Effects of Real Earnings Management: Evidence from Patents

Transcription

1 Effects of Real Earnings Management: Evidence from Patents Qin Lian a, Qiming Wang b, Zhaohui Xu c a Department of Finance, Louisiana Tech University, Ruston, LA 71272, , qlian@latech.edu b Department of Finance, Louisiana Tech University, Ruston, LA 71272, , qwang@latech.edu c Department of Accounting, University of Houston-Clear Lake, Houston, TX 77058, , xuzhao@uhcl.edu This draft: August, 2014 Abstract We examine the effect of R&D REM, a form of real earnings management (REM) by cutting research and development (R&D) expenditures abnormally to meet earnings targets, on firms subsequent innovation productivity measured by the scale and the novelty of patents. In the within-firm analysis, we find that R&D REM adversely affects the number of innovations, technological importance, and novelty of innovation in the subsequent periods. In cross-sectional analysis, after controlling for other related factors, we find that, relative to the matched sample firms that likely cut R&D expenditures for reasons other than manipulating earnings, R&D REM firms produce fewer innovations with significantly less novelty.

2 1. Introduction U.S. accounting rules require that R&D expenditures are immediately and fully expensed. This creates a potential managerial myopic problem, in which managers sacrifice R&D expenditures to maintain short-term earnings targets. Baber, Fairfield, and Haggard (1991) and Roychowdhury (2006) document that managers manipulate R&D expenditures to meet shortterm earnings. Porter (1992) suggests that this kind of myopic investment behavior has caused U.S. firms to fall behind German and Japanese firms in term of technological development. In this paper, we attempt to explore the effect of manipulating R&D expenses on innovation productivity. We focus on firms cutting R&D expenditures to meet earnings targets (R&D REM) one of the common types of real earnings management activities. 1 Graham et al. (2005) report that most of managers would willingly sacrifice positive NPV investment opportunities in order to meet earnings targets. 2 Bhojraj et al. (2009) document that real earnings management leads to suboptimal decisions on operations and has negative impact on future performance and firm value. On the other side, others argue that such deviation is temporary, which should not have significant effect on firms long-term value. This paper tests these two competing views by examining whether there is a decline in the quantity and quality of the firms patent production in the periods subsequent to the R&D REM activity. 1 Gunny (2010) groups REM activities into four categories: (1) decreasing R&D expense (R&D REM); (2) decreasing SG&A expense (SG&A REM); (3) timing sale of fixed assets (asset REM), and (4) boosting sales and/or decreasing COGS (production REM). 2 In their Appendix A, Graham et al. (2005) provide the following quote: A CFO at a research-intensive firm indicates the role of investment triggers based on whether the firm s actual EPS would fall within or outside the range of earning guidance. If the actual EPS comes in below the lower end of the guided EPS range, the disinvestment trigger goes off and the firm eliminates or postpones R&D spending (on positive NPV R&D projects) until a later time. We asked why the firm would not take all the positive NPV R&D projects, regardless of whether the reported EPS falls in the guided range or not. The CFO responded that the market has certain expectations about EPS growth from year to year and there is a trade-off between delivering EPS growth to the market and investing in R&D projects that would payoff in the long run. 1

3 Following prior literature (Roychowdhury 2006; Gunny 2010), we identify R&D REM firms as firms with negative abnormal R&D expenses and meeting one of the two earnings benchmarks (i.e. to avoid reporting a loss or a decrease over prior period earnings). We measure abnormal R&D expenditures as the change in R&D expenses not due to changes in investment opportunities. We use the number of patent applications and number of patent citations to measure the quantity and quality of the R&D output. Patents receiving more citations are considered to have more technological or economic importance. Patents that cite other patents in a broader array of technology areas are viewed as having more originality. Patents that are themselves cited by a more technologically dispersed array of patents are viewed as having greater generality (Hall et al. 2001; Lerner, Sorensen and Strӧnberg 2011). We conduct both within-firm analysis and matched sample analysis to investigate the effect of R&D REM on patent production. Firms may cut R&D expenditures for business considerations other than managing earnings, such as relying more on outsourcing and M&A than on in-house R&D, responding to decreasing returns on current R&D investments. To distinguish the effect of R&D REM from reduction of R&D due to changes in firms business strategy, we construct two matched samples: (1) firms that have negative abnormal R&D spending and miss earnings targets (i.e. the R&D Miss sample), and (2) firms that have negative abnormal R&D spending and beat earnings targets by a large margin (i.e. the R&D Beat sample). The firms are matched to the R&D REM firms by year, industry, and size to produce a group with similar business profiles. In the within-firm analysis, we find that the number of patent applications filed by R&D REM firms and their quality decrease in the three years following the REM event. However, there is no such pattern for Beat or Miss firms. Following R&D cut, Beat 2

4 and Miss firms file less patents and quality of patents doesn t change. In cross-sectional regressions that control for firm size, return on assets, prior year R&D expenditure, leverage, and Tobin s Q, the quantity and quality of patents filed by the matched sample firms exceeds that of R&D REM firms significantly. Specifically, the number of patents, the average number of citations, the originality, and the generality of patents filed by R&D REM firms in the three year period subsequent to the REM event are significantly and consistently lower than those measures for patents of R&D Beat and Miss firms. Overall, our test results indicate that managing earnings by cutting R&D expenditures abnormally leads to a decline in the quantity and quality of future patent applications and potentially undermines firms technological innovation capacity and long term competitiveness. Unlike prior studies that investigate the effect of REM by examining stock returns and operating performances, we focus on the effect of R&D REM on patent production for two reasons. First, prior studies (Bhojraj, Hribar, Picconi, and McInnis 2009; Gunny 2010; and Taylor and Xu 2010;) assess the impact of real earnings management on future performance by focusing on firms operating performance such as earnings and cash flows. These overall operating performance metrics are subject to the effect of many factors. Even if the studies attempt to control for the commonly known factors related to operating performances including industry, size, prior earnings and cash flows, it is not feasible to control for all of the factors that affect firms operations. As a result, the tests would be subject to the omitted variables problem and it is hard to tease out all the other factors and specifically identify the effect of real earnings management on firms future operations, especially when the proxies for real earnings management are found to be correlated to operating performance. 3

5 Second, patent production is a direct function of the amount of R&D expenditures and the quantity and quality of patents is a natural proxy for firms technological innovation capacity. Prior studies (Bound et al. 1984; Crepon et al. 1998; Czarnitzki et al. 2009) find a significant positive association between firms R&D spending and patent applications. Focusing on the quantity and quality of patents subsequent to R&D real earnings management activities, we are able to specifically identify the effect of cutting R&D expenses on firms ability to make technological innovation and maintain long term competitive edge. This paper contributes to the literature on managerial discretion over R&D expenditures (Baber et al. 1991; Bushee 1998; Bens et al. 2002). Baber et al. (1991) and Bens et al. (2002) show that managers cut R&D and capital expenditures to increase earnings in order to achieve earnings targets. Bushee (1998) finds that institutions with large, long-term holding discourages the behavior of reducing R&D to reverse a decline in earnings, which implies that this myopic investment behavior is hurting long-term firm value. We provide direct evidence about the adverse consequence of cutting R&D investment. We show that the short-termism of cutting R&D expenditures reduces the company s ability to innovate, hence undermining their competitiveness in the product markets. Our study also contributes to the real earnings management literature by providing new evidence on the impact of R&D REM on firms future business operations from their innovation output. Research on real earnings management has concentrated largely on accounting and stock returns. Prior literature has mixed empirical evidence on the effect of real earnings management. On one side, Taylor and Xu (2010) find no evidence of a significant decline in the subsequent operating performance. Gunny (2010) show that firms engaging in real earnings management to 4

6 just beat earnings benchmarks perform better subsequently, suggesting that real earnings management is not opportunistic. On the other side, Bhojraj et al. (2009) show firms that beat analyst forecasts through real earnings management have inferior operating performance and stock market performance in the subsequent three years compared to firms that miss analyst forecasts without earning management. To provide new evidences from a different perspective, we use a direct measure of innovation productivity to examine the effect of R&D REM and find that firms that reduce R&D spending to meet earnings targets experience a significant decline in the quantity and quality of their subsequent patents. The result indicates that real earnings management could have a significant negative effect on firms future competitiveness. Our finding suggests that this type of real earnings management is opportunistic with adverse effect on future operations, consistent with Bhojraj et al. s (2009) empirical finding and Graham et al. s (2005) survey evidence that management are willing to sacrifice long term economic value to meet short term financial reporting goals. The rest of the paper is organized as follows. Section 2 discusses relevant literature and research question. Section 3 describes data and research methodology. Section 4 presents the test results. Section 5 concludes the paper with discussions on implications for future research. 2. Hypotheses Development Patent data has been widely used to gauge the innovative output of firms R&D expenditures. Patenting offers firms an effective way to protect their innovative knowledge gained through R&D expenditures (Holbrook 1992; Papadakis 1993). Prior research has found a strong positive association between R&D input and the number of patent applications filed and granted (Bound 5

7 1984; Crépon et al. 1998; Prodan 2005). We test two hypotheses to assess the consequences on patent production by firms cutting R&D expenditures with different intentions. The first hypothesis tests the view that there is no long-term effect of the temporary R&D cut. While managers admit to decreasing R&D expense to meet earnings targets, managers often indicate that the cut in R&D is temporary to meet short-term earning goals and that firms would restore these expenditures to their normal levels in subsequent periods when possible. Such one time deviations would have insignificant effects on patent quantity and quality, if R&D REM firms could meet earning targets by either postponing R&D spending until later time or only eliminating wasteful R&D projects such as R&D projects with the lowest impact on the firm, i.e., the R&D projects with the lowest probability of turning into innovative products. Hypothesis 1a: R&D REM firms experience no decline in the number and quality of future patent applications in the subsequent period. On the other hand, there is growing evidence that managers are willing to sacrifice economic value to meet short-run earnings target and managers understand the consequence of myopic action. Jensen (2005) notes that when real operating decisions that would maximize value are compromised to meet market expectations, real long-term value is being destroyed. Bushee (1998) finds that managers are less likely to cut R&D to reverse earnings decline in the presence of high institutional ownership, which suggests that these activities are unlikely to increase longterm firm value. Bhojraj, Hribar, Picconi, and McInnis (2009) document that firms meeting earnings target with income-increase accruals and discretionary expenditure cuts underperform the firms missing earnings target and managers significantly increase insider selling following reducing discretionary expenditures (including R&D). If R&D REM firms could only meet 6

8 earning targets by eliminating not only wasteful R&D projects but also productive R&D projects, such actions could inflict a significantly negative impact on these firms future innovation output. We posit that, when managers use R&D cut as a tool to meet a short-term earnings target, such opportunistic manipulation of R&D expenditures may be more likely to lead to cut beyond wasteful R&D projects and hurt firms innovation ability. This view of myopic investment action leads to the competing hypothesis. Hypothesis 1b: R&D REM firms experience a significant decline in the number and quality of future patent applications in the subsequent period. Not all firms cut R&D for opportunistic purposes. Firms may cut R&D expenditure out of valid business considerations, such as decreasing investment returns and adjustment of the balance between in-house R&D and acquisition of outside technologies. Many firms beat by a large margin or miss earnings targets while reducing R&D expenditures. The R&D cut by the R&D Beat/Miss firms may reflect valid business decisions to dispose of some unproductive projects rather than to meet short-term earnings targets and thus would be unlikely to have negative impact on these firms long run innovation outputs. To distinguish between the alternative incentives to cut R&D investments, we compare the patent production of R&D REM firms with that of the R&D Beat and Miss firms following the R&D cut. We expect R&D REM firms to experience more adverse effect in future innovation outputs than the R&D Beat/Miss firms. Hypothesis 2: R&D REM firms experience a bigger decline in the number and quality of future patent applications than R&D Beat and R&D Miss firms. 7

9 In brief, Hypothesis 1 examines the changes in patent production around the R&D REM event, while Hypothesis 2 attempts to distinguish the effect of R&D cut on patent production between the R&D REM firms that cut R&D for opportunistic purposes and the matched sample firms that cut R&D likely for other reasons. 3. Research Methodology 3.1 Identifying firms engaging in R&D REM We use annual data since quarterly reporting on R&D is sparse. We follow prior research (Berger 1993; Roychowdhury 2006; Gunny, 2010) to estimate the normal level of R&D expenditures controlling for motivations to cut R&D that are unrelated to myopic investment behavior: (1) where (COMPUSTAT data items in brackets): RD = R&D expense [Data46], A = Total assets [Data6], MV = The natural log of market value [Data199*Data25], Q = Tobin s Q [((Data199*Data25) + Data130 + Data9 + Data34) Data6], and INT = Internal funds available [Data18 + Data46 + Data14]. We exclude financial firms (SIC ) and utility firms (SIC ) from our sample because firms in these regulated industries tend to follow different accounting rules. We use all other firm-years with information available in COMPUSTAT to estimate abnormal R&D expenditure. We estimate Equation 1 for each industry-year (defined by 2-digit primary SIC code) with more than 10 firms. Firms are defined as having negative abnormal R&D 8

10 expenditures if the actual R&D expenditures are lower than the normal level estimated with Equation 1. Following Gunny (2010), we then define R&D REM as firms that have negative abnormal R&D expenditures and meet one of the two earnings benchmarks: (1) just avoid losses and report small profits or (2) report small growth in earnings over the prior year. We define firms reporting small profits as those with the ratio of net income divided by beginning total assets equal to or larger than zero, but less than Firms reporting small earnings growth are defined as those with the ratio of the change in net income over prior year divided by beginning total assets equal to or larger than zero, but less than To facilitate a comparison of patent production pre- and post-r&d REM, we keep only those firms that do not engage in real earnings management in the three years prior to and the three years after the REM event. Further, we restrict our sample to firms with at least one successful patent application in the event year ±3 period. The NBER patent database only includes the patents that are eventually granted. Since the NBER patent database covers patent information from 1976 to 2006 and there is an average of two year lag from patent application to patent grant (Hall et al. 2001), we restrict our sample period for R&D REM firms in 1979 to 2000 to ensure each firm has patent information for three years of data before and after the event year. We end R&D REM sample at 2000 to ensure that the granted patent applications in the three years ( ) after the REM of a R&D REM event in 2000 are included in the NBER database. 5 We identify 1,126 firm years that meet this criteria during 1979 to If we end R&D REM sample at 2003, there would be an artificially lower number of patent applications in years +2 and +3 for REM event year 2002 and 2003 due to the two year lag from patent application to patent grant. 9

11 3.2 Measuring patent quantity and quality The patent variables are constructed from the latest version of the NBER patent database, which was initially created by Hall, Jaffe, and Trajtenberg (2001) and contains updated patent and citation information from 1976 to We follow the literature to use only utility patent grants and exclude other patent awards, such as design or reissue awards. Utility patents represent about 99% of all awards (Lerner, Sorensen, and Stromberg 2011). We use the number of patent applications to measure the quantity of patent production. Since patent filing varies over time and across technology classes, we calculate the scaled patent count following Bernstein (2012) and Bena and Li (2011). First, we compute the average number of applied patents of all firms for each technology class defined by U.S. Patent and Trademark Office (USPTO) in each year. Second, we scale a firm s applied patents in a technology class in a year by the corresponding average value for that technology class and application year obtained in the first step. 6 Third, for each firm-year, we sum the firm s scaled patent count from the second step across all technology classes. Not all patents have the same technological importance or economic value. Following the patent literature (Jaffe and Trajtenberg 2002; Lerner, Sorensen, Strӧnberg 2011), we construct three well adopted measures of patent quality. First, we count the number of citations a patent receives after its approval. If a patent is cited in numerous other patents, the technology revealed in that patent is apparently involved in many developmental efforts. Patents will keep receiving citations over a long period of time after they are filed, but we only observe citations up to the 6 Technology classes are defined by the United States Patent and Trademark Office (USPTO), and capture technological essence of an invention. Technological classes are often more detailed than industry classifications, consisting of about 400 main (3-digit) patent classes, and over 120,000 patent subclasses. 10

12 end of our sample period in 2006 in the NBER database. To correct for this truncation, we follow Hall et al. (2001) to adjust the citation counts for the effect of citation lag. Furthermore, we scale each of the adjusted citations received by a patent with the average number of adjusted citations received by all patents granted in the same year and the same technology class. In addition to the number of citations received, the distribution of citations is important. The other two quality measures are based on the dispersion of citations in different technology classes. Patents that cite other patents in a broader array of technology classes integrate more diverse body of technologies, break out of traditional areas and thus are viewed as having more originality. Patents that are themselves cited by a more technologically dispersed array of patents influence subsequent innovations in a variety of fields and would presumably have a more widespread impact, hence having greater generality. We calculate originality (generality) as one minus the Herfindahl index of the cited (citing) patents (Hall et al. 2001). Thus, a higher measure of originality or generality means that the patent is drawing on or being drawn upon by a more diverse array of patents. Similar to patent counts and citations, the originality and generality measures are scaled by the corresponding average originality and generality of the cohort of patents granted in the same year and technology class. 3.3 Empirical design to compare patent production before and after R&D REM To test whether R&D REM firms experience a decline in their patent production, we use a within-firm analysis that compares a firm s patent production before and after the REM event. The merit of the approach is that such an estimate of the impact of R&D REM on innovation is 11

13 not affected by the firm specific time-invariant characteristics. The regression model is specified as follows: (2) where EventYear ik is a dummy variable that indicates the year in which a patent was applied relative to the R&D REM event year (i.e. year zero, the omitted category). Y it is one of the scaled measures of patent production quantity and quality for firm i in year t, i.e. patent application counts, citations received, originality and generality. All model specifications are estimated using OLS regression and include firm ( and year ( ) fixed effects. 3.4 Empirical design to compare patent production of R&D REM and R&D Beat/Miss firms Innovation ability tends to deteriorate after firms cut R&D expenditure. Firms could cut their R&D expenditures for various reasons, not necessarily to manage earnings. Therefore, the within-firm analysis does not control for different motivations for cutting R&D expenditures across firms. To address the issue, we conduct matched sample analysis to compare innovation output of firms that engage in R&D REM with that of firms that reduce R&D expenditures for reasons other than meeting earnings benchmarks, which is actually a test of hypothesis 2. Similar to Gunny (2010), we construct two matched samples R&D Beat firms and R&D Miss firms. We classify firms that have negative abnormal R&D expenditures into three groups. As discussed in section 3.1, firms that report small profit or small earnings growth are identified as R&D REM firms. Firms that report scaled net income or earnings growth equal to or greater than 0.01 are classified as R&D Beat firms. Firms that report net income or earnings growth less than zero are classified as R&D Miss firms. For each R&D REM firm, we identify all possible R&D Beat and R&D Miss firms in the same year from the same two-digit SIC code industry. We 12

14 then choose a matching Beat firm and a matching Miss firm with the closest size (total assets). If no firm year fits the above criterion, we repeat the process by identifying firms without the industry restriction. We require all matching firms to have at least one successful patent application from three years before the REM-event year to three years afterwards. In addition, the matchting Beat and Miss firms should have never been classified as R&D REM firms during that seven year period. We are able to identify 1,126 matching Beat firms and 948 matching Miss firms. The baseline specification for the matched sample analysis is as follows: (3) where is an average patent production measure in the three years following the event year: the average scaled patent counts, average scaled citations, average scaled originality, and average scaled generality. is the equivalent patent measure in the three years before the event year. Beat (Miss) is the dummy variable that equals one if a firm belongs to the Beat (Miss) sample. X i is a vector of variables that represent firm characteristics that may affect the firm s innovation ability. X includes prior innovation ability, firm size (market value), profitability (ROA), investment in intangible assets (prior year R&D), leverage, and growth opportunities (Tobin s Q). We also control for the magnitude of the reduction in R&D expense, which should be negatively related to innovation output. These firm characteristics are measured in the year that firms abnormally cut R&D expenditures. This model controls for the industry ( and year ( ) fixed effects. 13

15 4. Description of Data and Test Results 4.1. Sample Characteristics Table 1 contains the statistics on the sample of 1,126 R&D REM firms. Panel A presents the yearly distribution of the R&D REM firms and their patents. The number of R&D REM firms ranges from 28 to 74 during 1979 to The number of patent applications for our sample firms increases from around four thousands to six thousands per year in 1980 s to around sixteen thousands per year in the late 1990 s. 7 Panel B illustrates the distribution of the sample of R&D REM firms across the 18 two-digit SIC code industries. The R&D REM firms concentrate in the Machinery (383 firms), Instruments and miscellaneous manufacturing (181 firms), Chemicals (144 firms), Services (86 firms), Primary and fabricated metals (77 firms), and Transport equipment (68 firms). There are no more than 50 R&D REM firms in any of other industries. We also follow Hall et al. (2001) to group the patents into six technology classes. As shown in Panel C, of the total 205,088 patent applications in the sample, 19 percent (38,283) are in Chemical class, 26 percent (54,061) in Computers and communications class, 7 percent (13,953) in the Drugs and medical class, 23 percent (46,542) in the Electrical and electronic class, 15 percent (31,217) in the Mechanical class, and 10 percent (21,032) in the Other class. The significant variation in the number of patent applications and grants across time and industry groups supports our use of the scaled measures of patent applications in the subsequent analysis 7 There are fewer patent applications in the and periods because these are the 3 year periods before the first event year and after and last event year in the sample. As an illustration, the numbers of patent applications in 1976 and 2003 contain patent applications made by the R&D REM firms identified in year 1979 and 2000 respectively, whereas the number of patent applications in 2000 includes patent applications made by all R&D REM firms identified in

16 to control for the inherent disparity in patent production across different industry and time period. <Insert Table 1> 4.2. Univariate analysis of relationship between R&D REM and innovation To illustrate the effect of R&D REM on innovation, Figure 1 plots the scaled number of patent applications for R&D REM, R&D Beat, and R&D Miss firms from three years before to three years after the year of REM event (the year of negative abnormal R&D expenditures). If the decline in the quantity of innovation after R&D REM is only due to experiencing negative abnormal R&D expenditures, all three groups of firms should have a similar trend. However, as shown in Figure 1, only the scaled number of patent applications for R&D REM firms demonstrates a monotonically decreasing pattern after the event year. In contrast, the scaled number of patent applications for the R&D Beat firms demonstrates a monotonically increasing pattern. The scaled number of patent applications for the R&D Miss firms does not change much around the REM event year. 8 <<Insert Figure 1>> Table 2 presents the statistics of the scaled measure of patent quantity at firm level in Panel A and patent quality measures at firm level in Panels B-D for the REM, Beat, and Miss firms from the three years before to the three years after the event year. The patent quality measures in Panels B-D are the average scaled number of adjusted citations, originality, and generality for all patent applications of a firm during that year. If there is no patent application in a year, these quality measures are assigned as zero. For R&D REM firms, the average scaled numbers of 8 The plot of the raw number of patent applications for R&D REM, R&D Beat, and R&D Miss firms from year +3 to year 3 around the REM event year is similar. It is not reported for brevity. 15

17 patents for the three years after the REM event is 6.66, significantly lower than 7.02 for the event year. In addition, all measures of the R&D REM firms patent quality, i.e. the scaled citations/originality/generality, decrease significantly in the three years after the REM event. We observe no such consistent declining patterns in patent production of the R&D Beat and Miss firms that also experience abnormal R&D reduction during the REM event year. For R&D Beat firms, after the abnormal R&D expenditure cut, the scaled number of patent applications actually increases significantly from 6.32 in the event year to an average of 7.0 in the subsequent three years; there is no significant change in any of the three measures of patent quality. For R&D Miss firms, both the scaled number of patent applications and the scaled originality in the three years following the R&D cut remain at a similar level to that in the event year, while the scaled citations and the scaled generality do have a significant decrease. The univariate analysis provides consistent evidence on the significant adverse effect of REM event on the quantity and quality of R&D REM firms patent applications. However, there is no evidence of adverse effect on R&D Beat firms. For R&D Miss firms, there is no evidence of adverse effect of the REM event on patent quantity while we find some evidence of adverse effect on two of the three patent quality measures. 9 <<Insert Table 2>> 4.3. Within-firm analysis of effect of R&D REM on innovation We estimate Equation 2 for REM, Beat, and Miss firms separately to examine the withinfirm changes in the quantity and quality of patents before and after the event year of abnormal R&D reduction, controlling for firm and year fixed effects. Table 3 Panel A reports results on the 9 To be consistent, we use firm level measures of both patent quantity and quality in all analyses. In untabulated analysis, we also use measures of patent quality at individual patent level. The results are similar. 16

18 scaled number of patent applications at firm level for REM (column 1 and 2), Beat (column 3 and 4), and Miss (column 5 and 6) firms. In column 1, 3, and 5, we have dummy variables representing each of the three years before and three years after the REM event year. The coefficients of these six yearly dummies reflect the difference in the number of patent applications between the event year and each of the six years. We include only two dummy variables representing the three year period before and the three year period after the event in column 2, 4, and 6 to compare the average annual patent applications during each of the two three-year periods to the number of patent applications in the REM event year. In column 1, the coefficients of all six yearly dummy variables are negative, with the dummy variables for Event year 2 and Event year 3 significant at 10 percent and 1 percent levels, respectively. The result indicates an increasing trend albeit insignificant in the scaled number of patent applications in the three years before the REM event. However, after the REM event, the scaled number of patent applications starts to decrease and the drop becomes significant two years after the REM event year. In other words, the upward trend in the scaled number of patent applications shifts to a downward trend after the REM event. In column 2, the coefficients for the two dummy variables representing the three year period before and the three year period after the event year are (insignificant) and (significant at 5 percent level), respectively, confirming that the REM event year is a turning point and the upward trend in the scaled number of patent applications shifts to a downward trend after the event year. Overall, we are inclined to conclude there is a significant negative effect of R&D REM on patent quantity. For R&D Beat and Miss firms, we also find a significant decline in the scaled number of patent applications after the REM event. However, the decline is less compared with R&D REM 17

19 firms. For R&D REM firms, the coefficients of Event year +2, Event year +3 in Column 1 are more negative than those in Column 3 for R&D Beat firms and column 5 for R&D Miss firms. Similarly, the coefficient of Event year 1 to 3 in Column 2 is more negative than the same coefficient in Columns 4 and 6. Further, for R&D Beat and Miss firms, the scaled number of patent applications start to decline significantly before the REM event year and the downward trend continues after the REM event. The coefficients of all dummy variables before the REM event year are positive in Columns 3-6. Overall, we find a decline in the number of patent applications after the abnormal R&D cuts for both R&D REM/Beat/Miss. But the number of patent applications is increasing/decreasing before the abnormal reduction of R&D expenditures for R&D REM/(Beat and Miss) firms and decreases significantly after the REM event. To examine changes in the quality of patent applications around the REM event, we repeat the same model specification with scaled citations/originality/generality at firm level as the dependent variable. In Panels B-D, the coefficients for all six yearly dummy variables are significantly negative in column 1, and the coefficients for the two dummy variables representing the three-year period before and the three-year period after the event year are also significantly negative in column 2. This suggests that the patent quality of R&D REM firms patent portfolio is increasing before the REM event and starts to decline significantly after the REM event. For R&D Beat and Miss firms, the patent quality does not change significantly around the R&D cut event as most of the coefficients are insignificant. Overall, Table 3 provides consistent evidence that there is a significant change in the trend of the quantity and quality of patent applications around the REM events. Specifically, the R&D REM firms patent production increases up to the REM event year and starts to significantly 18

20 decrease after the event year. However, there is no such change of pattern and consistently negative effect on the patent production for Beat firms and Miss firms that also experience abnormal negative R&D expenditures. The more pronounced adverse effect on the quantity and the quality of innovation for R&D REM firms suggests that firms cutting R&D expenditures to meet earnings targets may sacrifice their long-term competitiveness. <<Insert Table 3>> 4.4. Matched sample analysis of effect of R&D REM on innovation We estimate Equation 3 with the matched sample firms (i.e. REM, Beat, and Miss firms), controlling for various firm characteristics related to firms innovation capability that could attribute to the observed differences in patent production output across the R&D REM firms and the matched sample Beat and Miss firms. The model enables us to distinguish the effect of R&D real earnings management on subsequent innovation performance from the effect of cutting R&D expenditures due to other business considerations. Table 4 presents the results of the matched sample analysis. The dependent variables are the average scaled measures of patent quantity and quality in the three year period following the event year. Following prior literature (Seru 2012; Tian and Wang 2011), the regression model controls for factors related to R&D productivity, including prior patent production levels the average of scaled measures of patent quantity and quality during the three year period before the event year, firm size, firm s profitability as measured by return on assets (ROA), investments in innovative projects as measured by firms R&D expenditure, financial leverage, and the investment opportunities as measured by Tobin s Q. In addition, we also control for the magnitude of the R&D cut since the more the managers reduce R&D, the larger decrease in 19

21 innovation. The focus of the test is on the coefficients for the two dummy variables, Beat and Miss, which capture the difference in patent production between the Beat/Miss firms and the R&D REM firms after controlling for various patent related firm characteristics. All regressions are estimated with year and industry fixed effects to capture any macro effect in a given year or an average effect in an industry. Model specification 1 uses the scaled number of patent applications as dependent variable. The coefficients for the Beat and Miss dummies are all positive at (significant at five percent level) and (insignificant), suggesting that, after controlling for firm characteristics, the scale number of patent applications in the three years after the REM event for R&D Beat and Miss firms is higher (significantly higher for Beat firms) than that of R&D REM firms, confirming the result from within-firm analysis that the REM event has more pronounced adverse effect on the patent quantity for R&D REM firms. Model specifications 2 to 4 use measures of patent quality as dependent variables. In estimate of specification 2 that uses scaled citations received by patents as dependent variable, the coefficients for the Beat and Miss dummy variables are and 0.124, respectively, both significant at five percent level. Significant positive coefficients for the two dummy variables indicate that in comparison to patents filed by firms in the Beat and the Miss samples, the patents filed by R&D REM firms in the 3 year period after the REM event receive significantly fewer citations. Similarly, the coefficients for the Beat and Miss dummy variables in the estimate of specification 3 that uses scaled patent originality as dependent variable are and 0.079, respectively, both significant at one percent level. The coefficients for the Beat and Miss dummy variables in the estimate of specification 4 that uses scaled patent generality as dependent variable are and 0.095, respectively, both significant at one percent level. The results of 20

22 model specification 2 to 4 provide consistent evidence that the firms that cut R&D expenditures to meet earnings benchmarks experience a more significant decline in the quality of their subsequent patent production than firms that cut R&D expenditures for reasons other than meeting earnings benchmarks, confirming the similar conclusion from within-firm analysis. 10 <<Insert Table 4>> 5. Conclusions This study examines the effect of cutting R&D expenditures abnormally to meet earnings targets on firms innovation production measured by patent quantity and quality. We conduct univariate and regression analyses on the change in patent production of the R&D REM firms that cut their R&D spending to meet earnings targets, controlling for firm specific characteristics related to patent production ability. We also use matched sample analysis with two industry-size- R&D cut matched firm groups to explore the differential effect of abnormal R&D reductions on the patent production of the REM firms that likely cut R&D to meet earnings targets and of the matched sample firms that likely cut R&D for other business considerations. We find consistent evidence that cutting R&D expenditures to meet earnings targets adversely affects the quantity of patent production measured by the number of patent applications for R&D REM firms. Our test results also provide strong evidence that the quality of patent production in terms of technological importance and level of innovation declined significantly in the period following the R&D real earnings management activities. 10 In unreported results, we have examined the effect of R&D REM on subsequent operating and stock performances and found insignificant effects. The ROA of R&D REM firms in the post-event period is lower than that in the preevent period, but the difference is not significant. The buy-and-hold stock returns of R&D REM in the post-event period are not different from those in the pre-event period. The changes in ROA and abnormal stock returns of R&D REM, Beat, and Miss firms are not significantly different, after controlling for other factors. 21

23 There is a caveat with our study in the use of patent data. Not all output of firms R&D effort is in the form of patents. It involves a strategic decision for firms to decide whether to patent, or to rely on secrecy or other means to protect their R&D output. Even our paper cannot address the issue if firms keep their innovation in secrecy. However, the design of our tests helps to alleviate the concern on propensity to patent. Presumably, firms propensity to patent is unlikely to change. We use a within-firm estimator to compare the changes in innovation performance around real earnings management, which controls for the firm specific characteristics related to patent propensity and innovation capability. We also conduct matched sample analysis to compare the innovation performance of R&D real earnings management firms with groups of firms matched by year, industry and size that cut R&D expenditures likely for reasons other than meeting earnings targets. The consistent results produced by the different tests suggest that the decline in patent quality is more likely to be due to R&D real earnings management than due to variation in firms propensity to file patents. 22

24 Reference Baber, W., Fairfield, P.M., Haggard, J.A., The effect of concern about reported income on discretionary spending decisions: The case of research and development. Accounting Review 66 (4), Berger, P Explicit and implicit tax effects of the R&D tax credit. Journal of Accounting Research 31 (2), Bena, J., Li, K., Corporate innovations and mergers and acquisitions. Working paper, University of British Columbia. Bens, D.A., Nagar, V., Wong, M.H.F., Real investment implications of employee stock option exercises. Journal of Accounting Research 40 (2), Bernstein, S., Does going public affect innovation? Working paper, Stanford University. Bhojraj, S., Hribar, P., Picconi, M., McInnis, J., 2009, Making sense of cents: An examination of firms that marginally miss or beat analyst forecasts. Journal of Finance 64, Bound, J., Cummins, C., Griliches, Z., Hall, B.H., Jaffe, A.B., Who does R&D and who patents in R&D, patents and productivity. Chicago: University of Chicago Press. Bushee, B.J., The influence of institutional investors on myopic R&D investment behavior. Accounting Review 73, Cheng, S., R&D expenditures and CEO compensation. Accounting Review 79, Cheng, Q., Lee, J., Shevlin, T.J., Internal governance and real earnings management. Working paper, Singapore Management University. Crépon, B., Duguet, E., Mairesse, J., Research, innovation, and productivity: An econometric analysis at the firm level. Economics of Innovation and New Technology 7(2), DeFond, M., Park, C.W., Smoothing income in anticipation of future earnings. Journal of Accounting and Economics 23, Demski, J., Performance measure manipulation. Contemporary Accounting Research 15, Ewert, R., Wagenhofer, A., Economic effects of tightening accounting standards to restrict earnings management. Accounting Review 80,

25 Graham, J., Harvey, C., Rajgopal, S., The Economic implications of corporate financial reporting. Journal of Accounting and Economics 40, Gunny, K., The relation between earnings management using real activities manipulation and future performance: Evidence from meeting earnings benchmarks. Contemporary Accounting Research 27 (3), Hall, B.H., Jaffe, A.B., Trajtenberg, M., The NBER patent citations data file: Lessons, insights and methodological tools. Working paper, National Bureau of Economic Research. Healy, P.M., The effect of bonus schemes on accounting decisions. Journal of Accounting and Economics 7, Healy, P.M., and K. G. Palepu The effect of firms financial disclosure policies on stock prices. Accounting Horizons 7, Healy, P.M., and J.M., Wahlen A review of the earnings management literature and its implications for standard setting. Accounting Horizons 13, Holbrook, J.A., Basic indicators of scientific and technological performance evaluation of research and development organizations. Technological Forecasting and Social Change 47(2), 189. Jaffe, A.B., and M. Trajtenberg, Patents, Citations & Innovations: A Window on the Knowledge Economy. MIT Press. Legoria, J., Earnings management, the pharmaceutical industry, and health care cost reform: A test of the political cost hypothesis. Research in Accounting Regulation 14, Lerner, J., Sorensen, M., Strömberg, P., Private equity and long run investment: The case of innovation. Journal of Finance 66(2), Papadakis, M., Patents and the evolution of R&D. Evaluating R&D impacts: Methods and practice. Academic Press. Porter, M. E., and Wayland, R., 1992, Capital disadvantage: America's failing capital investment system. Harvard Business Review 70, Prodan, I., Influence of research and development expenditures on number of patent applications: Selected case studies in OECD countries and Central Europe, Applied Econometrics and International Development 5 (4), Roychowdhury, S., Earnings management through real activities manipulation. Journal of Accounting and Economics 42,

26 Sankar, M.R., Subramanyam, K.R., Reporting discretion and private information communication through earnings. Journal of Accounting Research 29, Seru, A., Firm Boundaries Matter: Evidence from Conglomerates and R&D Activity. Journal of Financial Economics. Forthcoming. Shivakumar, L., Do firms mislead investors by overstating earnings around seasoned equity offerings? Journal of Accounting and Economics 29, Song, M. H., and Walkling, R. A., 1993, The Impact of Managerial Ownership on Acquisition Attempts and Target Shareholder Wealth, Journal of Financial and Quantitative Analysis 28, Tan, H. C., Jamal, K., Effect of accounting discretion on ability of managers to smooth earnings. Journal of Accounting and Public Policy 25 (5), Taylor, G.K., Xu, Z., Consequences of real earnings management on subsequent performance. Research in Accounting Regulation 22, Teoh, S.H., Welch, I., Wong, T.J., 1998a. Earnings management and the post-issue performance of seasoned equity offerings. Journal of Financial Economics 50, Teoh, S.H., Welch, I., Wong, T.J., 1998b. Earnings management and the long-term market performance of initial public offerings. Journal of Finance 53, Tian, X., Wang, T.Y., Tolerance for failure and corporate innovation. Review of Financial Studies, forthcoming. 25

27 Appendix 1: Variable definitions Panel A: Innovation variables Scaled patents First, for each technology class defined by USPTO and patent application year, we compute the average number of applied patents of all firms. Second, we scale each firm s applied patents in each technology class in a year by the corresponding average number of patent applications in the same technology class and year as computed in the first step. Third, for each firm, we sum the scaled number of patent applications from the second step across all technology classes for each year. Scaled citations We follow Hall, Jaffe, and Trajtenberg (2001) to compute the scaled citations received by patents to control for citation lag. For each technology class defined by USPTO and patent grant year, we compute the average number of adjusted citations for all patents in the same technology class and grant year. Second, we scale the number of citations for that patent by the corresponding average value for that technology class and grant year computed in the first step. Originality Follow Hall, Jaffe, and Trajtenberg (2001), the originality of a patent is defined as one minus the Herfindahl index of all patents cited by that patent. Scaled originality For each technology class defined by USPTO and patent grant year, we compute the average originality for all patents in the same technology class and grant year. Second, we scale the originality of a patent by the corresponding average value from the first step. Generality Follow Hall, Jaffe, and Trajtenberg (2001), the originality of a patent is defined as one minus the Herfindahl index of all patents that cite the patent under question. Scaled generality For each technology class defined by USPTO and patent grant year, we compute the average generality for all patents in the same technology class and grant year. Second, we scale the generality by the corresponding average value from the first step. Panel B: Firm characteristics Ln(MV) The natural log of market value [Data199*Data25] in the event year ROA The ratio of net income (Data172) in the event year divided by beginning total assets (Data6) R&D/Assets The ratio of R&D expense [Data46] in the event year divided by beginning total assets [Data6] Leverage The sum of long term debt [Data 9] and debt in current liability [Data 34] in the event year divided by beginning total assets [Data6] Tobin s Q Market value of assets over book value of assets in the event year: [((Data199*Data25) + Data130 + Data9 + Data34) Data6] Size of R&D cut The absolute value of negative abnormal R&D expenditures. The abnormal R&D expenditure is computed as residual of regression of Equation 1. 26