2. METODOLOGÍA
3.4 JUVENTUD COMO CONCEPTO
3.4.1 Evolución del concepto de juventud 1900 a 1990
notes, it is also required to proxy for earnings management. The following section ex- plains the regression analysis. Further, it introduces several models to proxy for earn- ings management which were taken into account to conduct this research.
In general, the regression analysis aims to approximate the actual values as accurately as possible by the use of relevant independent variables. The regression analysis can
120 also be illustrated in a visual way by determination of a straight line which approxi- mates as close as possible to the point cloud of actual values. This straight line is speci- fied through the intercept and the slope. The calculation of regression coefficients is based on ordinary least squares method.
This research analyzes the empirical association between within-industry similarity of the notes and proxies for earnings quality. However, it must be taken into account that there are multiple aspects of disclosure quality (Dechow et al., 2010). Consequently, there are also various proxies to measure earnings quality. However, similarity of the notes should not be correlated with all earnings management proxies. Hence, the analy- sis only includes proxies that are expected to be affected by the similarity of the notes. Following many previous studies, abnormal levels of total accruals are used as a proxy for accrual-based earnings management. These abnormal accruals are measured using the cross-sectional modified Jones Model for each industry-year as proposed by Dechow et al. (1995) and Cohen et al. (2008). It has to be considered that this model is only applicable if at least 15 observations of each two-digit SIC code and year are avail- able. As opposed to Cohen et al. (2008), an intercept is included in the model. Moreo- ver, Kothari et al. (2005) suggest that including an intercept can help in controlling for heteroscedasticity, as well as mitigate concerns about an omitted size variable, see Ko- thari et al. (2005) for further explanation.
The abnormal accruals are also known as discretionary accruals. In particular, these are the total reported accruals divided by lagged total assets less the fitted values by using model (1) to proxy for normal levels of accruals. Normal levels of accruals are meas- ured as follows:
(1)
where for fiscal year t and firm i, TA is calculated as net income before extraordinary items less cash flow from operations. A is total assets, where PPE denotes gross proper- ty plant and equipment and ∆Rev-∆Rec represents changes in cash revenues from the preceding year (changes in total revenues less changes in accounts receivables). Throughout this paper, unsigned discretionary accruals are used as a proxy for accrual- based earnings management. This is due to the fact, that a certain direction of managed
121 accruals cannot be expected since there are various ways offered under US GAAP to manage the accruals either upwards or downwards (Ronen and Yaari, 2008). Further, accruals reverse over time. According to McNichols (2000), accounting information quality measured by similarity of accounting policy disclosures may reduce both in- come increasing and decreasing earnings management behaviour of firms.
The proxies for real earnings management are developed similar to Roychowdhury (2006). Following Zang (2012), overproduction of inventory in order to report lower costs of goods sold and reducing discretionary expenses are considered as tools for real earnings management by firms. In order to decompose normal levels and abnormal lev- els of production costs, the following equation (2) is measured, as suggested by Roy- chowdhury (2006):
(2)
where for fiscal year t and firm i, Prod refers to production costs as the sum of cost of goods sold and changes in inventories from the preceding year. A denotes total assets.
Rev is total revenues, and ∆Rev is defined as changes in total revenues from the preced- ing year. Abnormal production costs are measured cross-sectionally for each industry- year combination of at least 15 observations. Industries are defined by two-digit SIC codes.
Abnormal levels of discretionary expenses (Abdisx) are the difference between actual reported discretionary expenses and their estimated normal levels as shown in model (3). These normal levels are calculated cross-sectionally, in adaptation of Roychow- dhury (2006) and Zang (2012) if at least 15 observations for each industry-year are available. In particular, the normal levels were calculated as follows:
(3)
where for fiscal year t and firm i, Disx is the sum of R&D expenses, SG&A expenses and advertising expenses. Due to data limitations, advertising expenses were set to zero if missing. A is total assets and Rev represents total revenues.
122 Following prior literature (Zang, 2012), the abnormal level of discretionary expenses
Abdisx is multiplied by minus one so that higher amounts of discretionary expenses in- dicate higher magnitudes of income-increasing real earnings management. Since firms might use a range of real earnings management methods, both individual real earnings management proxies are aggregated to obtain a single real earnings management proxy. Furthermore, the firm-year accounting comparability score introduced by De Franco et al. (2011) is employed to validate the similarity scores of accounting policy and revenue recognition disclosures. In general, the degree to how similar firms translate economic performance measured by stock returns into accounting numbers proxied by earnings is the basis for their comparability measure. First, De Franco et al. (2011) estimate the following regression (4) for the prior 16 quarters:
(4)
where for fiscal quarter t and firm i, Earnings represents the ratio of quarterly net in- come before extraordinary items to the beginning-of-period market value of equity, and
Return is the stock price return during the quarter t. In contrast to the above described models, the following mathematical formulae (5) and (6) allow measurement in the cross-section. Thus, the following formulae differentiate between firm i and firm j by using α and β as the regression coefficients. With the obtained estimates of intercept α and slope coefficient β, they estimate the comparability of each firm pair by predicting firm j’s earnings using firm i’s returns (De Franco et al., 2011).
(5)
(6)
while the mathematical formula (6) compares firm i and firm j, De Franco et al. (2011) used formula (5) for control purposes. Finally, the results of both formulae are com- pared through subtraction as modeled by (7). The comparability between two firms is measured over the previous 16 quarters as follows:
123 (7)
If there are results for two firms being more comparable, the difference between the two predicted earnings of firm i and j for the same set of economic events is smaller. Based on this pairwise accounting comparability measure, De Franco et al. (2011) also pro- duce a firm-year comparability score by aggregating and calculating the mean value of
CompAcct for all pairwise combination within the same two-digit SIC code and year for each firm. As seen in model (7), De Franco et al. (2011) multiply their firm-year com- parability measure by minus one so that higher values indicate greater accounting com- parability. The aggregated mean comparability score Ind_acctcomp is utilized, since the similarity scores of the notes are similarly measured by taking the arithmetic mean of all pairwise firm cosine measure combinations for each firm in the same industry-year. This research obtained Ind_acctcomp from Rodrigo S. Verdi’s website and matched it with the sample. There is a free data access and the information is available under http://www.mit.edu/~rverdi/.
De Franco et al. (2011) used a univariate analysis in their empirical part. In particular, they performed preliminary benchmark tests for their accounting comparability meas- ure. In order to examine if there is an impact on the firm’s earnings, they focused on the quality, predictability and smoothness of earnings. The results of De Franco et al. (2011) show a significant relation between their adopted comparability measure and a profit or loss in the financial reporting. Hence, they conclude that there is a correlation of their measure with the firm’s performance. However, according to the previous re- search, the results of De Franco et al. (2011) could also be caused through the character- istics of the earnings response coefficients. The study of Hayn (1995) found out that the earnings response coefficients are related to reporting of a profit or loss. There are fur- ther studies pointing out similar associations. For example, Kormendi and Lipe (1987) state that predictability is associated with earnings response coefficients. According to Bushman et al. (2016), accrual quality is related with earnings response coefficients. None of these researchers conducted a comparability study to explore the association between these characteristics of earnings quality and earnings response coefficients. Therefore, it remains uncertain if the earnings attribute results calculated by De Franco
124 et al. (2011) are caused through comparability or if these are related to previous re- search findings.
In conclusion, this research recognizes that there may be an empirical association be- tween within-industry similarity of the notes and proxies for earnings quality. By draw- ing on Dechow et al. (2010), it is taken into account that there are multiple aspects of earnings quality. The similarity of the notes should not be correlated with all earnings management proxies. Hence, the analysis only includes proxies that are expected to be affected by the similarity of the notes. For instance, abnormal levels of total accruals are used as a proxy for accrual-based earnings management. More similar notes across firms within the same industry should lead to less variation of accruals. This is due to the fact that the accounting should be similar across firms within the same industry and extreme estimates are related to lower earnings quality. The adopted proxy attempts to measure abnormal performance or variation. By following Roychowdhury (2006), ab- normal levels of production costs and discretionary expenses were used to measure real earnings management. These included proxies cover various activities of a firm. They can be described as summary measures which is important for exploring similarity of the full set of accounting policy and revenue recognition disclosures. Accruals and earn- ings items are explained in the “Variable descriptions” included as Appendix 3.