University of Arkansas, Fayetteville ScholarWorks@UARK Theses and Dissertations 8-2013 Does Analyst Experience Affect Their Understanding of Non-Financial Information? An Analysis of the Relation between Patent Information and Analyst Forecast Errors Taiwhun Taylor Joo University of Arkansas, Fayetteville Follow this and additional works at: http://scholarworks.uark.edu/etd Part of the Accounting Commons, and the Finance and Financial Management Commons Recommended Citation Joo, Taiwhun Taylor, "Does Analyst Experience Affect Their Understanding of Non-Financial Information? An Analysis of the Relation between Patent Information and Analyst Forecast Errors" (2013). Theses and Dissertations. 833. http://scholarworks.uark.edu/etd/833 This Dissertation is brought to you for free and open access by ScholarWorks@UARK. It has been accepted for inclusion in Theses and Dissertations by an authorized administrator of ScholarWorks@UARK. For more information, please contact scholar@uark.edu.
Does Analyst Experience Affect Their Understanding of Non-Financial Information? An Analysis of the Relation between Patent Information and Analyst Forecast Errors
Does Analyst Experience Affect Their Understanding of Non-Financial Information? An Analysis of the Relation between Patent Information and Analyst Forecast Errors A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Business Administration By Taiwhun T. Joo Brigham Young University Bachelor of Science in Accounting, 2009 Brigham Young University Master of Accountancy, 2009 August 2013 University of Arkansas This dissertation is approved for recommendations to the Graduate Council. Dr. James Myers Dissertation Director Dr. Linda Myers Committee Member Dr. Vernon Richardson Committee Member Dr. Junhee Han Committee Member
ABSTRACT This study examines whether analyst experience affects the relation between patent information and analyst forecast errors. U.S. Generally Accepted Accounting Principles require that firms expense all in-house research and development (R&D) costs. This means that even when R&D activities produce intangible assets with future economic benefits, firms cannot capitalize R&D costs as assets. Consequently, financial statements are largely deficient in the information they provide regarding the output of R&D activities. However, patent information is one type of non-financial information about R&D output that is publicly available. Using updated patent data, I confirm the results of prior studies that find a positive association between patent citations and future firm performance. I also confirm the positive association between the absolute value of analyst forecast errors and patent citations. Next, in my main tests, I examine whether analyst experience affects the relation between patent information and analyst forecast errors (absolute value and signed). I find that analysts with more experience are not better at incorporating patent information to make more accurate earnings forecasts. Instead, they incorporate patent information to make more optimistic earnings forecasts than analysts with less experience. My findings should be of interest to standard setters in deciding whether to require firms to disclose patent information because this information should be useful to investors and analysts.
ACKNOWLEDGEMENTS Special thanks are due to my dissertation committee members and other professors in the Accounting Department for all of their comments and suggestions, as well as support. Also, a special thanks goes out to my family and classmates in the Accounting Department for their support, especially to Dr. E. Scott Johnson and Lauren Dreher for their friendship.
TABLE OF CONTENTS 1. INTRODUCTION...1 2. BACKGROUNDS AND HYPOTHESIS DEVELOPMENT...4 2.1 Patents...4 2.2 R&D and Patent Information...5 2.3 Patent Information and Future Firm Performance...8 2.4 Patent Information, Forecast Revisions, and Forecast Errors...9 2.5 Analyst Experience and the Relation between Patents Information and Forecast Errors..9 3. SAMPLE SELECTION AND DATA...11 3.1 Sample Selection...11 3.2 Patent Measures...11 4. METHODOOLGY...13 4.1 Patent Information and Future Firm Performance...13 4.2 Patent Information, Forecast Revisions, and Forecast Errors...15 4.3 Main Tests: Analyst Experience and the Relation between Patents Information and Forecast Errors Patent Information, Forecast Revisions, and Forecast Errors...17 5. RESULTS...20 5.1 Descriptive Statistics...20 5.2 Patent Information and Future Firm Performance...20 5.3 Patent Information, Forecast Revisions, and Forecast Errors...21 5.4 Analyst Experience and the Relation between Patents Information and Forecast Errors 22 6. ADDITIONAL TESTS...23 6.1 Alternative Scaling for Patent Measures...23
6.2 Other Types of Experience...24 6.3 R&D Capital...27 6.4 Regulation Fair Disclosure...28 6.5 Forecasting Horizons...29 6.6 Industry Effects...31 6.7 Truncating the Sample Period...31 7. CONCLUSION...32 8. REFERENCES...35 9. APPENDIX...39 10. TABLES...41
1. INTRODUCTION This study examines whether analyst experience affects their ability to understand the impact of patents on future earnings. Statement of Financial Accounting Standard (SFAS) No. 2, Accounting for Research and Development Costs, requires firms to expense all in-house research and development (R&D) costs as they are incurred. In other words, United States (U.S.) Generally Accepted Accounting Principles (GAAP) treat investments in R&D as expenses, and R&D costs are not capitalized on the balance sheets. This full expensing treatment of R&D investments can weaken the link between current and future earnings. Therefore, analysts may use non-financial information to supplement accounting information (Amir et al., 2003). Using a sample of firms with patents granted from 1994 through 2005, this study examines the relation between patent information and analyst forecast errors, and investigates whether analyst experience affects this relation. Some patents are more economically valuable than others. Trajtenberg (1990) finds that the number of citations that a firm s patents receive from follow-up patents is a better proxy for R&D output than is the number of patents. Patent citations arise when follow-up patents cite previous patents. 1 Prior studies find that firm value is positively associated with patent citations (Hirschey et al., 2001; Hall et al., 2005). Other studies find a positive relation between patent citations and future firm performance (Gu, 2005; Pandit et al., 2011). Before I examine the effect of analyst experience on the relation between patent information and forecast errors, I confirm that the findings of these prior studies exist for my sample. Using updated patent data, I find that the number of patent citations is positively associated with future earnings. 2 After confirming 1 This is similar to academic studies citing previous studies. 2 The methodology section provides more detailed descriptions of my patent measures and tests. 1
that firms with more patent citations have better future performance, I also confirm that analysts, on average, make less accurate earnings forecasts for firms with more patent citations. The inherent difficulty in estimating the value of intangible assets is the main reason that U.S. GAAP requires firms to expense the full costs of in-house R&D as incurred (FASB, 1974). Furthermore, active and transparent markets for intangible assets are limited (Gu and Wang, 2005). This lack of information about the value of R&D output might encourage financial analysts to use information not disclosed in financial reports, such as patent data available through the United States Patents and Trademarks Office (USPTO), when forecasting future earnings. However, even though patent information from USPTO is publicly available, analysts may be unable to fully understand the implications of patents for future earnings due to the uncertainty associated with new technologies (Aboody and Lev, 1998; Amir et al., 2003). Consistent with this, Amir et al. (2003) find that analyst forecasts are less accurate and more optimistic in industries that are more R&D intensive. Using the model from Amir et al. (2003), I confirm that the analyst consensus earnings forecast is less accurate for firms with more patent citations. I do not find evidence, however, that patent citations are related to forecast optimism. This leads to my main research question. Do analysts learn from their experience in using patent information to make more accurate earnings forecasts? Prior studies find that, on average, more experienced analysts make more accurate earnings forecasts (Mikhail et al., 1997; Clement, 1999). In addition, Drake and Myers (2011) find that analysts general experience is associated with less accrual-related over-optimism, even when controlling for analysts firm-specific experience. They measure general experience as the number of prior years in which an analyst makes earnings forecasts for any firm, and firm-specific experience is the number of prior years in which an analyst makes earnings forecasts for a given firm. 2
In this study, I focus on firm-specific experience because patents are unique across firms. 3 If patent information is difficult to incorporate into forecasts, even with experience, I expect analyst experience to have no effect on the relation between patent information and forecast errors. If analysts with more experience have a better understanding of the implications of patents for future earnings, I expect analysts with more experience to make more accurate earnings forecasts. I find, however, that analysts with more experience are not better at using patent information to make more accurate earnings forecasts. Interestingly, analysts with more experience appear to use patent information to make more optimistic earnings forecasts. Overall, I conclude that analysts with more experience have some understanding of the benefits of patents, but this understanding does not lead to more accurate earnings forecasts. My study contributes to the existing R&D literature. My finding suggests that analysts with more experience have some understanding of patent information. However, this understanding is limited in that it allows more experienced analysts to make more optimistic, but not more accurate, earnings forecasts. Regulators may consider requiring firms to disclose more information about the output of R&D activities because this information about R&D outputs appears to be useful and no other publicly available information on the output of R&D activities is widely available. Other voluntary disclosures on the benefits of R&D output, such as patent licensing income, is rare even among large firms investing heavily in R&D (Gu et al., 2004). Perhaps because of this lack of information, Cohen et al. (2013) find that the market does not value past R&D activities appropriately. My study also contributes to the analyst forecast literature by showing that financial analysts do not fully incorporate all of the available information about patents when making 3 In additional analyses, I explore the impact of general experience, industry-specific experience, and task-specific experience on analysts understanding of the information in patent citations. 3
earnings forecasts. Bradshaw, Richardson, and Sloan (2001) find that analysts are overoptimistic in valuing accruals. Given that prior studies observe a positive relation between patent information and future firm performance, one would expect analyst forecasts to incorporate patent information to compensate for the lack of other R&D output information. However, I confirm that patent citations are associated with less accurate earnings forecasts. Furthermore, I find that analysts with more experience are not better at making more accurate earnings forecasts by incorporating patent information, although they do seem to better understand that patents lead to improved future performance. The remainder of this paper is organized as follows. Section 2 discusses previous research and presents my hypotheses. Section 3 describes the data, and Section 4 explains my methodology. Section 5 presents the results from my analyses, and Section 6 presents results of additional tests. Section 7 concludes. 2. BACKGROUND AND HYPOTHESIS DEVELOPMENT 2.1. Patents A patent is intellectual property that gives an inventor the right to exclude others from making, using, offering for sale, or selling the invention throughout the United States or importing the invention into the United States for a specified time. 4,5 Intellectual property rights are critical to the modern economy. The Copyright and Patent Clause of the United States Constitution allows the United States Congress to exercise enumerated power to promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the 4 See 35 U.S.C. 154 (a)(1) (2006). 5 Other common types of intellectual property include copyrights, trademarks, and industrial design rights. 4
exclusive Right to their respective Writings and Discoveries (U.S. Constitution, Article I, Section 8). Not only are individuals guaranteed property rights for tangible assets, but also for intangible assets (i.e., the individuals ideas). This right for inventions gives individuals and firms incentives to invest time and effort in innovative discoveries. In exchange, the inventor discloses information about the invention to the public. The USPTO assigns patents to individual inventors or to the firms for whom the inventors work. Thus, the patent rights encourage firms to invest in innovative technologies by promising these firms exclusive rights to use the patents, given that the legal system is able to enforce these rights. Lieberman and Montgomery (1988) highlight success in patents as a mechanism by which a firm establishes technology leadership, which leads to a first-mover advantage. However, there is currently no requirement for firms to disclose information about patents or R&D activities in their financial statements. 2.2 R&D and Patent Information The prevalence of private R&D activities makes this study relevant. In the past half century, while the investment in R&D activities as a proportion of the U.S. Gross Domestic Product has remained steadily at 2-3%, the proportion of private R&D has increased and government-sponsored R&D has decreased (Cohen et al. 2013). Cohen et al. (2013) suggest that the responsibility for allocating resources in R&D is on the private sector rather than on the government. Therefore, the ability of market participants to understand the benefits of R&D activities and to allocate resources accordingly is important. The motivation for my study starts with patent information and its association with future firm performance. Consider a firm that incurs costs to conduct in-house R&D activities. These 5
activities produce intangible assets that have some economic value. 6 This paper refers to these intangible assets as R&D capital. Currently, U.S. GAAP does not allow firms to capitalize any in-house R&D costs (SFAS No. 2); therefore, even if a firm were successful in R&D, the financial statements would not reflect any R&D capital on their balance sheet. Instead, the full amount of R&D costs incurred during the year would show up on the firm s income statement as R&D expense. If U.S. GAAP allowed firms to capitalize R&D costs, there would be significant subjectivity in determining how much of R&D costs should be capitalized. This subjectivity and uncertainty about the future implications of R&D capital are the main reasons why the Financial Accounting Standards Board (FASB) requires full expensing treatment. Thus, financial statements contain only information about the input of R&D activities (in the form of R&D expense), and lack information about the value of the output from R&D activities. Rong (2012) suggests that market participants lack access to a considerable amount of information about R&D activities because management chooses to not disclose this information. Currently, there is no requirement for firms to disclose the estimated value of R&D capital, and there is no active market for many of these intangible assets. Perhaps because of this lack of information, Cohen et al. (2013) find that the market generally misvalues R&D investment. However, sophisticated market participants, such as analysts, may compensate for this lack of information about the value of R&D capital with non-financial information (Amir et al., 2003). One type of nonfinancial information analysts can use is patent information because patents are the output of R&D activities and are publicly available from the USPTO. 6 The value of R&D capital, as with other assets, would be the present value of future earnings generated by the asset. 6
Patent information may be informative to market participants in a number of ways. First, if a person can understand the potential impact of a patented technology, patent information may shed light on future sales of a product or on the reduced cost of manufacturing a product. Other information patents contain is the potential for royalty income. Second, market participants may use patent information to estimate potential royalty income. Gu et al. (2004) find that royalty income is more persistent than earnings. However, the authors state that voluntary disclosures about royalty income from patent licensing are uncommon. Lastly, Cohen et al. (2013) suggest that past information about R&D successes is informative about potential future success of R&D. They find that stock returns of firms with higher ratios of R&D expense to sales in the past perform better in the future than do firms with the same dollar amount of R&D expense, but lower ratios of R&D expense to sales. My patent measures count the number of citations that firms patents subsequently receive. More important patents are likely to be cited more often, while less important or obscure patents are likely to receive fewer citations. 7 Empirical studies find that the number of citations patents receive is a proxy for the patent s importance or innovativeness (Trajtenberg, 1990; Harhoff et al., 1999; Hall et al., 2005). Trajtenberg (1990) follow prices and subsequent patent citations for Computed Tomography (CT) scanners to examine the value of innovations. He finds that patents for CT scanners that command higher prices, receive more subsequent citations than patents for CT scanners that command lower prices. Harhoff et al. (1999) survey German firms with patents. The authors ask the firms to value the firms patents and follow the subsequent citations for those patents. They find that the number of patent citations is positively associated with the value of patents assigned by the firms. Rong (2012) also suggests that the number of 7 This is analogous to more important academic studies receiving more citations than less important studies. 7
patent citations is a signal of economic importance. I use patent citations as information about the firms R&D success. 2.3 Patent Information and Future Firm Performance Before examining whether analysts experience improves their understanding of patent information in making earnings forecasts, I test whether the number of patent citations is useful for predicting future earnings. This answers the question of whether a parsimonious count of patent information contains information related to the success of R&D activities and future firm performance. In prior research, Hirschey et al. (2001) and Hall et al. (2005) find a significant and positive association between patent citations and Tobin s q, and Gu (2005) and Pandit et al. (2011) find that patent citations are positively associated with future earnings. I construct three patent citation measures to proxy for information about the output of R&D activities. The first measure is the total number of citations received in the current year. The number of patent citations in the current year is a proxy for the economic importance of a firm s patents or a signal of potential royalty income from patent licensing. The second measure counts patent citations over the past five years. This measure is a proxy for the information about a firm s patent portfolio. The third measure is the number of citations received in the past years, in the current year, and in the future years. Although the number of citations a patent receives in the future is not observable ex ante, if analysts understand the future implications of specific patents, this measure may be a good measure of the overall economic importance of patents. I use future earnings to measure future performance and confirm that prior findings that patent citations are positively associated with future firm performance exist in my sample. Additional tests using return on assets (ROA) as a measure of firm performance also confirm these findings. 8
2.4 Patent Information, Forecast Revisions, and Forecast Errors Next, I test whether patent information is associated with analyst forecast revisions and forecast errors (both absolute value and signed). Patent information may provide analysts with additional information about R&D capital. If this is the case, I expect patent information to be associated with analyst forecast revisions. I also expect patent information to be associated with more accurate earnings forecasts. Finally, I expect patent information to be associated with more optimistic earnings forecasts, given the positive relation between patent information and future firm performance. Alternatively, if patent information is difficult to understand and does not provide analysts with additional information about R&D capital, I expect to find no association between patent information and forecast revisions. I also expect patent information to be associated with less accurate earnings forecasts, and do not expect any difference in signed forecast errors. A few prior studies examine the relation between patent information and analyst forecast errors. Using a sample of observations from 1983 through 1999, Gu (2005) finds that analysts, on average, underestimate the future implications of patent citations. In a follow-up study, Gu and Wang (2005) find that more innovative patents are associated with less accurate earnings forecasts. 8 2.5 Analyst Experience and the Relation between Patent Information and Forecast Errors This leads to my main research question. How does an individual analyst s experience affect the relation between patent information and analyst forecast errors? Learning-by-doing theory predicts that people learn to perform tasks better as they gain more experience related to the tasks (Arrow, 1962; Anzai and Simon, 1979). Arrow (1962) models how experience 8 The authors use patent citations as a proxy for the innovativeness of patents. 9
increases labor productivity over time and Anzai and Simon (1979) study how people learn to perform a task during a problem-solving process. Test subjects in Anzai and Simon s experiment are given the task of solving the Tower of Hanoi puzzle. The authors document how their subjects learn by producing strategies and adapting. In the accounting literature, an experimental study by Shelton (1999) finds that more experienced auditors (both audit managers and partners) place less weight on irrelevant information than do less experienced auditors (i.e., audit seniors). In addition, prior studies find that more experienced analysts make more accurate earnings forecasts (Mikhail et al., 1997, 2003; Clement, 1999). In my study, on the one hand, more experienced analysts may be able to better understand the future implications and benefits of patents. On the other hand, the implications and benefits of R&D capital are inherently difficult to estimate, which is why the FASB does not allow firms to capitalize in-house R&D costs in the first place. Thus, an analyst s experience may not allow him to overcome the inherent difficulty in estimating the future implications and benefits of patents. In this study, I focus on firm-specific experience because of the uniqueness of each firm s patents. My first hypothesis, stated in the alternative form, is: H1: More experienced analysts are better at incorporating patent information to make more accurate earnings forecasts. My second hypothesis relates to how analyst experience affects signed forecast errors. Given that patent information is positively related to future performance, I expect analysts who understand this relation to make more optimistic earnings forecasts than those who do not. Furthermore, analysts may learn to better understand this relation with experience. On the one hand, if more experienced analysts better understand the positive relation between patent information and future earnings, I expect their earnings forecasts to be more optimistic than 10
earnings forecasts of less experienced analysts. On the other hand, if more experienced analysts do not understand the positive relation between patent information and future earnings any better than less experienced analysts, I do not expect their earnings forecasts to be more optimistic. H2: More experienced analysts are better at incorporating patent information to make more optimistic earnings forecasts than less experienced analysts. 3. SAMPLE SELECTION AND DATA 3.1 Sample Selection I examine earnings forecasts from 1994 to 2005. Beginning in 1994 ensures that the analyst earnings forecast data are consistent throughout the sample period because I/B/E/S changed its method of calculating actual earnings per share (EPS) in the early 1990s (Abarbanell and Lehavy, 2007; Drake and Myers, 2011). I exclude from my sample industries and firms that do not have any patent. After eliminating observations with insufficient data from the Compustat Annual file, stock return data from the CRSP Monthly file, and analyst data from the I/B/E/S Detailed file between 1994 and 2005, my sample consists of 36,278 analyst-firm-year observations. This sample includes 4,364 unique analysts, 1,576 unique firms, and 206 unique industries (3-digit SIC code). 3.2 Patent Measures I construct my patent measures using publicly available data from the National Bureau of Economic Research (NBER) Patent Data Project Web site. Specifically, I use the following datasets: [1] the dataset of records for every patent (more than 3.2 million observations from 1976 through 2006), [2] the dataset that matches patents to assignee codes, [3] the dataset that matches assignee codes to firm codes (GVKEY in Compustat), and [4] the dataset that matches 11
cited patents to citing patents. I merge the four datasets into observations with firm identifiers, fiscal years, cited patent identifiers, and citing patent identifiers. I then construct three measures of patent information: 1. Number of new patent citations in the current year; 2. Number of patent citations in the past five years; 3. Number of total citations (over a five-year rolling period and in the future) related to patents granted in the past five years. In prior studies, to control for the number of citations patents usually receive in a given field, the number of citations is generally scaled by the average number of citations received in the industry (Hirschey et al., 2001) or in the technology classification (Gu, 2005; Pandit et al., 2011). I scale each measure by the median value for the 3-digit SIC code industry and year. 9, 10 This scaling adjusts the patent measures to reflect the number of patent citations compared to other firms in the same industry in the same year (Hirschey et al., 2001). The first measure counts the number of citations a firm s patents receive in a given year. Trajtenberg (1990) finds that the number of citations received is better for R&D output than the 9 I calculate industry-year medians only with firms that have at least one patent or citation during that particular year. Using the industry-year mean does not change my results. 10 Scaling by the industry benchmark is consistent with Hirschey et al. (2001). Other prior studies (Gu, 2005; Pandit et al., 2011) adjust their measures by means of each USPTO subcategory year. Using these means does not change my results. I use industry-year instead of subcategory-year for two reasons: (1) Table 1 of Gu (2005) shows an example of Patents with their IDs and Patent Subcategories. I find discrepancies between the Patent Subcategories presented and the USPTO Classification from the actual patent documents. Rather, the Patent Subcategory numbers seem to come from the manner in which the NBER categorized the patent, not the USPTO. (2) Patents can be listed under multiple subcategories. An example illustrates both of these reasonings: Table 1 Panel A (Gu, 2005) shows a patent ID of 4911173 with a Patent Subcategory of 32. The original patent document (available on the USPTO via Google Patent) shows that the correct U.S. Classification is 128/662.06 and 128/4. The subcategory number 32 is given by the researchers who compiled the data available on NBER, not the USPTO. 12
number of a firm s patents. In other studies, the number of citations captures the relative importance of patents a firm holds (Gu, 2005; Gu and Wang, 2005; Pandit et al., 2011). The reasoning is that more important or innovative patents receive more citations from follow-up patents than less important or obscure patents. The second patent measure is the sum of the number of all citations received in the past five years. I use this measure that includes citations received in the previous years as well as the current year because the information about previous R&D success may be relevant this year (Cohen et al., 2013), The third measure is the number of citations in the past five years and in the future related to the past five years patents. The reasoning for using patents granted in the past five years follows the finding in Lev and Sougiannis (1996) that R&D capital (the sum of R&D expenses assuming straight-line depreciation) is value relevant, on average, for five years. I use these three patent citation measures as information about the firms patents whether this information is a signal of economic importance for future royalty income or for future R&D success. 4. METHODOLOGY 4.1 Patent Information and Future Firm Performance Before examining how analyst experience affects the relation between patent information and forecast errors, I confirm the association between patent citations and future firm performance found by prior studies (Gu, 2005; Pandit et al., 2011). If patents guarantee firms exclusive rights to discoveries and inventions, I expect firms with more patents to have better future firm performance. I follow the future earnings model in Cao et al. (2011) and estimate all regressions in this study using ordinary least squares. I control for heteroscedasticity using Roger s standard errors and by clustering the residuals by firm (Petersen, 2009): 13
E i,t+1 = α 0 + α 1 PATENT i,t +α 2 E i,t + α 3 (D 1 * E i,t ) + α 4 ΔE i,t + α 5 (D 2 * ΔE i,t ) + α 6 ΔBV i,t-1 + α 7 (D 3 * ΔBV i,t-1 ) + α 8 ΔD i,t + α 9 BM i,t + α 10 ΔRDCAP i,t + α 10 RD_EXP i,t + ε i,t [1] where E i,t+1 PATENT E i,t = operating income after depreciation for firm i in t+1 scaled by the market value of equity at the end of year t; = one of the following six patent measures: 1. Number of new patent citations in the current year, 2. Number of patent citations in the past five years, 3. Number of total citations (over a five-year rolling period and in the future) related to patents granted in the past five years; = operating income after depreciation for firm i in t scaled by market value of equity at the end of year t; D 1 = an indicator set to 1 if E i,t is negative, otherwise 0; ΔE i,t = change in operating income after depreciation for firm i from t-1 to t scaled by market value of equity at the end of year t; D 2 = an indicator set to 1 if ΔE i,t is negative, otherwise 0; ΔBV i,t-1 = the change in book value of equity for firm i from year t-2 to year t-1 scaled by market value of equity at the end of year t-1; D 3 = an indicator set to 1 if ΔBV i,t-1 is negative, otherwise 0; ΔD i,t BM i,t ΔRDCAP i,t = change in dividends for firm i from year t-1 to year t scaled by market value of equity at the end of year t-1; = book value of equity divided by market value of equity for firm i in year t; = change in R&D capital. R&D capital is calculated as (R&D expense in year t * 0.9 + R&D expense in year t-1 * 0.7 + R&D expense in year t-2 * 0.5 + R&D expense in year t-3 * 0.3 + R&D expense in year t-4 * 0.1) scaled by market value of equity at the end of year t-1; RD_EXP i,t = R&D expense in year t. 14
I also use an alternative measure of future firm performance. ROA i,t+1 is defined as operating income after depreciation for firm i in t+1 scaled by total assets at the end of year t: ROA i,t+1 = α 0 + α 1 PATENT i,t +α 2 E i,t + α 3 (D 1 * E i,t ) + α 4 ΔE i,t + α 5 (D 2 * ΔE i,t ) + α 6 ΔBV i,t-1 + α 7 (D 3 * ΔBV i,t-1 ) + α 8 ΔD i,t + α 9 BM i,t + α 10 ΔRDCAP i,t + α 10 RD_EXP i,t + ε i,t [2] I also examine the earnings and ROAs for year t+2 because the benefits of patents may take more than a year to manifest. Lastly, I use gross margin percentage as the dependent variable. If patents grant exclusive rights to the firms holding them, the firms may be able to command higher prices for their products or reduce costs of productions through patented technologies. Thus, I use gross margin percentage as the dependent variable for an exploratory reason: GM% i,t+1 = α 0 + α 1 PATENT i,t +α 2 E i,t + α 3 (D 1 * E i,t ) + α 4 ΔE i,t + α 5 (D 2 * ΔE i,t ) + α 6 ΔBV i,t-1 + α 7 (D 3 * ΔBV i,t-1 ) + α 8 ΔD i,t + α 9 BM i,t + α 10 ΔRDCAP i,t + α 10 RD_EXP i,t + ε i,t [3] where GM% i,t+1 = sales minus cost of goods sold divided by sales for firm i in t+1; All other variables are as defined above. In order to maximize the number of observations, I extend the sample period back to 1988 because these tests do not include analyst characteristics variables. 4.2 Patent Information, Forecast Revisions, and Forecast Errors Next, I test whether patent information is associated with analyst forecast revisions and forecast errors. For analyst forecast revisions, I follow Cao et al. (2011): FR i,t+1 = α 0 + α 1 PATENT i,t +α 2 E i,t + α 3 (D 1 * E i,t ) + α 4 ΔE i,t + α 5 (D 2 * ΔE i,t ) + α 6 ΔBV i,t-1 + α 7 (D 3 * ΔBV i,t-1 ) + α 8 ΔD i,t + α 9 BM i,t + α 10 ΔRDCAP i,t + α 10 RD_EXP i,t + ε i,t [4] 15
where FR i,t+1 = the earnings forecast revision measure calculated as earnings forecast for t+1 for firm i following the announcement of year t earnings minus the earnings forecast for t+1 for firm i following the announcement of year t-1 earnings scaled by the absolute value of the latter; All other variables are as defined above. I expect the coefficient on PATENT to be statistically significant if patent information is informative to analysts in forecasting and revising future earnings estimates. A positive (negative) and significant coefficient means analysts revise their earnings forecasts optimistically (pessimistically) with patent information. A statistically insignificant coefficient means that analysts do not find patent information to be informative or relevant in revising their forecasts. Next, I use the model from Amir et al. (2003) to test for an association between patent information and analyst forecast errors. Following Amir et al. (2003), I use the absolute value of forecast errors (forecast accuracy) and signed forecast errors as my dependent variables. Larger absolute values of forecast errors represent less accurate forecasts, while small absolute values represent more accurate forecasts. More negative signed forecast errors represent more optimistic earnings forecasts, while more positive signed forecast errors represent more pessimistic earnings forecasts. Abs(FE i,t+1 ) = α 0 + α 1 PATENT i,t + α 2 R&DIntensity i,t + α 3 SIZE i,t + α 4 CVER i,t + α 5 ln(age i,t )+ ε i,t [5] FE i,t+1 = α 0 + α 1 PATENT i,t + α 2 R&DIntensity i,t + α 3 SIZE i,t + α 4 CVER i,t + α 5 ln(age i,t )+ ε i,t [6] where Abs(FE i,t+1) = forecast accuracy calculated as the actual t+1 earnings minus the earnings forecast for year t+1 earnings most immediately after the year t 16
earnings announcement scaled by the stock price at the year t fiscal year end; FE i,t+1 = signed forecast errors calculated as the actual t+1 earnings minus the earnings forecast for year t+1 earnings most immediately after the year t earnings announcement scaled by the stock price at the year t fiscal year end; RDIntensity i,t = the R&D intensity calculated as R&D capital divided by the sum of book value of equity and R&D capital. R&D capital is calculated as (R&D expense in year t * 0.9 + R&D expense in year t-1 * 0.7 + R&D expense in year t-2 * 0.5 + R&D expense in year t-3 *0.3 + R&D expense in year t- 4 * 0.1); SIZE i,t CVER i,t Ln(AGE i,t ) = the natural log of market value of equity; = the coefficient of variation for earnings calculated as the standard error of the past five annual earnings divided by the absolute value of mean earnings; = firm age calculated as the natural log of the number of continuous year observations on Compustat. If analysts misunderstand (understand) the positive future firm performance implications of patent information, I expect the coefficient on PATENT to be positive (negative) and significant for Equation [5]. For Equation [6], I expect the coefficient on PATENT to be negative (positive) and significant if analysts are optimistic (pessimistic) about patent information. 4.3 Main Tests: Analyst Experience and the Relation between Patents and Forecast Errors To test my main hypotheses, I follow the Drake and Myers (2011) model. I rank all continuous independent variables into deciles and standardize them to be between 0 and 1 (denoted by a superscript R ), consistent with prior studies (Bradshaw et al., 2001; Collins et al., 2003; Drake and Myers, 2011). I start with the forecast error variables as dependent variables. My independent variables are patent measure, analyst firm-specific experience and the interaction term between the two variables: 17
Abs(FE i,j,t+1 ) or FE i,j,t+1 = α 0 + α 1 R Patent i,t + α 2 R Patent i,j,t * α 3 R Fexp i,j,t + α 3 R Fexp i,j,t +ε i,j,t [7] where Abs(FE i,j,t+1) = absolute value of the actual t+1 earnings minus the earnings forecast for year t+1 earnings most immediately after the year t earnings announcement for firm i and analyst j scaled by the stock price at the year t fiscal year end; FE i,j,t+1 Patent i,t Fexp i,j,t = actual t+1 earnings minus the earnings forecast for year t+1 earnings most immediately after the year t earnings announcement for firm i and analyst j scaled by the stock price at the year t fiscal year end; = One of six patent measures as defined previously; = number of years analyst j has issued earnings forecasts for firm i in I/B/E/S. Next, I add control variables for financial information: where Abs(FE i,j, t+1 ) or FE i,j, t+1 = α 0 + α R 1 Patent i,t + α R 2 Patent i,t * α R 3 Fexp i,j,t + α R 3 Fexp i,j,t + α R 4 WCaccr i,t + α R 5 WCcfo i,t + α R 6 Abs(FE i,j, t ) or FE i,j, t + α 7 Loss i,t + α 8 EquityOff i,t + α R 9 BTM i,t + α R 10 Recom i,t + α R 11 Ret i,t + α R 12 Numest + α R 13 Size i,t + α R 14 Age i,t + ε i,t [8] WCaccr i,t WCcfo i,t = working capital accruals calculated as Increase in A/R + Increase in Inventory + Decrease in A/P and Accrued Liabilities + Decrease in Accrued Income Taxes + Increase (Decrease) in Other Assets (Liabilities), scaled by average assets; = working capital cash flow calculated as Earnings before Interest, Taxes, Depreciation, and Amortization minus working capital accruals, scaled by average total assets; Abs(FE i,j, t ) or FE i,j, t = the lag of either Abs(FE i,j, t+1 ) or FE i,j, t+1 ; Loss i,t = 1 if Income before Extraordinary Items is less than 0, otherwise 0; EquityOff i,t = equity offering indicator set to 1 if sales of common and preferred stock are greater than purchases of common and preferred stock by more than 5% of total assets, otherwise 0; 18
BTM i,t Recom i,t Ret i,t Numest i,t Size i,t Age i,t = the book-to-market ratio; = the analyst's outstanding recommendation for the firm at the time of earnings forecast from I/B/E/S; = 12-month buy-and-hold stock return up to the month prior to earnings announcement; = the number of analysts following the firm as of the earnings announcement month in I/B/E/S; = total assets; = firm age measured as the number of consecutive years to date the firm appears in the Compustat Annual Database. Lastly, I control for other analyst characteristics such as portfolio complexity, brokerage size, and forecast horizon. I also include industry and year dummies: where Abs(FE i,j, t+1 ) or FE i,j, t+1 = α 0 + α R 1 Patent i,t + α R 2 Patent i,t * α R 3 Fexp i,j,t + α R 4 Patent i,t * R Bsize i,j,t + α R 5 Patent i,t * R NoFirms i,j,t + α R 6 Patent i,t * R NoSIC2 i,j,t + α R 7 Patent i,t * R Fhor i,j,t + α R 8 Fexp i,j,t + α R 9 Bsize i,j,t + α R 10 NoFirms i,j,t + α R 11 NoSIC2 i,j,t + α R 12 Fhor i,j,t + α R 13 WCaccr i,t + α R 14 WCcfo i,t + α R 15 Abs(FE i,j, t )/FE i,j, t + α 16 Loss i,t + α 17 EquityOff i,t + α R 18 BTM i,t + α R 19 Recom i,t + α R 20 Ret i,t + α R 21 Numest i,t + α R 22 Size i,t + α R 24 Age i,t + Industry_dummies + Year_dummies +ε i,j,t [9] Bsize i,j,t = brokerage size measured as the number of analysts belonging to a brokerage firm in year t in I/B/E/S; NoFirms i,j,t = the number of firms followed by analyst j in year t; NoSIC2 i,j,t = the number of industries (2-digit SIC) followed by analyst j in year t; Fhor i,j,t = the forecast horizon measured as the number of days between the forecast issuance date and the fiscal year end date for year t+1. My main variable of interest is the interaction term Patent*Fexp. If more experienced analysts make more accurate (more optimistic) forecasts by incorporating patent information, I expect the 19
coefficient to be negative and significant when the dependent variable is the absolute (signed) value of the forecast error. 5. RESULTS 5.1 Descriptive Statistics Table 1 shows the descriptive statistics. The mean of Fexp is 4.03, which shows that analysts, on average, have firm-specific experience of 4.03 years. The three patent measures have means just below 2 and medians around 0.5. Considering that I scale these measures by the industry averages of patent citations, this indicates that there are firms in certain industries with many more patent citations than other firms in the industry, and the distribution is somewhat skewed. The statistics for analyst characteristics and other control variables are similar to those of Drake and Myers (2011). [Insert Table 1] 5.2 Patent Information and Future Firm Performance Table 2 shows the results from estimating the model, which tests for a relation between patents and future firm performance. Panel A of Table 2 reveals that the coefficients on all three patent measures are positively and significantly associated with future (year t+1) earnings at 1% significance level. This result suggests that, even after controlling for R&D expenses and R&D capital, information about patent citations is informative about future firm performance. In addition, Panel B shows that this association holds when I use t+2 earnings as the dependent variable. [Insert Table 2] 20
Additional tests using the future return on assets (ROA) as an alternative measure of future firm performance show that the patent measures are positively and significantly associated with year t+1 and t+2 ROAs at the 1% significance level (Panels C and D). The results in Table 2 are consistent with the positive relation between patent citations and future earnings documented by Gu (2005). In exploratory analyses, I also use the gross margin percentage (GM%) as a dependent variable. If patents give firms a first-mover advantage, I expect firms with more patents and patent citations to have larger gross margin percentages, because sales revenues are larger due to higher prices or because the costs of goods sold are smaller due to cost-saving processes. In Table 3, I find that the patent measures are also positively associated with the future gross margin percentage in year t+1 at 1% significance level. [Insert Table 3] 5.3 Patent Information, Forecast Revisions, and Forecast Errors Table 4 reveals that the patent measures are positively and significantly associated with analyst forecast revisions. This suggests that analysts use patent information to increase their forecast earnings. One interpretation is that the analysts understand that patents lead to better future firm performance, and thus revise their earnings forecasts upwards. Another interpretation is that analysts view the patent citations as a signal of economic importance or as a signal of potential royalty income. When I use a dependent variable that compares forecasted earnings to actual earnings in absolute value, I find that the coefficients on the patent measures are positive and significant. This suggests that patent information is associated with less accurate forecasts even after controlling for R&D intensity (Table 5, Panel A). One explanation is that even though patents 21
and patent citations are positively associated with future earnings, patent information is difficult to incorporate into earnings forecasts. I do not find any significant coefficients on patent measures when I use signed forecast errors as the dependent variable (Table 5, Panel B). [Insert Tables 4 and 5] 5.4 Analyst Experience and the Relation between Patent Information and Forecast Errors Table 6 presents the results from tests of my main hypotheses. The coefficient on the interaction between the patent measure and analyst firm-specific experience (Patent*Fexp) is the variable of interest. In Panel A, the dependent variable is forecast accuracy, calculated as the absolute value of forecasted EPS minus actual EPS, scaled by stock price. A larger (smaller) value of Abs(FE i, t+1 ) represents a less (more) accurate forecast. I expect a negative coefficient on Patent*Fexp if more experienced analysts are better at incorporating patent information to make more accurate earnings forecasts. Panel A of Table 6 reveals that the coefficient of Patent*Fexp is not statistically significant. Thus, I do not find evidence that more experienced analysts are better at making more accurate earnings forecasts with patent information. Furthermore, the coefficient on Fexp is negative and significant, confirming the findings of prior studies, which find that analysts with more experience make more accurate earnings forecasts. [Insert Table 6] Next, I use signed forecast error as the dependent variable. More negative (positive) FE i,t+1 means more (less) optimistic earnings forecasts (Table 6, Panel B). The coefficient on Patent*Fexp is again my variable of interest. Given the previous finding that patent citations lead to better future earnings, if more experienced analysts make more optimistic earnings forecasts using patent information, I expect the coefficient on Patent*Fexp to be negative and significant. Panel B of Table 6 reveals that the coefficient on Patent*Fexp is negative and statistically 22
significant. This suggests that more experienced analysts better understand the implications and benefits of patents so they make more optimistic earnings forecasts. Prior literature shows that analysts are generally optimistically biased (Bradshaw et al., 2001; Drake and Myers, 2011), which is confirmed by the negative and significant intercepts. Overall, Panel B of Table 6 seems to show that more experienced analysts make more optimistic earnings forecasts using patent information. The overall results of Table 6 suggest that analysts with more experience are not better at making more accurate earnings forecasts by incorporating patent information. However, they do make more optimistic earnings forecasts, possibly from understanding that patents lead to better future firm performance. 6. ADDITIONAL TESTS 6.1 Alternative scaling for patent measures As previously mentioned in Section 3, I scale my patent measures by the industry-year median values. There is a possibility that my patent measures are merely proxies for firm size because larger firms tend to have more patents. In order to mitigate this concern, I divide the firms of each industry by the median size (total assets) into large and small firms. I then calculate the alternative industry medians by industry, year, and size. The results of my main tests using patent measures with alternative scaling are presented in Table 7. [Insert Table 7] As previously predicted for the main test, if more experienced analysts make more accurate earnings forecasts using patent information, I expect the coefficient on Patent*Fexp to be negative and significant. Panel A of Table 7 presents that the coefficient on Patent*Fexp is 23
not statistically significant. This means that using the alternative scaling for my patent measures the accuracy of analyst earnings forecasts related to patent information is not better for more experienced analysts. Panel B of Table 7 presents the results of a test with signed forecast errors as the dependent variable. The coefficient on Patent*Fexp is negative and significant for two out of three patent measures, which is somewhat consistent with the results of my main tests. The P- value for the second patent measure is just outside of the 0.10 significant level at 0.11. Overall, the results of the test using the alternative scalar for my patent measures (Table 7) are consistent with the results of my main results (Table 6). 6.2 Other Types of Experience In addition to firm-specific experience, prior literature examines other measures of analyst experience such as general experience (Mikhail et al., 1997, 2003; Clement, 1999; Drake and Myers, 2011) and task-specific experience (Clement et al., 2007). I examine whether these two measures of analyst experience as well as industry-specific experience are correlated omitted variables. General experience measures the number of years an analyst has made earnings forecasts for any firm. Next, understanding patents may be industry specific. Prior literature generally does not focus on industry-specific experience, and I measure the industry-specific experience as the number of years an analyst has made earnings forecasts for a specific industry (3-digit SIC). For task-specific experience, Clement et al. (2007) examine whether analysts, who have experience on forecasting earnings for firms that undergo mergers and acquisitions, subsequently make more accurate earnings forecasts for firms that undergo mergers and acquisitions. Although patents are common in certain industries, I explore the possibility that using and understanding patents is a specific task for analysts. I measure the task-specific experience as the number of years an analyst has made earnings forecasts for firms with patents. 24