not combined with early guidance (i.e. if there is no value of*FNEWS*); one, if the announcement is combined with early guidance exceeding the analyst consensus estimate (i.e. if*FNEWS*> 0); minus one, if the announcement is combined with early guidance not exceeding the analyst consensus estimate (i.e. if*FNEWS*≤ 0).*Δ**CONSENSUS*change in the analyst consensus estimate relative to the earnings of year t + 1 that occurs after the earnings announcement of year t. We calculate the difference between the average of all outstanding analysts’ forecasts 5 days after the announcement and the average 5 days before the announcement._{it}=

## References

Aier, J.K., Chen, L. and Pevzner, M. (2014), “Debtholders’ demand for conservatism: evidence from changes in directors’ fiduciary duties”, Journal of Accounting Research, Vol. 52 No. 5, pp. 993-1027.

Ajinkya, B.B., Bhojraj, S. and Sengupta, P. (2005), “The association between outside directors, institutional investors and the properties of management earnings forecasts”, Journal of Accounting Research, Vol. 43 No. 3, pp. 343-376.

Balakrishnan, K., Watts, R.L. and Zuo, L. (2016), “The effect of accounting conservatism on corporate investment during the global financial crisis”, Journal of Business Finance and Accounting, Vol. 43 Nos 5/6, pp. 513-542.

Ball, R., Robin, A. and Sadka, G. (2008), “Is financial reporting shaped by equity markets or by debt markets? An international study of timeliness and conservatism”, Review of Accounting Studies, Vol. 13 Nos 2/3, pp. 168-205.

Barth, M.E. Landsman, W.R. Raval, V. and Wang, S. (2017), “Asymmetric timeliness and the resolution of investor disagreement and uncertainty at earnings announcements”, available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2348905

Basu, S. (1997), “The conservatism principle and the asymmetric timeliness of earnings”, Journal of Accounting and Economics, Vol. 24 No. 1, pp. 3-37.

Beaver, W.H. and Ryan, S.G. (2005), “Conditional and unconditional conservatism: concepts and modeling”, Review of Accounting Studies, Vol. 10 Nos 2/3, pp. 269-309.

Biddle, G.C. Ma, M.L. and Song, F.M. (2016), “Accounting conservatism and bankruptcy risk”, available at: SSRNhttp://papers.ssrn.com/sol3/papers.cfm?abstract_id=1621272

Burgstahler, D.C. and Dichev, I.D. (1997), “Earnings management to avoid earnings decreases and losses”, Journal of Accounting and Economics, Vol. 24 No. 1, pp. 99-126.

Callen, J.L. and Segal, D. (2010), “A variance decomposition primer for accounting research”, Journal of Accounting Auditing and Finance, Vol. 25 No. 1, pp. 121-143.

Callen, J.L., Segal, D. and Hope, O.-K. (2010), “The pricing of conservative accounting and the measurement of conservatism at the firm-year level”, Review of Accounting Studies, Vol. 15 No. 1, pp. 145-178.

Chen, L.H., Folsom, D., Paek, W. and Sami, H. (2014), “Accounting conservatism, earnings persistence and pricing multiples on earnings”, Accounting Horizons, Vol. 28 No. 2, pp. 233-260.

Chen, S., Matsumoto, D. and Rajgopal, S. (2011), “Is silence golden? An empirical analysis of firms that stop giving quarterly earnings guidance”, Journal of Accounting and Economics, Vol. 51 Nos 1/2, pp. 134-150.

Coller, M. and Yohn, T.L. (1997), “Management forecasts and information asymmetry: an examination of Bid-Ask spreads”, Journal of Accounting Research, Vol. 35 No. 2, pp. 181.

D’Augusta, C. (2018), “Does accounting conservatism make good news forecasts more credible and bad news forecasts less alarming?”, Journal of Accounting Auditing and Finance, forthcom.

D’Augusta, C., Bar-Yosef, S. and Prencipe, A. (2016), “The effects of conservative reporting on investor disagreement”, European Accounting Review, Vol. 25 No. 3, pp. 451-485.

D’Augusta, C. and DeAngelis, M.D. (2017), “Does accounting conservatism discipline qualitative disclosure? Evidence from tone management in the MD&A”, available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2910403

Dechow, P.M. and Ge, W. (2005), “The persistence of earnings and cash flows and the role of special items”, Review of Accounting Studies, Vol. 11 Nos 2/3, pp. 253-296.

Desai, N. (2016), “Impact of company level information on auditor conservatism”, Review of Accounting and Finance, Vol. 15 No. 4, pp. 518-532.

Dimitropoulos, P.E. and Asteriou, D. (2008), “Timeliness, conservatism and financial transparent firms under the greek accounting setting”, Review of Accounting and Finance, Vol. 7 No. 3, pp. 252-269.

Ding, Y. and Stolowy, H. (2006), “Timeliness and conservatism: changes over time in the properties”, Review of Accounting and Finance, Vol. 5 No. 2, pp. 92-107.

Dye, R. (1986), “Proprietary and nonproprietary disclosures”, The Journal of Business, Vol. 59 No. 2, pp. 331-366.

Fama, E.F. and French, K.R. (1997), “Industry costs of equity”, Journal of Financial Economics, Vol. 43 No. 2, pp. 153-193.

FASB (2010), “Conceptual framework for financial reporting”, in Statement of Financial Accounting Concepts, Vol. 8. Financial Accounting Standards Board.

Francis, J., Philbrick, D. and Schipper, K. (1994), “Shareholder litigation and corporate disclosures”, Journal of Accounting Research, Vol. 32 No. 2, pp. 137-164.

Frankel, R.R. and Roychowdhury, S. (2008), “Are all special items equally special? The predictive role of conservatism”, available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1001434

Garcìa Lara, J.M., Garcìa Osma, B. and Penalva, F. (2011), “Conditional conservatism and cost of Capital”, Review of Accounting Studies, Vol. 16 No. 2, pp. 247-271.

Givoly, D. and Hayn, C. (2000), “The changing time-series properties of earnings, cash flows and accruals: Has financial reporting become more conservative?”, Journal of Accounting and Economics, Vol. 29 No. 3, pp. 287-320.

Graham, J.R., Harvey, C.R. and Rajgopal, S. (2005), “The economic implications of corporate financial reporting”, Journal of Accounting and Economics, Vol. 40 Nos 1/3, pp. 3-73.

Guay, W. and Verrecchia, R.E. (2018), “Conservative disclosure”, Journal of Financial Reporting,

Houston, J.F., Lev, B. and Tucker, J.W. (2010), “To guide or not to guide? Causes and consequences of stopping quarterly earnings guidance”, Contemporary Accounting Research, Vol. 27 No. 1, pp. 143-185.

Hui, K.W., Matsunaga, S.R. and Morse, D. (2009), “The impact of conservatism on management earnings forecasts”, Journal of Accounting and Economics, Vol. 47 No. 3, pp. 192-207.

Jaggi, B. and Xin, H. (2014), Accounting conservatism and management earnings forecasts, available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2246168

Kellogg, R.L. (1984), “Accounting activities, security prices, and class action lawsuits”, Journal of Accounting and Economics, Vol. 6 No. 3, pp. 185-204.

Khan, M. and Watts, R.L. (2009), “Estimation and empirical properties of a firm-year measure of accounting conservatism”, Journal of Accounting and Economics, Vol. 48 Nos 2/3, pp. 132-150.

Kim, Y., Li, S., Pan, C. and Zuo, L. (2013), “The role of accounting conservatism in the equity market: Evidence from seasoned equity offerings”, The Accounting Review, Vol. 88 No. 4, pp. 1327-1356.

Kim, J.B. Nekrasov, A. Pervin, S.K. and Simon, A. (2018), “Valuation implications of unconditional accounting conservatism: evidence from analysts’ target prices”, available at: https://onlinelibrary.wiley.com/doi/abs/10.1111/1911-3846.12465

Kim, B.H. and Pevzner, M. (2010), “Conditional accounting conservatism and future negative surprises: an empirical investigation”, Journal of Accounting and Public Policy, Vol. 29 No. 4, pp. 311-329.

Kim, J.-B. and Zhang, L. (2016), “Accounting conservatism and stock price crash risk: Firm-level evidence”, Contemporary Accounting Research, Vol. 33 No. 1, pp. 412-441.

Krishnan, G.V., Pevzner, M. and Sengupta, P. (2012), “How do auditors view managers’ voluntary disclosure strategy? The effect of earnings guidance on audit fees”, Journal of Accounting and Public Policy, Vol. 31 No. 5, pp. 492-515.

LaFond, R. and Watts, R.L. (2008), “The information role of conservatism”, The Accounting Review, Vol. 83 No. 2, pp. 447-478.

Lee, S., Matsunaga, S.R. and Park, C.W. (2012), “Management forecast accuracy and CEO turnover”, The Accounting Review, Vol. 87 No. 6, pp. 2095-2122.

Li, Z. (2008), “Accounting conservatism and managers’ forecasting behavior”, Doctoral Dissertation, *University of Minnesota*.

Louis, H., Lys, T. and Sun, A. (2014), Conservatism, Analyst Ability, and Forecast Error: Evidence on Financial Statement Users ‘ Ability to account for Conservatism, available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1031981

Mensah, Y.M., Song, X. and Ho, S.S.M. (2004), “The effect of conservatism on analysts’ annual earnings forecast accuracy and dispersion”, Journal of Accounting, Auditing and Finance, Vol. 19 No. 2, pp. 159-183.

Mindak, M.P., Sen, P.K. and Stephan, J. (2016), “Beating threshold targets with earnings management”, Review of Accounting and Finance, Vol. 15 No. 2, pp. 198-221.

Narayanamoorthy, G.S. (2006), “Conservatism and Cross-Sectional variation in the post–earnings announcement drift”, Journal of Accounting Research, Vol. 44 No. 4, pp. 763-789.

Noravesh, I., Dianati Dilami, Z. and Bazaz, M.S. (2007), “The impact of culture on accounting: does gray’s model apply to Iran?”, Review of Accounting and Finance, Vol. 6 No. 3, pp. 254-272.

Penman, S.H. and Zhang, X.-J. (2002), “Accounting conservatism, the quality of earnings, and stock returns”, The Accounting Review, Vol. 77 No. 2, pp. 237-264.

Qiang, X. (2007), “The effects of contracting, litigation, regulation, and tax costs on conditional and unconditional conservatism: cross-sectional evidence at the firm level”, The Accounting Review, Vol. 82 No. 3, pp. 759-796.

Skinner, D.J. (1994), “Why firms voluntarily disclose bad news”, Journal of Accounting Research, Vol. 32 No. 1, pp. 38.

Skinner, D.J. and Sloan, R.G. (2002), “Earnings surprises, growth expectations, and stock returns or don’t let an earnings torpedo sink your portfolio”, Review of Accounting Studies, Vol. 7 No. 2/3, pp. 289-312.

Sunder, J., Sunder, S.V. and Zhang, J. (2018), “Balance sheet conservatism and debt contracting”, Contemporary Accounting Research, Vol. 35 No. 1, pp. 494-524.

Vuolteenaho, T. (2002), “What drives firm-level stock returns?”, The Journal of Finance, Vol. 57 No. 1, pp. 233-264.

Watts, R.L. (2003), “Conservatism in accounting part I: explanations and implications”, Accounting Horizons, Vol. 17 No. 3, pp. 207-221.

Weil, R.L., Schipper, K. and Francis, J. (2012), Financial Accounting: An Introduction to Concepts, Methods and Uses, 14th ed. South-Western College.