Detecting Earnings Management. An Analysis of Credit Institutions’ (Banks) Trading in Hungary

This paper examines evidence of Earnings Management (EM) in annual financial reports of foreign and domestic credit institutions’ trading in Hungary, an ex-communist country and a Central European economy, by applying an alternative approach, the Distribution of the Ratios method. Analyses were performed with 14 banking specific ratios for the period of 1999-2015, by applying Burgstahler and Dichev (1997), Degeorge et al. (1991) models, Kolmogorov-Smirnov, Monte-Carlo, Benchmark and Distributional tests. Primary findings confirm that (a) the Distribution of the Ratios method demonstrates that apart from significant evidence of EM presence, managers do not always manipulate the same variables, or one area of the financial statements, but at their discretion, choose different areas to engage EM and in different periods, and (b) Credit Institutions engaged in EM in the period Prior to and After the 2008 financial crisis. Additional Benchmark Analysis results present weak evidence of EM and should be read with caution; however, Benchmark comparison should not be excluded from research.<br>


Hypotheses
This paper test total of 14 variables to investigate evidence of EM of foreign and domestic credit institutions' (Banks) annual accounts trading in Hungary. Two hypotheses were tested: H0 (a) : Credit institutions' (Banks) in Hungary do not manage annual accounts. H0 (b) : Credit institutions' (Banks) in Hungary do not manage annual accounts 'Prior to ' and/or 'After' the 2008 financial crisis.

Methodology
Tests were performed with Burgstahler and Dichev (1997), Degeorge, Patel and Zeckhauser (1999) models, Kolmogorov-Smirnov, Monte-Carlo Method, benchmark and distributional tests. A total of 14 banking specific variables are tested on an annual basis. The two testing approaches are, see e.g., Beretka (2016, 129), the Distribution of Ratios Method, it has two testing designs, the EM1 and the EM2 models that test Hypothesis H0 (a) : and H0 (b) :. This paper calculates the 14 ratios by applying the EM1 model, or, where, EM1 is Earnings Management Model 1, and is equal to the Actual Observation (AO) in period (i) minus the Expected Observation (EO) in period (i). SD i = Standard Deviation, or, SD i = [Np i (1-p i ) + ¼ N (p i -1 + p i +1 ) (1-p i -1 -p i +1)] ½ , where, SD i = Standard Deviation of the difference in period (i); p i = probability of an observation in interval (i); N = number of total sample; Np i = total number of estimated Standard Deviation (SD) in interval (i); pi-1 = number in interval i-1; pi+1= number in interval i+1. For additional explanation see Beretka (2016, 300-304 where, i ϵ R, i ≠ n. p i is the ratio of the actual sample for year i of banks years, Δpn is the difference of p i -p i-1 . Mean (Δpi) is the average of Δp but excluding pi and s.d. (Δpi) is the standard deviation of Δp, excluding Δpi.
Test statistics with Kolmogorov-Smirnov (K-S) were performed, see, e.g., Massey (1951) for the Kolmogorov-Smirnov (K-S) test statistics, as well Lilliefors (1967, 399). K-S is applicable for ratio or interval data, a sample of N observations: D = max x∈ R | F(x) -F0(x) |, where, F(x) is the cumulative normal distribution, and F0(x) is the sample cumulative distribution, with µ = sample mean and σ 2 = s 2 sample variance with denominator n-1. F(x) = F0(x) for all x from −∞ to +∞; F(x) ≠ F0(x) for at least one value of (x). The K-S one sample test is non-parametric and distributionfree, an exact test, see, e.g., Panik (2005, 570). Additionally, benchmark analysis was calculated for each 14 ratio. Benchmark formula reads as: -Debt to Equity (DTE) = Total Liabilities (t) divided by Equity (t) .
-Loans to Total Assets (LTA) = Loans (t) divided by Total Assets (t) .
-Rate Paid on Funds (RPF) = Total Interest Expenses (t) divided by Total Earning Assets (t) .
-Return on Equity (ROE) = Profit After Tax (t) divided by Average Equity (t) .
-Return on Asset (ROA) = Profit After Tax (t) divided by Average Total Assets (t) . Where: Average Total Assets = (Total Assets (t) + Total Assets (t-1) ) divided by 2 -Equity to Total Assets (EtA) = Equity (t) divided by Average Total Assets (t) . See, e.g., Beretka (2016, 142) for ratio calculations. Ratios are calculated for the same companies within a time frame, for example, company 'z' in year (t) is calculated with the same company 'z' in year (t-1 ) or in (t-2 ).

Data
Audited data of publicly and privately owned foreign and domestic Credit Institutions -banks (operating as Joint-Stock Companies in Hungary) were obtained from the Central Bank of Hungary -CBH and from the Hungarian Financial Supervisory Authority (HFSA). Since 1 October 2013, HFSA has become part of the CBH and operates under the umbrella of CBH, CBH/HFSA thereafter. Tested data is from 1999 to 2015 and consist of one listed (on the Budapest Stock Exchange; OTP Bank) and non-listed banks. For the period 1999-2015, CBH/HFSA published data annually and per Hungarian Accounting Standards (HAS); including for the listed OTP Bank. See HAS vs. IFRS differences in Table 3.1, Beretka (2016, 113-116). From 2017, CBH/HFSA introduces IFRS reporting, however, as per 2017 published data by the CBH/HFSA, only few banks switched from HAS to IFRS. For research purposes, CBH/HFSA gather individual companies (audited) annual accounts and create their own simplified version of the annual reports for each financial company, shown in Hungarian Forint (HUF). Hungary is not member of the European Monetary Union. Tested annual data consists of Profit and Loss Account and Balance Sheet. Cash Flow is not prepared nor published by CBH/HFSA. Number of individual banks per year is between 37 to 45 and per variable varies from 502 to 682 for the sample. This paper performed tests of Credit Institutions -banks annual accounts that were prepared under the HAS.

Statistical results of Ratios tested on the Annual basis with the EM1 model
The Distribution of Ratio method is a hands on approach, based on statistical and distributional RAIS Conference Proceedings, November 6-7, 2019 Finance managers' skills have become so sophisticated in manipulating accounting entries that evidence of EM may not emerge only in one section of the financial statements, but evidence may arise in multiple, or in all areas of financial statements and in diverse time frames. Table 1 presents this evidence, and as it was expected, not every year shows statistically significant test results, which suggests that EM is not present in every financial year for each ratio. This is in line with practise, as managers in order to hide they action, as well to avoid high financial penalities, they will engage in EM in different periods. The same is true for 95% Confidence Interval and at p = 0.05 Significance levels for the EM1 model. Addional test statistics on annual basis for 95% Confidence Interval and at p = 0.05 Significance levels for the EM1 model are available from the author. The EM2 model in Table 2, for the same tested ratios, shows almost identical statistically significant results as for the EM1 model in Table 1. This would suggests, that managers in order to outplay the strict financial rules, engage in EM by manipulating different areas of financial statements with different techniques, see e.g., Beretka (2016, 22-24), and at their discretion, in years / periods of their choosing. The results suggest that the hypotheses H0 (a) : and H0 (

Benchmark Analysis
Earlier evidence suggests that research papers seldom test Benchmark when investigating EM. Tables 3 and 4 present Descriptive Statistics output for Benchmark and for Base ratios. Only a handful of papers argue, see e.g., Dechow, Ge and Schrand (2010, 351), in favour of including Benchmark analysis with test statistics and/or with a combination of histograms, see e.g., Beretka (2016, 189-199 Table 3 are positive thus suggesting 'ponty and heavily-tailed distribution, while negative values suggest 'flat' distribution, see e.g., Field (2009, 138). However, Benchmark results should be read with caution, due to the specifics of the Hungarian Accoiunting Standards and the lack of quality research availability. Table 3 Benchmark outputs were calculated from the same sample as the Base ratios statistics. Despite the lack of research material, Benchmark comparisons should not be excluded from research.

Discussion
The Distribution of Ratios research approach applies a technique to use all possible computable variables from the financial statements, thus ratios that do not contain assets and/or liabilities, as component in their formula, it eliminates the reversal accrual impact and increases the credibility and power of the statistical test results in comparison to accrual testing models. By rejecting Hypotheses  H0 (a) : and H0 (b) : for ratios that comprise current assets or total assets (i.e. ETA, LTA, ROA), or current / long term liabilities (i.e. DTE, ETL, LTA, LTD), which may have influenced the test statistics due to reversal accruals, a component of assets and/or liabilities in the balance sheet, this study could have made a Type I error.

Conclusion
This paper examined an alternative research approach, the Distribution of the Ratios method in search of evidence of Earnings Management of annual accounts of the Hungarian credit institutions (banks). Test outputs run on an annual basis with the EM1 and the EM2 models for each ratio present more details, specifically highlighting that evidence of Earnings Management, thus rejecting H0 (a) :. Furthermore, evidence for EM1 and EM2 models present an opportunistic approach by managers who manipulate accounting entries at their discretion in different time frames and in all areas of the financial statements, thus H0 (b) : is rejected. Benchmark analysis was performed with the same sample as the base ratios data. Due to lack of benchmark material for financials and for credit institutions in Hungary, the results should be read with caution. However, when industry benchmark data is available, it should be used for comparison. This paper also investigated the 2008 financial crisis, and concluded that 'Prior to' and 'After' the 2008 financial crisis there is evidence of Earnings Management, thus H0 (b) : is rejected. The EM1 and EM2 models test results on annual basis show statistical significant results for 99% confidence and p = 0.01 significant levels for all 14 tested ratios. Test results for 14 ratios for the EM1 and EM2 models tested with all sample and with annual data of 95% confidence interval and p = 0.05 significance level are statistically significant and they are available from the author.