3. Baseline model and empirical analysis

#### 3.1. Sample

status because they perceive an opportunity to maximize their own utility [10]. Consequently, agency theory holds that managers may take advantage of the information they have and their latitude in making accounting and reporting decisions to overstate financial information. They generally do this by acting in what they perceive to be in their own interest [11]. Reducing agency costs by imposing internal mechanisms of control should therefore encourage managers to behave in the best interest of shareholders instead of in their own interests. However, because controls are imperfect, we would expect some degree of opportunism to remain [10]. Since managers are widely paid based on firm's performance, it is plausible to expect that active earnings manipulation will occur in order to enhance managerial compensation packages [12]. This approach is highly focussed on bounded rational decision making around incentives, information and self-interest. Thus, it is a viewpoint that suggests that it may be necessary to limit managers' discretion with respect to accounting, since investors, as a consequence of asymmetrically distributed market information, cannot unravel the valuation effect

We suggest, however, that in addition to agency theory, a more cognitive viewpoint can also be used, to guide and further understand managerial behaviour. Social cognitive theory advocates that behaviour, cognitive and other personal factors and the external environment are the three main factors that drive the decision-making process [13]. These three factors are known to be asymmetrical, similar to the asymmetry of information in agency theory, in that they do not influence each other simultaneously, instantly or with equal strength, but they do influence each other multidirectionally. As a result, both investors and managers can be understood to be making decisions based upon a combination of factors that include a triad of perceptual, environmental and behavioural elements, all converging to ultimately produce an investment decision. Regarding cognition, two of the most relevant elements related to decision making are managerial biases and heuristics. The most common biases that managers revert to using include representativeness, availability and anchoring-and-adjustment [14] although there are now many other biases and heuristics that have been studied at length in a financial context [15]. The use of heuristics is considered a necessary way for humans to cope with our more limited capacity to process information [16]. More specifically to our study, many researchers have identified the biases and heuristics used in making financial decisions as highly relevant to understand the human and cognitive elements of the processes involved [6, 17]. Thus, according to the social cognitive approach, the market may wrongly perceive the actual firm performance disclosed in financial reports, as a consequence of biases and heuristics held perceptually and socially, in addition to behavioural and environmental elements. Thus, when managers overstate a firm's earnings, due to their bounded rationality and to information

of reported earnings in a timely manner under current reporting standards.

180 Corporate Governance and Strategic Decision Making

asymmetries, they can be easily misled to overprice the firm's shares.

the institutional context of Latin American countries, as is our case.

As described in Figure 1, by suggesting a complementary relation between agency and social cognition theories, we have produced a model that further explains how this process occurs and shows how the process is reinforced by the lack of sound corporate governance systems in

Financial markets in the region are still in a stage of early development, which allows managers to make use of accounting discretion to manipulate financial information. In immature financial The sample we use corresponds to 896 representative large non-financial firms from Argentina, Brazil, Chile, Colombia, Mexico and Peru. Data at firm level are collected from the Thomson Reuters dataset and data at the country level are collected from Worldwide Governance Indicators obtained from the updated work of Kaufmann et al. [21].<sup>1</sup> The sample corresponds to unbalanced panel data with a total of 9647 firm-year observations over the period from 1997 to 2013.

We use the Generalized Method of Moments (GMM) model to deal with the characteristic econometric problems of unobservable heterogeneity of individual firms and endogeneity [2, 22]. Several statistical contrasts are used as diagnostic tests for our panel data structures (e.g. the Hansen test for the validity of instruments, the second-order serial correlation test AR(2), the Wald test of joint significance of parameters, and the variance inflation factor (VIF) as a formal multicollinearity test). Additionally, non-linear restriction tests are used for those interacted (multiplicative) variables.

#### 3.2. Variable construction

Key to this study is the definition and analysis of our proxy independent variable of earnings manipulation, which corresponds to our measure of managerial discretion and quality of financial reporting. Two alternative estimations of earnings management are used based on absolute discretionary accruals. Since total accruals are known, the discretionary accruals must be estimated. Based on Dechow et al. [23], the total accrual (ACCit) denotes the component of earnings for each i firm during the t period computed as:

$$A\mathbb{C}\mathbb{C}\_{it} = (\Delta\mathbb{C}A\_{it} - \Delta\mathbb{C}ash\_{it}) - (\Delta\mathbb{C}L\_{it} - \Delta\text{STD}\_{it}) - Dep\_{it} \tag{1}$$

where CA denotes current assets, Cash is the cash and cash equivalent, CL are current liabilities, STD stands for short-term debt and the current proportion of long-term debt, and Dep is the annual depreciation expense.

Thus, once the total accruals are calculated, we have to split them into their non-discretionary and discretionary components. Non-discretionary accruals are aimed to improve the informational content of financial statements. According to the Jones [24] model, total accruals are affected by the firm's usual business (which can affect non-cash current assets and liabilities) and by fixed assets (which can affect the depreciation expense). Consequently, ACC are regressed depending on the change in sales (ΔSalesit) and the gross level of property, plant and equipment (PPEit) in the following equation:

$$\frac{A\text{CC}\_{it}^{\text{Mod}1}}{A\_{it-1}} = \beta\_0 + \beta\_1 \frac{\Delta Sales\_{it}}{A\_{it-1}} + \beta\_2 \frac{PPE\_{it}}{A\_{it-1}} + \varepsilon\_{it} \tag{2}$$

Regarding the expected signs for β1and β<sup>2</sup> it can be said that this is not trivial, except for β2, where a negative sign is expected because depreciation has been included with a negative sign in the definition of total accruals (ACC). However, there is no clear prediction for the sign of β<sup>1</sup> because, on the one hand, a higher level of sales might imply higher accounts receivables but, on the other hand, increase in sales usually imply increase in short-term debt too, so the net effect on working capital may not be determined a priori.

<sup>1</sup> The latest update took place in September 2014. Information can be downloaded from www.govindicators.org.

Hence, the value of (ACC) in Eq. (2) is the level of total accruals, depending on the firm's activity and the composition of the firm's assets. Therefore, the error term in the regression, which is the difference between observed and estimated accruals as stated in Eq. (3) would become the part of total accruals that is due to the discretionary behaviour of managers. So the first measure of discretionary accruals (DACC1it) should take the form:

Governance Indicators obtained from the updated work of Kaufmann et al. [21].<sup>1</sup> The sample corresponds to unbalanced panel data with a total of 9647 firm-year observations over the

We use the Generalized Method of Moments (GMM) model to deal with the characteristic econometric problems of unobservable heterogeneity of individual firms and endogeneity [2, 22]. Several statistical contrasts are used as diagnostic tests for our panel data structures (e.g. the Hansen test for the validity of instruments, the second-order serial correlation test AR(2), the Wald test of joint significance of parameters, and the variance inflation factor (VIF) as a formal multicollinearity test). Additionally, non-linear restriction tests are used for those interacted

Key to this study is the definition and analysis of our proxy independent variable of earnings manipulation, which corresponds to our measure of managerial discretion and quality of financial reporting. Two alternative estimations of earnings management are used based on absolute discretionary accruals. Since total accruals are known, the discretionary accruals must be estimated. Based on Dechow et al. [23], the total accrual (ACCit) denotes the component of

where CA denotes current assets, Cash is the cash and cash equivalent, CL are current liabilities, STD stands for short-term debt and the current proportion of long-term debt, and Dep is

Thus, once the total accruals are calculated, we have to split them into their non-discretionary and discretionary components. Non-discretionary accruals are aimed to improve the informational content of financial statements. According to the Jones [24] model, total accruals are affected by the firm's usual business (which can affect non-cash current assets and liabilities) and by fixed assets (which can affect the depreciation expense). Consequently, ACC are regressed depending on the change in sales (ΔSalesit) and the gross level of property, plant

> ΔSalesit Ait�<sup>1</sup>

Regarding the expected signs for β1and β<sup>2</sup> it can be said that this is not trivial, except for β2, where a negative sign is expected because depreciation has been included with a negative sign in the definition of total accruals (ACC). However, there is no clear prediction for the sign of β<sup>1</sup> because, on the one hand, a higher level of sales might imply higher accounts receivables but, on the other hand, increase in sales usually imply increase in short-term debt too, so the net

The latest update took place in September 2014. Information can be downloaded from www.govindicators.org.

þ β<sup>2</sup>

PPEit Ait�<sup>1</sup>

þ εit ð2Þ

ACCit ¼ ðΔCAit � ΔCashitÞ�ðΔCLit � ΔSTDitÞ � Depit ð1Þ

period from 1997 to 2013.

182 Corporate Governance and Strategic Decision Making

(multiplicative) variables.

3.2. Variable construction

the annual depreciation expense.

1

earnings for each i firm during the t period computed as:

and equipment (PPEit) in the following equation:

ACCMod<sup>1</sup> it Ait�<sup>1</sup>

effect on working capital may not be determined a priori.

¼ β<sup>0</sup> þ β<sup>1</sup>

$$\left| \frac{\text{DACC1}\_{it}}{A\_{it-1}} \right| = \frac{\text{ACC}\_{it}}{A\_{it-1}} - \left( \hat{\beta}\_0 \frac{1}{A\_{it-1}} + \hat{\beta}\_1 \frac{\Delta \text{Sales}\_{it}}{A\_{it-1}} + \hat{\beta}\_2 \frac{PPE\_{it}}{A\_{it-1}} \right) \tag{3}$$

where β^ <sup>0</sup>, <sup>β</sup>^ <sup>1</sup>, and <sup>β</sup>^ <sup>2</sup> are the estimators for β0, β1, and β<sup>2</sup> coefficients, respectively. Since the discretionary behaviour in earnings management may be used either to increase or reduce earnings, we follow Gabrielsen et al. [25] and calculate the absolute value for DACC to measure the extent of this discretionary behaviour instead of its direction.

Similarly, and as stated earlier, our second proxy measure of discretionary accruals also follows a cross-sectional model based on the Jones [24] model as described by Dechow et al. [23] as:

$$\frac{A\text{CC}\_{it}}{A\_{it-1}} = \beta\_0 + \beta\_1 \frac{\Delta \text{Sales}\_{it} - \Delta AR\_{it}}{A\_{it-1}} + \beta\_2 \frac{PPE\_{it}}{A\_{it-1}} + \varepsilon\_{it} \tag{4}$$

The coefficient estimates from Eq. (4) are used to estimate the firm-specific non-discretional accruals as:

$$\text{NDACC}\_{it} = \hat{\beta}\_0 \frac{1}{A\_{it-1}} + \hat{\beta}\_1 \frac{\Delta Sales\_{it} - \Delta AR\_{it}}{A\_{it-1}} + \hat{\beta}\_2 \frac{PPE\_{it}}{A\_{it-1}} \tag{5}$$

where ΔARit is the change in accounts receivable from the preceding year. Following Cohen et al. [26], while computing the non-discretionary accruals, we adjust the reported revenues on the sample of firms for the change in accounts receivable to capture any potential accounting discretion arising from sale credits. Then, the second measure of discretionary accruals is the difference between total accruals and the fitted non-discretionary accruals (DACC2it), defined as:

$$\left|\frac{\Delta \text{ACC} \mathbf{C}\_{\text{it}}}{A\_{\text{it}-1}}\right| = \frac{\text{ACC}\_{\text{it}}}{A\_{\text{it}-1}} - \left(\hat{\beta}\_0 \frac{1}{A\_{\text{it}-1}} + \hat{\beta}\_1 \frac{\Delta \text{Sales}\_{\text{it}} - \Delta AR\_{\text{it}}}{A\_{\text{it}-1}} + \hat{\beta}\_2 \frac{PPE\_{\text{it}}}{A\_{\text{it}-1}}\right) \tag{6}$$

Similar to the first measure, we also compute discretionary accruals in their absolute values.

In our models, the firm market performance as a dependent variable is computed through a number of alternative measures. First, we use the market return (MP1it) calculated as the annual change in the stock price for the firm i in the period t. The second measure of performance is based on the enterprise value (MP2it) calculated as the market capitalization, plus debt, minority interests and preferred shares, minus cash and cash equivalent for the firm i in the period t. To avoid the bias produced by scale issues, the enterprise value is computed in logarithms, which is the usual transformation applied to positive values with high dispersion. Finally, in our third measure of market performance we used the Tobin's Q. Due to this variable typically being unobservable by outsiders, a common practice is to rely on proxy variables. For doing so, we used the construct performed by Perfect and Wiles [27] which considers the reposition cost of total assets. Accordingly, the firm performance is:

$$\text{MP3}\_{it} = \frac{\text{MkCp} \text{ptz}\_{it} + \text{TD}\_{it}}{K\_{it}} \tag{7}$$

where MkCptzit is the market capitalization computed as the product between the year-end close price per share and the number of shares outstanding per i firm; TDit is the total liabilities at the year t; and Kit is the replacement value of firms' assets which is estimated by Perfect and Wiles [27] as follows:

$$K\_{\dot{t}t} = RNP\_{\dot{t}t} + RINV\_{\dot{t}t} + \left(TA\_{\dot{t}t} - BNP\_{\dot{t}t} - BINV\_{\dot{t}t}\right) \tag{8}$$

where RNPit is the replacement cost of net property, plant and equipment (net fixed assets); RINVit is the replacement value of inventories, TAit is the total assets; BNPit is the book value of net property, plant and equipment; and BINVit is the book value of inventories.

$$RNP\_{it} = RNP\_{it-1} \left[ \frac{1 + \phi\_t}{1 + \delta\_{it}} \right] + I\_{it} \tag{9}$$

For t > t<sup>0</sup> where t<sup>0</sup> is the first year of observations for a given company in this study; whilst RNPit<sup>0</sup> ¼ BNPit<sup>0</sup> . Moreover, ϕ<sup>t</sup> is the growth of capital good prices in year t which is defined by the Gross Domestic Product (GDP) deflactor. In other words, <sup>φ</sup><sup>t</sup> <sup>¼</sup> NomGDPt RealGDPt 100, where NomGDPt is the nominal GDP and RealGDPt is the real GDP, both reported by the National Institute of Statistics of Chile. <sup>δ</sup>it is the real depreciation rate defined as <sup>δ</sup>it <sup>¼</sup> Depit BNPit , where Depit is the annual book depreciation. Iit is the new investment in property, plant and equipment or capital expenditure which is defined as Iit ¼ BNPit � BNPit�<sup>1</sup> þ Depit.

$$RINV\_{it} = BINV\_{it} \left[ \frac{2WPI\_t}{WPI\_t + WPI\_{t-1}} \right] \tag{10}$$

where WPIt is the wholesale price index by country reported by the World Bank. This estimation for the replacement value of inventories assumes that the inventory accounting method is the average cost. For this method, the value of inventories reported at time t is approximately equal to the average of the prices at t � 1 and t.

The other independent variables correspond to control variables entered into the model in order to avoid problems of misspecification. The first control variable corresponds to the leverage at book value (LEVit) measured as the total liabilities over total assets, the company size (SIZEit) calculated as the logarithmic transformation of total assets, the firm's profitability (ROAit) measured as the earnings before interest and taxes over total assets, and finally we include the company's default risk (RISKit) which is measured through the alternative Altman [28] Z-Score which was specifically derived for developing countries computed as:

$$\text{RISK}\_{it} = 6.56 \text{WC}\_{it} + 3.26 \text{RE}\_{it} + 6.72 \text{EBIT}\_{it} + 1.05 \text{BVE}\_{it} + 3.25 \tag{11}$$

where WCit is the working capital over total assets, REit is the retained earnings over total assets, EBITit is the earnings before interest and taxes and BVEit is the book value of equity over total liabilities.

For country-level variables we use the Worldwide Governance Index<sup>2</sup> (GOVINDEXt) computed by Kaufmann et al. [21] as a measure of transparency across countries. This index is a composite of six dimensions of governance including: (i) Voice and Accountability, which are the process by which governments are selected, monitored and replaced; (ii) Political Stability and Absence of Violence/Terrorism, which measure the perceptions of the likelihood that the government will be destabilized or overthrown by unconstitutional or violent means, including politically-motivated violence and terrorism; (iii) Government Effectiveness corresponds to the quality of public and civil services, and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government's commitment to such policies; (iv) Regulatory Quality, which measures the perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development; (v) Rule of Law, which reflects the confidence that the agents will abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence; and finally (vi) the Control of Corruption, which measures the perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as capture of the state by elites and private interests. All of these six individual indicators are between �2.5 and 2.5 with increasing values as the governance indicator improves. GOVINDEX therefore corresponds to the average value among these six governance indicators by country and year.

Therefore, our estimation model would take the following form:

$$MP\_{it} = \beta\_0 + \beta\_1 DACC\_{it-1} + \beta\_2 CV\_{it} + \eta\_i + \mu\_t + \varepsilon\_{it\prime} \tag{12}$$

where MPit is the market performance, DACCit�<sup>1</sup> is the one-period lagged discretionary accruals measure, CVit is a vector of control variables (e.g. LEVit, SIZEit, ROAit, and RISKit), η<sup>i</sup> is the individual fixed effect, μ<sup>t</sup> is the time effect and εit is the stochastic error term. GOVINDEXt variable is used to split the sample and estimate separate regressions. Additionally, country, industry and time dummy variables are included in the model.

## 4. Empirical results

variable typically being unobservable by outsiders, a common practice is to rely on proxy variables. For doing so, we used the construct performed by Perfect and Wiles [27] which

MP3it <sup>¼</sup> MkCptzit <sup>þ</sup> TDit

where MkCptzit is the market capitalization computed as the product between the year-end close price per share and the number of shares outstanding per i firm; TDit is the total liabilities at the year t; and Kit is the replacement value of firms' assets which is estimated by Perfect and

where RNPit is the replacement cost of net property, plant and equipment (net fixed assets); RINVit is the replacement value of inventories, TAit is the total assets; BNPit is the book value of

For t > t<sup>0</sup> where t<sup>0</sup> is the first year of observations for a given company in this study; whilst RNPit<sup>0</sup> ¼ BNPit<sup>0</sup> . Moreover, ϕ<sup>t</sup> is the growth of capital good prices in year t which is defined by

NomGDPt is the nominal GDP and RealGDPt is the real GDP, both reported by the National

is the annual book depreciation. Iit is the new investment in property, plant and equipment or

where WPIt is the wholesale price index by country reported by the World Bank. This estimation for the replacement value of inventories assumes that the inventory accounting method is the average cost. For this method, the value of inventories reported at time t is approximately

The other independent variables correspond to control variables entered into the model in order to avoid problems of misspecification. The first control variable corresponds to the leverage at book value (LEVit) measured as the total liabilities over total assets, the company size (SIZEit) calculated as the logarithmic transformation of total assets, the firm's profitability (ROAit) measured as the earnings before interest and taxes over total assets, and finally we include the company's default risk (RISKit) which is measured through the alternative Altman [28] Z-Score which was specifically derived for developing countries computed as:

1 þ φ<sup>t</sup> 1 þ δit 

2WPIt WPIt þ WPIt�<sup>1</sup> 

RISKit ¼ 6:56WCit þ 3:26REit þ 6:72EBITit þ 1:05BVEit þ 3:25 ð11Þ

net property, plant and equipment; and BINVit is the book value of inventories.

RNPit ¼ RNPit�<sup>1</sup>

the Gross Domestic Product (GDP) deflactor. In other words, <sup>φ</sup><sup>t</sup> <sup>¼</sup> NomGDPt

Institute of Statistics of Chile. <sup>δ</sup>it is the real depreciation rate defined as <sup>δ</sup>it <sup>¼</sup> Depit

capital expenditure which is defined as Iit ¼ BNPit � BNPit�<sup>1</sup> þ Depit.

equal to the average of the prices at t � 1 and t.

RINVit ¼ BINVit

Kit

Kit ¼ RNPit þ RINVit þ ðTAit � BNPit � BINVitÞ ð8Þ

þ Iit ð9Þ

RealGDPt 100, where

, where Depit

ð10Þ

BNPit

ð7Þ

considers the reposition cost of total assets. Accordingly, the firm performance is:

Wiles [27] as follows:

184 Corporate Governance and Strategic Decision Making

#### 4.1. Univariate analysis

For the empirical results we proceed in two parts. As a starting point, in order to make the empirical analysis significant, we have to test the null hypothesis that the mean values of the discretionary accruals measures are statistically significant from zero. Previous literature suggests that managers of companies with weak governance structures have greater discretion to

<sup>2</sup> The latest update took place in September 2015. Information can be downloaded from www.govindicators.org.

engage in opportunistic earnings management [29]. A similar situation is observed when the regulatory environment does not efficiently constrain management's flexibility to misrepresent financial results [30]. The p-values reported in Table 1 suggest that the mean values of our alternative measures of discretionary accruals are significantly different from zero. In accordance with the previous literate, therefore, this preliminary finding may be used as evidence that listed firms in our sample opportunistically manipulate their financial reports.

Additionally, we split the sample into two big groups depending on the average value of the GOVINDEX variable by country. Although not reported, the average values were negative for all the countries, except for Chile and Brazil. Consequently, we can state that for the group of countries comprised of Chile and Brazil the transparency and corporate governance rules are relatively more efficient than for Argentina, Colombia, Mexico and Peru. Thus, the sample slip corresponds to these two groups, namely 'Transparent Countries' for those countries with relatively better transparency and corporate governance, and 'Opaque Countries' corresponding to those with relatively worse transparency and corporate governance. As observed in Table 2, the mean difference test was applied to verify if the extent of discretionary accruals as a measure of financial overstatement is the same across the two groups. The null hypothesis is that there is no difference in discretionary accruals between the two groups and the alternative hypothesis is that discretionary accruals are greater in the group of countries with relatively weaker transparency and governance. As tabulated, DACC1 is statistically greater in the set of countries which are less transparent and where corporate governance is weaker (e.g. Argentina, Colombia, Mexico and Peru, all of which we call the 'Opaque Countries') than in the set of countries such as Chile and Brazil where institutional transparency and corporate governance are better, and which we term the 'Transparent Countries'. This is shown to provide evidence that more financial statement manipulation is present when institutions and governance are weaker.

The descriptive statistics in Table 3 show that for our three measures of market performance (e.g. MP1, MP2 and MP3) there are positive average values for the companies included in the sample. Additionally, a typical company finances its total assets with about 48.73% of debt. In our sample, companies achieve an average rate of return of 4.27% on total assets. Finally, the average indicator of transparency and quality of corporate governance (GOVINDEX) is only 0.2146 with a maximum coefficient of 1.2482, which is still far away from its theoretical maximum achievable designated to be 2.5 [21].

The matrix of correlation coefficients is exhibited in Table 4. As would be expected, there is a high correlation for some measures of performance, such as the 0.533 correlation coefficient between MP2 and MP3. Similarly, high correlations are observed between the measures of discretionary accruals. On the other hand, we do not observe relatively high levels of correlation between the explanatory variables, with the exception of the correlation between the firm size (SIZE) and its leverage (LEV) (e.g. significant correlation of 0.408) and between firms' default risk (RISK) and the level of debt (LEV) (e.g. significant correlation of 0.691).3 These slightly high correlations might eventually cause problems of multicollinearity in the

<sup>3</sup> Although the tabulated correlation is negative, its interpretation is in the opposite direction as a consequence of the construction of the RISK variable where the firm risk increases as the variable decreases.

Earnings Quality and Market Performance in LATAM Corporations: A Combined Agency and Cognitive Approach… http://dx.doi.org/10.5772/intechopen.68485 187


This table shows the contrast to test the null hypothesis H0 that mean values for discretionary accruals measures are zero. The alternative hypothesis Ha is that such values are positive.

Table 1. One-sample t test.

engage in opportunistic earnings management [29]. A similar situation is observed when the regulatory environment does not efficiently constrain management's flexibility to misrepresent financial results [30]. The p-values reported in Table 1 suggest that the mean values of our alternative measures of discretionary accruals are significantly different from zero. In accordance with the previous literate, therefore, this preliminary finding may be used as evidence

Additionally, we split the sample into two big groups depending on the average value of the GOVINDEX variable by country. Although not reported, the average values were negative for all the countries, except for Chile and Brazil. Consequently, we can state that for the group of countries comprised of Chile and Brazil the transparency and corporate governance rules are relatively more efficient than for Argentina, Colombia, Mexico and Peru. Thus, the sample slip corresponds to these two groups, namely 'Transparent Countries' for those countries with relatively better transparency and corporate governance, and 'Opaque Countries' corresponding to those with relatively worse transparency and corporate governance. As observed in Table 2, the mean difference test was applied to verify if the extent of discretionary accruals as a measure of financial overstatement is the same across the two groups. The null hypothesis is that there is no difference in discretionary accruals between the two groups and the alternative hypothesis is that discretionary accruals are greater in the group of countries with relatively weaker transparency and governance. As tabulated, DACC1 is statistically greater in the set of countries which are less transparent and where corporate governance is weaker (e.g. Argentina, Colombia, Mexico and Peru, all of which we call the 'Opaque Countries') than in the set of countries such as Chile and Brazil where institutional transparency and corporate governance are better, and which we term the 'Transparent Countries'. This is shown to provide evidence that more financial statement manipulation is present when institutions and governance are weaker.

The descriptive statistics in Table 3 show that for our three measures of market performance (e.g. MP1, MP2 and MP3) there are positive average values for the companies included in the sample. Additionally, a typical company finances its total assets with about 48.73% of debt. In our sample, companies achieve an average rate of return of 4.27% on total assets. Finally, the average indicator of transparency and quality of corporate governance (GOVINDEX) is only 0.2146 with a maximum coefficient of 1.2482, which is still far away from its theoretical

The matrix of correlation coefficients is exhibited in Table 4. As would be expected, there is a high correlation for some measures of performance, such as the 0.533 correlation coefficient between MP2 and MP3. Similarly, high correlations are observed between the measures of discretionary accruals. On the other hand, we do not observe relatively high levels of correlation between the explanatory variables, with the exception of the correlation between the firm size (SIZE) and its leverage (LEV) (e.g. significant correlation of 0.408) and between firms' default risk (RISK) and the level of debt (LEV) (e.g. significant correlation of 0.691).3 These slightly high correlations might eventually cause problems of multicollinearity in the

Although the tabulated correlation is negative, its interpretation is in the opposite direction as a consequence of the

construction of the RISK variable where the firm risk increases as the variable decreases.

maximum achievable designated to be 2.5 [21].

186 Corporate Governance and Strategic Decision Making

3

that listed firms in our sample opportunistically manipulate their financial reports.


This table tests the null hypothesis H0 that the difference in mean values for discretionary accruals measures are the same between 'Other Countries' and 'Chile + Brazil' groups. The alternative hypothesis Ha is that this difference is positive.

Table 2. Two groups test with equal variances.


This table shows the descriptive statistics (e.g. mean value, standard deviation, minimum and maximum) for the variables used in the empirical analysis.

Table 3. Descriptive statistics of variables.


The table reports the pairwise correlation coefficient matrix. The significance level of each correlation coefficient is in parenthesis.

Table 4. Pairwise correlation coefficients.

regression estimates. Nevertheless, as reported in the subsequent regression tables, the variance inflation factor (VIF) test allows us to accept the hypothesis of the inexistence of this econometric problem.

#### 4.2. Multivariate analysis

Concerning the multivariate analysis, we interpret the outcomes of the model (12) for the whole sample according to the regression estimates shown in Table 5. This table includes nine regressions for our three alternative measures of the dependent variable (e.g. MP1, MP2 and MP3) explained by our three one-period lagged independent variables as measures of the quality of the financial reports and earnings manipulation (e.g. DACC1, DACC2 and DACC3). As a starting point for the interpretation of the coefficient estimates in our regression analysis, a battery of diagnostic tests is used to ensure the validity of our results in Tables 5 and 6. Robust standard errors were used in all the regression estimates. According to the Wald test, all the independent variables are jointly significant at the standard confidence levels. As mentioned

Earnings Quality and Market Performance in LATAM Corporations: A Combined Agency and Cognitive Approach… http://dx.doi.org/10.5772/intechopen.68485 189


This table includes the estimations of model (12). Variables construction is described in Section 3.2. Industry, time and country effects are included in the estimations but not tabulated. The Wald test of statistical significance of independent variables is reported at the bottom of the table. Similarly, the second-order autocorrelation test is reported (AR(2)). The Hansen contrast is used to test the hypothesis that the instruments are properly chosen. The VIF test is used to formally examine the multicollinearity problem. Standard errors are in parentheses.

\* Statistical significance at the 10% level.

\*\*Statistical significance at the 5% level.

regression estimates. Nevertheless, as reported in the subsequent regression tables, the variance inflation factor (VIF) test allows us to accept the hypothesis of the inexistence of this

The table reports the pairwise correlation coefficient matrix. The significance level of each correlation coefficient is in

GOVINDEX �0.050 0.535 �0.088 �0.065 0.060 0.042 �0.176 �0.115 0.057 0.139

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Variables MP1 MP2 MP3 DACC1 DACC2 DACC3 LEV SIZE ROA RISK

Concerning the multivariate analysis, we interpret the outcomes of the model (12) for the whole sample according to the regression estimates shown in Table 5. This table includes nine regressions for our three alternative measures of the dependent variable (e.g. MP1, MP2 and MP3) explained by our three one-period lagged independent variables as measures of the quality of the financial reports and earnings manipulation (e.g. DACC1, DACC2 and DACC3). As a starting point for the interpretation of the coefficient estimates in our regression analysis, a battery of diagnostic tests is used to ensure the validity of our results in Tables 5 and 6. Robust standard errors were used in all the regression estimates. According to the Wald test, all the independent variables are jointly significant at the standard confidence levels. As mentioned

econometric problem.

parenthesis.

MP2 0.040 1.000 (0.000) MP3 0.030 0.533 1.000 (0.004) (0.000) DACC1 �0.028 0.025 �0.065 1.000 (0.007) (0.019) (0.000) DACC2 �0.033 0.015 �0.068 0.934 1.000 (0.001) (0.168) (0.000) (0.000) DACC3 �0.036 �0.036 �0.121 0.859 0.839 1.000 (0.001) (0.001) (0.000) (0.000) (0.000) LEV �0.014 0.016 0.358 �0.125 �0.080 �0.103 1.000 (0.165) (0.135) (0.000) (0.000) (0.000) (0.000) SIZE 0.031 0.528 0.965 �0.085 �0.074 �0.136 0.408 1.000 (0.003) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) ROA 0.184 0.115 0.012 �0.003 �0.005 �0.029 �0.274 0.004 1.000 (0.000) (0.000) (0.266) (0.744) (0.626) (0.005) (0.000) (0.709) RISK 0.044 �0.015 �0.238 �0.041 �0.020 �0.0362 �0.691 �0.284 0.334 1.000 (0.000) (0.147) (0.000) (0.000) (0.050) (0.001) (0.000) (0.000) (0.000)

188 Corporate Governance and Strategic Decision Making

4.2. Multivariate analysis

Table 4. Pairwise correlation coefficients.

\*\*\*Statistical significance at the 1% level.

Table 5. Multivariate analysis for the whole sample.

above, the variance inflation factor (VIF) reported at the bottom of Tables 5 and 6 confirm that collinearity does not skew our estimation results since the VIFs are greater than 2. Regarding the moment conditions, the Hansen over-identification tests did not reject the over-identifying restrictions, meaning that we accept the null hypothesis of validity of the instruments in our estimations. Additionally, the AR(2) test proves the lack of second-order serial correlation. Consequently, our results are not biased by a possible incorrect choice of instruments or by autocorrelation and are robust, according to the standard diagnostic tests for panel data.



Variables DACC1t�1 DACC1t�1

\*SYS

MP1

7.059\*\* (0.9151) �4.9041\* (�0.7209)

DACC1t�1+DACC1t�1

DACC2t�1 DACC2t�1

\*SYST

DACC2t�1+DACC2t�1

DACC3t�1 DACC3t�1

\*SYST

DACC2t�1+DACC2t�1

LEV

SIZE

ROA RISK Constant

�1.6711

(�0.2296)

�0.7768

(�0.8334)

9.0907

(0.8169)

0.2831\* (0.5280)

2.5543

(0.2844)

 (

�1.1314)

 (

�0.9620)

 (13.4278)

 (15.6612)

 (10.2658)

 (

�3.9125)

 (

�1.8302)

 (

�2.2376)

�9.2934

�16.1757

 4.0527\*\*\*

 (2.4286)

 (1.0804)

 (

�1.4665)

 (6.2080) 4.5408\*\*\*

3.1268\*\*\*

�0.4891\*\*\*

�0.2535\*

�0.2807\*\*

 (4.0482)

 (2.1930)

 (0.3744)

 (1.5524)

0.8423\*\*

1.1136

�0.0159

 0.0709\*\*\*

 (2.1722)

 (2.0678)

 (

�0.6157)

 (

�2.1810)

 (6.4494) 0.0629\*\*\*

0.0143\*\*

0.0029

 0.0109

 (0.4093)

 (2.3561)

 (3.4411)

 17.2895\*\*

 (0.7225)

 (0.2922) 42.4076\*\*

�0.2355

 0.6890\*\*

 (25.2228)

 (27.2567)

 (25.3953)

2.4617\*\*\*

0.0619

 0.3948\*\*

 (64.1562)

 (63.9162)

 (75.0594)

0.5256\*\*\*

 0.6852

 0.3946

 1.0962\*\*\*

 (

�1.6045)

 (

�0.2746)

 (11.3608)

 (14.3559)

1.1786\*\*\*

1.0024\*\*\*

0.9617\*\*\*

0.9405\*\*\*

0.9377\*\*\*

 (1.2470)

 (10.4653)

 (11.5689)

 (10.5313)

�10.1574

�3.8463

 4.3005\*\*\*

5.2190\*\*\*

0.4523

 1.6498\*\*\*

1.7214\*\*\*

1.6522\*\*\*

\*SYS

\*SYS

\*SYS

 2.1549\* 3.5026\*\*\*

(3.2505) �1.9704\*\*\*

(�3.1764)

1.5322\*\*\* 2.0647\*\*\*

(3.0704) �1.1733\*\*\*

(�2.7950)

0.8914\*\*\*

 MP1

 MP1

 MP2 1.6018\*\*\*

(5.7779) �0.4964\*\*\*

(�12.9280)

1.1054\*\*\* 0.7596\* (9.2927) �0.6997\*\* (�15.7794)

0.0599\*\* 0.4148\*\*\*

(9.9215) �0.1247\* (�11.9847)

0.2901\*

MP2

MP2

MP3

1.7355\*

(1.8141)

�0.8013\*\*\*

(�5.6391)

0.9342\*

190 Corporate Governance and Strategic Decision Making

5.3086\*\*\*

(5.2294)

�4.2790\*\*\*

(�6.9984)

1.0296\*\*\*

3.2447\*\*\*

(3.7770)

�2.7271\*

(�3.6554)

0.5176\*\*

 MP3

 MP3 and country effects are included in the estimations but not tabulated. The Wald test of statistical significance of the independent variable is reported at the bottom of the table. Similarly, the second-order autocorrelation test is reported (AR(2)). The Hansen contrast is used to test the hypothesis that the instruments are properly chosen. The VIF test is used to formally examine the multicollinearity problem. Standard errors are in parentheses. \*Statistical significance at the 10% level.

\*\*Statistical

 significance at the 5% level.

\*\*\*Statistical significance at the 1% level.

Table 6. Multivariate analysis by levels of governance.

#### 4.2.1. Discussion of results of the whole sample

Concerning the findings, the results systematically show that higher manipulation of financial reports (DACC1, DACC2 or DACC3) leads to greater firm market performance (MP1, MP2 and MP3). Although in regressions (1), (6) and (8) our measures of discretionary accruals are not statistically significant, the direction of the relationship is still positive (e.g. see Table 5). For instance, in the second regression, we observe that an increase by one percentage point in our first one-period lagged measure of discretionary accruals (DACC1t�1) triggers an increase of 6.0404 times the market change in the stock price. Such a large change in market prices caused by a small change in earnings management is evidence of an elastic market performance. According to Leuz et al. [20], earnings management can be defined as the alteration of firms' reported economic performance by insiders to either mislead outsiders or to influence contractual outcomes. Our results provide evidence of this construct suggesting that when managers overstate or misreport financial statements by actively manipulating earnings, there is a market premium as a consequence of a general lack of transparency in the LATAM context [21], and investor biases, despite some distinctive levels of transparence, are observed in the region. We observe that the stock price change (MP1), the logarithmic transformation of the enterprise value (MP2), as well as the performance measure proxied by Tobin's Q (MP3), all serve to increase the manipulation of financial reports.

Our results also support the Lee et al. [9] model, where firms with higher accounting performance over-report earnings by a larger amount when looking for greater price responsiveness or market performance. In the Lee et al. [9] model, managers manage earnings to influence the stock price. This is a plausible explanation for our results. We suggest that under a rational setting that is free of market frictions, where information is symmetrically distributed and where there is complete alignment of interest between managers and shareholders, there is no room for managers to opportunistically manage earnings to increase market performance. Under these conditions, the market would be able to discriminate and choose a separating equilibrium as suggested by Akerlof [18], by rationally discounting for the over-statement of earnings. This supports the idea that buyers are guided by earnings but are unaware that earnings are inflated by the generous use of accruals, and that this is a consequence of individual biases, wrong perceptions and misuse of heuristics in making their financial decisions. Thus, investors are misled to pay too high a price [31], which triggers a greater change in stock price, enterprise value and Tobin's Q. Hence, when there is a lack of transparency and the agency conflict is not efficiently minimized, managers take advantage of their discretionary power to artificially boost the market performance of the company. The major motivation behind this is basically the improvement of contractual conditions and reward with better compensation packages. Or as stated in terms of Teoh et al. [31], managers manage earnings to exploit market credulity. This idea is consistent with investors naively extrapolating earnings without fully adjusting for the potential manipulation of reported earnings [32]. Consequently, the previous findings allow us to accept our research hypothesis that there will be a positive relationship between opportunistic manipulation of earnings and firms' market performance.

Concerning the control variables included in the model specifications, we observe that leverage (LEV) does not impact on the stock price change (MP1), but it has a positive relationship with our two other alternative measures of market performance, namely the enterprise value (MP2) and the proxy for Tobin's Q (MP3). As suggested by the capital structure literature, debt can be efficiently used to undertake profitable investment projects that the market interprets as positive growth opportunities by pushing up the market valuation [33]. Similarly, the size of the company (SIZE) is also positively related to its performance. According to our findings, larger firms take advantage of economies of scale and this dimension is rewarded with a premium in market valuation. The return on assets (ROA) is also positively associated with the market performance. Consequently, there is a direct correspondence between the bottomline net income and the firms' stock performance. Regarding our last control variable, the default risk (RISK) is found to be negatively associated with market performance. As mentioned earlier, by construction, the RISK variable increases as the default risk decreases, and consequently, the results reported in Tables 5 and 6 must be interpreted in the opposite direction. Hence, our findings suggest that the market discounts prices when the firm is approaching bankruptcy as suggested by the literature [28, 34].

#### 4.2.2. Discussion of results by levels of governance

4.2.1. Discussion of results of the whole sample

192 Corporate Governance and Strategic Decision Making

serve to increase the manipulation of financial reports.

manipulation of earnings and firms' market performance.

Concerning the findings, the results systematically show that higher manipulation of financial reports (DACC1, DACC2 or DACC3) leads to greater firm market performance (MP1, MP2 and MP3). Although in regressions (1), (6) and (8) our measures of discretionary accruals are not statistically significant, the direction of the relationship is still positive (e.g. see Table 5). For instance, in the second regression, we observe that an increase by one percentage point in our first one-period lagged measure of discretionary accruals (DACC1t�1) triggers an increase of 6.0404 times the market change in the stock price. Such a large change in market prices caused by a small change in earnings management is evidence of an elastic market performance. According to Leuz et al. [20], earnings management can be defined as the alteration of firms' reported economic performance by insiders to either mislead outsiders or to influence contractual outcomes. Our results provide evidence of this construct suggesting that when managers overstate or misreport financial statements by actively manipulating earnings, there is a market premium as a consequence of a general lack of transparency in the LATAM context [21], and investor biases, despite some distinctive levels of transparence, are observed in the region. We observe that the stock price change (MP1), the logarithmic transformation of the enterprise value (MP2), as well as the performance measure proxied by Tobin's Q (MP3), all

Our results also support the Lee et al. [9] model, where firms with higher accounting performance over-report earnings by a larger amount when looking for greater price responsiveness or market performance. In the Lee et al. [9] model, managers manage earnings to influence the stock price. This is a plausible explanation for our results. We suggest that under a rational setting that is free of market frictions, where information is symmetrically distributed and where there is complete alignment of interest between managers and shareholders, there is no room for managers to opportunistically manage earnings to increase market performance. Under these conditions, the market would be able to discriminate and choose a separating equilibrium as suggested by Akerlof [18], by rationally discounting for the over-statement of earnings. This supports the idea that buyers are guided by earnings but are unaware that earnings are inflated by the generous use of accruals, and that this is a consequence of individual biases, wrong perceptions and misuse of heuristics in making their financial decisions. Thus, investors are misled to pay too high a price [31], which triggers a greater change in stock price, enterprise value and Tobin's Q. Hence, when there is a lack of transparency and the agency conflict is not efficiently minimized, managers take advantage of their discretionary power to artificially boost the market performance of the company. The major motivation behind this is basically the improvement of contractual conditions and reward with better compensation packages. Or as stated in terms of Teoh et al. [31], managers manage earnings to exploit market credulity. This idea is consistent with investors naively extrapolating earnings without fully adjusting for the potential manipulation of reported earnings [32]. Consequently, the previous findings allow us to accept our research hypothesis that there will be a positive relationship between opportunistic

Concerning the control variables included in the model specifications, we observe that leverage (LEV) does not impact on the stock price change (MP1), but it has a positive relationship In this section on multivariate analysis, we aim to study the impact of different levels of transparency and efficiency of country-level governance systems on the relationship between earnings management and the firms' market performance. As has been widely supported in previous literature, governance systems in the Latin American region are comparatively weaker than in other more developed economies such as the US or Europe [35–37]. Consequently, such opaqueness in the financial markets and the weaker protection of investor rights determines how actively managers over-state financial information disclosed to the markets [38]. Moreover, differences in governance systems have also been observed across Latin American countries [39].

To disentangle this issue we add to our estimation model (12) a variable that allows us to control for cross-country differences in governance systems and transparency. To do so, we create a dummy variable (SYS) based on the subsamples of 'Transparent Countries' and 'Opaque Countries' described in Section 4.2. This dummy variable takes the value 1 if the country belongs to the subsample of 'Transparent Countries' and zero for the subsample of 'Opaque Countries'. Afterwards, we interact the SYS variable with our alternative measures of discretionary accruals and create a multiplicative variable, for instance, DACCt�1\*SYS. This variable allows us to measure the specific impact of discretionary accruals on firm performance moderated by the two different levels of cross-country transparency and governance defined in our sample.

The results reported in Table 6 indicate, on the one hand, that the one-period lagged variable of discretionary accruals is always positively related to market performance. On the other hand, the interacted or multiplicative variables between discretionary accruals and transparency and governance efficiency (see for instance DACCt�1\*SYS in the first regression in Table 6) consistently show a negative relationship with firm performance. The interpretation of these results are as follows. Taking the first regression in Table 6, for the subsample of 'Transparent Countries', namely Brazil and Chile, the SYS variable takes the value 1 and consequently the impact of discretionary accruals on the firm's performance corresponds to the addition of DACCt�<sup>1</sup> and DACCt�1\*SYS (=DACC<sup>t</sup>�<sup>1</sup> + DACCt�1\*SYS) which is reported in the table in italic characters. In the first regression, this addition of variables takes a value equal to 2.1549. Consequently, for the 'Transparent Countries', a marginal increase in earnings management (DACCt�1) causes more than twice an impact on the change in stock price (MP1). However, since SYS takes a zero value for the group of 'Opaque Countries', the impact of discretionary accruals on market performance for this subset of countries corresponds only to the coefficient estimate of the DACCt�<sup>1</sup> variable, which in the first regression goes up to 7.059. Thus, before any marginal change in opportunistic managerial behaviour is measured through discretionary accruals, the impact on the change in price will be more than seven times the change in discretionary accruals. The significance of the linear combinations of coefficients is tested and it is accepted in all cases that the addition of the discretionary accruals variables and the interacted or multiplicative variables are statistically different from zero (e.g. see italic characters in Table 6).

In all the subsequent regression estimates of Table 6, we observe that the impact on any measure of market performance (MP1, MP2 or MP3) as a consequence of a change in the discretionary accruals is systematically greater in the group of 'Opaque Countries' than in the group of 'Transparent Countries'. This may be used as robust evidence that managers take more advantage of market myopia when institutional settings are endowed with weaker governance systems and where greater gaps of information exist between insiders and outsiders. In other words, although we subscribe to previous literature on the fact that governance systems are relatively weak in the Latin American region [40], we also recognize that there are still some intraregional differences in transparency and governance, as supported by our findings. Thus, in more transparent financial systems and where the right of shareholders is relatively better protected, the impact on market performance caused by opportunistic manipulation of financial reports is not as large as in contexts of less transparency and governance. We can derive out of this finding that the market is fooled in order to increase the firm's valuation by mispresenting the financial information. And even more, the weaker the governance systems across countries in the Latin American region, the greater the changes will be to boost firm value by misleading the market towards making wrong investment decisions.

Finally, findings concerning the control variables listed in Table 6 remain consistent with those previously interpreted based on Table 5. Thus, we can conclude that our overall findings are robust to a battery of alternative test specifications and controls, as well as to elaborate dependent and independent variables.
