Software development capitalization and earnings management : a study of U.S. software firms

(1)

MSc Accountancy & Control

Thesis

Software development capitalization and

earnings management: A study of U.S.

software firms

Financial Accounting

Key words: R&D, capitalization, earnings management, software development, SFAS No. 86, ASC 985

Thomas Pacey - 10825080 22-06-2015

(2)

2

Statement of Originality

This document is written by Thomas Pacey who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

In this thesis, I study whether U.S. software firms utilize R&D accounting choice (expensing vs. capitalizing) for earnings management purposes and specifically whether capitalization serves as a substitute for reducing R&D expenditure. The option to capitalize software

development costs under U.S. GAAP ASC 985 is the only exception to ASC 730, which requires the expensing of R&D costs. This exception creates a unique opportunity to study R&D

capitalization and earnings management in a U.S. setting, and it appears no previous literature has focused on this topic. I do not find significant evidence that my sample firms cut total R&D expenditure to meet earnings benchmarks. I also do not find evidence that the capitalizers in my sample increase software development capitalization to achieve earnings benchmarks. Overall, my results indicate that U.S. software firms do not use R&D accounting choice for real or accrual-based earnings management.

Section 1: Introduction

The topic of earnings management has long been an area of focus in accounting research, due to the importance of earnings information to market participants. Given that discretionary R&D spending is one of the primary forms of real earnings management (Graham et al., 2005), the topic of R&D accounting as well as its relation to forms of earnings management and impacts on financial markets has also been well researched. Studies have been conducted in the context of U.S. GAAP, which requires expensing of R&D costs, providing substantial evidence that firms use R&D accounting choices (specifically real earnings management) to manipulate earnings by cutting R&D expenses (Baber, Fairfield, and Haggard 1991; Bushee 1998; Cooper and Selto 1991). Similar research of firms reporting under IFRS and other non-U.S. GAAPs concludes that the option or mandate to capitalize development costs (under IAS 38 or similar standards) also presents earnings management opportunities (Seybert 2010; Damak and Halioui 2013; Markarian, Pozza, and Prencipe 2008).

U.S. GAAP ASC 985 (formerly SFAS No. 86) is the only exception to ASC 730 (formerly SFAS No. 2), which requires expensing of research and development expenditure. ASC 985-20 requires capitalization under certain conditions, but the decision to capitalize costs remains

(5)

5

discretionary in nature and thus potentially vulnerable to manipulation by management. The option to capitalize software development costs under ASC 985-20 presents a limited and unique setting in which to study the impacts of different R&D reporting options (expensing vs. capitalization) in a U.S. setting and to specifically analyze whether U.S. software firms use capitalization as an earnings management tool.

Software development accounting under U.S. GAAP has been studied to a limited extent in the context of value relevance and earnings quality, and there is no clear agreement on

whether the option to capitalize costs is beneficial for investors. Further, it appears no studies have focused directly on the topic of software capitalization as a tool for earnings management. Cifti (2010) finds that the decision to capitalize software costs allows some firms to avoid losses, suggesting potential earnings management concerns. My thesis differs both in topic and in regards to the sample. First, Cifti’s study focusses on earnings quality (using the earnings response coefﬁcient as a measure) while I specifically test for earnings management towards benchmarks. Additionally, the data set used by Cifti represents the years 1981-1990 and is selected to study differences in the pre- and post SFAS No. 86 implementation period. My data is more recent, specifically from fiscal years 2012-2014. Cifti’s sample also included

non-software firms, while my sample consists exclusively of non-software firms with SIC 7372.

In addition to contributing to prior earnings management literature and research in the area of software capitalization, my study takes place in the context of convergence efforts between IFRS and U.S. GAAP as well as contrasting views over the best way to represent R&D costs (expensing immediately vs. capitalizing). The FASB requires immediate expensing of these costs, a position which supports the uncertainty of measuring future R&D benefits and also limits discretion that could be abused for earnings management activities. However, capitalization may allow signaling of useful R&D information to the market and give a more realistic

representation of firm positions on the Balance Sheet (Oswald and Zarowin, 2007, others). My research, especially using an example of capitalization in U.S. firms, may offer general insights into the debate on how to best account for R&D expenditures.

In my thesis, I follow the framework of Oswald and Zarowin (2004) and study the

(6)

6

and positive earnings growth serve as my earnings benchmarks. I use pre-tax earnings before R&D expenditure to divide firm observations into groups based on whether they could achieve the earnings benchmarks by cutting overall R&D spending. Logistic regression is used to test the relationship between each group and the likelihood of a cut in R&D spending as well as to compare results between capitalizers and expensers. To see if capitalization is used for earnings management, I then perform a similar test focusing only on capitalizers and using the increase in capitalization as the dependent variable rather than a cut in total R&D spending.

I find that neither expensers nor capitalizers cut overall R&D spending to meet benchmarks through real earnings management. Further, I find that capitalizers in my sample do not

increase software capitalization to meet benchmarks through accrual earnings management. To my knowledge these results are the first direct evidence regarding software capitalization and earnings management in a U.S. setting.

My thesis is structured as follows: Section 2 provides a more thorough review of previous relevant literature and describes my hypothesis development. Section 3 outlines my research methodology, including the sample selection process, regression framework used, and various tests. Section 4 reviews and analyzes the test results. Finally, Section 5 summarizes the study with a conclusion, followed by the references and tables in Section 6.

Section 2: Literature Review and Hypotheses

2.1 Accounting for Software Development

U.S. GAAP ASC 985 (previously known as SFAS No. 86) provides guidance for the accounting of costs of software products to be sold, leased, or otherwise marketed. The standard explains which of these costs should be classified as R&D costs to be expensed and which qualify as development costs that are capitalized. The costs are divided into three distinct stages along the development timeline:

1) All R&D costs incurred to establish the technological feasibility of a computer software product.

2) Costs of producing product incurred subsequent to establishing technological feasibility 3) Costs incurred after the product has become available for general release to customers

(7)

7

The costs in the second stage qualify as software development costs to be capitalized. The costs in the first and third stages are expensed as incurred. The stages of costs and their accounting treatments are shown in Figure 1, taken from Mulford and Roberts (2006).

The concept of technological feasibility is a key point in differentiating between costs to be expensed and costs to be capitalized leading up to product release. Costs incurred in the process of establishing technological feasibility are treated as R&D costs and expensed as required by ASC 730-10 (Accounting for Research and Development Costs). Costs incurred after technological feasibility is established are capitalized, and include allocated indirect costs and overhead and coding and testing before product release. According to ASC 985-25-2,

technological feasibility is established once the company:

“…has completed all planning, designing, coding, and testing activities that are

necessary to establish that the product can be produced to meet its design specifications including functions, features, and technical performance requirements.”

As evidence of technological feasibility, the firm must specifically meet the activity criteria in either (A) or (B).

(A) For computer software products created using a detail program design, costs are eligible for capitalization when:

a. Product design and detail program design are completed,

b. Completeness of detail program design and its consistency with product design have been confirmed by documentation, and

c. Detailed program design has been reviewed for high-risk development issues and issues have been resolved

(8)

8

(B) For computer software products created without a detail program design, costs are eligible for capitalization when:

a. A product design and working model are completed and

b. Testing has confirmed completeness of working model and its consistency with the product design

In theory, the standard requires capitalization when certain criteria are met, but in real practice, the decision to capitalize is essentially a discretionary accounting choice. Firms which prefer to capitalize do so, and those which wish to expense do so, given the flexibility of the requirements (Aboody and Lev 1998). Software projects with a detailed program design could allow for costs to be capitalized relatively early on, while projects without a detailed program design require a working model and testing and thus might delay capitalization (Mulford and Roberts 2006). This means that “. . . if a company wishes to capitalize, it draws up a detailed program design quickly. If it wants to expense lots of development costs, it simply holds off writing a detailed program design."1

ASC 985 is therefore the only exception to the requirement to immediately expense R&D costs as per ASC 730 (formerly SFAS No. 2). ASC 985 is quite similar to IAS 38, under which capitalization of development costs is mandated if certain circumstances are met. Within IFRS research, the impact of accounting policies on reported earnings and firm balance statement has been the cause of much debate over how best to account for R&D expenditures. In the relation between earnings management and R&D spending, U.S. firms typically only have the option of cutting R&D spending through real earnings management. However, firms which internally develop software also have the option to capitalize this portion of R&D to meet expectations through accruals-based earnings management.

Given this background, the purpose of this research is to more closely study the impact of R&D accounting choice to see if U.S. software firms utilize this discretion as an earnings management tool.

(9)

9

2.2 Prior Software Capitalization Literature

Since 1985, when FASB issued SFAS No. 86 as an exception to the SFAS No. 2 requirement to expense R&D costs, firms have been allowed to capitalize software development costs as

described in the previous section. Software capitalization research is a very specific and relatively small section of R&D accounting research, and therefore there are few examples of previous literature focusing on software development capitalization and the conclusions of these studies are mixed (Krishnan and Wang 2014). Previous research on this topic has mostly been in the area of information asymmetry, value relevance, and earnings quality, and serve as a foundation for this paper.

Motivated by the software industry's petition to abolish the capitalization option in SFAS No. 86, Aboody and Lev (1998) examined the relevance of software capitalization to investors and found no evidence that software capitalization hurt reported earnings quality. Specifically, they found that software capitalization data (annual capitalized costs and cumulative

capitalized software asset) was associated with stock returns, stock price, and future earnings and are thus value relevant. Eccher (1995) on the other hand, found results contrary to those of Aboody and Lev and concluded that capitalization of software is not value relevant for

investors, but that amortization of the cumulative software asset is value relevant. The software capitalization decision may also be used as an effective tool to mitigate information asymmetry problems related to R&D accounting. In a study using data from both pre- and post-implementation of SFAS No. 86, Mohd (2005) found that the new standard decreased information asymmetry by allowing for capitalization. Using bid-ask spread and share turnover as indicators, the results show significantly lower asymmetry for capitalizers compared to expensers, concluding that capitalization is a beneficial accounting policy.

Ciftci (2010) builds upon both Aboody and Lev (1998) and Mohd (2005), among others, but found that capitalization of software development under SFAS No. 86 does not improve

earnings quality. Instead, his results suggest that the software industry has lower earnings quality after the implementation of SFAS No. 86, and that expensers have higher earnings quality in relation to their capitalizer counterparts. Interestingly enough, while not the focus of his study, Cifti also concludes that a portion of capitalizers are able to avoid losses or earnings

(10)

10

decreases due to their capitalization decision. As stated in the introduction, my research differs in sample (only software firms, and more recent data) and focus (earnings management rather than earnings quality).

In a single-year survey of accounting practices in the software industry, Mulford and Roberts (2006) highlight trends regarding the decision to capitalize or expense software development. One third of their sample firms capitalized at least a portion of these costs, leading on average to a 27% increase in operating cash flows, compared to a direct decrease in operating cash flow for firms which only expensed. These results highlight the potentially high material impact of this choice on firm performance.

Outside of empirical research, there is direct evidence of earnings management through manipulated software development capitalization. Nelson, Elliot, and Tarpley (2003) review examples of attempted earnings management decisions as gathered from a survey of auditors and supported by Accounting and Auditing Enforcement Releases (AAERs) issued by SEC. One of the many earnings management strategies mentioned was the capitalizing or deferring of too much costs. The authors specifically refer to an income-increasing example of a company which internally developed software and over-allocated costs into capitalized software development. 2.3 Prior Earnings Management Literature Related to R&D

Earnings management has long been a relevant and important topic for accounting

research, given its impact on firm performance. According to Graham et al. (2005), along with advertising and maintenance, discretionary spending on R&D is the most frequent form of earnings management. Further, the relevance of R&D reporting to investors is clear, as these costs can impact stock price, returns, and informative value (Lev & Sougiannis, 1996; Chan et. al, 2001).

Much research has been done in the area of earnings management, especially in relation to the expensing of R&D. Baber, Fairfield, and Haggard (1991), Bushee (1998), and Cooper and Selto (1991), among others, provide evidence that mandatory expensing of R&D (as such under U.S. GAAP), leads to underinvesting in efforts to meet earnings benchmarks by real earnings management. R&D expenditure directly impacts pre-tax earnings and thus the decision to trim R&D costs can be used as a tool to manipulate earnings. Baber et al. (1991) found that R&D

(11)

11

costs were lower for firms when that spending interfered with the ability to report profit or profit increase. Bushee (1998) also found that R&D cuts are used to avoid earnings declines, but that the decision to cut R&D is less likely to be made in firms with high institutional ownership. Unlike other research on the topic, Cooper and Selto (1991) draw conclusions from

experimental data. In a scenario offering various investment decisions and where R&D projects were required to be expensed (and thus impact earnings), managers avoided R&D projects in favor of suboptimal alternatives that were not required to be expensed.

Perry and Grinaker (1994) find a strong association between unexpected earnings and unexpected R&D spending. In a more comprehensive study of various earnings management strategies, Roychowdhury (2003) finds that firms with small positive net income have

discretionary expenses that are unusually low. One of the main categories of these

discretionary expenses is R&D. The results of both of these studies adds to the body of research suggesting that when firms face earnings pressure, they manage earnings through R&D

expenditure.

2.4 R&D Capitalization and Earnings Management

Much of the research surrounding R&D and earnings management studies firms following U.S. GAAP and thus focusses on expensing. In addition to this, research outside of a U.S. setting has shown that the option to capitalization R&D can also lead to earnings management. Seybert (2010) shows that mandatory capitalization of R&D may not prevent earnings management, and may in fact cause overinvestment in capitalized projects to prevent impairment charges that negatively impact earnings. Damak and Halioui (2013) and Markarian, Pozza, and Prencipe (2008) show that firms can use capitalization of R&D expenditures as accounting-based

earnings management to smooth earnings or avoid debt covenant violations. Both expensing and capitalization of R&D can be used for opportunistic earnings management strategies (Dechow and Skinner, 2000), but capitalization can also be used to signal information to the market regarding expected success of a firm’s investment in R&D ventures.

(12)

12

2.5 Impact of Earnings Benchmarks On R&D Cuts

The focus of this study specifically surrounds the relation between R&D accounting policy and its use to meet or manage earnings towards earnings benchmarks. Garcia Osma and Young (2009) show that firms which fail to meet earnings targets during the previous year or face pressure to meet current targets are more likely to cut R&D expenses in the current year. They also relate these findings to stock performance, suggesting that investors notice increased earnings are associated with unexpected R&D changes.

Oswald and Zarowin (2004) perform research on U.K. firms which have the option to either capitalize or expense R&D costs and find trends of real and accrual-based earnings

manipulation. It follows that firms who fail to meet earnings targets may react differently depending on which options are available to manage earnings. From a theoretical standpoint, Oswald and Zarowin note that expensers who reduce R&D costs by 1 increase pre-tax earnings by 1. For capitalizers on the other hand, a portion of the current year R&D cost is in the form of amortization which cannot be managed. Thus a reduction of R&D costs by 1 yields less than 1 increase in pre-tax earnings. This, in addition to potential impacts on long-term firm growth, suggest that reducing R&D costs might not be as useful for capitalizers. The authors’ results support this claim, indicating that expensers tend to cut R&D if it allows them to achieve earnings goals but finding that capitalizers increase capitalization to achieve earnings goals.

Based on these results, in combination with a variety of previous literature linking R&D expensing policy to earnings management, I expect that the same will hold true in my study. For software firms who only expense development costs, their only option to use discretionary R&D to manage earnings is to cut overall R&D expenditure (in this case overall R&D expenditure is equal R&D expense) as a means of real earnings management. However, given that capitalizers also have the option to increase capitalization as a means to manage earnings in reaction to benchmark performance, I do not expect capitalizers who fail to earnings benchmarks to cut overall R&D spending. Specifically:

Hypothesis #1a: For expensers, a failure to meet current year earnings benchmarks is

associated with a decrease in total R&D expenditure (expense + capitalization) in the current year.

(13)

13

Hypothesis #1b: For capitalizers, a failure to meet current year earnings benchmarks is

not associated with a decrease in total R&D expenditure (expense + capitalization) in the current year.

I expect capitalizers to substitute accounting-based earnings management in place of real earnings management. This would be in line with Zang (2012) who concludes that real and accounting earning management activities are used interchangeably, depending on which strategy is less costly for the firm. In the case of R&D, using real earnings management is more costly because it impacts future performance and long-term value of the firm and thus other policies may be preferred (Bushee, 1998). Firms may be able to manage earnings towards benchmarks by using discretionary capitalization as a less costly substitute for discretionary R&D spending cuts (Dinh et al., 2014). This concept is supported by the results of Oswald and Zarowin (2004) who find UK firms capitalize R&D to meet benchmarks, and also to a lesser extent by implications from Cifti (2010) who raises the possibility that software capitalization can be used to manage earnings. Based on these prior studies, I predict my results will be similar. This leads to the formulation of my second proposition, structured below in two parts:

Hypothesis #2a: For capitalizers, a failure to meet current year earnings benchmarks is

associated with a decrease in R&D expenses in the current year.

Hypothesis #2b: For capitalizers, a failure to meet current year earnings benchmarks is

associated with an increase in the R&D capitalization rate in the current year.

Section 3: Methodology

3.1 Sample and Data Collective Process

My sample consists U.S. software firms with financial data present in the North American Compustat database. Following Krishnan and Wang (2014) and Mulford and Roberts (2006), I specifically focus on firms labelled with SIC code 7372 (Services-Prepackaged Software), as these present an appropriate setting to study the relationship between software capitalization and earnings management. This is the case because R&D intensity (R&D scaled by sales) is signiﬁcantly higher in SIC code 7372 relative to ﬁrms in other areas of the software industry,

(14)

14

namely SIC codes 7370, 7371, 7373, and 7374 (Krishan and Wang, 2014). Thus, the likelihood and magnitude of software development capitalization are highest for firms with code 7372 (Krishnan and Wang, 2014) and decisions to capitalize would potentially have greater material impact on earnings.

I first perform a data search on the North American Compustat database for U.S. firms with SIC code 7372, selecting data needed for my tests. For the capitalizers in my sample, annual software development capitalization costs is an important data item, but it is impossible to determine this item from Compustat data alone. Compustat includes a CAPSFT data item, but this refers to the carrying value of a firm’s total net software development asset, factoring in amortization. For those firms with values present for CAPSFT, I manually collect annual software capitalization costs from these firm’s 10-K filings, utilizing the Security and Exchange Commission’s EDGAR database. While there is no standard format for reporting these annual costs separated from the net software asset, the amount capitalized was typically found in the footnote regarding ASC 985, or broken down in the firm’s investing section of the cash flow statement.

I remove observations from the sample if the data is not found in the 10-K filing, or if the company is misclassified into SIC 7372 in Compustat. After removing observations that were collected for purposes of creating lagged variables, my final sample consists of 322

observations. I split my sample of observations into two groups: capitalizers (cases in which some or all R&D expenditure is capitalized) and expensers (cases in which no software capitalization occurs), with a total of 72 capitalizers and 250 expensers. My data covers fiscal years 2012-2014. Though a key data item was hand-collected, I wanted a reasonable sample size of at least 200 total firm observations years. Using multiple years achieved this, and also provided the opportunity to use lagged control variables.

3.2 Empirical Measurements and Regression Framework

My study’s empirical research framework relies heavily on the framework of Oswald and Zarowin (2004), which itself draws from Baber et al. (1991), Perry and Grinaker (1994), and Bushee (1998), among others. I also adopt aspects from Garcia Osma and Young (2009) and

(15)

15

from previous SFAS No. 86 research by Aboody and Lev (1998), Mohd (2005), Ciftci (2010), and Krishnan and Wang (2014).

The first objective of my research tests is to determine if broader conclusions from prior earnings management research holds true for the U.S. software industry. Specifically, my first set of tests analyze the relationship between pre-tax earnings benchmarks and current year total R&D spending. Additionally, I will explore whether this relationship differs between firms that capitalize a portion of R&D and firms who only expense, and if so, how. My second set of tests focus on the capitalizers in the sample and aims to determine if these firms substitute increased capitalization in place of R&D cuts in order to achieve earnings benchmarks. These tests will analyze the relationship between pre-tax-earnings benchmarks and software development capitalization rates.

My regression model follows directly from Oswald and Zarowin (2004) who in a UK setting studied how the choice to expense or capitalize R&D costs could be used as a tool to manage earnings towards benchmarks. The Oswald and Zarowin model itself builds upon research by Bushee (1998) and Burgstahler and Dichev (1997), which find that firms manage earnings to prevent losses and decreases in earnings. Therefore, it uses earnings benchmarks of zero earnings level and zero earnings change from the prior year and it uses a R&D benchmark of zero change from the prior year (Oswald and Zarowin 2004).

Test Set 1 – Earnings management via R&D expenditure cut? The foundation of my logistic regression equation is as follows:

RDCutt = β0 + β1C1t+ β2C3t + ε (1)

RDCutt is an indicator variable that equals 1 if a firm decreases R&D expenditure between year

t-1 and year t. I use indicator variables to divide my sample into target and non-target groups for earnings management (described below). C1 and C3 are indicator variables that equal 1 if a firm is in Group 1 and 3, respectively, and zero otherwise. For all of my equations to follow, I run a robust logit regression, clustered based on the fact that there are repeat firms across multiple years.

(16)

16

Following Baber, Fairfield and Haggard (1992), Garcia-Osma and Young (2009), Bushee (1998), and Oswald and Zarowin, I divide my sample of firms into three groups depending on their current year pre-tax earnings before total R&D expenditure (EBRDt), compared to their

total R&D expenditures in the previous year (RDt-1). I first divide the groups (shown below)

using the earnings benchmark of zero earnings, based on the assumption that firms manipulate earnings to avoid losses. In other words the grouping is based on EBRDt compared to RDt-1.

Zero/Positive earnings benchmark Group 1: EBRDt < 0

Group 2: 0 ≤ EBRDt ≤ RDt-1

Group 3: RDt-1 < EBRDt

My primary suspect group for earnings management is Group 2, as these firms would show negative earnings if they maintain previous levels of R&D expenditure but can cut R&D expenditures to achieve zero or positive earnings. On the other hand, Groups 1 and 3 do not have as strong of incentives to deviate from prior R&D levels. Group 1 contains firms who show losses even before considering the R&D impact on earnings (i.e. they will show a loss even if they decrease R&D expenditure to 0). Similarly, Group 3 consists of firms who experience positive earnings even while maintaining the prior year’s R&D expenditures.

With equation (1), I use a logit regression to test and compare results between the groups. There are dummy variables for Groups 1 and 3, and β0 (the intercept) captures the likelihood of a decrease in R&D for Group 2. β1 and β2 capture the “incremental probabilities for Groups 1

and 3 compared to Group 2” (Oswald and Zarowin 2004). This framework allows the measurement of each group’s deviation from our target Group 2.

With the firm grouping based on the zero earnings benchmark (using EBRDt), I expect that

expensers in Group 2 are more likely to cut total R&D expenditures than expensers in Group 1 or 3 (coefficients β1 and β2 should be negative). I expect that there are no differences in the

percentage of firms that cut total R&D expenditures between earnings groups for capitalizers (coefficients β1 and β2 should be insignificant).

I rerun equation (1) using a second grouping of my sample firms. This grouping assumes that firms manage earnings to avoid earnings decreases and thus I use zero earnings change

(17)

17

(∆EBRDt) as the benchmark (rather than zero earnings level as the benchmark used in the first

grouping). In other words the grouping is based on ∆EBRDt compared to RDt-1.

Zero/Positive earnings growth benchmark Group 1: ∆EBRDt < -RDt-1

Group 2: -RDt-1 ≤ ∆EBRDt ≤ 0

Group 3: 0 < ∆EBRDt

My logic and expectations remain the same for this second grouping. I expect expensers in Group 2 to be more likely to cut R&D expenditure compared to Groups 1 and 3. For capitalizers, I expect there to be no significant differences across groups.

Equation (1) is a naïve model, in that it assumes that meeting earnings benchmarks is the sole factor in a firm’s decision to cut R&D. Following Bushee (1998) and Oswald and Zarowin (2004), I add a number of control variables to equation (1) to account for other

possible determinants of R&D spending. The result is equation (2) below, which I run separately for expensers and capitalizers. This equation tests Hypothesis 1a and 1b.

RDCutt = β0 + β1C1t + β2C3t + β3CCAPEXt + β4CSALESt + β5SIZEt + β6LEVt

+ β7DISTt +β8RDINTt + YearDummies+ ε (2)

Where:

CCAPEXt = change in capital expenditures from prior year

CSALESt = change in sales from prior year

SIZEt = firm size based on assets

LEVt = leverage

DISTt = percentage R&D cut needed to achieve benchmark

RDINTt = R&D intensity

For expensers, I predict that the group variable coefficients β1 and β2 are significant and

negative, suggesting that Groups 1 and 3 cut R&D expenditure to a lesser extent than Group 2 (the target group). For capitalizers, I predict that there are no significant differences between the groups, so β1 and β2 should not be significant.

Change in capital expenditures (CCAPEXt) and change in sales (CSALESt) are included to

(18)

18

these variables and the decision to cut R&D, as firms with fewer funds available would be less likely to invest in R&D projects. Thus β3 and β4 should be negative.

Following Oswald and Zarowin, the variable firm size (SIZEt) is included as a proxy for two

things: 1) the firm’s information environment and 2) the firm’s likelihood of cash constraints. Information environment is relevant because larger firms tend to have fewer earnings management opportunities (Weidman 1996; Garcia Osma and Young 2009). Larger firms are also less likely to have cash constraint issues (Opler et al. 1999; Garcia Osma and Young 2009) is thus might have fewer reasons to cut R&D (Garcia Osma and Young, 2009). Based on this, I predict that larger firms are less likely to reduce R&D expenditure and thus the SIZEt variable

coefficient (β5) should be negative.

Like Mohd (2005), I measure leverage (LEVt) as long-term debt divided by the sum of equity

and long-term debt. The leverage variable serves to represent a firm’s pressure to debt covenants. Firms with a greater leverage ratio should more likely to manage earnings (due to fear of debt covenant violations). Therefore, I suspect a positive relationship between leverage and R&D decreases (β6 is predicted to be positive).

Following Oswald and Zarowin (2004), I use the variable DISTt to represent the percentage

cut in R&D expenditures needed to achieve the earnings benchmark (either positive earnings or positive earnings growth). The variable coefficient β7 should be positive because I predict the

greater amount of R&D cut needed, the greater the chance a firm will cut R&D.

Similar to Garcia Osma and Young (2009), I add in R&D intensity (RDINTt) as another control

variable. Firms with more intensive R&D activity tend to have greater market scrutiny regarding investment decisions and thus do not have as much flexibility to meet earnings benchmarks via R&D expenditure adjustment (Barth et al. 2001; Garcia Osma and Young 2009). I expect the coefficient of RDINTt (β8) to be negative based on this.

Test Set 2 – Earnings management via capitalization increase?

As mentioned previously, consistent with prior research, my prediction in Hypothesis 2a and 2b is that since the capitalizers have the option to increase the capitalized portion of overall R&D expenditures to manage earnings (and therefore can avoid an overall decrease in R&D spending), I expect their overall R&D expenditure (sum of expense and capitalized portions) to

(19)

19

not decrease. This would be in line with the results of Oswald and Zarowin (2004), among others, and reinforce Zang’s (2012) findings that real earnings management and accounting-based earnings management are substitutes. To test if this substitution is present in my sample, a second set of tests focusses on the capitalizing firms. The objective is to test the relationship between meeting earnings benchmarks and 1) a decrease in R&D expenses and 2) an increase in software capitalization.

To test Hypothesis 2a and see how the three groups of capitalizers cut R&D expenses, equation (2) is modified, changing the dependent variable from total R&D expenditures (RDCutt) to R&D expenses (RDExpCutt).

RDExpCutt = β0 + β1C1t+ β2C3t + β3CCAPEXt + β4CSALESt + β5SIZEt + β6LEVt

+ β7DISTt +β8RDINTt + YearDummies + ε (3)

The variable RDExpCutt is an indicator variable that equals 1 if the firm decreased R&D expense

relative to the prior year and 0 otherwise. I expect my results to be consistent with Zang’s (2012) substitution theory and Oswald and Zarowin’s findings and so I predict that capitalizers in Group 2 will be more likely to cut R&D expenses (β1 and β2 should be negative for both

earnings benchmarks). My expectations for the control variables are the same as in equation (2).

To test Hypothesis 2b and see how the three groups of capitalizers increase capitalization, I modify equation (3), changing the dependent variable to the firm’s increase in software

capitalization rate.

INC-CAP% = β0 + β1C1t+ β2C3t + β3CCAPEXt + β4CSALESt + β5SIZEt + β6LEVt +

β7DISTt +β8RDINTt + YearDummies ε (4)

The variable INC-CAP% is an indicator variable that equals 1 if the firm increased its software capitalization percentage (annual capitalized software cost as a percentage of total R&D expenditure) and 0 otherwise. I use capitalization percentage as the dependent variable rather than annual capitalized cost because it is a better measure of earnings management. Consider that an increase in capitalized cost from year t-1 to year t would not indicate earnings

(20)

20

capitalizers are more likely to increase their capitalization of software and thus β1 and β2 should

be negative for both earnings benchmarks.

Section 4: Results

4.1 Descriptive Statistics

Table 1 provides the descriptive statistics of my sample. Overall, expensers in my sample are significantly larger than the capitalizers and have greater average values in all categories (Total R&D, R&D Intensity, Earnings, etc.) except for leverage where they have lower average values than capitalizers. It is important to note that the characteristics of my sample firms are considerably different from those of the sample of Oswald and Zarowin (2004). Their firms have considerably higher leverage ratios and significantly lower R&D expense, earnings, sales, total assets, and market values. These characteristics offer potential insights into why my overall results contradict those of Oswald and Zarowin.

Table 2 provides a summary of the percentage of firms who reduce R&D expenditure, with the results categorized by groups. For both earnings benchmarks, the results for the expensing firms are not intuitive based on what my framework determines to be logical. I would expect that Group 2 has the highest percentage of expensers that cut R&D, but this is not the case. Using the zero earnings growth benchmark, for capitalizers the breakdown is more consistent with my expectation, as there is no clear pattern of difference across groups in terms of reducing R&D. However with the zero earnings level benchmark, Group 2 has a lower

percentage of firms who cut R&D firms compared to Groups 1 and 3 which is not intuitive given my logical expectations.

When comparing these basic statistics to those in Oswald and Zarowin (2004), a few key differences are apparent. First, their results were consistent with expectations while mine were not. In their sample, expensers in Group 2 had a higher percentage of firms who cut R&D and for capitalizers the results across groups were not significantly different. Additionally, a much higher percentage of firms in Oswald and Zarowin’s sample cut R&D expenditure. This

percentage varied from 30-47% across the different group breakdowns. In comparison, the percentage of my firms who cut R&D expenditure varied from 10-35%. Taken alone, these

(21)

21

percentages mean little, but do suggest that overall the firms in Oswald and Zarowin’s sample tend to reduce R&D to a greater extent. The next section contains my regression testing that further studies R&D cuts across groups.

4.2 Test Set 1 Results

Results for Expensers

For logit regression equation (2) using the zero level earnings benchmark, both β1 and β2 are

insignificant. β2 is negative (as expected) and near significance (p = .11, one tail). Still, I cannot

conclude that Group 2 significantly differs from either Group 1 or 3 in cutting R&D spending. Further, using the zero growth earnings benchmark, both coefficients go against my

predictions. β1 is again not significant, so I cannot conclude that Group 2 is significantly

different from Group 1, and β2 is significant but positive (β2 = +1.791, p = 0.00), suggesting that

Group 3 firms are more likely to cut R&D expenditure (to achieve the zero growth benchmark) than Group 2, the opposite of my expectation. Based on these results, I reject Hypothesis 1a and find no evidence that U.S. software expensers use R&D for real earnings management.

For expensers, only two of my controls variables are found to be significant in the decision to cut R&D spending: CSALESt (change in sales) and SIZEt (firm size). For the zero level earnings

benchmark, both variable coefficients are significantly negative (β4 = -10.289, p = .001; β5 =

-.578, p = 0.048). This result is consistent with my predictions. Firms with fewer funds available are less likely to invest in R&D, and larger firms have fewer incentives to cut R&D. For the zero earnings level benchmark, CSALESt is again significantly negative (β4 = -13.093, p = 0.003), but

SIZEt is not.

My results have some similarities with those of Oswald and Zarowin (2004), who also do not find significant differences in R&D reduction between Group 1 and Group 2 (the suspect group). They offer some explanation for this result, attributing the (higher than expected) cut in R&D for Group 1 firms to the fact that these firms have decreased profitability. Group 1 firms may be attempting to reduce the magnitude of their earnings loss. However, unlike my results, Oswald and Zarowin do find that Group 3 firms cut R&D to a lesser extent than Group 2. My sample’s Group 3 firms have a higher tendency than Group 2 to cut R&D based on the zero earnings

(22)

22

growth benchmark, despite the fact that these firms will meet the benchmark without cutting R&D. One possible explanation for this finding is that these firms may be motivated by other incentives to manage earnings through R&D cuts. Apart from the two earnings benchmarks I use, the other main earnings benchmark that incentivizes firm behavior is analyst expectations (Degoerge, Patel, and Zeckhauser 1999). Since Group 3 firms do not face as much pressure to achieve positive earnings or positive earnings growth (they are well above these thresholds), it follows that they might be more motivated by meeting analyst earnings forecasts (or meet internal benchmarks for that matter) and cut R&D to meet these expectations. Another explanation is simply that my regression model may not capture all characteristics of a firm or its business environment that affect R&D decisions.

Results for Capitalizers

For equation (2), using both earnings benchmarks, β1 and β2 are insignificant. Consistent

with my prediction, this shows that there is no significant difference between how firms in Group 1, 2, or 3 cut total R&D expenditure. I therefore fail to reject Hypothesis 1b and conclude that capitalizers in the U.S. software industry do not manage earnings through real R&D

transactions (i.e. cutting total R&D spending). This is consistent with the finding of Oswald and Zarowin (2004), who find the same lack of significant differences across groups of capitalizers.

In terms of control variables, CSALESt is significantly negative (β4 = -5.851, p = 0.056) when

using the zero earnings growth benchmark, as predicted (refer to above Expensers section). In addition, LEVt is significantly positive for both the zero earnings level benchmark (β6 = +7.694, p

= 0.072) and zero earnings growth benchmark (β6 = +8.573, p = 0.034), meaning that firms with

higher leverage ratios are more likely to cut R&D expenditure. This suggests that capitalizers cut R&D expenditure to avoid debt covenant violations. Capitalizers also have higher average and median leverage ratios than expensers, as shown in Table 1, which supports this idea. For the zero earnings level benchmark, DISTt is significantly negative (β7 = -0.249, p = 0.089), which goes

against my expectations. As DISTt represents the percentage cut in R&D expenditure needed to

achieve the earnings benchmark, these results are counterintuitive, as they suggest that firms which have a greater R&D cut needed are less likely to cut R&D.

(23)

23

4.3 Test Set 2 Results

For regression equation (3), using the zero earnings level benchmark, the C1 coefficient β1 is

significantly negative (β1 = -1.579, p = 0.147, one-tail) and the C3 variable β2 is insignificant. The

β1 result is as predicted and suggests that Group 2 firms are more likely cut R&D expenses than

Group 1 firms. However, the β2 result shows that the likelihood to cut R&D expenses is not

significantly different across firms in Group 2 and Group 3, against my expectations. Using the zero growth earnings benchmark β1 is insignificant, against my prediction, and β2 is significantly

positive (β2 = +1.922, p = 0.028), also against my prediction, suggesting that Group 3 firms

actually cut R&D expense more than Group 2 firms. As I mentioned in the section above, this could indicate that these firms manage earnings based on a different earnings benchmark (e.g. analyst forecasts). As only one of my predictions regarding the group variable coefficients is supported, I reject Hypothesis 2a and conclude that capitalizers do not cut R&D expenses as a form of real earnings management in the manner expected.

Of my control variables in equation (3), CSALESt, LEVt, and DISTt have significant explanatory

power over the decision to cut R&D expenses. For the zero earnings growth benchmark, CSALESt is significantly negative (β4 = -8.589, p = 0.067), consistent with my expectation. For

both earnings benchmarks, LEVt is significant and positive (β6 = +5.540, p = 0.002; β6 = +6.999, p

= 0.013) as expected. For both benchmarks DISTt is significant and negative (β7 = -0.377, p =

0.063; β7 = -0.496, p = 0.020) going against expectations.

For regression equation (4), neither my firm group variables nor control variables are significantly related to the dependent variable, INC-CAP%. I must reject Hypothesis 2b and conclude that there is no evidence that U.S. software firms use capitalization as a form of earnings management. While my results are not what I predicted, they are consistent with the results for capitalizers for equations (2) and (3). If these firms do not cut total R&D nor R&D expense, it follows that they would not increase capitalization (Total R&D = R&D expense + R&D capitalization).

Overall, my results for equations (3) and (4) are not consistent with the findings of Oswald and Zarowin (2004) who found that U.K. capitalizers in Group 2 were significantly more likely to cut R&D expenses to achieve earnings benchmarks. Further, they found that Group 2

(24)

24

capitalizers made up for this cut in R&D expense with an increase in capitalization. These two findings showed a substitution effect between real and accrual earnings management. My results are contrary in both cases. From this, I conclude that the results of Oswald and Zarowin do not hold true in my more focused sample, and that U.S. software capitalizers do not utilize R&D accounting choice to meet the two earnings benchmarks. As stated earlier, this could indicate that these firms instead manage earnings based on internal benchmarks or expectations set by analysts. Also, earnings may not be the most important measure of performance for software firms, who may care more about maintaining R&D levels and

ensuring the perceived future value of the firm and its projects remains high among investors. I also cannot discount the possibility that my results may be insignificant due to my model failing to adequately capture determinants of the capitalization choice (equation 4 has a Chi2

probability of 0.88) or due to the relatively small sample size of capitalizers (n = 72).

Section 5: Conclusion

My thesis investigates whether the decision to capitalize or expense R&D costs impacts if and how firms manage earnings through R&D spending. Previous earnings management literature regarding U.S. firms offers substantial evidence that firms cut R&D in order to

manipulate earnings. In settings outside the U.S. that allow for R&D capitalization, many studies have concluded that the option to capitalize also leads to earnings management. I specifically study software firms because ASC 985 allows for the capitalization of software development costs and therefore provides the only example of R&D capitalization within U.S. GAAP accounting. I find no evidence that U.S. software firms use either the reduction in R&D expenditure or the increase in software capitalization to manage earnings to achieve benchmarks, specifically positive earnings and positive earnings growth.

To my knowledge, my study is the first to focus on earnings management and R&D

accounting choice (expensing vs. capitalizing) in U.S. software firms. Overall, my results are not consistent with prior literature. Firms in my sample do not seem to use discretionary R&D spending as a means of real earnings management, going against previous conclusions by Baber, Fairfield, and Haggard (1991), Bushee (1998), and many others. Capitalizers in my

(25)

25

sample do not seem to use the capitalization option for earnings management purposes. This is inconsistent with Oswald and Zarowin (2004) and directly contradicts Cifti (2010) who provided the only previous evidence that suggests the possibility of U.S. software firms using

capitalization to avoid losses. In this way, my thesis contributes to the larger body of earnings management literature and suggests that conclusions from these prior studies, which had a larger scope of firms, may not hold true for this specific sections of the U.S. software industry.

There are a few possible explanations for the contradiction between my results and those from Oswald and Zarowin (2004) and Cifti (2010). The main difference with Cifti’s research is the age difference of the data. My research is much more recent (2012-2014 data compared to 1981-1990 for Cifti) and the industry and its environment may have changed. My sample is also very different from that of Oswald and Zarowin, who studied U.K. firms and did not focus on the software industry. As mentioned in section 4.1, the characteristics of their sample firms contrast greatly with those of my own sample, having much higher average leverage ratios and much lower average R&D, earnings, and sales (among other measures). These different

characteristics could create different incentives for those firms to manage earnings. Further, the nature of the software industry may present a different set of incentives regarding how or whether firms focus on earnings benchmarks as opposed to focusing on R&D activity and its perceived value to investors.

If my results are indicative of the software industry as a whole, concerns held by shareholders and regulators regarding earnings manipulation by managers can be at least partially mitigated, or focused instead on incentives other than beating the benchmarks I use. Further, in the larger context of possible convergence efforts between FASB and IASB, my results provide some evidence that the implementation of R&D capitalization in U.S. GAAP (akin to what is present in IAS 38) need not be met with the cynical view that such a change would allow for a new earnings management tool.

There are several potential limitations to my research. First, I note the possibility for errors in my hand-collected data. Due to the complicated nature of how software development accounting is reported, differences in practice across financial statements, and the potential for human error, my annual capitalized software cost data may contain mistakes (Mulford and

(26)

26

Roberts 2006). Additionally, because my thesis focused on a specific section of the software industry, my results might not be generalizable across the entire body of U.S. firms who capitalize software. I only use firms classified with SIC 7372 as their main activity. Even just within the software industry, there are many other firms with secondary activities that fall under the same category or with main activities classified as similar SIC codes. Expanding the scope of my research to include all software firms might make my results significant. Lastly, my sample size is also relatively small compared to prior research, specifically compared to Oswald and Zarowin (2004) whose model I use. My results may be more significant with an increased sample size.

(27)

27

Section 6: Reference List

Aboody, D. and Lev, B. (1998). The value relevance of intangibles: the case of software capitalization. Journal of Accounting Research, 36, pp. 161–191.

Baber, W.R., Fairfield, P.M. and Haggard, J.A. (1991). The effect of concern about reported income on discretionary spending decisions: The case of research and development. The

Accounting Review, 86(4), pp. 818-829.

Burgstahler, D. and I. Dichev, (1997). Earnings Management to Avoid earnings Decreases and Losses. Journal of Accounting and Economics, 99-126.

Bushee, B.J. (1998). The influence of institutional investors on myopic R&D investment behavior. The Accounting Review, 73(3), pp. 305-333.

Chan, H., Faff, R., Charghori, P. and Ho, Y.K. (2007). The relation between R&D intensity and future market returns: Does expensing versus capitalization matter? Review of

Quantitative Finance and Accounting, 29(1), pp. 25-51.

Ciftci (2010). Accounting Choice and Earnings Quality: The Case of Software Development.

European Accounting Review, 19:3, 429-459.

Cooper, J. C., and F. H. Selto. (1991). An experimental examination of the effects of SFAS No. 2 on R&D investment decisions. Accounting, Organizations and Society 16 (3): 227–242. Dechow, P. M. and Skinner, D. J. (2000). Earnings management: Reconciling the views of

accounting academics, practitioners, and regulators. Accounting Horizons, 14(2), pp.

235-250.

Degeorge, F., Patel, J., and Zeckhauser, R. J. (1999). Earnings Management to Exceed Thresholds. Journal of Business, Vol. 72, No. 1.

Dihn, T., Kang, H., and Schultze, W. (2014). Capitalizing Research & Development: Signaling or Earnings Management? Working paper, University of St. Gallen.

Eccher, E. (1995). The value relevance for software capitalized costs. Unpublished Working

Paper, Northwestern University.

García Osma, B. and Young, S. (2009). R&D expenditure and earnings targets. European

Accounting Review, 18(1), pp. 7-32

Graham, J.R., Harveya, C.R. and Rajgopal, S. (2005). The economic implications of corporate financial reporting. Journal of Accounting & Economics, 40(1-3), pp. 3-73.

Lev, B. and Sougiannis, T. (1996). The capitalization, amortization, and value-relevance of R&D.

(28)

28

Markarian, G., Pozza, L. and Prencipe, A. (2008). Capitalization of R&D costs and earnings management: Evidence from Italian listed companies. The International Journal of

Accounting, 43(3), pp. 246-267.

Mohd, E. (2005) Accounting for software development costs and information asymmetry. The

Accounting Review, 80, pp. 1211–1231.

Mulford, C., and J. Roberts. 2006. Capitalization of Software Development Costs: A Survey of Accounting Practices in the Software Industry. Working paper, Georgia Institute of

Technology.

Nelson, M., Elliot, J., and Tarpley, R. 2003. How Are Earnings Managed? Examples from Auditors. Accounting Horizons, Supplement 2003, pp. 17-35.

Opler, T., L. Pinkowitz, R. Stulz and R. Williamson (1999). The Determinants and

Implications of Corporate Cash Holdings. Journal of Financial Economics, 52(1), pp.3-46. Oswald, D.R. and Zarowin, P. (2004). Capitalization vs Expensing of R&D and Earnings

Management. Unpublished Working Paper, London Business School and New York

University.

Oswald, D.R. and Zarowin, P. (2007). Capitalization of R&D and the informativeness of stock prices. European Accounting Review, 16(4), pp. 703-726.

Seybert, N. (2010). R&D Capitalization and Reputation-Driven Real Earnings Management. The

Accounting Review, 85(2), pp. 671-693.

Wiedman, C. (1996). The Relevance of Characteristics of the Information Environment in the Selection of a Proxy for the Market’s Expectation for Earnings: An Extension of Brown, Richardson, and Schwager (1987). Journal of Accounting Research, 34(2),

pp.313-324.

Zang, A. Y. (2012). Evidence on the Trade-Off between Real Activities Manipulation and Accrual-Based Earnings Management. Accounting Review 87 (2), pp. 675-703.

(29)

29

Table 1 Descriptive Statisticsa

Expensers Capitalizers

Mean Median Mean Median

Total R&D 285.83 28.63 122.98 12.05

R&D expense 285.83 28.63 99.23 8.98

Capitalized software cost - - 23.75 1.31

Capitalization percentage - - 0.24 0.11

R&D intensity 0.27 0.19 0.21 0.19

Earnings before interest & tax 521.10 1.16 129.39 3.92 Earnings before I, T, & R&D 235.27 -19.40 6.41 -4.52

Total Assets 3620.72 282.04 1542.81 122.15

Market value of equity 7871.96 705.21 1953.25 231.81

Capital expenditure 80.37 7.04 15.63 1.77

Leverage 0.04 0.00 0.06 0.01

Sales 1866.38 170.82 668.59 100.53

Growth (% increase sales) 2.45 0.12 0.07 0.05

Number of observations 250 72

a_{My sample includes U.S. firms classified under SIC 7372. My observations are from fiscal years}

2012-2014. Observations are classified as expensers if the firm does not capitalize any software development costs in that year and are classified as a capitalizer if it does.

Total R&D consists of the sum of R&D expense and capitalized software costs. R&D intensity is

calculated as (Total R&D/Sales). Capitalization percentage is divided as (Capitalized software costs/Total R&D). Leverage is calculated as (Long-term debt) / (Equity + Long-term debt). Growth is measured as the percentage increase in Sales compared to the prior year.

(30)

30

Table 2

Summary of R&D cutting across groupsa Panel A: Zero Earnings Level Benchmarkb

C1 C2 C3

Expensers

Number of Observations 199 24 27

% which cut R&D 25% 21% 4%

Capitalizers

Portion which cut R&D 28% 10% 25%

Panel B: Zero Earnings Growth Benchmarkc

C1 C2 C3

Expensers

% which cut R&D 19% 13% 35%

Capitalizers

Portion which cut R&D 25% 22% 30%

a_{This table summarizes the percentage of firms which reduce R&D expenditure from the prior year}

(Oswald and Zarowin 2004).

b_{Firms in Panel A are grouped based on the following criteria mentioned in Section 3 for the zero}

earnings level benchmark. This grouping is the same in all of the regression equations to follow: Group 1: EBRDt < 0

Group 2: 0 ≤ EBRDt ≤ RDt-1 Group 3: RDt-1 < EBRDt

c_{Firms in Panel B are grouped based on the following criteria mentioned in Section 3 for the zero earnings}

growth benchmark. This grouping is the same in all of the regression equations to follow: Group 1: ∆EBRDt < -RDt-1

Group 2: -RDt-1 ≤ ∆EBRDt ≤ 0 Group 3: 0 < ∆EBRDt

(31)

31

Table 3

Logit Regression 2 for Expensersa

VARIABLES Level Benchmark Growth Benchmark

RDCut . . (.) (.) C1 -0.163 -1.094 (0.787) (0.396) C3 -1.336 1.791*** (0.220) (0.000) CCAPEX -0.725 -0.452 (0.230) (0.523) CSALES -10.289*** -13.093*** (0.001) (0.003) SIZE -0.578** -0.420 (0.048) (0.229) LEV 2.482 -0.091 (0.346) (0.974) DIST -0.104 -0.255 (0.304) (0.191) RDInt -0.131 -0.384 (0.948) (0.868)

YearDummies Yes Yes

Constant 0.424 -0.883

(0.694) (0.532)

Observations 250 250

Pseudo R2 0.3103 0.3852

Robust pval in parentheses *** p<0.01, ** p<0.05, * p<0.1

(32)

32

Table 3 – Continued Logit Regression 2 for Expensers

a_{This table displays the coefficients and p-values for the following regression equation (Oswald and}

Zarowin 2004):

RDCutt = β0 + β1C1t + β2C3t + β3CCAPEXt + β4CSALESt + β5SIZEt + β6LEVt

+ β7DISTt +β8RDINTt + YearDummies+ ε

Where RDCutt is an indicator variable that equals 1 if a firm decreases R&D expenditure between year t-1

and year t. C1 and C3 are indicator variables that equal 1 if a firm is in Group 1 and 3, respectively, and zero otherwise.

And where the control variables are defined as follows and are relevant for equations 2, 3, and 4:

CCAPEX (change in Capital expenditure) = log of current capital expenditures – log of capital

expenditures in t-1

CSALES (change in Sales) = log of current year sales – log of sales in t-1

SIZE = log of market value of equity at fiscal year end (shares outstanding * share price) LEV (leverage) = (Long-term debt) / (Equity + Long-term debt)

DIST (% R&D cut needed to achieve benchmark)

= (Earnings before tax and R&D / previous year R&D) – 1 OR for growth benchmark

= (Change in earnings before tax and R&D / previous year R&D) – 1

(33)

33

Table 4

Logit Equation 2 for Capitalizersa

RDCut . . (.) (.) C1 0.139 -0.274 (0.929) (0.811) C3 1.897 0.912 (0.405) (0.290) CCAPEX -0.975 -0.638 (0.339) (0.502) CSALES -5.002 -5.851* (0.215) (0.056) SIZE -1.017 -0.891 (0.115) (0.254) LEV 7.694* 8.573** (0.072) (0.034) DIST -0.249* -0.206 (0.089) (0.135) RDInt -0.328 -0.364 (0.784) (0.735)

YearDummies Yes Yes

Constant -1.041 -1.242

(0.698) (0.511)

Observations 72 72

Pseudo R2 0.2778 0.2695

a_{This table displays the results for the same regression equation and variables as Table 3, but for}

(34)

34

Table 5 Logit Equation 3a

RDExpCut . . (.) (.) C1 -1.579 -2.324 (0.147)*b (0.374) C3 0.351 1.922** (0.774) (0.028) CCAPEX -1.791 -0.682 (0.186) (0.604) CSALES -3.697 -8.589* (0.389) (0.067) SIZE 0.012 0.473 (0.977) (0.397) LEV 5.540*** 6.999** (0.002) (0.013) DIST -0.377* -0.496** (0.063) (0.020) RDInt 1.844 1.692 (0.157) (0.213)

YearDummies Yes Yes

Constant -1.216 -4.542**

(0.358) (0.030)

Observations 72 72

Pseudo R2 0.2043 0.2579

a_{This table displays results for the following logit regression equation (relevant only for capitalizers):}

RDExpCutt = β0 + β1C1t + β2C3t + β3CCAPEXt + β4CSALESt + β5SIZEt + β6LEVt

Where RDExpCut is an indicator variable that equals 1 if the firm decreased R&D expense relative to the prior year and 0 otherwise and the remaining variables are as defined in Table 3.

(35)

35

Table 6

Regression Equation 4a

INC-CAP% . . (.) (.) C1 -1.078 -0.169 (0.236) (0.863) C3 -0.635 0.174 (0.633) (0.692) CCAPEX -0.961 -0.917 (0.276) (0.366) CSALES -4.373 -5.226 (0.220) (0.155) SIZE -0.100 0.030 (0.745) (0.929) LEV -1.595 -1.576 (0.550) (0.564) DIST -0.035 -0.007 (0.814) (0.963) RDInt -1.302 -1.468 (0.318) (0.286)

YearDummies Yes Yes

Constant 1.768 0.601

(0.247) (0.628)

Observations 72 72

Pseudo R2 0.0704 0.0554

a_{This table displays results for the following logit regression equation (relevant only for capitalizers):}

INC-CAP% = β0 + β1C1t + β2C3t + β3CCAPEXt + β4CSALESt + β5SIZEt + β6LEVt

Where INC-CAP% is an indicator variable that equals 1 if the firm increased its software capitalization percentage and 0 otherwise. and the remaining variables are as defined in Table 3.

Software development capitalization and earnings management : a study of U.S. software firms

MSc Accountancy & Control

Thesis