Identifying and Evaluating Contributions to Value within a Firm Assessing the Relative Economic Importance of Activities and Investments Economics Partners, LLC White Paper Series White Paper 2015-01 Tim Reichert and Erin Hutchinson Table of Contents I. II. Introduction ................................................................................................................................... 1 Decomposing Value: Overview of Economics Partners’ Research Strategy ....................... 4 A. Overview of Approach .......................................................................................................... 4 B. Tobin’s Q.................................................................................................................................. 4 III. Review of Relevant Literatures ................................................................................................... 7 A. Introduction............................................................................................................................. 7 B. Intangible Asset Investments ................................................................................................ 7 1.Early Studies of R&D Investments, Productivity, and Market Value ..................................... 8 2.Studies of the Rate of Return on R&D Investments ................................................................... 9 3.Studies of Marketing and Sales Investments ............................................................................ 13 4.Summary ........................................................................................................................................ 17 C. Multinational Enterprises .................................................................................................... 17 D. Cost Efficiency ...................................................................................................................... 20 IV. Econometric Model ..................................................................................................................... 22 A. Sample Strategy .................................................................................................................... 22 B. Model Specification .............................................................................................................. 22 1.Dependent Variable ...................................................................................................................... 22 2.Independent Variables ................................................................................................................. 23 3.Econometric Model ....................................................................................................................... 24 C. Model Results........................................................................................................................ 24 1.Descriptive Statistics ..................................................................................................................... 24 2.Econometric Results ..................................................................................................................... 25 3.Conclusion ..................................................................................................................................... 26 V. Attribution of Profits within a Firm ......................................................................................... 27 A. Overview of Statistical Attribution .................................................................................... 27 B. General Procedure: Two Steps .......................................................................................... 28 C. Attribution of Profits to Variables ...................................................................................... 28 D. Example of Attribution of Variables to Entities ............................................................... 29 1.Factors Attributable to IPCo ........................................................................................................ 30 2.Factors Attributable to OpCo ...................................................................................................... 30 3.Unattributed Factors..................................................................................................................... 30 E. Summary of Results ............................................................................................................. 30 VI. Conclusions Reached .................................................................................................................. 32 Economics Partners, LLC www.econpartners.com Page | 1 I. Introduction The past several years have seen an increased emphasis by transfer pricing regulatory bodies (tax authorities and quasi-regulatory bodies such as the United Nations and the OECD) on the economic determinants of value within the multinational enterprise. Arguably, this has corresponded with a de-emphasis, by the same transfer pricing regulators, of intercompany contracts and the comparability of controlled and uncontrolled transactions. We have argued elsewhere that an emphasis on the causes of shareholder value, while perhaps useful in principle, suffers from an inherently high degree of subjectivity. It is easy for both tax authorities and taxpayers to assert, without reliable quantitative evidence, that one or another activity or asset is the primary “value driver” within a business. The subjectivity of these sorts of assertions, or “value statements” as we call them in this paper, is one reason for the prevalence of seemingly intractable audits and multi-year transfer pricing negotiations. Such audits tend to devolve into arguments over the relative importance of things like R&D versus marketing and sales, or of recent R&D versus “early-stage and high risk” R&D, or of marketing and sales versus manufacturing, and so on. While it is certainly true that R&D generates value, and it is also true that early stage R&D which has proven successful is often more valuable than follow-on R&D (and for that matter it is true that marketing and sales give rise to value), the real question is “what empirical evidence exists regarding the relative values – or relative claims over operating profit – of these purported sources of value?” Said differently, “which value statements are credible?” Is it credible, for example, to argue that manufacturing or assembly is the primary claimant of profit and the primary value driver, and if so under what conditions and in what industries? The purpose of this paper is to describe a framework for identifying the primary determinants of shareholder value and operating profit within a firm, and for estimating an allocation of operating profit among them.1 Our objective is admittedly a minimalist – i.e., to provide a framework that can produce empirical benchmarks for assessing the credibility of competing assertions regarding value creation within the MNE. Our methodology is straightforward and will be familiar to economists. Following a substantial economics literature that measures the relationship between “Tobin’s Q” (which is the ratio of market enterprise value to book value) and various enterprise value drivers, we first show how to devise a structural model that explains the Q as a function of key value creating activities. We focus on Tobin’s Q, or simply “Q,” because the ratio of market enterprise value to book asset value is a direct indicator of the productivity of the firm’s investments. The higher is Q, the more profitable the firm’s investments are expected to be. In fact, as we show later, there is For purposes of this study, the term “residual profit” will refer to operating profit earned by a business in excess of the required operating profit return to physical and financial capital. In short, residual profit, for purposes of this document, is the return to intangible capital. 1 Economics Partners, LLC www.econpartners.com Page | 2 a direct arithmetic linkage between Q and operating profit, which means that a model that relates Q to various value drivers is also a model that relates operating profit to those value drivers. Within our model, the structural determinants of Q are chosen because specific academic literatures exist to support the claim that these are in fact determinants of shareholder value and operating profit. That is, literatures present and test the hypothesis that these “factors” constitute investments or activities that are net present value positive (the cost of these activities or investments is less than the present value of the operating profits that they generate). Once we have a reasonably accurate and reliable structural model, we then use a technique called “variance decomposition” to determine the relative importance of our chosen factors. The term variance decomposition is a synonym in this context for “value attribution.” Variance decomposition is a way that economists and statisticians use to attribute explanatory power, or more loosely causality, to individual variables. Variance decomposition gives us the relative contributions of each variable in an econometric model to the variable of interest – in this case, operating profit. As an intuitive example of the way in which variance decomposition works, consider the team members of an American football team. Across the National Football League, it will almost always be the case that a team’s kicker will lead the team in total scoring. For example, if a team scores 23 points in a game, the kicker will perhaps have scored 11 out of the 23 (three field goals and two extra points, or nearly half of the team’s points). If we measure a kicker’s relative contribution to the team’s success by looking only at this statistic, we will grossly overstate the importance of the kicker. Why is this? Because losing a kicker is not a serious problem for most teams. By imagining what would happen to the team’s success if the kicker left the team, we see that the kicker is in fact not very important relative to players in other positions – a kicker can easily be replaced. Continuing with the example, left tackles score very few points. Left tackles never lead their team in total scoring, and often go their entire careers without scoring any points. Thus, if we measure the left tackle’s relative contribution to the team’s success by looking only at this statistic, we will grossly understate his importance. But losing a left tackle is almost always a serious problem for a football team. It also explains why left tackles are often taken in the first round of the NFL draft and are paid tens of millions of dollars, while kickers often go undrafted and many are paid near the league minimum. In other words, by imagining what would happen to the team’s success if the left tackle were injured or replaced, we see that the left tackle is in fact very important on a standalone basis. This approach – i.e., measuring the effect of losing a “factor” – is exactly what variance decomposition does. Variance decomposition measures the relative importance of each variable by examining the effect of not having that variable, or factor, as part of the group of variables that explain the dependent variable. That is, variance decomposition measures the effect of not Economics Partners, LLC www.econpartners.com Page | 3 having a factor as part of the “team,” in order to determine the relative share of the dependent variable (in this case, economic value) that should be attributed to that factor. By applying variance decomposition to our structural model we are able to estimate each variable’s relative importance to Tobin’s Q (and therefore for operating profit). While this could arguably be used to devise a profit split, we propose that the modest objective of providing a benchmark against which value statements can be assessed is the more appropriate use of this approach. This research paper follows a three step research strategy. First, we review economics literature related to value creation. Second, we develop an econometric model that relates Q (and therefore operating profit) to a set of “value drivers,” using a set of comparable publicly-traded services firms.2 Finally, we use variance decomposition to determine high level value shares among these value drivers. This paper proceeds as follows. Section II outlines our research strategy. Section III provides a review of the relevant literatures. Section IV describes our econometric model. Section V then describes our use of variance decomposition to estimate a high level attribution of shareholder value and operating profit to the factors identified in Section IV. 2 By “comparable,” we mean that these firms are comparable to one another. Economics Partners, LLC www.econpartners.com Page | 4 II. Decomposing Value: Research Strategy A. Overview of Approach Our research strategy comprises three primary steps. First, we survey the fields of economics and finance for literatures that examine the determinants of productivity and profitability. Second, relying on insights from the literatures surveyed, we develop an econometric model that predicts a firm’s market enterprise value to book asset value ratio (i.e., its capitalized residual profit) based upon factors that research economists have identified as value-relevant over time. This ratio is known in economics as the Q ratio, or Tobin’s Q. Finally, we apply variance decomposition to the econometric model developed in Step 2 to attribute profit to the value drivers obtained in Step 2. B. Tobin’s Q Later sections of this paper describe the results of our research for each of the three steps described above. However, before proceeding to these some discussion of our reasons for focusing on the market-to-book ratio (Q) are in order. Within the fields of economics and finance, much emphasis is given to a measure called “Tobin’s Q.” Tobin’s Q was developed by Nobel Laureate James Tobin in 1969, prompted by the following passage from John Maynard Keynes’ General Theory. [The] daily revaluations of the Stock Exchange, though they are primarily made to facilitate transfers of old investments between one individual and another, inevitably exert a decisive influence on the rate of current investment. For there is no sense in building up a new enterprise at a cost greater than that at which a similar existing enterprise can be purchased; whilst there is an inducement to spend on a new project what may seem an extravagant sum, if it can be floated off on the Stock Exchange at an immediate profit.3 Recalling that Keynes’ core macroeconomic concern and research focus was investment. That is, his observation was that investment would often not match savings. Given this, the idea behind Tobin’s Q is that corporate investment can be predicted with some degree of accuracy by examining the Q ratio. Whenever Q exceeds 1, the market value of existing assets exceeds their replacement cost. This should drive new investment, as it is cheaper to create new assets (including new firms) than to buy existing assets. Correspondingly, whenever Q is below 1, the opposite pattern holds (i.e., investment should fall or be negligible, as it is cheaper to buy existing assets through the stock market than to invest anew). 3 The General Theory of Employment, Interest, and Money. John Maynard Keynes, 1936. p. 151. Economics Partners, LLC www.econpartners.com Page | 5 In practice, Tobin’s Q is generally measured as the market enterprise value of a firm (its market capitalization plus debt) divided by its book value of assets. While the measure should, technically, have the replacement cost of assets in the denominator, in practice, the replacement cost of a firm’s existing asset base is very difficult to estimate reliably. For our purposes, Tobin’s Q is of interest for many reasons. These are as follows. First, Tobin’s Q involves measures of capitalized value, rather than transitory profit measures. The numerator of Tobin’s Q is the market capitalization of the firm, which is the net present value of the market’s expectations for all future profits (available to equity holders) into perpetuity. Thus it is not dependent on profits in the current or recent past period, which could be influenced by one-time, non-recurring events. This means that econometric measurements of the correlation between Tobin’s Q and certain causal factors should be more reliable than econometric measurements that examine the relationship between profitability and causal factors – because Tobin’s Q influenced by long term systemic considerations regarding the firm’s prospects, in a way that annual or quarterly profitability data is not. Second, Tobin’s Q is market-driven. Therefore, it is the result of an information aggregation mechanism rather than a function of corporate forecasts. The market capitalization represents the aggregation of the beliefs of thousands of informed market participants about the future prospects for a firm (including both level and riskiness of profits) rather than any specific forecasts prepared by the company. Thus Tobin’s Q represents an objective market expectation, and cannot be manipulated by firm management. Third, Tobin’s Q is forward looking. That is, Q capitalizes future profits rather than examining only the current period or past periods. This means that conclusions regarding the relationship between Tobin’s Q and a given causal factor are applicable to future time periods. Fourth, Tobin’s Q corresponds closely to residual profit, and to the value of intangible assets. Because Tobin’s Q values above (or below) 1 are measuring market enterprise values that are above (or below) the value of a firm’s physical and financial capital, this is capturing the market’s value of the firm’s intangible assets (i.e., those assets that are not recorded on the balance sheet). Since physical and financial assets typically earn their required rate of return (rarely much higher or lower), they will also tend to trade at or near the present value of the expected profits associated with that required return. Therefore, a Tobin’s Q above 1 implies that the market expects total profits in excess of the required return to the physical and financial capital. In other words, the market expects residual profit. And that in turn implies that the market recognizes that the firm owns intangible assets, even though those assets do not appear on the balance sheet. Fifth, for any given period, Tobin’s Q can be converted from a measure based upon stock values (i.e., observed market values) into a corresponding measure involving operating profit flows. This means that if we understand the determinants of Tobin’s Q, we also understand the determinants of a firm’s steady state operating and residual profit flows. To see this, note that Economics Partners, LLC www.econpartners.com Page | 6 ∞ Q = TEV / Assets. Therefore, Q*Assets = TEV. TEV = ∫0 ππππ΄ππ‘ × π −ππ‘ ππ‘, and NOPAT = EBIT*(1-t). Therefore, π∗π΄π π ππ‘π (1−π‘) ∞ = ∫0 πΈπ΅πΌππ‘ × π −ππ‘ ππ‘. The right-hand side of this equation is simply the present value of operating profits into perpetuity. Finally, Tobin’s Q is heavily relied upon by investors. For example, see Smithers & Co. in the UK, which discusses Tobin’s Q at length in its discussion of its investment philosophy. Finance and other research analysts have also written extensively about Tobin’s Q. See for example Professor Aswath Damodaran’s “Investment valuation: Tools and techniques for determining the value of any asset valuation”, or Robert Huebscher of Advisor Perspectives, “The Market Valuation Q-uestion”. Economics Partners, LLC www.econpartners.com Page | 7 III. Review of Relevant Literatures A. Introduction As outlined in the previous section, the first step in our research strategy is to review available literature related to determinants of profit and productivity. Our goal is to identify economics and finance literatures that can be used as a guide to the identification of assets, or activities, that create value for multinational firms in the IT services industry. We identified two primary categories of research that pertain to value creation. The first is research related to Tobin’s Q, which was discussed in the prior section, and the second is research related to what economists call “Total Factor Productivity” (“TFP”). Productivity, at its most basic level, is the ratio of output per unit of input. From an economic perspective, productivity measures output per unit of capital and/or labor. The term “total factor productivity” refers to output (generally the output of a firm) that cannot be directly attributed to either labor or capital inputs, as traditionally measured. That is, TFP is the amount of output that results from combining capital and labor inputs. In plain English, it is the difference between the “whole” and the “sum of the parts.” TFP is thus a “residual,” sometimes referred to as the “Solow Residual,” that equals the “extra” output produced when a firm combines labor and capital to create something more than the sum of the labor and capital “parts.” Obviously, there is a close correspondence between the idea of total factor productivity and the concepts of intangible assets and residual profit. It is obvious that the firm’s organizational, technological, and customer-based, intangible assets are an important reason why the value of what a firm produces is often greater than the cost of the inputs that it uses. It should also be obvious that there is a close relationship between the concept of residual profit and residual productivity (the Solow Residual). Through our review of the literature pertaining to TFP, we were able to classify the determinants of residual value and/or Tobin’s Q into the following three broad categories: ο· ο· ο· Intangible asset investment; Contributions of MNEs; and Cost efficiency. In the following subsections, we summarize several key articles that examine the determinants of residual value, as measured by estimates of Tobin’s Q and/or TFP. B. Intangible Asset Investments A key focus of the TFP literature has been on the relationship between investments in intangible assets and productivity. This literature seeks to explain the relationship between productivity Economics Partners, LLC www.econpartners.com Page | 8 and investments in intangible capital, which are not directly reflected in traditionally measured inputs (i.e., physical capital and labor). A large portion of this literature on intangible assets and productivity has focused on the relationship between technology intangible capital, as measured by R&D investment, and productivity. These studies yield valuable insights into the relationship between intangible capital and productivity, and provide a useful basis of studying the relationship between intangible capital and Tobin’s Q. 1. Early Studies of R&D Investments, Productivity, and Market Value One of the first studies to analyze the relationship between intangible asset investment and productivity at the firm level was conducted by Griliches and Mairesse (1984). Griliches and Mairesse develop a sample of R&D-performing manufacturing and non-manufacturing firms over time, and examine the relationship between a firm’s productivity and its physical capital stock, labor, and the unobserved R&D capital stock (i.e., its technology intangible capital). Specifically, this study analyzes the relationship between productivity and R&D investment across a broad sample of firms that perform R&D. The study also examines this relationship over varying time periods, and between different types of firms. While scientific firms (i.e., firms operating in the chemical, drug, computer, electronics, and instruments industries) exhibit higher R&D intensity and productivity growth compared to non-scientific firms, the authors find that the relationship between R&D investment and productivity is relatively consistent between scientific and non-scientific firms. Moreover, this study highlights the importance of R&D capital relative to physical capital in explaining a firm’s productivity and residual profit.4 While Griliches and Mairesse (1984) find that there is a positive relationship between R&D investment and productivity, Cockburn and Griliches further develop the role of knowledgebased intangible capital by examining the relationship between knowledge-based intangible capital and stock market valuation. Cockburn and Griliches measure the stock market valuation of knowledge-based intangible capital by examining examine the relationship between a firm’s knowledge-based intangible capital, as measured by a firm’s capitalized R&D expenses and stock of existing patents, and stock market valuation, measured using Tobin’s Q. The authors find a stronger relationship between R&D expenditure and value than between a company’s stock of patents and its market value. Markets view R&D expenditures as a stronger measure of a firm’s ability to create technical innovation than patents as an ex post observation of technical innovation’s output. In other words, markets do tend to attribute value to R&D, the input a firm uses to produce technical innovation.5 More fundamentally, Cockburn and Griliches, Z., and Mairesse, J., (1984), Productivity and R&D at the Firm Level. In R&D, Patents, and Productivity. Ed. Zvi Griliches. University of Chicago Press, 339-374. 5 Cockburn, I., and Griliches, Z., (1988), Industry Effects and Appropriability Measures in the Stock Markets Valuation of R&D and Patents, American Economic Review, Proceedings Issue, 419-423. 4 Economics Partners, LLC www.econpartners.com Page | 9 Griliches find that stock market valuations, and Tobin’s Q specifically, do reflect a firm’s knowledge-based intangible capital. Griliches and Mairesse (1984) examine the relationship between R&D expenditures and productivity, while Cockburn and Griliches (1988) seeks to explain the relationship between R&D expenditure and stock market valuation using Tobin’s Q. Griliches and Mairesse (1984) find that R&D expenditures are a key determinant of a firm’s productivity, and Cockburn and Griliches (1988) further demonstrate that there is a positive relationship between R&D and firm value. Taken together, Griliches and Mairesse (1984) and Cockburn and Griliches (1988) highlight the relationship between R&D, productivity, and Tobin’s Q, and demonstrate that R&D expenditures have a positive relationship with a firm’s productivity and its market valuation. This is not surprising in light of research by Antonelli and Colombelli (2009), which shows that TFP generally accounts for Tobin’s Q. 2. Studies of the Rate of Return on R&D Investments Economists have been measuring the returns to R&D investments for over 50 years. The first serious empirical investigation of the returns to R&D investments appeared in the Journal of Political Economy in 1958 (Griliches, 1958), and most of the literature since that time has followed the same basic analytical framework. The methods for evaluating the return to R&D investments fall into three categories. ο· ο· ο· The production function approach. The production function approach begins from the premise that a firm can be represented as an output function, or production function, F, of inputs such as capital, K, labor, L, and R&D. The firm’s output is then simply modeled as F = F(K,L,R&D), where R&D is the accumulated stock of R&D expenditures, net of depreciation.6 Given a specific functional form,7 publicly available data are used across large samples of firms and long time series to econometrically examine the relationship between changes in the value of firms’ outputs and the investments made in inputs (such as R&D). The market value approach. This approach is very similar to the production function approach, except that the relationship modeled is between firms’ market valuations and K, L, and R&D. The accounting approach. This approach uses econometric techniques to examine the relationship between accounting measures of firm performance (e.g., operating margin, return on equity, return on assets) and R&D expenditure. It is customary in the economics literature to treat the depreciation rate of R&D as 15 percent per year. However, Hall (2007) studies this empirically, and finds that depreciation rates are more likely between 20 and 40 percent per year. 7 Usually Cobb-Douglas. 6 Economics Partners, LLC www.econpartners.com Page | 10 In general, the accounting approach does not measure all of the inputs into a firm’s production of goods and services, and therefore does not hold constant these inputs when measuring the standalone contribution of R&D. In part as a result of this, the accounting approach tends to find a weak correlation between R&D investments and accounting measures of performance. For example, a recent study by Booz Allen Hamilton found that the correlation between R&D and return on equity and R&D and operating margin were both near zero, and that the relationship between margin growth and R&D was only weakly positive. In general, academic economists tend to shy away from the accounting approach. Instead, economists have focused their efforts on the production function approach and the market value approach. These approaches have the distinct advantages of identifying and holding constant other basic inputs such as capital and labor, and of greater realism. The production function and market value approaches both yield a rate of return to R&D by measuring the effect of changes in the values of inputs (capital, labor, and R&D) on changes in the value of a firm’s output or market-measured enterprise value. The rate of return to R&D is then simply the change in output or market value divided by the change in the firm’s net R&D stock. Importantly, the production function and market approaches attempt to measure the private returns to R&D, rather than the social returns. R&D creates social benefits that firms “externalize,” meaning that they are unable to capture the full social value of the technology that they create. However, R&D investment decisions are made based upon the “private” benefits of the R&D, meaning the benefits that firms can capture in their profits. It is clear from the literature that the gap between private and social returns to R&D is large. Technology spillovers from one firm to another mean that the measured value of a firm’s output is not only a function of the firm’s investments in its own R&D, but also a function of the benefits of R&D by other firms that has “leaked” into the public domain.8 Hall (2009) provides a comprehensive survey of the results of both the production function and market approaches to the measurement of the rate of return to R&D. The exhibit below summarizes the findings of 51 studies surveyed by Hall. As we discuss later, this has important implications for the Investor Model framework. Specifically, if a share of the value of the firm’s output is generated by the R&D of other firms, then “enterprise value” methods such as the market capitalization method or the income method that value technology by subtracting tangible assets from either the firm’s total enterprise value (market capitalization method) or total operating profit (income method) are including, among other things, the value of other firms’ technologies in the valuation of the subject firm’s technology. 8 Economics Partners, LLC www.econpartners.com Page | 11 Exhibit III-1 The exhibit also shows a wide dispersion in the measured rate of return to R&D. This is caused by variation in time periods covered, industries studied, and econometric models used. However, the exhibit also shows that the vast majority of the studies find that the rate of return to R&D is between 10 percent and 50 percent – with 34 of the 51 studies surveyed falling within this range. It is also clear that almost none of the studies found negative or single digit rates of return, and only 12 of the 51 studies found rates of return above 50 percent. The median study found that the rate of return to R&D was 29 percent. The average from the studies surveyed was 31 percent. According to Hall, these represent reasonable estimates of the rate of return to R&D. The exhibit below provides a summary of the studies surveyed by Hall, and their rate of return findings. Summary statistics are given at the bottom of the exhibit, including the interquartile range. Economics Partners, LLC www.econpartners.com Page | 12 Exhibit III-2 Economics Partners, LLC www.econpartners.com Page | 13 3. Studies of Marketing and Sales Investments While the relationship between investments in technology and productivity is well-studied, technology is not the only type of intangible asset that creates value and leads to higher productivity. There is also literature, though not as copious, dedicated to examining the relationship between customer-based intangible assets and Tobin’s Q. Ayanian (1983) examines the fact that accounting rates of return tend to be positively correlated with the advertising-to-sales ratio. That is, firms with higher advertising-per-dollar-of-sales tend to have persistently higher rates of return over time. This fact seems to contradict the theory of competitive markets, which states that risk adjusted rates of return of competitors tend to equality over the long term.9 There are two contending hypotheses for explaining this phenomenon. One is that the positive correlation between advertising and rates of return is due to the fact that high advertising creates barriers to entry that allow for possibly persistent “above average” rates of return for firms that advertise heavily. Put differently, potential competitors will avoid entering industries that require heavy advertising, thus increasing the likelihood that firms already operating in the industry earn above normal returns. The second hypothesis is that accounting bias explains the correlation. The argument here is that if advertising is a long-lived investment, then expensing of advertising spending (and the concomitant exclusion of the “advertising asset” from the balance sheet) will bias the rate of return upward by reducing the asset base on which the return is measured. Thus, firms with high advertising spending have excluded a relatively large asset from the balance sheet, causing a significant overstatement of the rate of return. Ayanian tests the second hypothesis by estimating the advertising rate of economic obsolescence. If it is found that the annual rate of economic obsolescence is at or near 100 percent, then advertising would not be a long-lived investment and the “accounting bias” argument would not be supported. If, on the other hand, annual advertising rates of economic obsolescence are found to be well below 100 percent, then advertising would in fact be a longlived investment and the “accounting bias” hypothesis would be supported to some degree. Support for the theory of competitive markets does not require that rates of return be identical over the long term. Limitations in the data and the simple fact that random, uncorrelated factors affect returns on an annual basis will inevitably cause observed rates of return to differ somewhat. Nonetheless, economists would expect returns to be “close” to each other and to the average of all companies over time. On the other hand, economists would not expect one firm to have a systematically and persistently higher return than another competing firm operating in a similar fashion and under similar circumstances (unless one firm has a monopoly position of some sort). It turns out, however, that the observed differences in accounting rates of return are so pronounced that economists have sought some explanation for these differences. 9 Economics Partners, LLC www.econpartners.com Page | 14 Ayanian describes the statistical estimation of an equation that relates industry average accounting rates of return on stockholders equity to industry advertising-to-sales ratios, measures of capital requirements, and industry growth. The model uses data for the years 1957-59 and it includes observations for 39 industries representing beverages, food, clothing, appliances, motor vehicles, cosmetics, furniture, and soaps and detergents as well as other consumer products industries. The results indicate that the average durability of advertising in the sample is 6.8 years.10 Hirschey (1982) and Hirschey and Weygandt (1985) extend prior research on whether advertising expenditures should be viewed as a long-lived capital good that depreciates over time. Hirschey (1982) contains an interesting critique of earlier studies in which economists have explored the relationship between sales and advertising. First, Hirschey argues that the emphasis of product sales and product advertising has ignored the potential impact of “institutional” advertising on overall aggregate demand for firm product. Institutional advertising works by improving the firm’s image or aids the firm in interactions with the public. Thus, institutional advertising, while not directed at promoting specific products, will potentially have a spillover effect according to which better firm image improves all product sales. Ignoring such advertising, which (it is claimed) earlier studies have done, will tend to understate the durability of advertising. Both papers specify virtually the same econometric model, which posits that a firm’s ratio of market value to replacement cost11 is explained by advertising spending, R&D spending, the market concentration ratio, growth, and company beta. The model is estimated using 1977 data for 390 firms taken from the Fortune 500. The idea behind the econometric procedure is the following. Firm market value represents the discounted present value of future expected cash flow. If advertising has a strong positive effect on market value, then it follows that advertising in the current year must be expected (by the market) to impact future cash flows for some period of time. The larger is the statistical relationship, the larger is the impact of advertising, and the longer the life of advertising spending. The model provides estimates of a highly statistically significant relationship between firm market value and advertising. The results hold for both durable and nondurable goods as well as for the sample as a whole.12 Thus, it appears from the results that current year advertising is Page 358. Ayanian uses data taken from Comanor and Wilson (1974). Comanor and Wilson find that advertising has no durability whatsoever: annual economic obsolescence is 100 percent. Ayanian identifies and corrects a misspecification that is found in the Comanor and Wilson paper, however, and reaches a conclusion quite different from that of Comanor and Wilson. Ayanian contains various statistical tests that provide compelling support for its conclusions. 11 Replacement cost and book value are virtually identical. 12 Page 331. 10 Economics Partners, LLC www.econpartners.com Page | 15 perceived as increasing market value, and therefore future cash flow. That is, the market believes that advertising imparts a durable effect on cash flow. Thomas (1989) identifies three primary causes of persistently high profits in heavily advertised consumer goods: 1) durability of advertising, 2) scale economies, and 3) heterogeneity for advertising. Thomas suggests that studies showing high durability of advertising use statistical models that are poorly specified in that these models focus only on buyer behavior and ignore advertiser behavior. The basic idea is that, while advertising induces buyers to purchase branded product, greater purchases of branded product induce companies to increase advertising. Thomas provides evidence that the premium profits earned by advertising companies are due to the economies of scale in advertising and not to the durability of advertising. As brand revenue rises, advertising per dollar of revenue required to maintain sales and margins declines, yielding higher margins and the observed premium profits. The evidence is embodied in brand advertising to brand sales ratios in the U.S. soft drink and cigarette industries. Brands with the highest sales tend to exhibit the lowest advertising to sales ratios. The data on brand advertising comes from Advertising Age, however, and we doubt that these data capture all the brand advertising performed by the owners of the brands. Thomas estimates a model using data for 26 cigarette brands covering the years 1955-70, and 15 soft drink brands cover the years 1962-80. The results show that advertising exhibits a high degree of durability with an annual rate of economic obsolescence of between 2 percent and 10 percent. Once she augments these equations with a set of advertiser response equations, the rate of economic obsolescence for cigarettes is virtually 100 percent per annum while for soft drinks the rate of economic obsolescence is over 80 percent per annum. Landes and Rosenfield (1994) is a direct response to the Hirschey papers. Landes and Rosenfield are concerned that the advertising rates of economic obsolescence revealed in other studies are much too low. The authors provide estimates of the annual rate of economic obsolescence of advertising by two-digit SIC code.13 The study examines data for 417 companies covering the years 1982 thru 1986. Rates of economic obsolescence for advertising derived from this first model range from 26.5 percent in Motor Vehicles (#37) to zero percent in several industries such as General Merchandise stores (#53). The industry of interest here is Chemicals and allied products (#28) with an estimated rate of economic obsolescence of 10 percent.14 Landes and Rosenfield then alter the model to include a firm-specific dummy variable. The notion that led to the inclusion of the firm-specific dummy is that firms with higher quality products will advertise more because the productivity of advertising increases with the quality 13 14 Page 268, Table I. Page270, Table II. Economics Partners, LLC www.econpartners.com Page | 16 of the good. If quality could be included as a separate variable, the measured effect of advertising on sales might decrease. Product quality is not generally observable, but the firm dummy variable would in theory reflect differences in quality (if there are no other unobservable differences that affect advertising and sales). The firm-specific dummy variable is intended to capture unobserved variables (such as quality of product) that may be positively correlated with both advertising and sales. Because of this positive correlation, Landes and Rosenfield argue, omission of the dummy variable from the model results in too high an attribution of observed sales to lagged advertising expenditures.15 Inclusion the dummy variable raises overall rates of economic obsolescence considerably. The lowest rate of economic obsolescence estimated is now 25.6 percent in the Paper and Allied Products industry (#26), while certain other industries reveal 100 percent rates of economic obsolescence: Holding and other Investment Offices (#67). For Chemicals and Allied Products, the rate of economic obsolescence rises to 55.4 percent.16 Graham and Frankenberger (2000) examine the asset value of advertising for a sample of 320 firms that reported advertising expenditures for each of the 10 consecutive years ending in 1994. They estimate econometric equations for three different industry groups: consumer products, industrial products, and sales and services companies. They find that advertising expenditures are significantly associated with earnings up to five years following the year of the expenditure. They further find that asset values are longest lived in the consumer and industrial products industries, and shorter lived in the sales and services industry. There are several notable aspects of this article. First, they estimate parameters for all consumer products. Second, they estimate first an earnings equation whereby firm earnings are related to present and lagged advertising expenditures; and then, using parameters from the earnings equation, estimate a market value equation whereby firm market value is related to advertising asset, among other variables. Third, they include R&D in the model, thus controlling for R&D’s influence on profitability and value. Their key result is that the advertising asset does not depreciate over the first two years after the expenditures take place. In years three, four and five, however, the effect of advertising drops off significantly and then, presumably goes to zero in year six. One cannot compute a simple parameter for the rate of economic obsolescence from the Graham and Frankenberger paper nor do the authors attempt to compute rates of economic obsolescence. The key result from this paper, however, is that advertising has an effect on earnings for five years from the date of the advertising spending. Thus, a 20 percent straightline economic obsolescence pattern is roughly consistent with these results. 15 16 Page 272. Page 270, Table II. Economics Partners, LLC www.econpartners.com Page | 17 4. Summary In summary, research has shown that investments in intangible assets, both technology and customer-based, have a positive impact on productivity and therefore create intangible value. It is clear that the research on the impact of R&D is both broader and deeper than the research on sales and marketing investments. It is also clear that the research on R&D is quite uniform in its conclusion that research and development expenditures exhibit a high average rate of return, and are productive of residual value. By contrast, the research on marketing and sales investments is less broad, and less clear. Marketing and sales investments are generally found to give rise to a long-lived asset, but this is not a uniform result. Moreover, the rate of return to marketing and sales investments is in some studies very low (Landes & Rosenfield). C. Multinational Enterprises Given the critical role of globalization in the IT services industry, the growing research into the value creation of multinational enterprises is of particular interest to us. Specifically, we reviewed literature that examines the TFP of MNEs as compared to local firms. As discussed earlier, MNE’s introduce a number of important advantages and efficiencies to a business, including capital budgeting, improved management and oversight, economies of scale, cost savings through offshoring, and other organizational intangibles. Research in the finance and economics literature has shown that these unique capabilities of MNE’s allow them to increase productivity and create long-term intangible profits. One important article that examines the link between globalization and productivity is Carrado, Lengermann, and Slifman (2009), who found that MNEs reportedly have been able to achieve significant efficiencies by reorganizing the way they conduct their operations. They attempted to measure the contribution of MNEs to the aggregate productivity record of the United States. In particular, they examined the contribution of MNEs to overall labor productivity growth in the United States. The authors found that, although the MNEs account for only 40 percent of the total output of nonfinancial corporations (“NFCs”) between 1977 and 2000, these multinational companies appear to have accounted for more than three-fourths of the increase in NFC labor productivity over this same period of time. Moreover, MNEs account for all of the NFC sector’s increase in labor productivity growth in the late 1990s. Therefore, MNEs account for more than half of the much-studied acceleration in aggregate labor productivity during this time period.17 Carrado, C., Lengermann, P., and Slifman, L., (2009), The Contribution of Multinational Corporations to U.S. Productivity Growth, 1977-2000, International Trade in Services and Intangibles in the Era of Globalization, University of Chicago Press, 331-360. 17 Economics Partners, LLC www.econpartners.com Page | 18 Criscuolo and Martin (2003)18 document a similar “MNE effect” in the U.K. manufacturing sector, while Griffith, Redding, and Simpson (2004)19 provide evidence of an MNE productivity advantage in the U.K. service sector. Much of the literature in this area attempts to explain why MNEs achieve superior productivity. This work has focused mostly on two main factors—economies of scale and operational integration across multiple geographies (i.e., economies of scope). Research by Doms and Jensen (1998) showed that MNEs tend to be larger than local firms, and they are also characterized as being more capital intensive and using more sophisticated technology. They found that, all else equal, MNE characteristics such as size, capital intensity, and technology use lead to higher labor productivity. This increase in productivity is partly explained by the greater amount of capital per worker, and partly explained by the fact that the size of a firm and the technology that it employs can help enhance its organizational efficiency.20 Research by Mataloni (2004) showed that MNEs may also be able to enhance their organizational efficiency by cross-border integration of their operations. Mataloni’s research indicates that intra-company trade within MNEs has risen steadily. In 2002, these cross-border trades within companies accounted for 22 percent of total U.S. exports and 16 percent of total imports. This vertical integration between parents and foreign affiliates is the phenomenon known as offshoring or outsourcing. Such offshoring allows MNEs to take advantage of international factor price differentials and keep costs down.21 One final reason that helps explain the higher productivity enjoyed by MNEs is the transfer of information and knowledge between parents and affiliates that fosters innovation. Research by Criscuolo, Haskel, and Slaughter (2005) suggests that MNEs generate more ideas than their purely domestic counterparts. This is due to the fact that MNEs can draw on an “intra-firm worldwide pool of information” that is typically larger and more diverse than the stock of ideas generated by most local companies.22 A similar point is made by Coe, Helpman, and Hoffmaister (1997) regarding the productivity benefits of international trade.23 Criscuolo, C., and Ralf Martin., “Multinationals and U.S. Productivity Leadership: Evidence from Great Britain,” CEP Discussion Paper no. dp0672, 2005. 19 Griο¬th, R., Redding, S., and Simpson, H., (2003), Productivity convergence and foreign ownership at the establishment level, Center for Economic Policy Research Working Paper no. 3765. 20 Doms, M. E., and Jensen, J.B., (1998), Comparing wages, skills, and productivity between domestically and foreign-owned manufacturing establishments in the United States. In Geography and ownership as bases for economic accounting,.ed. R. E. Baldwin, R. E. Lipsey, and J. David Richardson, Studies in Income and Wealth, Volume 59, University of Chicago Press, 235-258. 21 Mataloni, R. J. Jr., (2004), U.S. multinational companies: Operations in 2002, Survey of Current Business, U.S. Commerce Department: Bureau of Economic Analysis, 10-29. 22 Criscuolo, C., Haskel, J.E., and Slaughter, M.J., (2005), Global engagement and the innovation activities of firms, NBER Working Paper no. 11479. 23 Coe, D. T., Helpman, E., and Hoο¬maister, A.W., (1997), North-South R&D spillovers. Economic Journal, Royal Economic Society, 134–49. 18 Economics Partners, LLC www.econpartners.com Page | 19 MNEs are also able to easily share firm-specific intangible assets (e.g., R&D) across geographical boundaries. Research published by Lipsey, Blomström, and Ramstetter (1998) demonstrates this.24 The sharing of intangible assets and operational know-how across borders occurs prominently, for example, in service and technology industries where factors of production are easy to transport. Bloom and Van Reenen (2010) found further evidence that characteristics of MNEs help explain persistent differences in productivity at both the firm and national level. The focus of their research was on management practices, and the authors developed a measure of management effectiveness to test their hypothesis that differences in management practices contribute to differences in productivity. Their research showed that management practices do have a significant impact on firm productivity. Furthermore, in all countries, there research showed that MNEs were better managed than domestic firms. Their findings are summarized in the exhibit below, which displays average management scores (based on their measure of management effectiveness) by country for MNEs and domestic firms.25 Lipsey, R. E., Blomström, M., and Ramstetter, E.D., (1998), Internationalized production in world output. In Geography and ownership as bases for economic accounting, ed. R. E. Baldwin, R. E. Lipsey, and J. D. Richardson, Studies in Income and Wealth, vol. 59. University of Chicago Press, 83-135. 25 Bloom, N. and Van Reenen, J., (2010), Why Do Management Practices Differ across Firms and Countries?, Journal of Economic Perspectives, 203-224. 24 Economics Partners, LLC www.econpartners.com Page | 20 Exhibit VII-4 Management Effectiveness: MNEs vs. Domestic Firms In summary, there is substantial evidence from research regarding the productivity of MNEs that suggests that they are more productive than their purely domestic counterparts. Various factors, including superior management, increased-knowledge sharing, cross-border integration, and advantages related to economies of scale, may contribute to the increased productivity observed in MNEs. However, for purposes of our research, what matters is the fact that, for whatever reason(s), MNEs display higher levels of intangible value. Therefore, our econometric model should include a variable to capture the value creation attributed to firms that are part of a MNE. D. Cost Efficiency Other factors beyond intangible capital and a company’s status as an MNE may affect productivity and ultimately a firm’s market value. In addition to its absolute productivity, a firm’s productivity relative to competitors in the industry may be expected to drive market value. For example, a firm may generate a competitive advantage from its ability to control costs relative to competitors, or otherwise able to generate greater output per unit input than the average company. While Cockburn and Griliches (1988) focus on the relationship between Tobin’s Q and knowledge-based intangible capital, their model has subsequently been extended to include other factors that may affect Tobin’s Q, such as a firm’s productivity relative to competitors. Economics Partners, LLC www.econpartners.com Page | 21 Following the same approach as Cockburn and Griliches (1988), Dwyer (2001) analyzes the role of relative cost efficiency in productivity and market value. Dwyer (2001) extends the model developed by Cockburn and Griliches (1988) to treat the productivity of a firm’s manufacturing plants as an intangible asset. If highly productive plants truly represent a competitive advantage to the firm, then this competitive advantage should be reflected in higher stock market valuations. In other words, productivity should have a price, and firms with high relative productivity should have higher market valuations, as measured by Tobin’s Q. To examine the relationship between relative productivity and market value, Dwyer (2001) measures relative productivity as the difference between a firm’s own productivity and the average productivity in its industry. In effect, relative productivity measures the additional output that a firm is able to produce relative to the industry average given the same factor inputs.26 In order to estimate the impact of relative productivity on market value, Dwyer (2001) employs the same basic model developed by Cockburn and Griliches (1988), and controls for the impact of other intangible capital created by R&D and advertising on market value. This study finds that firms with higher relative productivity do have higher market valuations as measured by Tobin’s Q. Thus, relative productivity does have a price, and acts like an intangible asset to the firm. Moreover, Dwyer’s conclusions regarding relative productivity are important for developing an empirical model of the factors that affect Tobin’s Q for two reasons. Dwyer shows that a firm’s relative productivity (i.e., low cost of production) creates a comparative advantage and functions like an intangible asset. However, this value of relative productivity is not reflected on the company’s balance sheet, and is fundamentally a proxy for the sources of comparative advantage that are not measured in other forms of physical and intangible capital. Thus, in order to capture the value of this relative productivity, it is necessary to measure the factors that determine a company’s comparative advantage. Dwyer’s relative productivity measures the ability of a manufacturing firm to produce more output for a given level of factor inputs. Similarly, an IT services company may derive a comparative advantage from the cost efficiencies that allow it to deliver the same level of service at a lower cost. Dwyer, D.W., (2001), Plant-level Productivity and the Market Value of a Firm, Census Bureau CES working paper. 26 Economics Partners, LLC www.econpartners.com Page | 22 IV. Econometric Model The term “econometric model” is used to describe a statistical technique that allows an economist to measure functional relationships between economic variables. Econometric models are also referred to, loosely, as “regression models.”27 The reader with a basic understanding of mathematics will recall that a functional relationship is one in which one variable (usually denoted “Y”) depends upon another (usually denoted “X”). An econometric model is a statistical method for “finding” functional relationships in which one variable that is interesting to a researcher (“Y”) depends on other observed variables (“X1,” “X2,” and so forth). We begin our summary of our econometric analysis by describing our sampling approach. That is, we describe how we obtained the statistical sample from which we developed our econometric model. Next, we describe the “specification” of the model. “Specification” is nothing more than “specifying,” or choosing, the variables that are of interest to us. Finally, we describe the results that we obtained. A. Sample Strategy The primary objective of our model is to identify variables that explain enterprise value (and by extension profitability) in the Japanese IT services industry. We conducted a search in the CapitalIQTM database for Japanese firms providing IT services. After applying various screens for industry, geography, financial results, and data availability, we conducted a review of business descriptions and arrived at a sample of 41 Japanese IT services firms. We then constructed a “cross-sectional” data set using the five-year weighted average of market and financial data from 2007 – 2011, for the 41 firms in our sample. While a cross-section of the data does not control for unobserved year-over-year differences in a firm’s financial performance, utilizing the five-year weighted average mitigates this concern. In short, we endeavor to explain the key contributors to average profitability in Japan, not fluctuations driven by differences in the business cycle. B. Model Specification 1. Dependent Variable As we have discussed at length, Tobin’s Q is our chosen measure of value, as it corresponds closely to residual profit, and directly reflects the value of intangible assets. Further, Tobin’s Q can be converted from a stock measure (i.e., observed market values) into a corresponding measure involving operating profit flows. This means that if we understand the determinants of Regression is a more general term than “econometric,” given that regression is used in fields other than economics, whereas “econometrics” is specific to economics and finance. 27 Economics Partners, LLC www.econpartners.com Page | 23 Tobin’s Q, we also understand the determinants of operating profit. For the purposes of this analysis, we defined Tobin’s Q as the five-year weighted average of the ratio of total enterprise value to the firm’s reported capital (“K”).28 2. Independent Variables Based upon our review of the existing literature, we determined that Tobin’s Q is driven largely by five key independent variables: research and development, intangible assets and goodwill, marketing and advertising, relative cost efficiency, and the firm’s status as a multinational enterprise. The subsections that follow expand on each of these variables. a) Research and Development As noted in our literature review, investments in technology, such as R&D, have been tied to the growth of TFP, and in turn, TFP is positively linked with Tobin’s Q. Thus, R&D expense is included in our analysis. R&D is measured as the five-year weighted average of the ratio of R&D expense to capital. b) Goodwill and Intangible Assets Firms generally report the fair market value, net of amortization, of goodwill and intangible assets for other companies that they have acquired. In general, the category goodwill and intangible assets is composed primarily of goodwill, which is the excess of the acquisition price for an acquired company over the fair market value of the acquired company’s assets. Goodwill is included in our model because, as discussed earlier, goodwill is the value of the firm (or an acquired company) over and above the value of its identifiable assets. That is, goodwill and going concern are “assets” that result from management decisions to structure the firm so that it can exploit the sales channel across many products and/or exploit the management asset itself across geographies and products. That is, goodwill and going concern are closely related to economies of scope – i.e., they are the value associated with exploiting many assets together. Intangible assets and goodwill are measured as the five-year weighted average of the ratio of the book value of acquired intangible assets to capital. c) Marketing and Advertising As discussed earlier, marketing and advertising may drive increases in Tobin’s Q through enhancements to brand-related equity. Marketing and advertising is measured as the ratio of marketing and advertising expense to capital. 28 Capital is defined as the sum of net PP&E, accounts receivable, and inventory. Economics Partners, LLC www.econpartners.com Page | 24 d) Relative Cost Efficiency As noted, TFP has a value that is not captured on the balance sheet. The construction of a relative cost efficiency variable is meant to reflect the intrinsic value of cost-related advantages. Relative cost efficiency is measured as the five year weighted average of the difference between the average ratio of cost of service to capital across all 41 firms and the ratio of cost of service to capital for a given firm. e) Multinational Enterprise Our review of the literature showed a strong positive link between multinational enterprises and TFP. Most of this increase can be explained by improved management efficiency. A firm’s status as a multinational enterprise is defined by a dummy variable (i.e., a binary 0-1 variable), with “1” indicating that the firm is part of a multinational enterprise. 3. Econometric Model We posit the following mathematical (i.e., functional) relationship between Tobin’s Q and the independent variables. π &π· πΈπ₯ππππ π πΌππ‘ππππππππ & πΊππππ€πππ (1) ln(πππππ′ π π) = π½0 + π½1 ( ) + π½2 ( ) πΎ πΎ π&π΄ πΈπ₯ππππ π + π½3 ( ) +π½4 (πππΈ)+π½5 (π ππππ‘ππ£π πΆππ π‘ πΈπππππππππ¦)+π½6 (πππΈ ∗ π ππππ‘ππ£π πΆππ π‘ πΈπππππππππ¦) + π πΎ Equation (1) represents a “semi-log” model. In a semi-log model, the dependent variable is measured as a natural logarithm. This form of estimating equation incorporates the economic concept of diminishing returns. For example, a semi-log model captures the idea that the marginal benefit to Tobin’s Q from increases in R&D expenditure at high levels of existing R&D should be less than the marginal benefit at lower levels of R&D. Further, semi-log models allow the coefficient estimates on the independent variables (i.e., β1, β2, β3, β4, β5, and β6) to be interpreted as the percentage change in Tobin’s Q given an absolute change in the independent variable, all else equal. The “interaction term” (MNE*Relative Cost Efficiency) represents the contribution to Tobin’s Q through improvements in relative cost efficiency for multinational enterprises. Interaction terms are used when the change in the dependent variable (i.e., Tobin’s Q) with respect to one independent variable (i.e., relative cost efficiency) depends on the level of another independent variable (i.e., status as a multinational enterprise). C. Model Results 1. Descriptive Statistics Descriptive statistics for each of the variables are presented in the table below. Economics Partners, LLC www.econpartners.com Page | 25 Exhibit VIII-1: Descriptive Statistics Standard Variable Tobin's Q R&D Expense GW & Intangibles Marketing & Advertising Multinational Enterprise Relative Cost Efficiency 2. Deviation Mean 2.5623 0.0491 3.9427 0.1067 0.1848 0.0842 0.451 8.34E-08 0.2414 0.1126 0.5 1.231 Econometric Results The econometric results from our OLS regression are presented in the exhibit below. Exhibit VIII-2: OLS Regression Results Number of observations 41 F-Statistic (6 , 34) 11.77 Adjusted R-squared 0.6176 Variables R&D Expense GW & Intangibles* Marketing & Adv. Expense** MNEdummy Relative Cost Efficiency MNE*Relative Cost Efficiency** Intercept ** - significant at 99% level * - significant at 95% level Coefficient Standard Error T-Statistic Prob. > |T| 1.1170 0.8363 3.5900 0.1009 -0.1324 0.4896 -0.0906 0.8017 0.4010 0.8631 0.2029 0.1051 0.1558 0.1425 1.39 2.09 4.16 0.5 -1.26 3.14 -0.64 17.3% 4.5% 0.0% 62.2% 21.6% 0.3% 52.9% 95% Confidence Interval -0.5122 0.0212 1.8359 -0.3115 -0.3459 0.1730 -0.3803 2.7463 1.6213 5.3441 0.5133 0.0811 0.8063 0.1990 Several observations regarding our econometric results are in order. First, of the statistics shown in the preceding exhibit, the adjusted R-squared is perhaps the most important. The adjusted R-squared represents the percentage of the variation in Tobin’s Q that is explained by our econometric model – i.e., the percentage of variation that is explained by research and development, intangible assets and goodwill, marketing and advertising, relative cost efficiency, the firm’s status as a multinational enterprise, and the relative cost efficiency of a multinational enterprise. The econometric model clearly has significant amount of explanatory power, with an adjusted R-squared of 61 percent. Another key summary statistic to examine is the F-statistic. The F-statistic represents a hypothesis test of overall significance of the econometric model. Essentially, the F-statistic is a test to verify that the entire set of estimated coefficients in each model – or said differently, the model as a whole – is statistically significant. If an econometric model “fails the F-test,” the model is indistinguishable from randomness, and is therefore a meaningless model. In our Economics Partners, LLC www.econpartners.com Page | 26 model, the F-statistic is quite high, and we can say with 99 percent confidence that our model is in fact meaningful. Recall that coefficient estimates in a semi-log model are interpreted as percentage change. For example, a one unit increase in the five-year weighted average ratio of goodwill and intangibles to capital would result in an 83.6% increase in Tobin’s Q. A similar chain of logic suggests a 112 percent increase in Tobin’s Q for a one unit increase in R&D, and similarly 359 percent for marketing and advertising expense. While these coefficients may seem alarmingly high, one must remember that this is reflects a one unit change in the ratio of these variables to capital. Increases in these items are likely to increase the ratio to capital by a factor closer to 0.1, rather than 1. A key point of focus here is the statistical significance of the coefficient estimates. There were three highly significant variables in the model: marketing and advertising, the relative cost efficiency of a multinational enterprise, and goodwill and intangibles. Marketing and advertising and the relative cost efficiency of a multinational enterprise are significant with 99 percent confidence, and the coefficient on goodwill and intangibles is significant with 95 percent confidence. R&D and relative cost efficiency are significant at the 80 percent and 75 percent confidence levels, respectively. The statistical significance of the interaction term and relative statistical insignificance of the individual components of the interaction term may seem odd at first glance.29 However, this result suggests that the market only rewards multinationals for cost efficiency. Said differently, the gains of improving relative cost efficiency are only realized when the firm is part of a multinational enterprise, and the gains of being part of a multinational enterprise are not realized unless there are improvements in relative cost efficiency. 3. Conclusion It is important that the reader keep in mind that econometric models are by their nature imperfect. They make strong demands of often limited data. It is often difficult for the researcher to obtain both economic and statistical significance across the board. In light of these facts, our statistically significant coefficient estimates all produced signs supported by the literature. Furthermore, the econometric model presented in this section can be characterized as both reliable and robust, given the strong adjusted R-squared and large Fstatistic for a relatively small sample size. Recall that the interaction term is the product of relative cost efficiency and the firm’s status as a multinational enterprise (a binary 0-1 variable). 29 Economics Partners, LLC www.econpartners.com Page | 27 V. Attribution of Profits within a Firm A. Overview of Statistical Attribution Without further modification, an econometric model such as the one described in the previous section does not give us a model that attributes profitability or value to specific factors and legal entities within a firm. The reason for this is that the coefficients of the model – that is, the β1, β2, β3, β4, β5, and β6 variables – are essentially the same thing as slopes. That is, each of these coefficients measures the rate of change in the dependent variable (Tobin’s Q), per unit change in the independent variable to which the coefficient is attached. While this information is important, it only tells us how important one independent variable is given that the other variables are also contributing to the dependent variable. Said differently, each of the coefficients is a measure of the impact of one of the independent variables working in conjunction with all of the other variables. Thus, the econometric model does not provide us with much information about the standalone importance of each variable. In order to develop this information from an econometric model, we need something called “variance decomposition.” The term “variance decomposition” is a fancy name for “attribution.” Variance decomposition is a way that economists and statisticians use to attribute explanatory power, or causality, to individual variables. Variance decomposition gives us what we are looking for in this case – the standalone influence of each variable in an econometric model. As an example, consider an American football team. There are many contributors to the points scored during a game or over a season. Across the National Football League, it will almost always be the case that a team’s kicker will lead the team in total scoring. For example, if a team scores 23 points in a game, the kicker will have scored 11 out of the 23 (three field goals and two extra points, or nearly half of the team’s points). If we measure a kicker’s relative contribution to the team’s success by looking only at this statistic, we will grossly overstate the importance of the kicker. Even a mediocre kicker scores a large number of points during a season. Losing a kicker is not a serious problem for most teams. In other words, by imagining what would happen to the team’s success if the kicker left the team, we see that the kicker is in fact not very important on a standalone basis. A kicker can easily be replaced. Continuing with the example, left tackles score very few points. Left tackles never lead their team in total scoring, and often go their entire careers without scoring any points. Thus, if we measure the left tackle’s relative contribution to the team’s success by looking only at this statistic, we will grossly understate his importance. But losing a left tackle is always a serious problem for a football team. It also explains why left tackles are often taken in the first round of the NFL draft and are paid tens of millions of dollars, while kickers often go undrafted and Economics Partners, LLC www.econpartners.com Page | 28 many are paid near the league minimum. In other words, by imagining what would happen to the team’s success if the left tackle were injured or replaced, we see that the left tackle is in fact very important on a standalone basis. This approach – i.e., measuring the effect of losing a “factor” – is exactly what variance decomposition does. Variance decomposition measures the “standalone” importance of each variable by examining the effect of not having that variable, or factor, as part of the group of variables that explain the dependent variable. That is, variance decomposition measures the effect of not having a factor as part of the “team,” in order to determine the amount of the dependent variable that should be attributed to that factor. B. General Procedure: Two Steps The attribution of profits to specific entities within a firm follows two steps. First, we attribute Tobin’s Q, and thus residual profits, to specific variables, or factors. Second, we attribute those factors to specific entities. C. Attribution of Profits to Variables The attribution technique that we use is called “proportional marginal variance decomposition” (“PMVD”).30 When the dependent variable (i.e., the variable being explained by the regression) is residual profit (or, more precisely, Tobin’s Q exceeds 1 when capitalized residual profit is positive), then the PMVD analysis can be interpreted as producing a predicted residual profit split. There are two reasons for this. First, PMVD provides each variable’s average contribution to the model – or how much of what the model explains (in total) is in turn explained by each variable (separately). As we alluded to earlier, PMVD does this by examining the change in the explanatory power of our econometric model (our equation) when we eliminate each of the independent variables, one at a time. The second reason why the PMVD procedure can be interpreted as a profit split is that it is exactly the procedure developed by the Nobel Prize winning economist and game theorist Lloyd Shapley for determining how gains should be split in a cooperative “game” or cooperative context. Lloyd Shapley developed a solution to the question of how a set of participants that contribute to a jointly-created surplus (e.g., in a cooperative game setting) would split the resulting surplus. The Shapley value is a measure of an individual participant’s marginal contribution to the surplus created by a group (of which the participant is a member). This value, compared to the Shapley value for all other participants, is also the predicted share See for example Feldman, Barry E., “A Theory of Attribution” (May 29, 2007). Available at SSRN: http://ssrn.com/abstract=988860 or http://dx.doi.org/10.2139/ssrn.988860; “Relative importance and value”, Barry Feldman (working paper, 2005); “Estimators of Relative Importance in Linear Regression Based on Variance Decomposition”, Ulrike Gromping, The American Statistician, May 2007, Vol. 61, No. 2. 30 Economics Partners, LLC www.econpartners.com Page | 29 of the surplus that will be captured by each participant. In the same way, the PMVD value for the regression described in the previous section is the marginal contribution of each variable to the creation of operating profit. As such, it is also the share of operating profit that would be expected at arm’s length in a profit split when multiple parties are contributing to the total operating profit in the system. One thing that is important to consider is the fact that the sum of the PMVD values for each of the independent variables in our econometric model equals the R2 of the regression. Recall from Section IV of this paper that the R2 statistic is always less than one, and represents the proportion of the variation in the observed values for the dependent variable that is explained by the independent variables, taken together. This means that the PMVD values (i.e., how much of the total explanatory power of the model to attribute to each “factor”) sum to the total explanatory power of the regression model. This is very similar to the way that so-called “Shapley values” sum to the total surplus created by the group in a cooperative game. However, since R2 is a measure of the total explanatory power of the model, and the model does not explain all of the variation in the dependent variable, application of the PMVD procedure leaves open the question of the remaining, or unexplained portion of the residual profit. There are two possibilities we consider with respect to this unexplained portion. First, the unexplained portion may be due to factors that were not considered, or that cannot be measured with precision. In these cases, assuming that the relative contributions of the entities over the explained portion would apply in the same proportion to the unexplained portion is an unbiased estimate of the total contribution to operating profit for each entity. Second, it is possible that the unexplained portion is due to factors outside of the entities considered – such as exogenous forces (e.g., economic conditions in the market) that neither party could claim at arm’s length. For purposes of this analysis, we assume that one entity’s contribution to (and thus claim over) residual profit for the explained portion would apply equally to the unexplained portion, acknowledging that we are potentially overstating the compensation that the entity could expect to earn at arm’s length (i.e., bargaining on a standalone basis). D. Example of Attribution of Variables to Entities Here, as an example, we will use the regression model to allocate profits among two entities within a hypothetical firm – Company A’s IP owner (“IPCo”) and an operating affiliate provider of IT services in Japan (“OpCo”). The regression model outlined in the prior section attributes residual profit to six factors. We now must attribute these six factors to the entity that would claim that factor (and therefore its share of residual profit) at arm’s length. Each of these factors, and the entity that owns or controls the factor, are described below. Economics Partners, LLC www.econpartners.com Page | 30 1. Factors Attributable to IPCo Four of the six factors are attributable to IPCo. These are: ο· ο· ο· ο· R&D – IPCo manages, controls, and develops the technology and know-how. MNE – IPCo provides access to the global intellectual property, and multi-national structure, which provides for efficiencies of scope and scale. MNE*Relative Cost Efficiency – Similar to above, this variable measures how the MNE status impacts cost efficiency Goodwill & Intangibles – Intangible assets are developed and maintained by IPCo, which provide access to and facilitate efficient sharing among individual entities. 2. Factors Attributable to OpCo Two of the six factors are attributable to the OpCo in Japan. These are: ο· ο· Marketing and Advertising – OpCo provides marketing and advertising in the local market. Relative Cost Efficiency – Attributed to OpCo to explain the non-MNE (local) portion of the measure of cost efficiency. 3. Unattributed Factors As noted, the PMVD analysis attributes proportional contributions to the factors included in a model up to the total variance explained by the model. The remaining, unexplained portion is initially unattributed to either IPCo or OpCo. For purposes of this analysis, we assume that the relative contributions of IPCo and OpCo apply proportionally to the unexplained portion of the variance. However, the unexplained portion is arguably the result of factors actually owned and controlled outside of IPCo and OpCo. Given that the unexplained portion is at the very least exogenous to the model as specified, an alternative approach would be to assume that the claim over residual profit for IPCo and OpCo is limited to the share of the residual profit actually explained by factors owned and controlled by IPCo and OpCo. This would be the preferred approach if there were other claimants to the residual profit in the system, such as a parent company or another entity within the firm that owned additional IP. In our hypothetical model, there are only two entities, and for purposes of our example we simply allocate all residual profit to these two entities. E. Summary of Results In summary, the six factors, their relative importance, and their attribution to either IPCo or OpCo is shown below. Economics Partners, LLC www.econpartners.com Page | 31 Exhibit IX-1: PMVD Results Variable R&D Expense Goodwill & Intangibles Marketing & Advertising Expense MNE Dummy Relative Cost Efficiency MNE*Relative Cost Efficiency Contribution to Residual Profit Attribute to IPCo? Attribute to OpCo? 4% 20% 42% 10% 3% 21% Implied Residual Profit Split: Yes Yes No Yes No Yes No No Yes No Yes No 55% 45% As can be seen, the four factors attributable to IPCo have a combined contribution of 55 percent, while the two factors attributable to OpCo have a combined contribution of 45 percent. This analysis therefore suggests that, at arm’s length, OpCo would capture approximately 45 percent of the system residual profit, while IPCo would capture approximately 55 percent. Economics Partners, LLC www.econpartners.com Page | 32 VI. Conclusions Reached Recognizing the importance of understanding the primary sources of value creation within an enterprise when evaluating transfer pricing policies, the purpose of this paper was to provide a framework for identifying and evaluating the primary determinants of operating and residual profit within a firm. To that end, we developed a research strategy and completed the following analyses. ο· ο· ο· We performed an extensive review of literature related to the determinants of productivity and profitability. We developed an econometric model that predicts a firm’s Q ratio (i.e., its capitalized residual profit) based upon factors that economists have identified as value-relevant over time. We used the econometric model to attribute profit to within a hypothetical firm to particular legal entities within that firm. Specifically, we converted the econometric model to a profit attribution model to allocate profit between a hypothetical firm in the IT services industry with a US IP owner and a local Japanese affiliate. For our hypothetical firm, the results of our analysis indicated the IP Owner would capture approximately 55 percent of the residual profit in the system, and the local affiliate would capture approximately 45 percent of the residual profit, assuming no other residual profit claimants exist in our simple example. While the analysis presented here was for an example MNE in the IT services industry with just two entities, the approach outlined in this paper can be applied to other industries and to other MNE structures. Economics Partners, LLC www.econpartners.com