ownership affect the
relationship between fair value accounting and
earnings quality?
Master’s Thesis 30 credits
Programme: Master’s Programme in Accounting and Financial Management Specialisation: Management and
Control
Department of Business Studies Uppsala University
Spring Semester of 2021
Date of Submission: 2021-06-02
Emma Ericsson Göran von Essen
Supervisor: Joachim Landström
This thesis examines the effect government ownership has on the relationship between fair value accounting and earnings quality when “mark-to-model” techniques are used in fair value estimations. To this end we compare 36 real estate companies controlled by private interests with 32 real estate companies controlled by the state and municipal governments in Sweden. We find that the relationship between fair value accounting exposure and aggregated earnings quality is negatively affected by government ownership when unobservable inputs are used in the fair value estimation. Previous research treats government ownership simply as having a first order relationship with earnings quality. Our results indicate determinants of earnings quality can affect one another, and that these effects should be considered.
Keywords: earnings quality, fair value, investment properties, government ownership, corporate governance, mark-to-model valuation, unobservable inputs, thin markets, smoothing, predictive ability, conservatism, earnings management,
1. Introduction 2
2. Literature review and hypothesis development 4
2.1 Accounting regulation for investment properties in Sweden: IAS 40 and K3 4
2.2 Earnings quality: properties and determinants 6
2.2.1 Properties of Earnings Quality and a Composite Measure 7
2.2.2 Determinants of Earnings Quality 9
Financial reporting practices: Fair value accounting 10
Governance and controls: Ownership structures and government ownership 11 2.3 Corporate Governance: How ownership can affect the relationship between fair value accounting and earnings quality 13
2.4 Model and hypothesis development 15
3. Research design 18
3.1 Sample selection 18
3.2 Model and variables 19
3.2.1 Aggregated Earnings Quality 21
Predictive ability 21
Volatility or Smoothing 22
Conditional conservatism 23
Aggregated Earnings Quality: A composite measure 23
3.2.2 Exposure to fair value accounting 24
3.2.3 Government ownership 25
3.2.4 Control variables 25
3.3 Statistical analysis 26
4. Empirical results 26
4.1 Descriptive statistics and correlations 26
4.2 Results of regressions 29
4.2.1 The individual components of aggregated earnings quality (AEQ) 29 4.2.2 The relationship between Fair value exposure and Aggregated Earnings Quality 31
4.2.3 The moderating effect of government ownership 32
5. Analysis and discussion 36
5.1 Limitations 39
6. Summary and conclusions 41
References 43
Appendix 1:Outputs 46
Appendix 1.1: Residuals plot 54
1. Introduction
The purpose of this thesis is to examine the effect government ownership has on the relationship between fair value accounting and earnings quality when “mark-to-model” techniques are used in fair value estimations. To this end we compare real estate companies controlled by private interests with real estate companies controlled by the state and municipal governments in Sweden.
When fair value accounting was introduced under IFRS 13 (2018) the stated purpose was to increase the decision usefulness of financial reporting (Hitz, 2007). Decision usefulness is also what earnings quality is supposed to encapsulate; high-quality financial reports should improve decision making and thereby capital market efficiency (Šodan, 2015). As fair value accounting has become more dominant, interest in measuring the quality of financial reporting, the earnings quality, has naturally increased (Perotti and Wagenhofer, 2014). The reliability of fair values and their effect on earnings quality has been much debated, particularly after the 2008 financial crisis (Laux and Leuz, 2009). However, there are still areas where research is lacking. One such example is the effect of fair value accounting on earnings quality in so-called “thin markets” (Lind and Nordlund, 2019) where fair values are by necessity estimated using “mark-to-model”-techniques, rather than “mark-to-market”.
This study contributes to existing research in several ways. First, it extends knowledge of the effect of fair value accounting on several properties of earnings quality. Previous research has mainly focused on the relationship between changes in fair values and the future operating performance or the value relevance of fair value estimates (Šodan, 2015; Yao et al., 2018). This study on the other hand takes a broader approach, incorporating more than one aspect of earnings quality into an aggregated multi-dimensional measure.
It also contributes to knowledge about fair value accounting and earnings quality in thin markets by looking at a sample of Swedish real estate companies. The real estate market for investment properties is a perfect example of such a “thin market” where the actors are few and objects are unique (Lorentzon, 2011). The market for investment properties accounts for about 40 percent of Sweden's GDP, a much larger portion than in other European countries (Finansinspektionen,
2019). This makes understanding what affects the earnings quality is not only of theoretical significance but also of high practical importance. Despite the impact this sector has on the economy the previous studies which exist on “mark-to-model” valuation are more or less entirely focused on banks (Bratten et al., 2016; Šodan, 2015).
Third, it contributes to knowledge of how government ownership relates to earnings quality. In existing research, determinants affecting the quality of earnings are usually studied in isolation and their mutual effect is usually not considered (Dechow et al., 2010; Liceran-Gutierrez and Cano-Rodriguez, 2019). Government or state ownership is studied in relation to earnings quality as a determinant in several studies (see for example Capalbo et al., 2014; Chen et al., 2010; Gaio and Pinto, 2018; Wang and Yung, 2011). Research indicates that weak corporate governance negatively affects the implementation and quality of fair value accounting. companies with weaker corporate governance mechanisms show greater information asymmetry (Song et al., 2010).
Furthermore, state or government ownership leads to weaker and less efficient corporate governance (Al-Janadi et al., 2016). If corporate governance and ownership structures influence the quality of the financial reporting practice, then government ownership might not have a first- order relationship with earnings quality. Government ownership could instead have a moderating function on the relationship between the financial reporting practice and earnings quality. In comparing government controlled real estate companies with those controlled by private interests, we can study the possible effect of government ownership on the relationship between earnings quality and fair value accounting estimates. Sweden has one of the largest proportions of state- owned enterprises as percentage of GDP in the EU (European Commission and Directorate- General for Economic and Financial Affairs, 2016) of which real estate companies make up a considerable portion. In 2009 the Swedish state owned, either in corporate form or via direct ownership, more than 33 000 properties (Sveriges regering, 2009). Today the taxation value of the properties owned by Regeringskansliets corporate group (including for instance Jernhusen and Akademiska Hus) is approximately 110 billion SEK according to the real estate registry Datscha.
The performance of state-controlled companies may have a significant impact on government budgets, as they are necessary for a working fiscal consolidation which makes the quality of their earnings of great public interest (Gaio, 2010). In conclusion, we believe this study is of interest
The rest of this thesis is constructed as follows: Section 2 contains a literature review, summarizing previous research and establishing the theoretical framework for the study. Section 3 describes the sampling and variables. Section 4 contains the results of the study. Section 5 analyses and discusses the results related to previous research as well as the limitations of the study. Finally, section 6 consists of the conclusions and a brief discussion on possible avenues for future research.
2. Literature review and hypothesis development
This section contains a description of the relevant fair value accounting regulations, a review of previous research on the determinants and properties of earnings quality and finally the hypothesis development and framework of the study.
2.1 Accounting regulation for investment properties in Sweden: IAS 40 and K3
All publicly traded companies in Sweden adhere to IFRS 13, (2018) accounting standard established by the International Accounting Standards Board (IASB). The standard defines fair value as the price that would be received when selling an asset in an orderly transaction at the measurement date under current market conditions regardless of whether that price is directly observable or estimated using another valuation technique (§ 24). IFRS 13 makes a clear distinction between entry price (the price paid to acquire an asset) and exit price (the price that would be received when selling the asset) (§ 57). There is also a distinction between fair value and value in use which makes it clear the fair value is not supposed to include specific competitive advantages or private information (Hitz, 2007). IFRS 13 details a three-tiered hierarchy for estimating fair value. The hierarchy is based on the idea that market prices are the most accurate estimates for fair value, if available. Level 1 is based on observed market prices. Level 2 on market prices for comparable assets. Level 3 on inputs, called unobservable inputs, which are used when mark-to-market information is unavailable. These estimates are usually determined through company-specific models estimating the value of future cash flows; this practice is called mark- to-model (Hitz, 2007; Lorentzon, 2011). Most of the discussion concerning the reliability of fair
value is centered around Level 3 inputs since they leave the management a large degree of discretion and are inherently the least reliable fair values (Šodan, 2015). These are also the fair value estimates we focus on in this study.
Many studies on the relationship between fair value accounting and earnings quality focus on the banking sector because their balance sheets consist mostly of financial instruments reported at fair value (Šodan, 2015). The same might be said for the real estate sector where companies’ balance sheets consist mostly of investment properties reported at fair value.
The fair value models under IFRS 13 apply to the valuation of investment properties, as detailed in IAS 40. It requires companies to either recognize the annual fair value of investment properties on the balance sheet or disclose them in the footnotes of the annual report. All changes in fair value should be recognized in profit or loss, and all changes in the valuation of the property are changes in the estimated fair value of the asset with no recorded depreciation. An investment property is defined under IAS 40 as land or a building/part of a building (or both) held by the owner or through a financial lease for the purpose of earning rentals or for capital appreciation (or both). An investment property is not held for (1) use in the production or supply of goods or services or for administrative purposes or (2) sale in the ordinary course of business (IAS 40, § 5). This definition applies to commercial real estate as well as housing. The fair value of investment properties is almost always based on Level 3 inputs. One European survey (PwC, 2014) indicates 97 percent of the reviewed companies identify inputs in the valuation of investment properties as Level 3.
Investment properties are rarely directly comparable to one another, so market prices are unavailable (Lind and Nordlund, 2019). The real estate market can therefore be described as a so- called “thin market” where there are few transactions and the heterogeneous actors’ reservation prices might differ significantly (Lind and Nordlund, 2019) In a thin market, the entry price paid after a bidding process can therefore not automatically be seen as an expression of fair value defined as an exit price (Lind & Norlund, 2019).
Most of the state-owned (fully or in part) companies are managed by Regeringskansliet (Swedish Government Offices), the accounting responsibility for these companies are placed with
companies managed by other parts of the government. The accounting standards applied are detailed in the government’s owners’ policy (Ekonomistyrningsverket, 2018) Accounting for the state is detailed in Budgetlagen (2011:203). For investment properties the principle follows those stated in International Public Sector Accounting Standards (IPSAS). According to IPSAS some assets and liabilities are valued at fair value, this also includes investment properties (IPSAS 16).
IPSAS is based on IFRS and fair value is supposed to be understood as an exit price (Ekonomistyrningsverket, 2018).
Private companies which are not publicly traded usually adhere to the K3 accounting standard by Bokföringsnämnden (BFN). This is basically a simplified version of the IFRS standard and is mandatory for companies who fulfill two out of three following points for more than two consecutive years: a) more than 50 employees b) a net worth over 40 million SEK or c) total annual revenue over 80 million SEK. Smaller companies use a further simplified standard (K2), only using historical cost accounting, and will not be included in this study. For investment properties K3 (ch 16) allows companies to choose whether to use fair value or the acquisition value in traditional historical cost accounting in the financial reports, but in the latter case the fair value must be disclosed in the notes. Once fair value has been adopted, investment properties can no longer be valued at historical cost. Municipal companies generally follow K3 accounting standards and should therefore disclose fair value if historical cost is used in financial reports. Companies using K3 have been obligated to disclose fair value in the notes since 2014 (Ekonomistyrningsverket, 2018)
In conclusion, fair value estimates of investment properties can in principle be considered comparable for any larger company on the Swedish real estate market for the purpose of this study, whether controlled by the government or by private interests, listed or unlisted, though modifications are of course required to directly compare several items between companies using K3 and IFRS.
2.2 Earnings quality: properties and determinants
Though earnings quality is considered as a key characteristic of financial reporting there is no
Dechow et al., (2010) define high earnings quality simply as earnings that provide better information about the actual financial performance of a firm. The earnings should convey a relevant picture by providing accurate, specific information for the decision maker, thereby enabling better decisions. This definition stems from what is referred to as the decision usefulness perspective and defines earnings quality from the standpoint of its usability in making decisions.
There is also a definition which stems from an economic perspective, based on Hicksian income (Kamarudin and Ismail, 2014). However, the decision usefulness perspective is emphasized in most major journals and publications on earnings quality. This is also the perspective we adopt going forward.
2.2.1 Properties of Earnings Quality and a Composite Measure
Earnings quality is specific to the decision being made and who is making it, which means it is impossible to determine a single best way to measure it for every possible situation (Dechow et al., 2010). Since it is not possible to observe earnings quality directly, several different proxies or properties have been developed (Dechow et al., 2010).
These proxies can be sorted into four general categories: earnings management (accruals quality, abnormal earnings), earnings smoothness, time-series properties (predictability, persistence) and conservatism (conditional/unconditional) (Liceran-Gutierrez and Cano-Rodriguez, 2019).
Comprehensive literature reviews by Liceran-Gutierrez and Cano-Rodriguez (2019) and Dechow et al. (2010) find several issues with the state of current research on earnings quality. Though there is a strong theoretical consensus on earnings quality as a multidimensional construct most studies (90%) use single property measures. By measuring a multidimensional concept through one dimension researchers risk biased estimates of the relationships between the concept and other variables, such as determinants or consequences. The different measures capture different aspects which can lead to contradictory conclusions when a single measure is supposed to represent the entire concept. On the other hand, the composite measures which are used in current research suffer from the subjective selection of the proxies included, their respective weights and lack of control over how the proxies correlate. In conclusion, a gap between the theoretical concept of earnings quality and the empirical research focused on measuring it can be clearly observed. Both reviews
mentioned above call for further research using composite measures where correlations between the proxies are accounted for and weights between the proxies are considered.
This study contributes to bridging this gap, by using a composite measure of earnings quality incorporating three properties: predictability, smoothness and conditional conservatism which will be discussed below. In line with the analysis by Liceran-Gutierrez and Cano-Rodriguez, (2019) we incorporate proxies from as many of the four categories as possible, avoiding more than one measure from the same category. Persistence and predictability are for instance highly correlated in their analysis, so only predictability is included. The common measures of earnings management correlate positively with earnings smoothness in several studies (for example Boterenbrood, 2014;
Yeo et al., 2002) which is consistent with the theory suggesting smoother earnings can be achieved through earnings manipulation (Leuz et al., 2003) This leads us to exclude earnings management in order to avoid different proxies which essentially measure the same thing.
The predictive power of earnings quality simply means higher quality earnings are more useful in predicting future earnings (Perotti and Wagenhofer, 2014). Predictability enhances the decision usefulness of earnings quality because sustainable earnings are expected to better indicate future cash flows (Dechow et al., 2010).
Smoothness is more complicated as there are two completely opposing views. The first sees smoothness as having a negative relationship with earnings quality (Leuz et al., 2003). From this point of view smoothness is considered a result of earnings management where managers are altering the results by judgment in reporting to achieve different objectives (Trueman & Titman, 1988). The aim of this manipulation is to deceive a stakeholder with respect to the underlying performance of the firm (Healy and Wahlen, 1999), the most common way to do this is by adjusting accruals (Kjaerland et al., 2020). There are three main reasons for why managers engage in earnings management: capital market motives, contract motives and regulation motives (Healy and Wahlen, 1999). Earnings management reduces the information value of reported earnings, making them less useful and thereby also reducing the earnings quality (Perotti and Wagenhofer, 2014). An alternative view is that smoothness should be positively associated with earnings quality. Smoother earnings are both more likely to be persistent and more predictive (Šodan, 2015).
If the objective of accounting is to determine earnings (operating cash flows plus accounting accruals) then the purpose of accruals could be said to smooth cash flows by filtering out volatility.
Some smoothing must be desirable from a decision-making perspective otherwise users would not be interested in earnings but just cash flows (Perotti and Wagenhofer, 2014). Management discretion is from this point of view likely a good thing. Smoothing simply adds information which only the management has access to about future cash flows into present earnings. Earlier studies show investors view smoothness as a desirable attribute of earnings (Rountree et al., 2008). We adopt this latter point of view in this study.
Finally, the principle of conservatism (Gaio and Pinto, 2018) enhances the quality of financial reporting by recognizing potential losses directly upon discovery, while gains are not recognized until fully realized or sufficiently certain (Šodan, 2015). Studies distinguish between conditional conservatism, which is the tendency to require higher verification when you recognize good news (gains) than when you recognize bad news (losses), and unconditional conservatism, which describes a policy that results in lower book values of assets (higher book values of liabilities) in the early periods of an asset or liability life. Whether unconditional conservatism increases or decreases the decision usefulness of earnings is a controversial issue (Dechow et al., 2010). In this case we concentrate on conditional conservatism where the expectation is that losses will be recorded in a timelier fashion than gains for better earnings quality (Liceran-Gutierrez and Cano- Rodriguez, 2019).
2.2.2 Determinants of Earnings Quality
Earnings quality fundamentally depends on two things: the firm’s financial performance and the accounting system which measures that performance (Dechow et al., 2010). Research in the field usually focuses on the accounting system and related aspects (Dechow et al., 2010). Accounting system research can in turn be divided into research on the determinants of earnings quality and research into the consequences of earnings quality.
The determinants of earnings quality most commonly discussed in literature are: firm characteristics, financial reporting practices, governance and controls, auditors, equity market
determinants are usually studied in isolation from one another and the potential interaction between variables is rarely considered (Liceran-Gutierrez and Cano-Rodriguez, 2019). Below are two of those determinants discussed further as they relate to this study: financial reporting practices, or specifically fair value accounting, and governance and controls, specifically government ownership.
Financial reporting practices: Fair value accounting
Previous research on fair value accounting and earnings quality mainly focuses on the value relevance of fair value estimates and the relationship between changes in fair values (unrealized gains and losses) and future operating performance (such as operating cash flows) while there is very little direct evidence on the earnings persistence of fair value estimates (Šodan, 2015; Yao et al., 2018). Hitz, (2007) states that theoretically, fair value changes should be transitory since market values should incorporate all available information. As a result, gains and losses could be correlated over time, despite the market’s efficiency. Yao et al. (2018), one of few studies in this area, conclude that fair value Level 1 estimates are positively associated with earnings persistence, while estimates made at Level 2 and Level 3 are not. On the other hand, Song et al. (2010) show that the reliability of Level 3 estimates increases with strong corporate governance when measuring value relevance.
Fair values are usually considered to hold higher predictive power concerning the future realization of cash flows and earnings compared with historical cost (Yao et al., 2018). Fair value estimates represent the present value of expected future cash flows, so if fair values are reliable measures of asset values, then changes in those values (unrealized gains and losses) should be reflected in the firm’s future performance. Several studies support this theory, such as Bratten et al. (2016) who show that reliable fair values of financial instruments in the banking industry increase the predictive power of future earnings. However, the opposite was true for fair values which were not reliable such as net unrealized gains and losses on derivative contracts classified as cash flow hedges. Ehalaiye et al. (2020) support these latter conclusions by showing increased predictive power only where level 1 or 2 inputs were used. There are also studies with more conflicting results, where changes in fair values stated in net income or other comprehensive income are in fact transitory and are not found to increase the predictive power (for example Chen et al., 2010).
The relationship between smoothness and fair value accounting is generally considered to be negative (Šodan, 2015). Whether this is indicative of better or worse earnings quality depends on the perspective adopted (see discussion above). Evidence is mixed, however. Yao et al. (2018) find fair value accounting offered smoother earnings, but see no evidence that factors commonly thought to encourage opportunistic earnings management behavior is associated with the persistence of earnings, even though smoother earnings are more persistent. Fair value accounting is considered to result in less conditional conservatism, since losses are recognized in the same timely fashion as gains, something empirically observed in several studies (Goncharov and Hodgson, 2011). However, loss recognition still differs between companies with the same standards. This suggests the existence of several other factors which are relevant besides the accounting system that seem to vary between studies and regions (Dechow et al., 2010).
Overall previous research on the relationship between fair value accounting and earnings quality offers somewhat conflicting evidence. Most studies focus on a single measure or proxy of earnings quality, and usually relate this to a single determinant. The studies are mostly from common law countries such as the UK or the US (such as Ball and Shivakumar, 2005), with some studies in transitional economies such as China (for example Chen et al., 2020) or Eastern Europe (Šodan, 2015). Research also shows that country specific institutional factors affect the reliability of fair value accounting which makes results hard to generalize between countries in international comparisons (Yao et al., 2018). One multidimensional study, Šodan (2015), does find that increased exposure to fair value accounting leads to a decrease in overall earnings quality. In that study persistence, predictability, smoothness, accrual quality, value relevance and conditional conservatism are aggregated into a single measure of earnings quality.
Governance and controls: Ownership structures and government ownership
Though there are numerous studies on the effect of ownership structures on earnings quality, these all treat ownership as having a first-order relationship with earnings quality. Though this study takes a different approach, we will still briefly look at the state of current research.
That ownership structure affects the quality of earnings has been established through several studies, for example Dempsey et al. (1993) who define three main types of ownership structures:
companies managed directly by the owner; companies that have professional managers but are controlled by a clearly defined outside party or parties; and companies run by professional managers but with dispersed ownership, for example a public company without majority owners.
Management in the last category are not only more prone to manipulate earnings but also have incentives to cover up bad management decisions from the owners (Dempsey et al. 1993).
Government ownership would fall into this last category.
Previous studies offer several reasons to expect that government ownership might affect earnings quality negatively, though actual empirical evidence is mixed. For instance, Shleifer and Vishny, (1997) argue that state ownership leads to poor corporate governance. Since state ownership does not provide a clear owner and the firm is managed by bureaucrats with no rights to the cash flows generated, politicians and the bureaucrats could potentially run the firm in ways which are favourable for their own personal purposes (Shleifer and Vishny, 1997). Laswad et al. (2005) see evidence of politicians influencing financial reporting in the public sector and Ben-Nasr et al.
(2015) suggest state-owned companies could have incentives to engage in earnings management to hide expropriation of a firm's resources for political purposes. Government-owned companies might also have less incentive to improve earnings quality as they do not need to compete for investors and capital the way private companies must and often have stable contracts. For example, Chaney et al. (2011) find that politically connected companies have lower quality accounting information than non-connected companies. Furthermore, Capalbo et al. (2020) show political pressures can drive earnings management by observing an increase in earnings management within municipality owned companies during election periods. This conclusion is supported by Al-Janadi et al. (2016) who find evidence for what they call “the grabbing hands-hypothesis” where bureaucrats in state-owned companies manipulate earnings to hide the expropriation of the firm’s resources for political purposes.
However, some of the pressures which motivate earnings management are less present in government owned companies. Dempsey et al. (1993) propose the worry for hostile takeover is a naturally occurring factor in externally owned and professionally managed companies which might
induce earnings management. Government controlled companies do not face the risk of hostile takeover to the same degree as other privately owned companies. This would suggest government owned companies are relieved of some of the pressures which drive earnings management.
Studies show government owned companies are less conservative when it comes to accounting, because its ownership banks and institutions are less concerned about the firm (Chen et al., 2010).
There might therefore be fewer reasons for government owned companies to manage earnings for debt contracts. Chaney et al. (2011) show that politically connected companies do not need to display the same level of quality in accounting as non-connected. These companies are not punished for the lower quality of reporting with higher cost of debt as is the case in companies without political connections. These companies do not need to respond to market pressures in the same way as normal companies. Gaio and Pinto, (2018) also argue that government owned companies do not need to be as conservative. Furthermore, these companies engage in earnings management to a lesser extent compared to other companies due to government protection.
Nonetheless, they also find that government-owned companies are not immune to capital market pressures. These companies are more likely to have higher levels of abnormal accruals and worse accruals quality which lowers their earnings quality. An additional Italian study investigates the degree to which state-owned companies engage in earnings management and finds that these companies significantly manage earnings but to a smaller extent than their private counterparts (Capalbo et al., 2014). Several other studies show that government ownership has a negative relationship with earnings management (Guo and Ma, 2015; Wang and Yung, 2011) leaving us with a very mixed bag of empirical results. These mixed results could be the result of different cultural or political factors, considering they are from different countries with different pre- existing conditions.
2.3 Corporate Governance: How ownership can affect the relationship between fair value accounting and earnings quality
Corporate governance can very simply be defined as the means by which the shareholders, the owners, of a firm align the managers’ interests with their own (Shleifer and Vishny, 1997). This
perspective is grounded in the principal-agent theory, or the issue of separation of ownership and control (Shleifer and Vishny, 1997).
Corporate governance differs widely between different parts of the world, depending on aspects such as the legal system and the political system (Shleifer and Vishny, 1997). Thomsen, (2016) examines the Nordic governance model by rating it on four different criteria: legal system, ownership structure, board structure and executive incentives. Overall, the Nordic model has elements of stakeholder governance, but with a concentrated ownership structure where shareholders are empowered to make decisions. There appears to be less formality and more decentralized decision making than in the US. At the same time, major blockholders are more sensitive to stakeholder concerns, for example because they have ties to unions or the government.
The Nordic countries all have strong corporate governance in general, according to Thomsen (2016)
Government ownership, on the other hand, has been found to result in significantly poorer monitoring of firm performance because the government is too detached from the firm (Al-Janadi et al., 2016). Firth et al. (2007) finds government owners pressure companies into implementing governmental objectives, which is done at the expense of other shareholder objectives (maximizing profit).
Al-Janadi et al. (2016) argue that an independent board of directors is a key factor in an effective monitoring system since their role is to minimize the agency problem between shareholders and management. Ordinarily independent directors have strong incentives to monitor management and provide quality information. However, concentrated government ownership weakens the independence of the directors as they interfere in director selection. Agency theory, for instance, also argues that independent directors are effective for monitoring management and overcoming agency problems (Shleifer and Vishny, 1997).
Previous research also indicates weak corporate governance negatively affects the implementation and quality of fair value accounting so companies with weaker corporate governance mechanisms show greater information asymmetry (Song et al., 2010). Information asymmetry theory argues
that majority shareholders, such as the government, will try to control companies by not sharing relevant information with minority shareholders (Shleifer and Vishny, 1997). This lack of information will not only affect minority shareholders but also other stakeholders or anyone who might want to use the information (Al-Janadi et al., 2016). Information asymmetry arises from different sources such as concentration of ownership, poor or irresponsible managerial performance and lack of regulations (Al-Janadi et al., 2016). Previous studies also show that the quality of the board of directors and internal audit quality are both important corporate governance mechanisms to counteract earnings management (Bajra and Cadez, 2018).
Though results from studies in other countries cannot of course be directly applied to Swedish conditions, there are certain observations that might be relevant, such as the importance of independent boards of directors, which has been observed in several different settings as seen above. In Sweden there are several documents which detail how government owned companies should be governed. Apart from general Swedish rules and legislation, state owned companies are run according to the principles laid out in Statens ägarpolicy (2020) while municipal owned companies are regulated through Principer för styrning kommun- och regionägda bolag (2020). In state owned companies that are not publicly traded, the directors are appointed by the Department of Finance through a recruitment process. In municipal owned companies the directors are appointed by the municipal board, and many are recruited from the municipal council members.
In both cases the principle of independent directors is noted as not being particularly relevant for companies owned completely by the government since there are no minority shareholders to protect, though it is noted the director in question should not have strong ties to the firm or to its management.
2.4 Model and hypothesis development
As previously stated, this study aims to fill a gap between theory and empirical research by using a multidimensional measure of earnings quality consisting of three properties: predictability, smoothness and conditional conservatism (see Figure 1 below).
fair value accounting regime is to increase decision usefulness (Hitz, 2007) and earnings quality measures the properties that determine decision usefulness in accounting information (Dechow et al., 2010). On the other hand, research indicates that fair value accounting negatively affects several properties of earnings quality, such as smoothness (Šodan, 2015) and conditional conservatism (Goncharov and Hodgson, 2011) which we use as proxies in our composite measure.
Level 3 (unobservable) inputs which are almost exclusively used in the real estate sector are negatively associated with persistence and predictability (Yao et al., 2018). This leads us to our first hypothesis:
H1: Fair value accounting exposure is negatively related to aggregated earnings quality when unobservable inputs are used in the fair value estimation
In current research determinants of earnings quality are usually studied in isolation from one another (Liceran-Gutierrez and Cano-Rodriguez, 2019). Interaction effects occur when the effect of one variable depends on the value of another variable, in this case the effect of one determinant on earnings quality might depend on the value of another determinant. In almost all existing studies all determinants are treated as having first-order relationships with earnings quality properties, but it might be fruitful to consider that some determinants could have a primarily moderating function instead.
In the case of government ownership, research indicates weak corporate governance negatively affects the implementation and quality of fair value accounting so companies with weaker corporate governance mechanisms show greater information asymmetry (Song et al., 2010). Since fair value estimates, at least Level 3 estimates, leave a lot of room for managerial discretion, it results in information asymmetry between managers and investors (Thesing and Velte, 2021).
From the perspective of the principal-agency theory managers could exploit the value estimates opportunistically which would decrease the reliability of the accounting information (Yao et al., 2018). These principal-agency conflicts would be mitigated through corporate governance (Shleifer and Vishny, 1997) which explains why weaker corporate governance would lead to greater information asymmetry as in the study by Song et al. (2010). Since government ownership has been shown to lead to weaker and less efficient corporate governance (Al-Janadi et al., 2016;
Shleifer and Vishny, 1997) these issues would conceivably be exacerbated in government owned companies.
If corporate governance and ownership structures affect the quality of the financial reporting practice, then government ownership might not have a first-order relationship with earnings quality as Gaio and Pinto (2018) indicate, but instead a moderating function on the relationship between the financial reporting practice and earnings quality. From the reasoning above it also seems likely this effect would be negative in the case of government ownership. This leads to our second hypothesis:
H2: The relationship between fair value accounting exposure and aggregated earnings quality is negatively affected by government ownership when unobservable inputs are used in the fair value estimation.
In conclusion, the aim of this study is to examine how government ownership affects the relationship between fair value accounting and earnings quality, as expressed by our model.
Figure 1 Model describing the moderating effect of government ownership on the relationship between Fair Value Accounting Exposure and the multidimensional concept Aggregated Earnings Quality
3. Research design
This section describes the sample selection process as well as the statistical analysis and the sample. The model and measures for each variable are explained.
3.1 Sample selection
To test the hypotheses stated in the previous section, we compare 36 listed and unlisted real estate companies with 32 real estate companies controlled by the state and municipal governments in Sweden that use either IFRS or K3 accounting standards. In existing literature, it is most common to use complete samples, for example every listed company in a country or in a sector (see for example Al-Janadi et al., 2016; Šodan, 2015). There are also examples of convenience sampling or purposive sampling resulting in small sample sizes (<50 sample companies) (see Yasa, 2020 for example). Our sample would therefore ideally consist of every real estate company using IFRS or K3 accounting regulations operating on the Swedish market, which is not possible due to time constraints. A convenience sampling technique comparable with previous studies such as Yasa, (2020) is therefore used. The companies are selected by using sector search in the Retriever database (code 138325, sni codes 668.100, 68.201, 68.202, 68.203). From this sector search the companies are sorted by size (considering revenue and number of employees). The sampling is then done from the top of the list as far down as possible considering the time constraints which result in an initial sampling of 140 companies. Of 34 listed companies, 15 are discarded due to either being mislabeled, or due to not being in operation for a minimum of 5 years, or due to incomplete data. 19 are retained for the final sample. None are government controlled. Of 106 unlisted companies, 52 are government controlled and 54 controlled by private interests.
Government controlled companies are identified by selecting companies owned by at least one shareholder with a minimum of 20 percent ownership of the following type: public authorities, state government or municipal government. This threshold is chosen in line with previous research (Gaio and Pinto, 2018), and implies the government has significant influence on the decisions of the firm. Of the 52 government-controlled companies, 20 are discarded due to either being mislabeled, not being in business for a minimum of 5 years or due to incomplete data. 32 government-controlled companies are retained for the final sample. Of the remaining 54 unlisted
privately controlled companies, 33 are discarded due to being mislabeled, not being in operation for a minimum of 5 years, or not having comprehensive accounting data (i.e lack of cash flow statements, or lack of mandated fair value estimates). 17 companies are retained for the final sample which totals 68 companies.
Table 1. Structure of sample.
Government
controlled companies
Companies controlled by private interests
Total sample
IFRS 11 28 39
K3 21 8 29
Total 32 36 68
Listed 0 19 19
Unlisted 32 17 49
Total 32 36 68
An issue to note is that size usually has a positive relationship with earnings quality, as larger companies generally have higher earnings quality (Gaio, 2010). Therefore, the sampling process conceivably means the companies included in the sample will demonstrate higher earnings quality than the population in general. However, since the study is not meant to draw any conclusions about the level of earnings quality of the industry in comparison with any other group this is not a problem for this study. Proxies of earnings quality were calculated using at least five years of accounting data, in line with previous studies (Šodan, 2015). Part of the data is secondary data, from the Retriever database, and part is primary data gathered from the annual reports of the companies in the sample. The annual reports are also downloaded from the Retriever database in their entirety.
3.2 Model and variables
Earnings Quality and the independent variables are Fair value accounting exposure and Government ownership as well as two control variables. An interaction variable is constructed to test the effect government ownership has on the relationship between fair value exposure and aggregated earnings quality. The individual proxies that are part of the composite measure of earnings quality; predictability, smoothness and conservatism are in turn dependent variables in their individual models.
The measures for earnings quality and the individual properties are presented in a format which implies higher values mean lower quality earnings. The measures for dependent, independent and control variables have all been established through previous studies. Šodan, (2015) is to a large extent used as a model for operationalising the concepts of aggregated earnings quality and fair value exposure, while the measures for smoothness, predictability and conservatism as well as the control variables are adapted from Dechow et al. (2010), Liceran-Gutierrez and Cano-Rodriguez (2019) and Gaio and Pinto, (2018). The following model is used for testing the first hypothesis (H1):
𝐴𝐸𝑄 = 𝛽0+ 𝛽1𝐹𝑉𝐸 + 𝛽2𝐺𝑂𝐸 + 𝛽4𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠 +𝛽5𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 + 𝜖𝑖 (1)
AEQ signifies Aggregated Earnings Quality and FVE means Fair Value Exposure. Government Ownership is captured with the dummy variable GOE which is equal to 1 when the company is owned by the government and 0 otherwise. Total assets is a control variable capturing the size of the company and Leverage is the ratio of total liabilities to total assets in the firm.
A second model is created to investigate the second hypothesis (H2). In this model an interaction variable for government ownership and fair value exposure is added:
𝐴𝐸𝑄 = 𝛽0+ 𝛽1𝐹𝑉𝐸 + 𝛽2𝐺𝑂𝐸 + 𝛽3𝐺𝑂𝐸 × 𝐹𝑉𝐸 + 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠 +𝛽6𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 + 𝜖𝑖 (2)
Each variable is described in detail below.
3.2.1 Aggregated Earnings Quality
Earnings quality is estimated using predictive ability, smoothness and conditional conservatism.
In the section below, each individual measure is described, ending with an explanation of how these measures are aggregated into a composite measure. All proxies for earnings quality are constructed in accordance with earlier works (see Dechow et al., 2010; Gaio, 2010; Liceran- Gutierrez and Cano-Rodriguez, 2019; Šodan, 2015). Before these measures are described, net income (NI), that is used to compute several of the variables below, is explained.
As in previous literature, net income is used to measure the concept of earnings (Gaio, 2010;
Šodan, 2015). Net income is useful as it captures the changes in property valuation in companies reporting according to IFRS. The companies reporting under the K3 standard, however, disclose fair value estimates in the notes, and there is no impact from unrealized gains and losses on the income statement. Therefore, the net income of the K3 companies is adjusted to incorporate these unrealized value changes. A portion of these companies disclose the unrealized gains and losses explicitly in the notes making this a straightforward process, however some companies do not. For these companies the unrealized gains or losses are calculated manually. The change in fair value estimates for each year is adjusted for new acquisitions and sales which would otherwise skew the value. For this purpose, the real estate transaction database Datscha was used to pull information for every sale or acquisition from each company for the relevant years. The K3 companies net income is thus adjusted as follows:
𝑁𝐼𝑖 = 𝑁𝑒𝑡 𝐼𝑛𝑐𝑜𝑚𝑒𝑖+ 𝑉𝑎𝑙𝐶ℎ𝑎𝑛𝑔𝑒 × (1 − 𝑡𝑖)
Where Net Income is the net income reported on the income statement ValChange is the change in property values during the year and t is the effective tax rate.
Predictive ability
Previous research on the predictability of earnings has to do with equity valuation (Dechow et al., 2010). For us it is of special interest to see if earnings generated by asset revaluations are a good
property values are motivated, they need to be reflected in the future cash flows generated by those properties. We therefore measure the predictive ability of earnings in line with Šodan, (2015) as:
𝐶𝐹𝑂𝑖,𝑡 = 𝛽0+ 𝛽1𝑁𝐼𝑖,𝑡−1+ 𝜖𝑖,𝑡
Where CFOi,t is the cash flow from operations from company i at time t and NIi,t-1 is the net income from time t-1. In line with previous literature we scale the cash flow with total assets (Dechow et al., 2010; Šodan, 2015) As Gaio, (2010) and Šodan, (2015) we define the variable of predictive ability as the variance in earnings shocks. A higher variance would then imply earnings are less predictable. The predictive ability of earnings is therefore measured as the standard deviation of the estimated error from the equation.
𝑃𝑅𝐸𝐷𝑖 = √𝜎2𝜖𝑖,𝑡
Volatility or Smoothing
Smoothing can be said to result in earnings which are more informative and have a higher persistence, and for this reason lower volatility in earnings is generally understood to be attractive to investors (Dechow et al., 2010; Gaio, 2010). The most common way to measure smoothness is as the standard deviation of earnings divided with the standard deviation of the cash flows from operation (Dechow et al., 2010; Gaio, 2010; Šodan, 2015).
𝑆𝑀𝑂𝑂𝑇𝐻𝑖 = 𝜎(𝑁𝐼)𝑖,𝑡
𝜎(𝐶𝐹𝑂)𝑖,𝑡
A major criticism of smoothness as a proxy for earnings quality is that it is hard to determine what smoothness represents. Smoothness is treated the same whether it is a result of better representation of earnings, a result of accounting rules or a result of earnings management (Dechow et al., 2010).
Nevertheless, low volatility is seen as an important sign of legitimacy both for the companies themselves and investors (Lorentzon, 2011:78).
Conditional conservatism
Conditional conservatism is understood as timely loss recognition (losses are recognised earlier than comparative gains) or that companies are more conservative when recognising unexpected gains than unexpected losses (Šodan, 2015). Dechow et al. (2010) note that conditional conservatism is valued by investors because management might be overly optimistic, higher conditional conservatism therefore means higher quality earnings. Conditional conservatism is commonly measured through stock returns (Basu, 1997). As our sample includes companies which are not publicly traded, we use an alternative, suggested by Basu, (1997):
𝛥𝑁𝐼𝑖,𝑡 = 𝛽0+ 𝛽1𝐷𝑖,𝑡+ 𝛽2𝛥𝑁𝐼𝑖,𝑡−1+ 𝛽3𝐷𝑖,𝑡× 𝛥𝑁𝐼𝑖,𝑡−1+ 𝜖𝑖,𝑡
Where ΔNIi,t is the change in net income from year t-1 to year t and ΔNIi,t−1 the change from year t-2 to year t-1, scaled with the ingoing value of total assets. The variable Di,t is a dummy variable which is equal to 1 when ΔNIi,t−1is negative. Lastly an interaction variable between the two independent variables is used (Di,tΔNIi,t). Under high conditional conservatism, good news will be more persistent than bad news. If good news is more persistent, as is implied, then β3<0 (Šodan, 2015). Our measure for conditional conservatism is therefore:
𝐶𝑂𝑁𝑆𝑖 = 𝛽3,𝑖
Dechow et al. (2010) note a potential downside of the model as having an unknown net effect on earnings quality. The reasoning is that during periods with large write-offs conservatism will generate lower persistence as compared to periods with better news, which in turn means lower earnings quality.
Aggregated Earnings Quality: A composite measure
Aggregated earnings quality is estimated using predictive ability, smoothness and conditional conservatism. These measures are then ranked using percentile ranks and aggregated into a composite measure (Gaio, 2010; Gaio and Pinto, 2018; Šodan, 2015). All measures are ranked so
conservatism and worse predictability) which in turn equals lower aggregated earnings quality.
This is in line with established research but there are several problems associated with this practice.
In creating a composite measure of earnings quality several, sometimes contradictory, perspectives on usefulness are incorporated into one value and caution is required when interpreting the results (Liceran-Gutierrez and Cano-Rodriguez, 2019). The theory behind the construction of a composite measure of earnings quality was discussed in section 2.2, resulting in three properties being included for this study.
A second issue is related to the weights assigned to each of the proxies when constructing a composite variable. The most common method is to assign equal weights through percentile or decile ranking (see Gaio and Pinto, 2018; Šodan, 2015 for example). But in doing this, it is implicitly assumed all variables have equal importance. Unfortunately, there is no guarantee those weights truly represent the relative importance of each property on the multidimensional concept of earnings quality (Liceran-Gutierrez and Cano-Rodriguez, 2019).
Liceran-Gutierrez and Cano-Rodriguez, (2019) suggest solving this issue by using structural equation modeling (SEM) instead of multiple regression analysis, but this is not possible in this case due to the comparatively large sample size necessary to use SEM effectively (N>150). The common approach of assigning equal weights is therefore used despite associated problems.
The following equation is therefore applied to calculate aggregated earnings quality:
𝐴𝐸𝑄𝑖 =𝑅𝐴𝑁𝐾(𝑃𝑅𝐸𝐷𝑖) + 𝑅𝐴𝑁𝐾(𝑆𝑀𝑂𝑂𝑇𝐻𝑖) + 𝑅𝐴𝑁𝐾(𝐶𝑂𝑁𝑆𝑖) 3
3.2.2 Exposure to fair value accounting
Exposure to fair value accounting is measured in line with previous works (Šodan, 2015) through the income statement approach (Bratten et al., 2016). The gains and losses as a component of income are separated from net income, this component is called fair value income (FVI). We capture the relative importance of FVI, referred to as fair value exposure (FVE). The fair value
exposure can then be seen as the percentage of FVI make up of net income. The absolute values of net income and of FVI for multiple years are used:
𝐹𝑉𝐸𝑖 =𝐹𝑉𝐼𝑖 × (1 − 𝑡) 𝑁𝐼𝑖,𝑡
Where FVI is the fair value income, t is the tax rate and NI is the net income. The companies reporting under the K3 standard disclose the fair values in the notes, these are however not impacting the income statement and so we use the adjusted net income discussed earlier.
3.2.3 Government ownership
Government ownership is included in the models as a dummy variable (1= government owned, 0
= not government owned), and as previously stated, government owned companies are defined as companies owned by at least one shareholder of the following type: public authorities, state government or municipal government, with a minimum of 20 percent ownership. This threshold is chosen in line with previous research (Gaio and Pinto, 2018) and implies the government has significant influence on the decisions of the firm.
3.2.4 Control variables
The most common control variables in similar studies are mean company size, mean accounting leverage, country and industry (Gaio, 2010; Goncharov and Hodgson, 2011). In this case industry and country specifics are already accounted for. Size usually has a positive relationship with earnings quality, where larger companies have higher earnings quality (Gaio, 2010).
Leverage is more debated but is sometimes found to have a negative relationship with real earnings management. Therefore, a higher leveraged firm means less earnings management and higher earnings quality (Zamri et al., 2013). Other studies show no effect of leverage on earnings quality, or even the opposite effect (Yasa, 2020), but the variable is still included here. Size is defined as the natural logarithm of mean total assets over the sampled period (years 2014-2019), leverage is the ratio of total liabilities to total assets over the same period which is in line with previous studies
3.3 Statistical analysis
The sample is first tested for normality. Several variables are measured using percentile ranks (predictability, conservatism, smoothness and aggregated earnings quality) which means these values are uniformly distributed. This is corrected using a z-score. Of the remaining variables Total assets and Leverage both show skewness and kurtosis outside of the acceptable (-2,2) range, using natural logarithms for these two variables leads to acceptable skewness and kurtosis values. As these are both control variables, this correction is not of any direct importance when interpreting the results. We check for heteroskedasticity by plotting the residuals for FVE (see Appendix 1.1) and see no evident signs in the shape in the plot.
The bivariate correlations are checked through a correlation matrix and none of the variables show correlations over 0.8 which is a common threshold. VIF values are below 4 in tests for multicollinearity. Several ordinary least square (OLS) regression models are used. Model 1 tests the bivariate relationship between fair value accounting exposure and aggregate earnings quality (H1). Model 1a, 1b, and 1c test the bivariate relationship between fair value exposure and each individual measure of earnings quality (predictability, smoothness and conservatism). Model 2 adds the control variables to model 1. Model 3 adds an interaction effect which tests the effect government control has on the relationship between fair value accounting exposure and aggregate earnings quality (H2). Model 4 adds the control variables to Model 3.
4. Empirical results
The section below is organized as follows: First descriptive statistics and correlations are described, then the results from the regression models are analysed in the order stated in the previous section.
4.1 Descriptive statistics and correlations
To better understand the sample, we use descriptive statistics. Table 2 provides mean values and standard deviations for fair value exposure, total assets and leverage. The sample shown in the table is diverse with some companies being much larger than others and leverage shows a large
spread between maximum and minimum values. Since both skewness and kurtosis is outside of standard acceptable range (-2,2) they are transformed through the use of natural logarithms. Since these are control variables, there is no real issue with interpretation of the coefficients later. There is also a wide spread in the level of exposure to fair value accounting. Some companies have an exposure above 100 percent of their earnings while others have more moderate figures around 25 percent. It is worth noting that for all companies the revaluation of properties make up a considerable portion of the earnings figure.
Table 2: Description of sample. FVE is Fair Value Exposure. Variables re-calculated as percentile ranks are not included in the sample description.
The fair value exposure variable only has one distinct outlier (see below), which is kept in the sample. The outlier is not an error, which is verified by checking the initial data, and the sample is small so every data point is critical. Furthermore, running the regressions excluding the outlier did not change the significance nor the directions of the coefficients.
Figure 2: Boxplot of Fair value exposure (FVE).
The correlation matrix below shows the observed significant correlations. There is a significant positive correlation between aggregated earnings quality (AEQ*) and Fair value exposure (FVE), which is expected since a higher rank of earnings quality means worse earnings quality.
Table 3: Correlation matrix with Pearson coefficients. AEQ* is the aggregated Earnings Quality, FVE is Fair Value Exposure, GOE is Government control.
AEQ* is the aggregated earnings quality excluding conservatism, explained below
The significant positive correlation between aggregated earnings quality and fair value exposure displayed in the correlation matrix is plotted below. The scatterplot for these two variables illustrates their relationship visually. Considering the results above there seems to be a relationship between the two variables. Since the aggregated earnings quality is measured in a way where higher values indicate lower earnings quality, the relationship is negative as predicted in hypothesis one.
Figure 3: Scatterplot of Fair value exposure (FVE) and Aggregated earnings quality (AEQ*).
AEQ* is the aggregated earnings quality excluding conservatism, explained below
4.2 Results of regressions
To further test our first hypothesis, that exposure to fair value accounting is negatively related to aggregate earnings quality when unobservable inputs are used in the fair value estimation, we run several regression models as described in section 3.3. Higher values of aggregated earnings quality (AEQ) imply lower earnings quality. Therefore, we expect to find a positive relationship between the variable AEQ and exposure to fair value accounting (FVE).
4.2.1 The individual components of aggregated earnings quality (AEQ)
First, we study the bivariate relationship between FVE and the individual components of earnings quality. Model 1a, 1b and 1c test the bivariate relationship between the individual proxies of earnings quality (1a, smoothness, 1b, predictability and 1c conservatism). In the table below we summarize individual tests for smoothness, predictive ability and conditional conservatism.
Table 4: Summary of the individual proxies of earnings quality and their relationship with the independent variable, fair value exposure (FVE).
AEQ* is the aggregated earnings quality excluding conservatism
Smoothness is the only proxy which shows a positive significant relationship with fair value exposure. This should be interpreted as higher fair value exposure meaning higher volatility or lower smoothness. However, the explanatory power for smoothness is low, the R-square is 6.8%, which is much lower than the final aggregated model. For predictive ability the coefficient is also positive, however it is not significant, and the level of variance explained remains low. For conditional conservatism (1c) we observe a R-square which is close to zero. This implies the variable has no relationship with the independent variable, detracts from the explanatory power and does not fit a linear model. The data series needed for calculating this variable was relatively short for most companies which indicates low quality data may play a role. As the companies in the study are real estate companies, the main effect on earnings aside from rentals is property revaluations. The conservatism concept may therefore be of lower importance when comparing companies in the sample. Also, large increases in property prices occurred during the sampling period so revaluations during this period generated large positive earnings which might influence the measure. Very few years of negative results are observed, and these are followed by quick bounce backs. This leads to high persistence of earnings which biases the measure to yield high levels of conditional conservatism for all companies in the sample.
For these reasons conditional conservatism is excluded from our aggregated measure moving forward. Aggregate earnings quality (AEQ*) excluding conditional conservatism is used going forward. For comparison, we also create a series of models with conditional conservatism included in the aggregated earnings quality, and overall, it’s inclusion or exclusion does not change any of the models significantly. The explanatory power decreases for all models when it is included, but
neither direction nor significance of the coefficients of the different coefficients are affected by including or excluding the measure. This means our aggregated measure is not capturing all aspects of earnings quality as described by Dechow et al. (2010) and we cannot claim complete coverage of the aggregated earnings quality as a multidimensional concept. The aggregated variable with the two remaining proxies combined, however, has a much stronger relationship with the independent variable than the individual proxies have by themselves. This indicates there is explanatory power to be gained from the multidimensional concept which is not fully captured by the individual parts.
4.2.2 The relationship between Fair value exposure and Aggregated Earnings Quality
Our first model (Model 1) tests the bivariate relationship between fair value exposure (FVE) and the aggregated earnings quality (AEQ*). A significant relationship positive can be seen in the simple model. In line with our hypothesis the direction of the coefficient is positive. This implies higher levels of fair value exposure result in higher ranking of earnings quality (lower earnings quality). The level of the variance explained, R-squared is low which implies there could be unobserved variables that might improve the model's explanatory power. We therefore use an expanded model with added control variables (Model 2).
Table 4: Model 2 testing H1. FVE is Fair value exposure, GOE is a dummy variable indicating government control, Total assets and Leverage are control variables.
𝐴𝐸𝑄∗= 𝛽0+ 𝛽1𝐹𝑉𝐸 + 𝛽2𝐺𝑂𝐸 + 𝛽3𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠 +𝛽4𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 + 𝜖𝑖
