1. Corporate Financial Reporting in the post Sarbanes Oxley Act era Panos Patatoukas

2. Motivation and Research Question The Sarbanes Oxley Act of 2002 is a US federal law passed in response to a series of major corporate scandals. The SOX act aims to improve the reliability of corporate financial statements by modifying the governance, reporting and disclosure requirements for public companies. Can we explain Income Statement reporting choices? What, if anything, has changed with respect to earnings management, earnings quality and reporting conservatism in the post SOX period?

3. Data Sources: COMPUSTAT annual industrial file and CRSP (WRDS) The full sample includes 421 different firms. Unique dataset in the sense that it includes values for twelve popular (including a new one) measures of earnings quality, earnings management and reporting conservatism. These are listed firms in NASDAQ, AMEX and NYSE. A firm enters the final sample if it exists in both the pre and post SOX samples.

4. Measures Earnings Management occurs when managers use judgment in financial reporting to mislead some stakeholders about the underlying economic performance of the firm Discretionary Accruals (MJ1, MJ2, MJ3) Performance adjusted discretionary accruals (PMJ) Special Items (special) The smoothing ratio (Smooth) Accruals (A) and Scaled Total Accruals (TA) A high-quality earnings number reflects current and future operating performance and accurately annualizes the intrinsic value of the firm. Measures of earnings quality attempt to capture the impact of earnings management Reporting conservatism impacts the quality of earnings but it should not be confused with earnings management. Conservatism as the manager’s tendency to require a higher degree of verification to recognize good news than bad news. Conservatism-Score (C) Quality-Score ( Q) Market to Book ratio (MB) The Basu measure of conditional accounting conservatism (Basu)

5. Preliminary data analysis Identify extreme observations Criteria for MB & Basu: Negative MB Extremely high MB (>1000) Basu should be within the range [-2,+2] The final sample of firms includes 388 firms with values for all the different measures for both the pre and post SOX years. Who are they? Utilities, Heavy Construction & Mining Industry

6. How do the data look like? 2001

7. How do the data look like? 2003

8. Correlations and PCA Correlations are not as high as I would expect… Some of the measures are highly correlated with each other: Discretionary accruals, total scaled accruals PCA could potentially allow me to create a composite index and capture most of the variability in my data. Not really For 2001 (2003) the first six principal components account for almost 82% (80%) of the total variability in the data.

9. How do the different measures cluster with each other? No major swifts in the grouping of the different measures in the pre and post SOX years The different clusters closely follow the theoretical grouping of the measures. Earnings Management cluster of measures Reporting Conservatism and Earnings Quality cluster of measures.

10. Conditional Conservatism Predict the Income Statement reporting choices in 2001/2003 For both the pre and post SOX years I group firms in aggressive, unbiased and conservative reporters based on first and third quartile of the cross sectional distribution of the Basu measure in the pre SOX year. Predict the grouping of firms based on the earnings management, earnings quality and unconditional conservatism variables.

11. K-Means Clusters and Reporting Conservatism From the confusion matrices of the Basu grouping versus the k- means clustering solutions (k=3) for the pre and post SOX period, I observe that the k-means clustering does not really capture the grouping of firms in aggressive, unbiased and conservative For 2001 (2003) for only 38% (36%) of the cases, the Basu grouping agrees with the k-means clustering. Given that the k-means clustering of firms in three groups is based on all the twelve measures the question that arises is to what extent I will be able to predict the Basu-grouping of firms based on these measures.

12. K-means clustering and PCA A commonly used method to determine the number of clusters is to look at the clusters in the Principal Components space. I plot the first principal component versus the second principal component and over-impose the k-means clustering solution for different k. I adjust the number k of clusters and choose the clustering which gives homogenous groups. I observe that for k>=4 the k-means groups are not well separated. For k=3 I am able to identify three well separated groups for both the pre and post SOX data. The grouping of firms in the three different clusters remains surprisingly stable in the pre and post SOX years.

13. K-means Clustering and PCA plots of the first versus the second principal component and k- means clustering for different k for 2003

14. Classification results Predictive AccuracyPre SOXPost SOX Classification Tree62.47% Random Forests47.98%45.60% Linear SVM51.07%52.02% Non Linear SVM (radial Basis Kernel)69.67%69.12% LOTC for non linear SVM67.70%67.50% Reporting Choices are predictable based on earnings management and reporting quality In support of theoretical claims that reporting conservatism is linked to reporting quality and earnings management.

15. Paired sample t-tests The question that naturally arises is what if anything has changed in the post SOX year? Paired sample t-tests could be used to detect significant changes in the post SOX year. The null hypothesis is that the mean paired sample difference is zero and the two sided alternative is that the mean paired sample difference is not equal to zero.

16. Paired sample t-tests Results for final sample paired sample t-testsMean differencest statp-value 95% Confidence Interval Special Items0.011726152.63970.0080.0029920.020459958 Basu0.0016941560.29020.7718-0.009790.013173374 log(MB)0.12647554.07560.000050.0654620.1874887 log(C-Score)0.051587962.06790.039310.002540.100636201 Q-Score0.082095921.29750.1952-0.04230.20649478 MJ10.0061960091.15030.2507-0.004390.016785975 MJ20.0058848571.29990.1944-0.003020.014785659 MJ30.010715072.14240.032780.0008820.020548373 PMJ0.016299622.12860.033920.0012440.031355033 TA0.011312722.12810.033960.0008610.021764562 log(A)0.012817650.77720.4406-0.019830.04546328 log(Smooth)-0.01667563-0.47680.6338-0.085440.05208985

17. Discussion In the post SOX period unconditional reporting conservatism has significantly increased and this is expected in light of the more stringent capital expensing requirements introduced by the SOX. The surprising finding is that even though firms have become more conservative, at the same time they report significantly higher discretionary accruals and special items and experience a lower smoothing ratio consistent with opportunistically managing earnings.

18. What’s wrong? These findings appear to be contradicting each other; on the one hand firms seem to be reporting more conservatively in the post SOX year and on the other they are reporting more opportunistically, managing earnings more extensively. Firms in the post SOX era have become more conservative in their reporting choices: they tend to more timely recognize bad news as opposed to good news and understate their book value relatively to their market value, but they do so excessively. The implementation of excessively conservative choices is consistent with the opportunistic use of financial reporting with the intention to build up reserves that could be used to smooth out future bad performance.

19. Robustness In order to test the robustness of my results I trim my final sample excluding firms according to the following criteria: The log Market to Book paired sample difference is more than 2 (4 firms) The log C-Score paired sample difference is more than 2 or less than -2 (6 firms). The Q-Score paired sample difference is more than 5 or less than -5 (8 firms). The trimmed sample of firms includes 370 firms. From the ticker identification codes I pin down that all the excluded firms are small capitalization firms, operating in the high tech sector of internet providers.

20. Full sample

21. Trimmed sample

22. Paired sample t-tests Results for trimmed sample paired sample t-testsmean of differencest statp-value 95% Confidence Interval Special Items0.010578182.33720.019960.0016780330.019478330 Basu0.0020014330.34290.7319-0.013479780.009476914 log(MB)0.10214763.82830.00015160.049678650.15461655 log(C-Score)0.060802052.3860.017540.010692350.11091175 Q-Score 0.0073372520.53630.5921-0.034239580.01956508 MJ10.0053610881.01770.3095 0.015720171 MJ20.0046466281.07290.284-0.003869550.013162811 MJ30.010514032.12340.03440.001204960.017557907 PMJ0.015789522.11750.03490.001737520.027880301 TA0.010311632.11920.03470.001341040.018961763 log(A)0.009907240.60010.5488-0.022558890.04237337 log(Smooth)-0.01247773-0.35130.7255-0.082318200.05736274 The surprising finding that in the post SOX year firms are not only reporting more conservatively but also more opportunistically, remains intact for the trimmed sample and thus it is not driven by few extreme observations.

23. Conclusions For both the pre and post SOX years, earnings management, earnings quality and unconditional reporting conservatism are useful variables in predicting Income statement reporting choices. Firms have become more conservative in the post SOX year, but at the same time they report significantly higher discretionary accruals and special items consistent with an increase in earnings management after the passage of the new regulation. Even though these findings seem to be pointing to different directions they clearly give credit to the view held by the SEC that the excessive use of conservative financial reporting methods is another form of opportunistic earnings management.