Political Risk & Earnings Quality: An analysis of political effects on earnings management

(1)

Political Risk & Earnings

Quality - An analysis of political effects on earnings management

Master’s Thesis 30 credits

Department of Business Studies Uppsala University

Spring Semester of 2019

Date of Submission: 2019-05-29

Simon Hawborn Dahlstedt

Supervisor: Jan Lindvall

(2)

i

Abstract

The high level of political risk might enhance the information asymmetry between managers and stakeholders, therefore leading to increased opportunity for earnings management activities, which depress the usefulness of financial information. On the other hand, times of high political uncertainty possibly increase the demand for information among stakeholders, consequently leading to enhanced scrutiny and fewer earnings management activities. By examine 625 firms listed in the United States between 2002- 2016, I make use of a firm-level measurement of political risk to identify the possible impact on earnings quality. I identify that political risk exposure measured on a firm-level is negatively associated with earnings management. Therefore I can conclude that firm- level political risk increases earnings quality. I further show how firm-level political risk better predicts earnings management activities than an aggregated measurement of political risk. Finally, I provide evidence that suggests that accrual-based earnings management is affected by the past level of political risk exposure. Real earnings management activities show no such indications.

Keywords: Political risk, Uncertainty, Earnings quality, Earnings management

(3)

ii

I. Introduction

Concerns about policy uncertainty and political risk have been intensified in recent years. From the decision by UK´s voters to leave the European Union to a rise in popularity towards political parties that polarize our society. In these times of high uncertainty, the importance of genuinely understand its effects and complications to the business environment intensifies. The subject of how the effects of risks emanating from the political system affect the growth of the economy and the efficiency of our financial system are debated among economists, business leaders, and politicians. Current research has mainly focused on how political risk effect macroeconomic conditions. For example, Alesina et al. (1996); Julio and Yook (2016); Gourio et al. (2015); Handley and Limao (2015); Boutchkova et al.

(2012) all identify a negative association between political risk and macroeconomic growth indicators.

A smaller and more recent part of studies has directed their attention to how political risk impact aspects of managerial decision making, such as investments and cash holding preferences (Gulen and Ion, 2016; Julio and Yook, 2012; Bonaime et al., 2018). This paper aims contribute to this relatively new area of research by examining how political risk impacts another dimension of managerial decision making, namely the action of earnings management, and consequently what this can imply for earnings quality.

The importance of earnings quality has been conventional wisdom since the revolutionary papers by Ball and Brown (1968); Beaver (1968), who first acknowledged how decisions and beliefs among investors changed when faced with new earnings figures. Complimentary papers identify a strong association between earnings quality and cost of capital (Lev and Zarowin, 1999; Skinner and Sloan, 2002; Bharath et al., 2008; Lang and Lundholm, 1996). Thus the efficiency of capital markets is highly dependent on the quality of financial information provided by firm management. A collective agreement of how managerial incentives related to the reporting and management of accounting earnings are affected by increased exposure to political risk is absent. Schipper (1989) suggest that increased information asymmetry between managers and stakeholders during times of uncertainty initiate the opportunity to conceal earnings management activities. Therefore, reduce earnings quality.

A contrarian argument is presented by Mitton (2002), who means that times of uncertainty imply an increased demand for transparency among stakeholders. Thus, the opportunity to conceal earnings management diminishing with the rise of political risk. Which of these two effects dominates is an open empirical question.

I aim to fill a gap in our understanding of the determinants of earnings quality but also broaden our knowledge about political risk, and its influence on how managers use their discretion over

accounting numbers. More specifically, I aim to answer the following research question:

(5)

2

Does political risk affect earnings quality by changing managerial incentives for earnings management activities?

Research that investigates the impact of political risk is dependent on a measurement that can capture the complexity of the term; this can be problematic since the term is a subject of high subjectivity (Sottilotta, 2013). The EPU index is a widely recognized measurement of aggregated political risk, measuring the level of political risk a specific market is exposed to (Baker et al., 2016). However, firm-level decisions will be based on managers perception of the current business climate; by using the EPU index, one would assume that each firm in the same market is equally exposed to political uncertainty. Hence, implicitly assume that managers perception (regarding political risk) of the current business climate is homogenous. To be able to capture the individual firm management´s perception of the firm´s exposure to political risk, an alternative measurement is required. By analyzing how managers communicate with investors through quarterly earnings conference calls, Hassan et al. (2017) capture the individual firm management´s perception and conduct a firm-level measurement of political risk.

To the best of my knowledge, no other study has used a firm-level specific measurement for political risk to exam the relationship between political risk and earnings quality. This study is based on data provided by Hassan et al. (2017), including 9,478 firms listed in the United States between 2002 and 2016. Hence, the hypotheses will just be tested on this data set; I do not aim to make any conclusions about political risk and earnings quality outside the borders of the United States.

Further, I only investigate in the discretionary part of earnings quality, i.e., I only make conclusions regarding how political risk affects managerial behavior and judgment when it comes to financial reporting. However, the innate part of earnings quality is justified to be covered in the literature review because of the implications it introduces to the measurement of earnings management.

The remainder of the paper is organized as follows. Section Ⅱ is a literature review of political risk and earnings quality and concludes in hypotheses. Section Ⅲ provides my data sample and research methodology. Section Ⅳ documents the main empirical ﬁndings and the testing of my hypotheses.

Section Ⅴ is conclusion and suggestion for future research.

(6)

3

II. Literature Review Political Risk: Definition & Measurement

Political risk is a commonly used term in business literature. However, an agreement among

researchers about the terms true meaning is not yet conducted. A common definition of political risk is the implication of unwanted government interference with business operations (Kobrin, 1979). A representative description is provided by Weston and Sorge (1972):

"Political risks arise from the actions of national governments which interfere with or prevent business transactions, or change the terms of agreements, or cause the confiscation of wholly or partially foreign owned business property.”

However, a caveat to that definition is that it assumes that government interference is necessarily harmful to the business environment. Sottilotta (2013) argues that this is not always true and that political risk assessments must be analyzed by observing the divergence of objectives between businesses and the hosting governments.

Another part of the literature defines political risk in terms of events, either as a single event that can be classified as political (e.g., presidential elections) or constraints that are forced upon the business environment (e.g., regulatory actions) (Kobrin, 1979). Further, distinctions are made by Robock (1971), that distinguish between macro political risk (political changes that influence the global market) and micro political risk (political changes that affect specific business fields or single business). Sottilotta (2013) means that in the absence of a clear definition, uncertainty surrounds researchers on how political risk should be assessed. This uncertainty holds for all level of analysis (i.e., macro- and micro-level).

The multitude of meaning attached to the term “political risk” makes the quantification of a

measurement that truly captures the term complex. A common theme among current literature is the dependence of event studies or an aggregated measurement (macro-level) for political risk. Event studies can provide a detailed analysis of the impact of a political event. However, a limitation is that it implicitly implies that political risk is constant during periods of non-political events. To be able to capture the dynamics of political uncertainty, an alternative measurement must be used (El Ghoul et al., 2018). A frequently used and widely recognized measurement for political risk is the aggregated Economic Policy Uncertainty (EPU) index developed by Baker et al. (2016). By a textual analysis of newspapers, Baker et al. (2016) measure the frequents of politically related words in conjunction with words that imply risk or uncertainty. The EPU index correlates with political events that by nature enhance the political uncertainty, e.g., presidential elections, the 9/11 attacks, and the failure of Lehman Brothers. Hence Baker et al. (2016) use single points of political events to increase the

(7)

4

validity of EPU but compared to event studies the EPU is a dynamic measurement that also varies over periods of non-political events.

Moreover, by conducting textual analysis of quarterly earnings conference-call transcripts, Hassan et al. (2017) quantify political risk on a firm-level basis (micro-level). Thus, constructing an alternative measurement to EPU for political risk on firm-level.

Hassan et al. (2017) collected the transcripts of conference calls in conjunction with the earnings release of 9,478 firms listed in the United States between 2002 and 2016. The objective was to analyze how risks associated with political topics were covered in the conversation between the analyst and the firm´s management during quarterly earnings conference calls. Hassan et al. (2017) first identify word combinations (“bigrams”) that are commonly used in political language. Then by using a pattern-based sequence-classification method, they count the number of “bigrams” used in each conference call in conjunction with synonyms for “risk” or “uncertainty,” which conduct the measurement of firm-level political risk.

The main findings from Hassan et al. (2017) are that the incidence of political risk across firms is significantly more volatile and heterogeneous than what is predicted by previous theory. Thus, the problem with relying on EPU index and any other aggregated measurement for political risk is that it can mask a significant part of the variation in political risk by assuming homogeneity among

industries and firms.

Effects of Political Risk

A growing literature examines how aggregated political risk affects macroeconomic conditions.

During periods of political uncertainty, countries will experience significantly lower growth of GDP per capita, according to Alesina et al. (1996). Moreover, Julio and Yook (2016) identify a reduction of cash flow to foreign affiliates by US companies during election periods. Boutchkova et al. (2012) argue that markets exposed to high political risk experience higher return volatility, which suggests an explanation for Gourio et al. (2015) findings, were investor withdrawal capital from the market to protect themselves from the enhanced risk that higher return volatility implies. By examining the effects of political uncertainty related to trade policies, Handley and Limao (2015) identify a negative relation between political uncertainty and international trade. The consensus is that increased political risk and uncertainty will affect economic growth negatively.

Another part of the literature attempt to investigate how firm-level decision making is affected by political risk. Gulen and Ion (2016) find evidence of a negative relationship between policy

uncertainty and capital expenditures. Consistent with Gulen and Ion (2016) findings, Julio and Yook (2012) acknowledge the same negative relationship by analyzing managerial investment decisions during election years. Further, Julio and Yook (2012) additional identify an enhanced preference for

(8)

5

holding cash during periods of high political risk. Bonaime et al. (2018) investigate how political uncertainty affect managerial appetite for M&A activities; their findings suggest that political uncertainty negatively influence M&A activities both in aggregate and on firm-level. This result can partly be explained by the preference for cash during uncertain times suggest by Julio and Yook (2012). The results from previous literature evidently suggest that increased exposure to political risk cause increased managerial prudence. Consequently, questions arise regarding other dimensions of managerial decision making and if the increased prudence is a general effect.

Earnings Quality

Krishnan and Parsons (2008) state that the manager of the firm has an ethical obligation to report high-quality information to shareholders in a timely matter. High-quality information leads to higher quality judgments and decisions both among capital markets participant and in contract negotiations.

A common term describing the quality of financial information is earnings quality. Dechow at el.

(2010) provides the following definition of earnings quality.

” Higher quality earnings provide more information about the features of a firm’s financial performance that are relevant to a specific decision made by a specific decision-maker.”

Dechow at el. (2010) distinguishes three features of their definition of earnings quality. First, the term

“earnings quality” is meaningless on its own; what they mean is that earnings quality is defined in the context of a specific decision model. Second, the quality of the earnings numbers is dependent on whether it is informative about the real financial performance of the firm. Third, the quality of the earnings is jointly measured by the relevance of financial performance to the decision and by the ability of the accounting system to measure real performance. By Dechow et al. (2010) definition, earnings quality is not constrained to the usefulness of the information for any single purpose. Instead, it is measured as the usefulness of the information for any decision model relying on earnings figures.

This definition suggests that any action that reduces the usefulness of the information from any specific decision model imply lower earnings quality. Moreover, important to acknowledge is that each decision model generates its own proxy for earnings quality, by focusing on different elements of decision usefulness the proxies do not measure the same underlying construction. Thus, one cannot expect them to work equally under all circumstances.

The drivers of earnings quality are usually distinguished between innate and discretionary

determinates. The innate factors are related to business models and exogenous factors like operating environment and economy-wide forces. There is usually little that stakeholders and managers can do about them, except understand and acknowledge their exitance and impact. For example, the

profitability of R&D-activities is by nature hard to predict, making the reported earnings inherently volatile for companies and industries that rely on R&D projects. The discretionary part of earnings

(9)

6

quality is more volatile than innate factors and driven by managerial reporting decisions; information systems; and regulations (Kothari et al., 2002).

The pure knowledge of the distinction between innate and discretionary drivers of earnings quality is essential for this paper. Dichev et al. (2013) suggest that innate factors drive 50% of earnings quality, this is problematic when one wants to examine the discretionary factors in isolation, which is the purpose of this study. Prior research has already concluded that political risk have an adverse effect on macroeconomic conditions (Alesina et al., 1996; Julio and Yook, 2016; Gourio et al., 2015; Handley and Limao, 2015; Boutchkova et al., 2012), suggesting that political risk can affect earnings quality through the innate drivers if the impact on the economic condition leads to a change in the operating environment for the business. Therefore, the identification of through which driver political risk influence earnings quality can be puzzling.

Earnings Management

About 20% of firms manage earnings to misrepresent underlying performance (Dichev et al., 2013).

Healy and Wahlen (1999) describes earnings management as an event when managers use discretion over financial reporting to either mislead stakeholders about the actual economic performance of the company or to influence contractual outcomes that are dependent on accounting numbers.

Opportunistic behavior by the management reduce the usefulness of the reported earnings since they no longer reflect actual financial performance; hence, earnings management activities depress earnings quality. Further, Cheng and Warfield (2005) state two conditions that must hold for the opportunity of earnings management. First, the expected benefits of earnings management must be higher than the expected costs of earnings management. Second, stakeholders are not able to “see through” the earnings management.

The literature distinguishes between accrual-based earnings management and real earnings management. The first is conducted by opportunistic or misleading using of accruals, example aggressive revenue/expense recognition policies, or changes in accounting choices. Real earnings management is accomplished by using real activities to depress long-term growth for short-term profit, hence affecting cash flows. Roychowdhury (2006) find evidence that real earnings management is generally more costly for the firm. However, Graham et al. (2005) suggest that managers are more willing to engage in real earnings management, which appears inconsistent with the higher cost. Lo (2008) provide an explanation by arguing that managers are more willing to engage in real earnings management because it is harder to detect. The business environment is inherently uncertain, which makes real earnings management attractive. Zang (2012) acknowledge that managers use accrual-based and real earnings management as substitutes. Thus, suggesting that examine one of the two methods of earnings management in isolation cannot lead to a definitive conclusion.

(10)

7

El Ghoul et al. (2018) examine how political uncertainty affect managerial judgment related to both accrual-based earnings management and real earnings management. By using EPU index as a

measurement for political uncertainty and a cross-country sample of 27,888 firms over the 1990-2015 period, El Ghoul et al. (2018) suggest that increased policy uncertainty will reduce earnings

management activities. The result is consistent with the hypothesis that a high level of political risk exposure is followed by enhanced prudence among managers. A possible explanation for this phenomenon is provided by Mitton (2002), who suggests that during a period of high uncertainty outside investors increase the demand for transparency. The enhanced scrutiny that follows from the increased demand for financial information reduces any opportunity for earnings management, especially when managers need access to external capital.

To the contrary, Ghosh and Olsen (2009) identify increased use of accrual-based earnings

management when firms operate in high uncertainty. They argue that managers use accrual-based earnings management as a tool to offset the increased variability in reported earnings caused by environmental uncertainty. Further, Schipper (1989) suggest that the information gap between managers and stakeholders is likely to increase when firms are exposed to a high level of uncertainty.

Thus, managers ability to conceal earnings management activities increase during times of high uncertainty, which possibly is an additional explanation for the findings by Ghosh and Olsen (2009).

Hypothesis Development

The result from Cheng and Warfield (2005) suggest that if political risk affects at least one of the two conditions of earnings management, a relationship between political risk and earnings management can be acknowledged. Moreover, since earnings management has a depressing effect on earnings quality through discretionary drivers, a connection between earnings quality and political risk can be identified.

Schipper (1989) argues that without full communication, and increased information asymmetry, investors are exposed to a higher risk of expropriation by managers through earnings management activities. During times of uncertainty, the information asymmetry between managers and

stakeholders is likely to be higher. Thus, the ability among stakeholder to “see through” earnings management activities is restrained; hence, Cheng and Warfield (2005) second condition for

incentivizing earnings management holds. Therefore, a high level of political uncertainty might allow managers to conceal earnings management and incentives them to engage in more earnings

management activities.

On the other hand, Mitton (2002) suggest that during times of political uncertainty, investors become more prudent and increase their demand for transparency. The intensified scrutiny depresses the opportunity for the management to act opportunistically due to enhanced difficulty in concealing earnings management. McInnis and Collins (2011) find evidence that increased scrutiny due to policy

(11)

8

uncertainty, enlarge the cost of earnings management activities, hence suggesting that a high level of political uncertainty will reduce earnings management activities.

By taking advantage of the firm-level measurement of political risk developed by Hassan et al.

(2017), I present my first hypothesis:

H1: Earnings management is unrelated to firm-level political risk.

The findings of Zang (2012) emphasize a comprehensive analysis of the methods of earnings management. Therefore, I decompose my first hypothesis into two parts.

H1a: Accrual-based earnings management is unrelated to firm-level political risk.

H1b: Real earnings management is unrelated to firm-level political risk.

Further, as suggested by Hassan et al. (2017), an aggregated measurement for political risk will fail to acknowledge a significant amount of the variation in political risk that plays out on a firm-level.

Hence, a firm-level measurement better represents the exposure to political risk each firm face. I expect that a firm-level measurement of political risk will capture a more significant part of the variation in the dependent variable (earnings management measurements) because earnings management activities are firm-level decisions.

H2: Firm-level political is not a better predictor of the variance in earnings management than the EPU index.

Moreover, by following the approach from the first hypothesis, the second hypothesis can be decomposed into two parts.

H2a: Firm-level political is not a better predictor of the variance in accrual-based earnings management than the EPU index

H2b: Firm-level political is not a better predictor of the variance in real earnings management than the EPU index

(12)

9

III. Data & Methodology

Data & Sample Construction

The data set developed by Hassan et al. (2017) consists of 9,478 firms listed in the United States between 2002 and 2016. In the data set all firms get a score based on their exposure to political risk for each quarter; this implies a data set of 209,093 quarterly firm observations. To obtain an annual score for each firm, I calculate the mean based on the score for each quarter. If a quarterly observation is missing in a year, I remove that firm-year observation from the sample, this to assure that each firm-year observation consists of a complete set of quarterly observation. After removing missing observations, my data set consists of 36,235 firm-year observations.

I use the US Monthly EPU Index as an aggregated measurement for political risk (Baker et al., 2016).

The annual EPU score is calculated as the average EPU score over 12 months. I merge the annual EPU data with the data set obtained from Hassan et al. (2017).

I obtain financial data for all firms in Compustat North America, and then merge it with the data set obtained from Hassan et al. (2017) and Baker et al. (2016). Observations that do not identify a matching observation in the merger is dropped. I exclude firms from the financial industry (SIC 6000- 6999) because their financial structure and high regulatory scrutiny make a significant difference in managerial decision making for these firms. Extreme observations can distort the result of the study.

To mitigate the impact of outliers, I winsorize all continuous variables at the 1st and 99th percentiles for each year. Implying that each variable is compared against the whole data set to identify and replace extreme observations for each year.

Several control variables are calculated based on the standard deviation of a specific variable during all observed years. Hence, firms with less than ten firm-year observation are dropped from the data set, this to increase the validity of the control variables and mitigate the impact of single year observations. Further, data covering macroeconomic conditions such as GDP growth and inflation rate, I obtained from the World Bank’s World Development Indicators (WDI) database. Table 1 shows the derivation of my sample. My final data set consists of 625 different firms and 7,461 firm- year observation; all firms are listed in the United States between 2002 and 2016.

By only observing one regulatory market (United States), I isolate the discretionary impact on earnings quality to managerial judgment. Moreover, a homogenous sample also reduces the risk of introducing innate drivers in my measurements.

(13)

10 Table 1: Derivation of Sample, 2002 to 2016

Obs Total quarterly observations of firm-level political risk from the dataset by Hassan et al. (2017)

(uncleaned). 209,093

Firm-years observations consisting of a complete set of quarterly observations (i.e.,

four quarterly observations for each year). 36,235

Firms-year observations after winsorize all continuous variables, and with available cash from operations, earnings, sales, cost of goods sold, and changes in accounts receivable, inventory and total assets (other missing variables is assumed as zero).

14,121

Firms with 10 or more annual observations 625

Firm-years available for the 625 firms used 7,461

Variables

Earnings Management Accrual-based

Accrual-based models are conventional for detecting earnings management, thus also used as a proxy of earnings quality (Dechow et al., 2010). By modeling the accrual processes (changes in working capital and depreciation), I distinguish between “abnormal” and “normal” accruals. Normal accruals are meant to be explained by real business activities and are primarily driven by innate factors, while abnormal accruals are accounted as products of discretionary factors. The general interpretation is that if normal accruals are modeled accurately, then the abnormal component represents a distortion of lower quality. Misclassification errors weaken the explanatory power of the model. Type Ⅰ error, a false positive, i.e., classify accruals as abnormal when they are justified by fundamental performance.

Type Ⅱ, a false negative, i.e., classify true abnormal accruals as normal (Dechow et al., 2010).

Total accruals can be observed in financial statements by subtracting cash flow from operations from net income. However, normal accruals cannot be observed directly. Thus the first step is to estimate the normal level of accruals. The purpose of the estimation model is to capture a sort of constant relationship between the observable accounting numbers and the unobservable normal level of accruals. Jones (1991) solves this problem by assuming that accruals are a function of revenue growth and that depreciation is a function of PPE. Hence, normal level of accruals is expected to have a constant relationship with revenue growth and PPE.

Dechow et al. (1995) criticized the Jones (1991) model by questioning the model's implicit assumption that revenue is nondiscretionary, i.e., type Ⅱ error. Dechow et al. (1995) suggest a modified Jones model, where revenue growth is adjusted by excluding the change in receivables in years identified as manipulation years.

(14)

11

To detect accrual-based earnings management activity, I use discretionary accruals (𝐷𝐴_𝑖𝑡) estimated as actual total accruals (𝑇𝐴_𝑖𝑡) subtracted by the normal level of accruals (𝑁𝑜𝑟𝑚𝑎𝑙_𝑙𝑒𝑣𝑒𝑙𝑇𝐴_𝑖𝑡) for firm 𝑖 in year 𝑡.

𝐷𝐴_𝑖𝑡 = 𝑇𝐴_𝑖𝑡 − 𝑁𝑜𝑟𝑚𝑎𝑙_𝑙𝑒𝑣𝑒𝑙𝑇𝐴_𝑖𝑡 (1) To identify 𝑇𝐴_𝑖𝑡 I use a cash flow approach originally developed by Hribar and Collins (2002) where the variables are observable in financial statements. 𝑇𝐴_𝑖𝑡 is defined as the difference between net income from the income statement (𝑁𝐼_𝑖𝑡) and cashflow from operations from the statement of cash flows (𝐶𝐹𝑂_𝑖𝑡).

𝑇𝐴_𝑖𝑡 = 𝑁𝐼_𝑖𝑡− 𝐶𝐹𝑂_𝑖𝑡 (2) When 𝑇𝐴_𝑖𝑡 is observed I use the modified Jones model (Dechow et al., 1995) to estimate

𝑁𝑜𝑟𝑚𝑎𝑙_𝑙𝑒𝑣𝑒𝑙𝑇𝐴_𝑖𝑡. Equation (3) is the estimation model.

^𝑇𝐴^𝑖𝑡

𝐴𝑇_𝑖𝑡−1 = 𝛽₁ ¹

𝐴𝑇_𝑖𝑡−1+ 𝛽₂^{∆𝑆𝐴𝐿𝐸}^𝑖𝑡

𝐴𝑇_𝑖𝑡−1 + 𝛽₃ ^𝑃𝑃𝐸^𝑖𝑡

𝐴𝑇_𝑖𝑡−1+ 𝛽₄ ^𝐼𝐵^𝑖𝑡

𝐴𝑇_𝑖𝑡−1+ 𝜀_𝑖𝑡 (3) Where ∆𝑆𝐴𝐿𝐸_𝑖𝑡 is the change in net sales from year 𝑡 − 1 to year 𝑡 observed from the income

statement; 𝑃𝑃𝐸_𝑖𝑡 is gross property, plant and equipment from the balance sheet year 𝑡; and 𝐼𝐵_𝑖𝑡 which is income before extraordinary items year 𝑡. All variables are scaled by lagged total assets (𝐴𝑇_𝑖𝑡−1).

By an OSL regression analysis of equation (3) I obtain the coefficients from the estimation model, they are then used in a prediction model to obtain finally 𝑁𝑜𝑟𝑚𝑎𝑙_𝑙𝑒𝑣𝑒𝑙𝑇𝐴_𝑖𝑡.

𝑁𝑜𝑟𝑚𝑎𝑙_𝑙𝑒𝑣𝑒𝑙𝑇𝐴𝑖𝑡 = 𝛽1 1

𝐴𝑇_𝑖𝑡−1+ 𝛽2

∆𝑆𝐴𝐿𝐸_𝑖𝑡−∆𝐴𝑅_𝑖𝑡 𝐴𝑇_𝑖𝑡−1 + 𝛽3

𝑃𝑃𝐸_𝑖𝑡 𝐴𝑇_𝑖𝑡−1+ 𝛽4

𝐼𝐵_𝑖𝑡

𝐴𝑇_𝑖𝑡−1 (4) Where ∆𝐴𝑅𝑖𝑡 is the change in accounts receivables from year 𝑡 − 1 to year 𝑡 observed from the balance sheet; and all 𝛽 is equal to all 𝛽 from equation (3). When 𝑇𝐴_𝑖𝑡 and 𝑁𝑜𝑟𝑚𝑎𝑙_𝑙𝑒𝑣𝑒𝑙𝑇𝐴_𝑖𝑡 is identified and estimated, 𝐷𝐴_𝑖𝑡 is easy to generate based on equation (1). As stated by Healy and Wahlen (1999), earnings management activities can both be income-increasing and income- decreasing depending on the managerial motive. One would expect a negative (positive) value for 𝐷𝐴𝑖𝑡if income-increasing (decreasing) earnings management is present, thus higher (lower) earnings realization. However, I aim to investigate the existence of earnings management, the direction (income-increasing or income-decreasing) is not of significance for this study. Hence, any

discrepancy from 𝑁𝑜𝑟𝑚𝑎𝑙_𝑙𝑒𝑣𝑒𝑙𝑇𝐴_𝑖𝑡 is acknowledge as discretionary accruals which increase the risk for earnings management. To make sure that no distinctions is made between income-increasing and income-decreasing earnings management I use the absolute value of 𝐷𝐴_𝑖𝑡, where a higher value means more earnings management hence lower earnings quality.

(15)

12 Real Earnings Management

Roychowdhury (2006) introduces the measurement of real earnings management. He identified three possible activities through which the management could use their discretion to act opportunistically.

First creating abnormal cash flow levels from operations by providing discounts or changing

consumer credit terms; second decrease (increase) discretionary expenses (e.g., R&D and marketing) to inflate (deflate) short-term profit; and third overproduction of goods to spread out fixed cost over more units hence decrease cost of goods sold (COGS). By acknowledging an abnormal level of activity in one of these areas, real earnings management can be identified.

I use abnormal cash flow as a proxy for real earnings management activities. I adopt the approach by Roychowdhury (2006) to estimate normal cash flow from operation (𝐶𝐹𝑂_𝑖𝑡).

^𝐶𝐹𝑂^𝑖𝑡

𝐴𝑇_𝑖𝑡−1= 𝛾₁ ¹

𝐴𝑇_𝑖𝑡−1+ 𝛾₂^{𝑆𝐴𝐿𝐸}^𝑖𝑡

𝐴𝑇_𝑖𝑡−1+ 𝛾₃^{∆𝑆𝐴𝐿𝐸}^𝑖𝑡

𝐴𝑇_𝑖𝑡−1 + 𝜀_𝑖𝑡 (5) Where 𝐶𝐹𝑂_𝑖𝑡 is cash flow from operations from the statement of cash flow in year 𝑡; 𝑆𝐴𝐿𝐸_𝑖𝑡 is net sales from income statement in year 𝑡; and ∆𝑆𝐴𝐿𝐸_𝑖𝑡 is the change in net sales from year 𝑡 − 1 to year 𝑡 observed from the income statement. The residual (𝜀_𝑖𝑡) of equation (5) is the abnormal cash flow from operations (𝐴𝑏𝐶𝐹𝑂_𝑖𝑡). By reducing discretionary expenses (e.g., R&D and advertising expenses), managers can inflate current earnings. Further, a reduction of discretionary expenses reduces cash outflow from the operation, this results in abnormally higher 𝐶𝐹𝑂_𝑖𝑡. Alternatively, managers can increase 𝑆𝐴𝐿𝐸_𝑖𝑡 by providing discounts or changing consumer credit terms. Such actions would enhance 𝑆𝐴𝐿𝐸_𝑖𝑡 temporarily but would lead to abnormally low 𝐶𝐹𝑂_𝑖𝑡. To control for both cash flow increasing and cash flow decreasing earnings management I use the absolute value of 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡 as a proxy for real earning management.

Overview of Earnings Management proxies Table 2: Descriptive Statistics

Variable Mean Std.Dev. Min Max

𝐷𝐴_𝑖𝑡 ^0.155 ^0.177 ^0.000 ^1.393

𝐴𝑏𝐶𝐹𝑂_𝑖𝑡 ^0.175 ^0.108 ^0.000 ^2.441

𝑇𝐴_𝑖𝑡 ^-0.058 ^0.082 ^-2.362 ^0.657

𝐶𝐹𝑂_𝑖𝑡 ^0.108 ^0.097 ^-1.153 ^1.323

𝑆𝐴𝐿𝐸_𝑖𝑡 ^1.164 ^0.913 ^0.053 ^15.809

𝑃𝑃𝐸_𝑖𝑡 ^0.603 ^0.403 ^0.021 ^2.800

𝐼𝐵_𝑖𝑡 ^0.049 ^0.113 ^-2.639 ^0.523

All variables are scaled by lagged total assets (𝐴𝑇_𝑖𝑡−1)

Table 2 summarize 𝐷𝐴_𝑖𝑡 (a proxy for accrual-based earnings management) and 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡 (a proxy for real earnings management); and, the variables used to estimate the two proxies for earnings

management. All variables in table 2 are deflated by lagged total assets (𝐴𝑇_𝑖𝑡−1) to make them comparable. The mean 𝐷𝐴𝑖𝑡 is 0,155; and the mean of 𝐴𝑏𝐶𝐹𝑂𝑖𝑡 is 0,175, suggesting that the existence

(16)

13

weighted by the impact of both accrual-based earnings management and real earnings management is almost equally. However, the mean of 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡 is inflated by the higher maximum value. The maximum value of 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡 (2,441) is generated by company eMagin in the year 2007. A quick look at the firm’s financial statement of 2007 suggest that high sales growth (∆𝑆𝐴𝐿𝐸_𝑖𝑡) of 115% and a reduction of cash flow from operation (𝐶𝐹𝑂_𝑖𝑡) of -82% is the drivers of the high value of 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡 identified (eMagin, 2007). To further investigate the impact of single observations I count the numbers of observations with a 𝐷𝐴_𝑖𝑡 above 1, and then do the same with 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡. In my sample of 7,461 observations, 47 observations have 𝐷𝐴_𝑖𝑡 value of above 1; and only 5 observations have 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡 value of above 1. This result is consistent with the higher standard deviation observed in 𝐷𝐴_𝑖𝑡 (0,177) compared to 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡 (0,108). The higher standard deviation of accrual-based earnings management compared to real earnings management suggest that accrual-based earnings management might be more affected by differences in circumstances.

Another exciting value from table 2 is the maximum value of 𝑃𝑃𝐸_𝑖𝑡. All variables are scaled by 𝐴𝑇_𝑖𝑡−1, this imply that if 𝑃𝑃𝐸_𝑖𝑡 is above 1, the current level of 𝑃𝑃𝐸_𝑖𝑡 is higher than last year’s total assets (𝐴𝑇_𝑖𝑡−1). This is only possible if the firm have acquired very high amounts of 𝑃𝑃𝐸_𝑖𝑡 assets during the year. To validate the observation, I use the corporate financial statement for that specific year. An observation of company Goldcorp in the year 2005 suggest a value of 𝑃𝑃𝐸_𝑖𝑡 scaled by 𝐴𝑇_𝑖𝑡−1 at 2,800. To control for this high value, I turn to the financial statement of Goldcorp in the year 2005. Some new mining interest contracts mainly in South America justify the observed value of 2,800 (Goldcorp, 2005).

Both proxies for earnings management are suffering the risk of measurement error. To reduce the risk of type Ⅰ and Ⅱ errors, I use a homogenous sample to mitigate the influence of other factors apart from managerial judgment.

Firm-level Political Risk

By adopting a pattern-based sequence-classification method developed in computational linguistics, Hassan et al. (2017) distinguish between political and non-political language by the participants in the conference call. To be able to distinguish between political and non-political language Hassan et al.

(2017) used an undergraduate political science textbook and transcripts of political speeches as a base to identify two-word combinations (“bigrams”) that are frequently used in political language. Then by using their pattern-based sequence-classification method, they were able to correlate language

patterns used in the conference call with their political language base. How frequently these political

“bigrams” could be found in the conference-call transcript in conjunction with synonyms for “risk” or

“uncertainty” and relation to the length of the call conducts the measurement.

(17)

14

𝑃𝑅𝑖𝑠𝑘_𝑖𝑡 = ^∑ (1[𝑏∈ℙ\ℕ]×1[|𝑏−𝑟|<10]×^𝑓𝑏,ℙ

𝐵ℙ 𝐵𝑖𝑡

𝑏

𝐵_𝑖𝑡 (6) The first term [𝑏 ∈ ℙ\ℕ] count the number of bigrams in the conference call transcript that are

political related in connection with the political language base. The second term [|𝑏 − 𝑟| < 10]

acknowledge how many words it is between the bigrams and a synonym for “risk” or “uncertainty”

(maximum within 10 words). 𝑓_𝑏,ℙ is the frequents of a specific bigram in the political language base, and 𝐵_ℙ is the total number of bigrams in the political language base. Hence, the last term ^𝑓^𝑏,ℙ

𝐵_ℙ weight each bigrams with a score that reflect how significant that specific bigram is to the political language base (Hassan et al., 2017). 𝑃𝑅𝑖𝑠𝑘_𝑖𝑡 is then the measurement for firm-level political risk for each quarter.

It is problematic to quantify political risk for the simple reason that the term does not have a precise definition. The lack of consensus around the definition of political risk will introduce subjectivity in the measurement of the term. Hassan et al. (2017) aspire to remove much of the subjectivity by adopting a pattern-based sequence-classification method that is run by python, hence mitigate human interaction with the measurement. However, the source of the political language base and the choice of synonyms for “risk” or “uncertainty” are still subjects of subjectivity.

I define firm-level political risk (𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘_𝑖𝑡) as the natural logarithm of the average of the 𝑃𝑅𝑖𝑠𝑘_𝑖𝑡 over four quarters, this to mitigate skewness in the variable distribution (El Ghoul et al., 2018).

Aggregated measurement for political risk (EPU)

The economic uncertainty index is developed by Baker et al. (2016) by using a computer-automated search of newspapers. The first component of the index is a search result from ten large US

newspapers; “USA Today”, “Miami Herald”, “Chicago Tribune”, “Washington Post”, “Los Angeles Times”, “Boston Globe”, “San Francisco Chronicle”, “Dallas Morning News”, “Houston Chronicle”, and “WSJ”. To construct the index, Baker et al. (2016) search monthly for articles in above mentioned newspapers related to economic and policy uncertainty. An article will only influence the index if it includes following triple: “uncertainty” or “uncertain”; “economic” or “economy”; and one of these policy terms: “Congress”, “deficit”, “Federal Reserve”, “legislation”, “regulation” or “white house”.

The raw data of the index is simply the monthly number of articles contain the triplet by the search result from the ten large US newspapers. To control for the volume of articles over time and across newspapers, Baker et al. (2016) scale the raw counted articles by a total number of articles in each newspaper monthly. Then the scaled result for each newspaper is standardized to a unit standard deviation between 1985 and 2009, and then take the average across the newspapers by month. Finally, Baker et al. (2016) sum the standardized values of the ten newspapers and normalize the 10-paper series to a mean of 100 from 1985 to 2009.

(18)

15

I obtain an annual index score of the US EPU by calculating the twelve-month average score. I take the natural logarithm of the annualized EPU score to generate my aggregated measurement for political risk (𝐸𝑃𝑈_𝑡).

Control Variables

To assure the validity of the relationship between political risk (𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘𝑖𝑡 and 𝐸𝑃𝑈𝑡) and my earnings management proxies (𝐷𝐴𝑖𝑡 and 𝐴𝑏𝐶𝐹𝑂𝑖𝑡) I isolate the effects of political risk by including a comprehensive set of variables previous literature shown have significant effects on earnings quality.

Studies by Gulen and Ion (2016); Julio and Yook (2012) among others, identify that firm-level decision are influenced by aggregated economic conditions. Further, Jenkins et al. (2009) even identify a significant connection between accounting quality and the business cycle. Thus, I include real annual GDP growth (𝐺𝐷𝑃_𝐺𝑟_𝑡) to control for the influence of the general macroeconomic condition. Next, I follow the suggestions by Dechow and Dichev (2002), to control for firms size (𝑆𝐼𝑍𝐸_𝑖𝑡) calculated as the natural logarithm of total assets in US dollars reported on the balance sheet;

and the operating cycle (𝑂𝑃𝐶_𝑖𝑡) calculated as the natural logarithm of the sum of days receivables and inventory stay on the balance sheet before realization. Hribar and Nichols (2007) acknowledge that operating volatility enhance the risk of rejecting the existence of earnings management activities.

Thus, I control for operating volatility by including volatility variables also recommended by Dechow and Dichev (2002). Accordantly I control for cash flow volatility (𝑆𝑡𝑑_𝐶𝐹𝑂_𝑖), sales volatility

(𝑆𝑡𝑑_𝑆𝐴𝐿𝐸_𝑖), earnings volatility (𝑆𝑡𝑑_𝑒𝑎𝑟𝑛_𝑖) and working capital volatility (𝑆𝑡𝑑_𝑊𝐶_𝑖). The operating volatility variables are all scaled by total assets and calculated as the standard deviation of the specific variable during ten to thirteen years depending on how many yearly observations each firm have.

Furthermore, Sweeney (1994) identify that debt covenants incentives earnings management activities, because breaching them imply high costs and loss of control for the management. Hence, a variable that controls the capital structure is needed. I use leverage (𝐿𝐸𝑉_𝑖𝑡), calculated as the ratio between long-term debt obligations and total assets. Consistent with Kothari et al. (2005) I also control for operating performance by including return on assets (𝑅𝑂𝐴_𝑖𝑡) defined as the ratio of net profit and total assets. Finally, I control for weather a firm report a loss in net income (𝐿𝑂𝑆𝑆_𝑖). By again following Dechow and Dichev (2002) I calculated 𝐿𝑂𝑆𝑆_𝑖 as the ratio between the number of years with an observed net loss and the total number of observed years for each firm. Table 3 provide the full set of control variables and their descriptive statistics.

All variables ( i.e. earnings management proxies; measurments for political risk; and control varibles) are summarised and briefly described in the first section of the appendix.

(19)

16 Table 3: Descriptive Statistics of Control Variables

Variable Mean Std.Dev. Min Max

𝐺𝐷𝑃_𝐺𝑟_𝑡 ^-0.003 ^0.017 ^-0.041 ^0.027

𝑆𝐼𝑍𝐸_𝑖𝑡 ^7.847 ^1.731 ^1.894 ^13.590

𝑂𝑃𝐶_𝑖𝑡 ^9.014 ^1.713 ^4.751 ^15.421

𝑆𝑡𝑑_𝐶𝐹𝑂_𝑖 ^0.049 ^0.036 ^0.005 ^0.592

𝑆𝑡𝑑_𝑆𝐴𝐿𝐸_𝑖 ^0.187 ^0.174 ^0.014 ^2.050

𝑆𝑡𝑑_𝑒𝑎𝑟𝑛_𝑖 ^0.044 ^0.038 ^0.004 ^0.607

𝑆𝑡𝑑_𝑊𝐶_𝑖 ^0.034 ^0.024 ^0.004 ^0.238

𝐿𝐸𝑉_𝑖𝑡 ^0.194 ^0.169 ^0.000 ^1.395

𝑅𝑂𝐴_𝑖𝑡 ^0.045 ^0.111 ^-2.708 ^0.758

𝐿𝑂𝑆𝑆𝑖 0.022 0.121 0.000 1.000

Omitted Variable Bias

The omission of variables that positively correlates with the dependent variable can induce bias in my predictive model. In this case, omitted variables that are positively correlated with my proxies for earnings management (𝐷𝐴_𝑖𝑡 and 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡) introduce the risk of incorrectly attribute the effects to the independent variable (𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘_𝑖𝑡 and 𝐸𝑃𝑈_𝑡). A common way to deal with the problem is to introduce dummy variables for space or time units, thus control for potential unobserved heterogeneity (Allison 2009). Following El Ghoul et al. (2018) I include fixed effects of time represented as a year dummy variable in main analysis. Conventional literature (Bertrand and Mullainathan, 2003; Angrist and Pischke, 2009; Khan et al., 2016) also suggests controlling for firm or industry fixed effects, however, I decide that the loss in predictability that a dummy variable for firm or industry would imply is not worth the small gain in accuracy. Hence, I exclude to control for firm or industry fixed effects in my main analysis.

(20)

17

IV. Empirical Analysis & Results Main Analysis

To test my hypotheses, I estimate the following model:

𝑃𝑟𝑜𝑥𝑦𝐸𝑀𝑖𝑡 = 𝛽0+ 𝛽1𝑃𝑜𝑙𝑅𝑖𝑠𝑘𝑖𝑡+ 𝛽2𝑋𝑖𝑡+ 𝛼𝑡+ 𝜀𝑖𝑡 (7) Where 𝑃𝑟𝑜𝑥𝑦𝐸𝑀_𝑖𝑡 is one of the two proxies for earnings management (i.e. 𝐷𝐴_𝑖𝑡 or 𝐴𝑏𝐶𝐹𝑂_𝑖𝑡) and 𝑃𝑜𝑙𝑅𝑖𝑠𝑘_𝑖𝑡 is one of the two measurments for political risk (i.e. 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘_𝑖𝑡 or 𝐸𝑃𝑈_𝑡). 𝑋_𝑖𝑡 is a vector that comprise all control varibles that are previously defined. Further, 𝛼_𝑡 is included to control for year fixed effects.

Earnings Management & Political Risk

Table 4 report the result of the regression between both earnings management proxies and 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘. The dependent variable in regression (1) and (2) is 𝐷𝐴, the proxy for accrual-based earnings management. And the dependent variable in regression (3) and (4) is 𝐴𝑏𝐶𝐹𝑂 which is the proxy for real earnings management activities. Regression (1) and (3) only includes 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 as independent variable and regression (2) and (4) additionally includes control variables (i.e., vector 𝑋_𝑖𝑡). The main coefficient of interest is 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘. A negative (positive) coefficient indicates a negative (positive) relationship between 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 and earnings management.

The general conclusion of table 4 is that a negative coefficient is identified for all four regressions and that the negative relation between 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 and all forms of earnings management is statistically signiﬁcant. For instance, a negative coefficient of 0.009 identified in regression (2) is equivalent to firms decreasing accrual-based earnings management with 5,8% (0.009/0,155) of the sample mean.

To assure the independence of the independent variables in regression (2) and (3), I test for

multicollinearity. 𝑆𝑡𝑑_𝐶𝐹𝑂 and 𝑆𝑡𝑑_𝑒𝑎𝑟𝑛 reports VIF values above 5. A correlation analysis between the two variables indicate a correlation of 91%, suggesting that at least one of the variables is

superfluous. The main variable of interest 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 present a VIF value of 1,048, this is of importance since it indicates that the variance of the dependent variable captured by 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 is not influenced by other variables included in the model (see appendix 1). Further, I use the Breusch- Pagan / Cook-Weisberg test for heteroskedasticity on regression (2) and (4) in table 4. The result indicates the existence of heteroskedasticity in both regressions and hence complicates the validity of the results (see appendix 2 and 3). To mitigate the impact of heteroskedasticity, I run the four

regressions presented in table 4 one more time. However, this time I use robust standard errors in the regressions. The negative association between 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 and both proxies of earnings management holds (see appendix 4). Hence, one can conclude that both accrual-based earnings management and real earnings management is related to firm-level political risk.

(21)

18 Table 4: Regression Results

This table reports the regression results of earnings management proxies in relation to political risk. The dependent variable in regression (1) and (2) is the proxy for accrual-based earnings management, 𝐷𝐴^,defined by following the modified jones model developed by Dechow et al., (1995). The dependent variable in regression (3) and (4) is the proxy for real earnings management, 𝐴𝑏𝐶𝐹𝑂, defined by following Roychowdhury (2006) approach to identify abnormal cash flow from operation.

𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 is the natural logarithm of the average firm-level political risk score developed by Hassan et al. (2017) over the four-quarter period ending in the month of the fiscal year-end. All regression (1), (2), (3), and (4) are controlled for year fixed effects. Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1

𝐷𝐴 𝐴𝑏𝐶𝐹𝑂

(1) (2) (3) (4)

𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 -0.015*** -0.009*** -0.005*** -0.006***

(0.002) (0.002) (0.001) (0.001)

𝐺𝐷𝑃_𝐺𝑟 -0.542 -1.848

(2.719) (1.663)

𝑆𝐼𝑍𝐸 -0.012*** -0.001

(0.001) (0.001)

𝑂𝑃𝐶 -0.011*** 0.006***

(0.001) (0.001)

𝑆𝑡𝑑_𝐶𝐹𝑂 1.574*** 0.724***

(0.160) (0.098)

𝑆𝑡𝑑_𝑆𝐴𝐿𝐸 -0.082*** 0.033***

(0.013) (0.008)

𝑆𝑡𝑑_𝑒𝑎𝑟𝑛 -0.788*** 0.267***

(0.138) (0.085)

𝑆𝑡𝑑_𝑊𝐶 0.558*** -0.708***

(0.123) (0.075)

𝐿𝐸𝑉 0.034*** -0.042***

(0.012) (0.008)

𝑅𝑂𝐴 0.182*** -0.024*

(0.021) (0.013)

𝐿𝑂𝑆𝑆 -0.012 -0.099***

(0.020) (0.012)

Constant 0.227*** 0.346*** 0.223*** 0.182***

(0.073) (0.076) (0.044) (0.046)

Observations 7461 7461 7461 7461

Adj R-squared 0.61% 10.50% 1% 10.10%

Year Dummy YES YES YES YES

The result is consistent with previous literature covering how political risk effects managerial judgment. Gulen and Ion (2016); Julio and Yook (2012); and Bonaime et al. (2018) suggest that political risk is relevant for firm-level decision making, they also find evidence of enhanced prudence in managerial judgment related to the level of political risk exposure. The significant negative

coefficient of 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 reported in table 4 indicate the same prudent behaviour among managers exposed to political risk in my sample. The interpretation of the result is that increased exposure to firm-level political risk reduce the amount of earnings management, both accrual-based and real earnings management activities. A possible explanation for the negative relation between firm-level political risk and earnings management is suggested by Mitton (2002). He points out that risk and uncertainty in general make investors more prudent, thus the demand for information and

(22)

19

transparency increase. The enhanced scrutiny that follows from the increased demand for information among investors reduce the opportunity form managers to conceal earnings management activities.

Further, McInnis and Collins (2011) conclude that the increased scrutiny enlarges the cost of earnings management. The difficulty to conceal earnings management suggested by Mitton (2002), and the increased cost acknowledged by McInnis and Collins (2011) capture both the two conditions of earnings management stated by Cheng and Warfield (2005). The result from my study do not provide any evidence supporting the perspective of Schipper (1989), but the possible existance of the effects cannot be neglected. However, I can conclude that it is not a dominant factor in my study.

The result is also consistent with the findings of El Ghoul et al. (2018). However, my firm-level measurement of political risk reports a significantly lower coefficient of absolute value to compare to the EPU measurement used by El Ghoul et al. (2018). Thus, suggesting that my measurement of political risk capture less of the variation in earnings management. A reasonable explanation of this divergence is the significant differences of characteristics in the data samples.

The definition of earnings quality by Dechow et al. (2010) suggest that any action that reduces the usefulness of the information reduce the quality of earnings; thus, earnings management activities depressed earnings quality. A consequence of the result that earnings management is negatively related to firm-level political risk is that earnings quality is adversely affected. A logical conclusion is that earnings quality is positively associated with firm-level political risk. The interpretation is that firm-level political risk reduces earnings management activities; hence, financial information better represents true economic performance. Thus, make the information a more useful input in decision making processes among stakeholder.

Firm-level Political Risk vs. EPU Index

Next step is to include 𝐸𝑃𝑈 as an aggregated measurement of political risk in the regression. Table 5 reports regression results relating both earnings management proxies to both 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 and 𝐸𝑃𝑈.

Regression (1) and (3) are the same regressions reported in table 4 as regression (2) and (4), where control variables are included. Regression (2) and (4) in table 5 examines both earnings management proxies to 𝐸𝑃𝑈. A positive coefficient of 𝐸𝑃𝑈 can be identified in both regression (2) and (4), which imply that a higher level of aggregated political risk is associated with increased earnings

management. Hence, inconsistent with the negative relation found between 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 and earnings management, and the findings of El Ghoul et al. (2018). However, the relation between 𝐸𝑃𝑈 and earnings management is not statistically signiﬁcant in either regression (2) or (4). Thus, justifying the inference that 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 better predicts earnings management variation than 𝐸𝑃𝑈 in this context.

The result is surprising based on the findings of Hassan et al. (2017), they identify a significant correlation of 80% between the two measurements of political risk. Hence one would expect that

(23)

20

𝐸𝑃𝑈 should capture more of the variation identified by 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘. I conduct a correlation analysis between my two measurements of political risk to identify possible divergence from the findings of Hassan et al. (2017). In my sample, the two series of data only present a correlation coefficient of 0,142 (14,2%) (see appendix 5). The significant reduction of correlation between 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 and 𝐸𝑃𝑈 is best explained by the sample size. The original dataset by Hassan et al. (2017) consists of 9,478 firms, after the sample constructing process, my final data set include 625 firms. Since the 𝐸𝑃𝑈 index is an aggregeded measurment for political risk, it is more likely to correlate with a

comperhensive data set of 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘, i.e., the original data set. A smaller sample is likely to be more sensible to variation of individual firms, hence a bad aggregeded measurement.

𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 evidently capture more of the variation in earnings management than 𝐸𝑃𝑈 according to table 5. Thus, support the findings of Hassan et al. (2017), that an aggregated measurement of

political risk mask a lot of the variation that plays out on a firm-level. El Ghoul et al. (2018) suggest a significant association between 𝐸𝑃𝑈 and earnings management, however, my result do not provide any support for their findings. Instead, the lack of a statistically significant coefficient indicate that managerial judgment is not affected by aggregated political risk. Thus, inconsistent with previous literature that acknowledge more prudent behaviour among managers as a result of increased exposure to aggregated political risk. Gulen and Ion (2016); Julio and Yook (2012); and Bonaime et al. (2018) investigate how aggregated political risk affect different dimensions of managerial judgment, i.e., not related to earnings management. One can imagine that the effects of political risk is not homogenic along with all aspects of the manageial judgment. I compare the coefficient (-0,168) from Gulen and Ion (2016) (political risk related to investments) to the coefficient (-0,047) obtained by El Ghoul et al.

(2018) (political risk related to earnings management). The comparison indicate that political risk explains more of the variation of investment decisions than earnings management, hence supports the premise that political risk affect the different dimensions of managerial judgment differently.

Important to acknowledge is that no inferences should be based on the comparison alone, since the studies are constructed by different data sets and methodologies. Hence, I only use it as an indication and a possible explanation for my result. However, the results inconsistency with the findings of El Ghoul et al. (2018) cannot be related to the same assumption, instead, the explanation is more likely related to differences in sample characteristics. A small sample is more dependent on individual observations. Since an aggreged measurement of political risk fail to identify part of the variation of each individual observation, an aggregated measurement is less usable in the analysis of small samples. Thus, the size of my sample in relation to the sample of El Ghoul et al. (2018) provides an explanation for why I cannot identify a significant coefficient of 𝐸𝑃𝑈. Therefore, one cannot conclude that 𝐹𝑖𝑟𝑚𝑃𝑅𝑖𝑠𝑘 are a better predictor of earnings management than 𝐸𝑃𝑈 in all circumstances. It is possible that a firm-level measurement only outperforms an aggregated measurement when the sample size is relatively small.

Political Risk & Earnings Quality: An analysis of political effects on earnings management