ased on the postulate of the positive theory Zimmermann, 1986-1990), which considers that the directors of financially distressed companies, exploit the accounting information in their interests, we adopted an econometric approach to detect accounting manipulations by the income management method and by estimating, under ordinary least squares, the Modified Generalized Jones model, it was possible to confirm the existence of discretionary accounting manipulations in the accounting results published by the companies forming the sample studied throughout listing from 1999 -2014, on the Tunis Stock Exchange. The analysis of the significance and relevance of the model used allowed us to validate empirically this hypothesis concerning the management of the result. Other significant returns relating to the residual of the estimate were revealed by the regression that was conducted on the econometric model used. Indeed, the terms of the residual of the estimate, which summarize all the discretionary accruals or all the other exogenous variables not taken into account in the modeling, do indeed satisfy the stochastic and structural assumptions (relating to the bias and the convergence), in other words, these terms are governed by a normal centered, reduced law, and therefore, the sample studied perfectly induces the characteristics of the population it represents in terms of mean and variance. What led us in this article to push the analysis of the residual by distinguishing the companies that handle discretionary accruals upward from those that manage it downward, this approach is interesting, in the sense that it allowed us to divide the sample into two groups and to reveal the characteristics of the defensive strategy through exogenous explanatory variables of the adjustments of the accounting items of total regularization (total accruals). This process allowed us to verify the correlation between financial failure and the upward management of the result (defensive strategy).

Author: Ph.D in Management Sciences, Finance and Accounting Department -ESSEC, University of Tunis. e-mail: [email protected] For the specification of the characteristics of the discretionary adjustments by the econometric Model our approach must take into account the constraint relating to the unavailability of information (lack of database relating to companies in financial difficulty) at the level of the TMB or the CMF concerning companies that have not complied with the NPCGA (managing their results outside the limits of generally accepted accounting principles and standards). Indeed, the provision by the directors of discretionary accounting and financial information would make it possible to identify difficulties and thus to resort to an informal reorganization. This is why one of the privileged fields of research on the accounting/business failure interface is that of prediction models of financial failure vs. fraud through the accounting and financial variables characterizing the incentives to the management of the result. In this respect, the predictions of the positive theory (Dechow, Kothari, Watts, 1998) announce that executives of companies experiencing difficulties make artificially increase the published net result. To highlight this assumption of the positive theory of accounting at the Tunisian context level, we have borrowed an approach that focused on a selection of a sample of 18 companies for a listing at the BVMT over the period 1999 -2014. This choice is thus set to have two subsamples of the same overall size and size at the level of the selected activity sectors (industry, trade, and service delivery -Table 1). This led us to dismiss the E21 company from the global sample as it is the only one in its sector that of air transport.

The relative variation (see Table 6) of the exogenous variation in cash flow generated by the operation is on the average negative of -0.30% for the group that managed its earnings upwards (+4 % on average of discretionary accruals) versus a positive variation of 1.94% in the other group (-4% on average of discretionary accruals). This failure was found again at the level of the performance indicator variable, namely that of the change in turnover adjusted for any abnormal increase at the level of the receivables item, where there was a clear difference (7.73% for the sample F versus 8.76% in sample C). This finding is corroborated by research work such as that conducted by Kothari, Leone, and Wasley, (2001) . Finally, in this chapter we managed to make an important step in the exploratory study of the financial statement fraud's phenomenon by specifying the determining variables of the companies' financial failure that manages the result upwards and identifies at the same time the two samples (F and C). This approach can be used to perform a logistic regression of a fraud prediction model at the level of a new econometric Fraud Detection process research perspective at published financial statement levels.

The model (see Table 2) is relevant in its entirety, even though the observations concerning company E21 have been removed. Indeed, the adjusted R2 has improved (76.2568% instead of 75.9369%), and the coefficients of the regression are very significant at the 1% threshold, therefore, the exogenous variables are relevant.

The residual of the estimate corresponds to the discretionary accruals thus to all the other exogenous variables not taken into account in the modeling. As shown in Figure 3 -4, the residual of the estimate is highly dispersed between both positive and negative directions, so it is important to distinguish the two meanings of variation of the residual to identify the firms that manage discreetly their results upward (financially failing companies) of those who manage their results downwards (companies not financially failing). This approach is interesting insofar as it makes it possible to better understand the accounting and financial variables characteristic of the home of discretionary accruals. The generation of the two directions of variation is carried out by E Views software 9 as indicated in the table (3) with identification of the companies according to these meanings of variation of the average of the discretionary accruals (residuals of the estimate of the model (1).

accounting choices to improve the result by carrying out discretionary accounting manipulations that tend to

Following the econometric analysis carried out previously, which proved the effectiveness and relevance of the model tested (equation 1), to detect the management of the result, we were able to push its exploitation towards the distinction and the identification of the companies that manage the upward result (defensive strategy) of those who adopt a completely divergent strategy (offensive strategy). This procedure is justified in the work on the topic of accounting data management. Indeed, Kothari, Leone and Wasley (2001) were able to verify that firms manipulating their results take into account the past and current performance of their economic activities. In other words, profitable firms use discretionary increments (offensive strategy) differently than unprofitable businesses (defensive strategy). This finding, confirmed in the research work on results management (Dechow, 2010), was found in our empirical work. Indeed, the results (Tables 4, 5 and 6) of our study conducted on the sample of Tunisian companies listed in the BVMT, reveals that the companies that manage their results upwards (sample F) manipulate towards the increase in discretionary accruals up to (on average) + 4% of total assets delayed versus -4% for companies that manage their results downwards (sample C). Another revealing result of the characteristic of financial failure among the companies that manage the result upward, is that they display an average negative cash flow variation of about -0.30% of the total assets delayed against a positive variation of + 1.94% of total assets delayed among those who manage their results downward. This failure is also reflected in the variable change in turnover adjusted for any abnormal increase at the level of the receivables item (7.73% in the sample F versus 8.76% in the sample C). This procedure allowed us to identify two sub-samples "F" and "C" of the same size ("F": 9 companies that manage the result upward and "C": 9 other companies that manage the result towards the decline), which features the manipulations of discretionary accruals summarized by, The total accruals are obtained by the following estimation model ( 2):

And the discretionary accruals then correspond to the difference between the observed value of the total adjustments and the calculated value of so-called normal adjustments

We therefore measure earnings management from accruals explaining the difference between net income and cash flow from operations (equation 1). Total accruals are based on normal accruals and abnormal or discretionary accruals. Previous literature (Dechow, 2013)

In Accounting Theory Literature, Operating cash flow is considered as a determinant of the level of non discretionary accounting adjustment variables. Indeed, the relevance of this variable is justified by its negative correlation (-0.402415) that is to say between the variation of the operating cash flow of two consecutive periods (t -1 and t) and the level of current accruals. The small variation (-0.30%) of the flows in the sample F, explains that a large part of the accruals level seems predetermined by the operating cash flow of the previous year, it is logical to think that it is a non discretionary portion. We also note (Table 6) that the level of net fixed assets in the F sample represents approximately 51% of the net assets delayed versus 47% in the C sample, which suggests the sample F tends to minimize the calculated costs (low depreciation allowances or taken from large provisions) or to an under-exploitation of the economic assets and consequently a low economic and financial profitability. This work is part of an early preparation of a future research track on the interface accounting/failure of companies and the modeling of prediction of financial failure vs. fraud through the accounting and financial variables characterizing the incentives for manipulation of Accounting data.