The estimated reading time for this post is 20 Minutes

In this article, the author used the same methodology employed by Almeida et al. (2012) with some modifications in relation to, particular aspects of Mozambique. Control variables were also introduced to make the tests more robust. Therefore, initially, the coefficient of variation, the method developed by Eckel (1981) to determine the presence of smoother firms and non-smoother was applied. Subsequently, both the Basu (1997) primary model and C_Score metric[vii]. to gauge the degree of conservatism was applied and, finally, the regression to test the hypothesis of the research were applied.

## 3.1 Data and sample selection

The sample employed in this study consists of the top 100 companies in Mozambique according to the KPMG – Mozambique index for the period 2010 to 2015. All financial reporting data were extracted from the KPMG – Mozambique database. We supplement these data with financial data from the BVM-Mozambique Stock Exchange and the National Institute of Statistics (INE) of Mozambique and/or manually collected from the financial statements on the company website respectively.The data were processed and were analyzed using SAS software. The sample is composed only of non-financial companies since financial and investment institutions have specific accounting standard and in agreement with the study of Leuz et al. (2003). The initial sample was reduced to 10335 observations, classified in a first stage, as smoothing firms and non-smoothing firms, based on the Eckel (1981) criterion. First, the ratio is calculated for each company in the sample over a period of 6 years (study period corresponded to the years 2010 to 2015), if at least five years of data are available. In a second stage, the Basu (1997) primary model and C_Score metric to gauge the degree of Conservatism were applied.

Consistent with the studies of Almeida et al. (2012) and Valipour et al. (2011) required observations (net income and sales revenue) were obtained to sort the companies into smoothing firms and non-smoothing firms. Then, variables with extreme values based on Eckel`s model were eliminated, lowering the sample to 8378 observations. In order, to minimize the effect of the outliers in the regressions, the upper and lower extreme values in each variable were disqualified (1% standard deviations in each tail). This treatment is in agreement with the works of Basu (1997), Ball et at. (2000), and different from Almeida et al. (2012), which disqualified from the sample 3% standard deviations of the upper and lower extremes in each variable applied to the models. In this way, the final database used to sort the firms in smoothing companies and non-smoothing companies in econometrics tests included 6.524 observations.

## 3.2 Income smoothing measure

The Eckel`s model (1981) is applied in several studies to measure the existence of smoothing firms and non-smoothing firms, this model assumes that net income is related to sales by a linear function. According to Eckel`s model (1981), *Ceteris paribus*, the coefficient of variation of sales should be lower than the coefficient of variation of net income. If this is not the case, Eckel` model (1981), documented that the company is artificially smoothing its net income.

The model is represented by the following expression:

*CV Δ% **Net income * ≤ *CV Δ **% **Sales * ⇒ * **Smoothing** * *(1)*

Where: CVΔ% *Net income* = *Net income*t – *Net income*t-1 / *Net income*t-1; CVΔ% Sales = Revenuet – Revenuet-1 / Revenuet-1;

*CV*(x) = σ(x)/μ(x)

Where: *CV*(x) = coefficient of variation of a random variable; σ(x) = the standard deviation of random variable and μ(x) = the mean of random variable.

After that, the Eckel`s Index Model (EI) was calculate using the following expression:

EI=CVΔ%Net IncomeCVΔ%Sales

(2)

According to several authors such as Almeida et al. (2012), Valipour et al. (2011), Martinez and Castro (2009b) establishes a range between 0.9 and 1.1 as “gray area” where it is not possible to classify firms as smoothing firms and non-smoothing firms. This procedure reduces the sorting error, according to this method 228 observations was disqualified in this study. The following formula expresses the Eckel`s Model applied in the analysis:

0.9≤CVΔ%Net IncomeCVΔ%Sales≤1.1 (3)

⇒* Smoothing ≤Gray area ≤ Non-Smoothing *

In equation 3 CVΔ% *Net Income* = *Net Income*t – *Net Income*t-1 / *Net Income*t-1. CVΔ% Sales = Revenuet – Revenuet-1 / Revenuet-1.

## 3.3 Conditional conservatism measure

In this sub-section, the Basu (1997) primary model was applied to assess the existence of conditional conservatism, consistent with previous studies such as Almeida et al., 2012; Younes et al., 2012; LaFond and Roychowdhury, 2008; Ball et al., 2000; Khan and Watts, 2009 and Lara et al., 2005, and represented by the following expression ** (Model 1)**:

Nini,t=β0+β1Dumi,t+ β2Reti,t+β3Dum*Reti,t+ Ɛi,t (4)

In equation 4, Nin_{it} = The scaled net income for year t; D_{it} = dummy variable will be 1 if the economic return is negative and zero in the otherwise; Ret_{it} = annual stock return of the company i in year t; β_{1} = reflects the accounting profit opportunity. β_{2} = reflects the good economic news. β_{3} reflects the incremental timeliness of bad economic news over the good economic news (conservatism); β2 + β_{3} reflect the total bad economic news timeliness; Ε_{it} = White`s robust error regression coefficient.

According to Basu (1997), will be conservatism if the negative return (bad news) was recognized more quickly than the positive return (good news). Therefore, it is expected that the β3 coefficient will be positive, significant and higher for non-smoothing firms subsample, that’s means, the opportunity to recognize the future economic losses in profits disclosed is expected to be greater, relative to smoothing firms subsample.

Then, we applied Khan and Watts (2009) metric, to compare the degree of conditional conservatism between both subsamples. Represented by the following expressions:

G_Score ≡ β2 = μ0 + μ1 Sizei,t + μ2 MTBi,t + μ3Levei,t (5) C_Score ≡ β3 = λ0 + λ1 Sizei,t + λ2 MTBi,t + λ3 Levei,t (6)

Where *C_S**c**ore*reflects the incremental timeliness of bad news; *G**_S**c**ore*reflects the timeliness of good news; *Size*is the natural logarithm of total assets; *M**T**B*is the market-to-book ratio; *Leve *is leverage, measured as total debt deflated by total assets. We predict that Mozambican firms are conservative on overall. Following Khan and Watts (2009) it`s expected that the mean of C_Score will be higher than the mean of G_Score. Therefore, we obtain **annual****cross-sectional****regression****model,**equation (7) below, by replacing equations (5) and (6) in the regression model equation (4) including additional (*G**_**Sc**o**r**e*+ *C**_S**c**o**r**e*) terms in the last parenthesis to separately control the characteristics of the company[viii]:

Nini,t = β0 + β1Duni,t + Reti,t (μ0 + μ1Sizei,t + μ3MTBi,t + μ3Levei,t) + Dumi,t*Reti,t (λ0 + λ1Sizei,t + λ2MTBi,t + λ4Levei,t) + (δ0Sizei,t + δ1MTBi,t + δ2Levei,t + δ3Dumi,t*Sizei,t + δ4Dumi,t*MTBi,t + δ5Dumi*Levei,t) + εi,t (7)

Following previous literature (khan and Watts, 2009 and Khalifa et al., 2014), to make our analysis more robust, we have included four control variables in our model to respond the sub-hypothesis the H2.1: The effect of contracting. Sub-hypothesis H2.2: The effect of risk litigation. Sub-hypothesis H2.3: The effect of taxation. Finally sub-hypothesis H2.4: The effect of firm`s age on conditional conservatism. We estimate the impact of each of these four factors as follows:

Nini,t=β0+β1Dumi,t+ β2Reti,t+β3Dumi,t*Reti,t+ β4Contri,t+ β5Contri,t*Dumi,t+ β6Contri,t*Reti,t+ β7Contri,t*Dumi,t*Reti,t+Ɛi,t

(8)

In equation 8: Nin_{i,t}, Dum_{i,t} and Ret_{i,t} are defined in sub-section 3.3 above,* Contr** _{i,t}*represents each of these four factors of conditional conservatism: leverage, auditing litigation risk, taxation and firm age. The significant negative (positive) values of β

_{7}coefficient mean that lesser values of the identified factor lead to lower (higher) levels of earnings timeliness. Models 2, 3, 4, 5 and 6 are applied to test H2.1, H2.2, H2.3 and H2.4. We run the model separately for each subsample.

We test the effect of Leverage (Leve) on conservatism running the following regression (Model 2):

Nini,t=β0+β1Dumi,t+ β2Reti,t+β3Dumi,t*Reti,t+ β4Levei,t+ β5Levei,t*Dumi,t+ β6Levei,t*Reti,t+ β7Levei,t*Dumi,t*Reti,t+Ɛi,t

*(Model 2)*

In other to investigate the impact of litigation risk on conservatism, we run the model 3 shown above.

Nini,t=β0+β1Dumi,t+ β2Reti,t+β3Dumi,t*Reti,t+ β4BigAuditi,t+ β5BigAuditi,t*Dumi,t+ β6BigAuditi,t*Reti,t+ β7BigAuditi,t*Dumi,t*Reti,t+Ɛi,t (Model 3)

To study the effect of taxation on earnings timeliness, we run the model 4 shown above.

Nini,t=β0+β1Dumi,t+ β2Reti,t+β3Dumi,t*Reti,t+ β4TaxTi,t+ β5TaxTi,t*Dumi,t+ β6TaxTi,t*Reti,t+ β7TaxTi,t*Dumi,t*Reti,t+Ɛi,t

*(Model 4)*

Finally, to test the impact of the age of the company (*FirmAge*) on earnings timeliness, we run the model 5 shown above.

Nini,t=β0+β1Dumi,t+ β2Reti,t+β3Dumi,t*Reti,t+ β4FirmAgei,t+ β5FirmAgei,t*Dumi,t+ β6FirmAgei,t*Reti,t+ β7FirmAgei,t*Dumi,t*Reti,t+Ɛi,t

*(Model 5)*

Where Nin_{i,t}, Dum_{i,t} and Ret_{i,t} are defined in sub-section 3.3 above,* Leve** _{i,t}*represents a dummy variable equal to 1 if leverage is higher than the median sample, and zero otherwise. It is expected that the high values of leverage will induce the demand for conditional conservatism. BigAudit is a dummy variable that equals 1 if the firm is audited by Big4 and zero otherwise. We expect a positive association between the demand of conditional conservatism and the degree of auditor litigation risk.

*TaxT*

*represents a dummy variable equal to 1 if Taxable earnings are higher than the median sample, and zero otherwise. The tax cost drives incentives for earnings timeliness to decrease tax liabilities to the extent that taxable income and book income are associated. We predict a positive association between the demand of conditional conservatism and the taxable earnings. Finally, FirmAge represents the age of the ﬁrm i at the end of year t, measured as the number of years firm have been listed in KPMG – Mozambique database. We predict a positive (negative) association between age of the firms and conditional conservatism.*

_{i,t}Figure 1 illustrates the results of the classification process performed to sort smoothing companies and non-smoothing companies using Eckel’s model (1981). The results show that the number of non-smoothing companies was higher in relation to smoothing companies (4408 vs 1846), and finally, (4%) 278 observations correspond to the gray area.

**[Insert Figure about here]**

## 4.1 Descriptive statistics

Table 1 summarizes the descriptive statistics of the main variables of interest for the object of the study. The descriptors are calculated for each of the two subsamples (6,254 firm-year observations) of smoothing and non-smoothing firms, respectively. Underlying the previous literature on income smoothing (e.g. Almeida et a., 2012 Valipour et al. (2011), Martinez and Castro (2009b), we observed that the mean and standard deviation of the Eckel index (EI) presented highest values (mean = 4.07 and standard error = 214.927), as expected. We can highlight some testing variables. The mean of *Earn*is higher on non-smoothing than smoothing firms (0.022 VS -0.012). But the standard error of *Earn*is also higher in smoothing relative to the non-smoothing firms (0.074 VS 0.191).On overall, the standard error of returns is greater than the standard error of earnings, consistent with the argument that net income is a function of past and present returns, (Ball et al. 2000). The mean *and the *standard error of the variable *Dum**** Ret**is -0.259 VS 0.581 and 0.394 VS 0.458. This implies, that more than 25% of smoothing firms have reported negative returns and close to 60% of non-smoothing firms subsample have reported positive returns during the scoop of our study.

__I____n____t____e____r____ms of____c____ont____r____ol ____v____a____r____i____a____bl____e__** s** on firm characteristics, the mean of leverage (Leve) is higher on non-smoothing relative to the smoothing firms (0.192 VS 0.107), suggesting that in overall, the non-smoothing firms are higher leveraged relative to smoothing firms. The mean of

*Size*is 3.201 VS 2.947 for smoothing and non-smoothing firms subsamples, with

*the*standard error is 1.091 VS 0.007 respectively. For market-to-book (MTB), the mean and

*the*standard error for smoothing (non-smoothing) is 1.251 and 2.094 (0.077 and 0.105).

**[Insert Table 1 about here]**

## 4.2 Correlation between variables

A Pearson (top triangle) and Spearman (bottom triangle) correlation matrix of the main variables of both samples during the scoop period of study are showed in Table 2. We can observe the significantly positive Pearson correlation between the dependent variable Earn and the independent variable Return (Pearson correlation = 0.504, p <0.01). This evidence supports the argument of Almeida et al. (2012), that there is the greater expectation with the information contained in the profits before the disclosure of the financial statements. Also, can be observed the significant positive correlation between the dependent variable Earn and following independent variables Leve, Size, BigAudt and firmAge for both Pearson and Spearman correlation, confirming the expectation based on the theory. On the other hand, we can observe negative and significant Pearson correlation and Spearman correlation between Earn, MTB, and FirmAge. As expected.

**[Insert Table 2 about here]**

## 4.3 Analysis of the results of the regressions

In analyzing the regression of Panel A, Table 3, it was observed the degree of conditional conservatism in the full sample applying the Basu (1997), primary model. The coefficients β_{0}, β_{1}, β_{2}, and β_{3} were statistically significant and on the other hand, the coefficients β_{0} (β_{0 }= 0.035, P-value <0.01), showing the incorporation of unconditional conservatism into the model. Also, the coefficient β_{2 }(β_{2 }= 0.021, P-value <0.01) showing the effect of good news in this period, is very small, this is consistent with the previous study. On the other hand, the β_{3} coefficient presents a positive sign (β_{3} = 0.109, P-value <0.01), confirming the expectation based on the theory. This means that the Mozambican firms are conservative on overall, that is, they recognize timely the bad news embodied in the profits and the market identifies this information on the return of shares. The statistics t (26.01), with significant (P-value <0.01), indicates that the variable β_{3} explains the profit behavior in the model. It is worth noting that the adjusted coefficient of determination (R² = 11.6%) represents the explanatory power of profit generation by the Basu (1997) primary model. According to Ball et al. (2000), the adjusted R² ranges from 4.2% to 12.6%, which can be observed in this research and is consistent with previous studies by Lara et al. (2005),__Lopes et al. (2007)__.

Panel B of Table 3. shows the results applying Khan and Watts (2009) metric, the values of C_Score the incremental timeliness of bad news is higher than the average of G_Score, which reflects the timeliness of good news (mean = 0.198 vs 0.067 and median = 0.098 vs 0.055). As expected, this suggesting that our sample still conservative on overall, which still consistent with our first findings applying the Basu (1997) primary model. And Panel C Table 4 display the respective correlations matrix (Pearson top triangle; Spearman bottom triangle). The correlations are similar to those reported in the prior literature (e.g. Khan and Watts, 2009).

These results jointly lend support our H1 thatthere is a positive relationship between income smoothing and accounting conditional conservatism.

**[Insert Table 3 about here]**

Thirdly, we re-run the Basu (1997) primary model to confirm our hypothesis. As shown in Table 4 Panel A, the coefficient β3 (tow-way interaction term between variables Dum*Ret), that show the reflection of the asymmetric timeliness in the recognition of bad versus good news, is higher for the non-smoothing firms subsample comparative to the smoothing firms subsample (mean = 0.248 vs 0.195), as predicted. But, the bad news difference is insignificant (diff = 0.053). Additionally, the degree of unconditional conservatism is significantly lower in non-smoothing firms subsample compared to smoothing firms subsample (β_{0} = 0.017 vs β_{0} = 0.052), and this difference is significant (P-value < 0.01), as expected. Fourthly, we re-run the C_Score model. In Panel B Table 4, we can observe that the mean of C_Score is higher in non-smoothing firms subsamples relative to smoothing firms subsamples (mean = 0.351 vs 0.058; P-value < 0.01). This result suggests that conditional conservatism is more important in non-smoothing firms subsamples than smoothing firms subsamples, as expected.

**[Insert Tablet about 4 here]**

Table 5 presents the results applying the regression model. In the first column, we can observe the results applying the generalized model for both subsamples (smoothing and non-smoothing), the coefficient β_{2} which captures the effect of economic gains, is significantly positive (β_{2} = 0.042, P-value < 0.01), as expected. There is evidence that Smoothing firms subsample is interfering with the figures disclosed, distorting the economic reality of the business, thus making it impossible for the recognition of anticipated economic losses in profits. However, for non-smoothing firms subsample the coefficient β_{2} is significantly negative (β2 = -0.076, P-value < 0.01). This means that good news is negatively associated with accounting earnings for non-smoothing firms subsample, consistent with Khan and Watts (2009) study. When we analyze the coefficient β_{3} (Dum*Ret), which captures the recognition of economic losses, it is observed that for both subsamples (smoothing and non-smoothing firms) is significantly positive (β3 = 0.064, P-value < 0.01 and β3 = 0.565 P-value < 0.01). This shows that our samples are conservative on average. In addition, this result indicates that the association between earnings and bad news is stronger than between earnings and good news. In another hand, the coefficient β_{3} for smoothing firms subsample was less significantly positive than the non-smoothing firms subsample. This means that the degree of conditional conservatism in non-smoothing firms subsample is more important relative to the smoothing firms subsample. This suggests that non-smoothing firms subsample is more conditionally conservative. These results jointly lend support our H2.

The second column, Table 5, are displayed the estimation results for Model 2. This model investigates the effects of leverage on conservatism for both subsamples (smoothing vs non-smoothing firms). When we analyze the coefficient β3 (the two-way interaction between Dum*Ret), which captures the recognition of economic losses, it is observed that for both subsamples are still significantly positive (β3 = 0.033, P-value < 0.01 vs β3 = 0.069, P-value < 0.01). This means that less levered firms are conditionally conservative. Additionally, when we analyze the incremental effect of leverage on conditional conservatism, the coefficient β7 (three-way interaction term between variables Leve*Dum*Ret), in our model 2. The results display that the coefficient β7 are positive and significant for both subsamples (smoothing vs non-smoothing), as expected. However, the non-smoothing firms subsample are more significant relative to the smoothing firms subsample (β7 = 0.056, P-value < 0.01 vs β7 = 0.038, P-value < 0.01). Consistent also with prior studies (LaFond and Watts, 2008 Khan and Watts, 2009) that argue that more levered firms are more conservative relative to less levered firms this means that the degree of leverage is a significant factor of earnings timeliness. These results jointly lend support our hypothesis *H2.1 **a*nd indicate that the *level of leverage* of company has a great impact in conditional conservatism for non-smoothing firms subsample.

Further, in respect of control variables in Model 3, our hypothesis *H2.2 *predict that the higher degree of conditional conservatism in non-smoothing firms is driven by their level of auditor litigation risk, in Table 5 third column, are displayed the result of coefficient estimates and the respective P-values. We can observe that the coefficient β7 (the three-way interaction term, BigAudt*Dum*Rete) the incremental effect of auditor litigation risk is marginally significant for the non-smoothing firms subsample (β7 = 0.254, P-value < 0.10), which offers evidence that a larger (greater) level of auditor litigation risk appears to induce additional conditional conservatism in non-smoothing firms subsample. According to Khalifa et al. (2014), this happens when companies have more tangible assets; therefore, the risk of overvaluation of net assets is greater. Following St-Pierre and Anderson (1984), the risk of auditor litigation will be higher since companies are more expect to be sued when they overstate their assets than understate them. In this way, the auditors are motivated to be more conservative. Nevertheless, when we observe the coefficient β7 is not statically significant for the smoothing firm`s subsample (β7 = 0.133, P-value > 0.10), suggesting that the level of auditor litigation risk do not have an impact on conditional conservatism in smoothing firms subsample. Thus, these results jointly provide evidence support our hypothesis H2.2.

Table 5 fourth column, are displayed the result of control variables in Model 4, and in order to test our Hypothesis H2.3, we can observe that the coefficient β7 (the three-way interaction term, TaxT*Dum*Rete) is not significant in both subsample smoothing and non-smoothing firms (β7= -0.001, P-value > 0.10 vs β7= 0.019, P-value > 0.10). This means that conditional conservatism is less likely to be desirable in tax arrangement. Supporting the Qiang (2007) argue that taxation induces singly unconditional conservatism because the conditional conservatism is irrepressible, instability, and unnatural but undesirable in tax arrangement. Thus, we have not evidence to confirm our hypothesis H2.3.

In respect of control variables in Model 5, and in order to test our Hypothesis H2.4 that predicts that the demand of non-smoothing firms for conditional conservatism is not related to their firm`s age, we analyzed our mode 5. In Table 5 fifth column, are displayed the result of coefficient estimates and the respective P-values. We can observe that the coefficient β7 (the three-way interaction between FirmAge*Dum*Rete) is not significant in both subsamples (smoothing and non-smoothing firms). For smoothing firms subsample β7= -0.008, P-value > 0.10 and for non-smoothing firms subsample β7= 0.025, P-value > 0.10 respectively. This means that conditional conservatism is less likely to be desirable in the age of the firm. Thus, these results jointly lend support our hypothesis H2.4 and indicate that the age of company does not interfere in conditional conservatism in our sample.

Finally, in Table 5, the last column, we re-run our model including the prior factors in the full model, for both subsamples (smoothing and non-smoothing) separately. We can observe that the coefficient β_{2} which captures the effect of economic gains, is significantly positive only for smoothing firms subsamples (β_{2} = 0.34, P-value < 0.01 vs -0.207, P-value < 0.01), and the β3 coefficient, which capture the conditional conservatism is significantly positive and higher for non-smoothing firms than the smoothing firms (β3. = 0.293 vs 0.070), as expected. This result suggests that the non-smoothing firms are more conservatism than smoothing firms. In addition, leverage generate a demand for conditional conservatism in smoothing firms subsample (β7= 0.051, P-value < 0.01), however auditor litigation risk (β7= 0.017, P-value > 0.10), taxation (β7= 0.005, P-value > 0.10) and firm age (β7= 0.009, P-value > 0.10) does not engine the earnings timeliness in smoothing firms subsample. However, for non-smoothing firms, the results display that the leverage and auditor litigation risk generate a demand for earnings timeliness (β7= 0.060, P-value < 0.01), and (β7= 0.299, P-value < 0.10), however taxation (β7= 0.028, P-value < 0.10) and firm age (β7= 0.030, P-value < 0.10) have marginal positive effect on earnings timeliness. In summary, the β3 coefficients in all six model, which capture the conditional conservatism is significantly positive and higher for non-smoothing firms than the smoothing firms (coeff. = 0.565 vs 0.064 for model 1; coeff. = 0.069 vs 0.033 for model 3; coeff. = 0.158 vs 0.109 for model 3; coeff. = 0.081 vs 0.031 for model 4; coeff. = 0.093 vs 0.065 for model 4 and coeff. = 0.293 vs 0.070 for model 6). These results suggest that our sample is conservatism on overall. These results jointly lend support our hypothesis H1, hypothesis H2, hypothesis H2.1, hypothesis H2.2 and hypothesis H2.4 but do not support our hypothesis H2.3.

**[Insert Table 5 about here]**

## 4.3 Robustness test

In this sub-section, we examine the distribution of ROA[ix] in C_Score decile presented in table 7 and the results of the Variance Inflation Factor – VIF and Durbin-Watson tests applied to our model for the full sample are presented in Table 9, before segregation of the sample between income smoothing and non-smoothing firms.

Table 6 displays the results applying Basu (1997) primary model to measure the C-Score decile. The coefficient Β_{3}, (Dum*Ret), which captures or recognizes economic losses, shows a tendency of a higher level of conservatism to higher deciles. However, Β_{3}, for the eighth C-score decile is greater than the nineteenth C-score decile, which is in divergence with a measure of monotonicity. The rank correlation is significant and positive (β3 = -0.627, P-value < 0.05) as expected. However, the coefficient β_{2} which captures the effect of economic gains, is negative as predicted although the same is insignificant (β2 = -0.269, P-value > 0.01). Thus, these results jointly provide the efficacy of C-score in capturing the conditional conservatism.

**[Insert Table 6 about here]**

With respect to our first robustness test, we can observe the results in Table 7; the standard deviation of ROA is decreasing up to C-score decile rank 6, which is in agreement with Khan and Watts (2009) study. However, it starts decreasing after decile rank 6, which is in abeyance with predicted relationship between C-score decile rank and Standard deviation. The bottom line shows the classification correlation between the C_Score decile and the moments of the ROA distribution. The rank correlation is significant, negative and more monotonically decreasing in non-smoothing firms than smoothing firms (-2 vs -0.009), which is in agreement with Khan and Watts (2009) study, who argues that most conservatism companies have the negative mean of ROA. According to Prencipe et al. (2001), this means the manager for non-smoothing firms do not have the incentive to practice income smoothing because they profitability is high enough to satisfy managers and investors. In additional, we can observe that the mean of ROA is negative, more variable and more negatively skewed for the non-smoothing firm relative the smoothing firms, show the news driven conditional conservatism is still valid for our sample. The results in table 7 are consistent with the results of the main tests in Table 5.

**[Insert Table 7 about here]**

Regarding verify the second robustness tests, were carried out to verify the existence of multicollinearity and autocorrelation tests in the residual terms related to the variables in the model. The result reported in Table 9, shows the maximum value of VIF was 2.276 so there is no evidence of multicollinearity problems in the variables of the model applied in the research. Additionally, the Durbin-Watson test was performed; the null hypothesis being tested is that the regression residuals are autocorrelated. The values of the Durbin-Watson test show a maximum level of 2.321, as observed in Table 5. We can, therefore, reject the residual autocorrelation hypothesis, confirming the robustness of the models applied in the research.

**[insert Table 9 about here]**

#dissertation #thesis #essay #assignment #phd #phdstudent #phdlife #essaywriting #research #phdjourney #dissertationlife #gradschool #assignments #phdproblems #academicwriting #assignmenthelp #thesiswriting #gradstudent #dissertationproblems #university #academiclife #gradschoolproblems #academia #researchpaper #gradschoollife #graduateschool #college #doctoralstudent #dissertationdone #writing