Intangible Assets and the Market Valuation of UK Companies: Evidence from Fixed Effects Models

Derek Bosworth and Alex Wharton

Manchester School of Management UMIST

Christine Greenhalgh
St. Peters College, Oxford

Working paper No 2: April 2000

Copyright in the authors, 2000

Intangible Assets and the Market Valuation of UK Companies: Evidence from Fixed Effects Models
Derek Bosworth and Alex Wharton
Manchester School of Management UMIST

Christine Greenhalgh
St. Peters College, Oxford

This publication is sponsored by

St. Peters College, Oxford OX1 2DL Phone: 01865 278900 Fax: 01865 278855 www.oiprc.ox.ac.uk

Intangible Assets and the Market Valuation of UK Companies: Evidence from Fixed Effects Models
Derek Bosworth and Alex Wharton Manchester School of Management UMIST Christine Greenhalgh St. Peters College, Oxford April 2000

Abstract: This paper extends the tradition of applied econometric research on the market valuation of intangible assets initiated by Griliches in 1981. Unlike most of the previous research on the subject, which is predominantly US-based, it uses UK data drawn from 146 publicly-traded production companies for the period 1990 to 1994 inclusive. In addition, unlike the earlier body of research, it includes not only research and development and patent data, but also trade mark information. The initial analysis of the data follows established practice by estimating a market value equation, which uses Tobins q to measure company performance, and firm dummies to capture the effects of unobserved firm-specific factors. The parameter estimates are used primarily to infer the quality of the relationship between companies market value and their own intangible assets. The first result is rather surprising in the light of previous research: the stock of measured intangibles fails to explain variations in market values, both within and across firms. Nevertheless, there is evidence of systematic differences in the performance of firms that are captured by the firm fixed effects and not by the measured variables. This suggests two lines of enquiry: first, to examine the returns to unobserved firm-specific factors for evidence that the markets have assimilated the intangible stocks with unobserved factors; and second, to control for the effects of unexpected changes in the stocks of measured intangibles. The results indicate that the distribution of returns to unobserved, firm-specific factors is highly skewed, with the bulk of the observations clustered around the lower end of the distribution and a Paretian upper tail a finding consistent with other work in this area. The subsequent analysis of the data is novel in the sense that it estimates the relationship between the firm-specific effects and the within-firm averages of measured intangibles. The results show that the variation in firm-specific effects, and thus the variation in the firms q ratio, is positively related to both the average stock of trade mark applications and the average stock of patent publications. The main conclusion, therefore, is that the value of firm-specific factors is linked to the firms accumulated patenting and trade mark portfolio. [Comments welcome: derek.bosworth@umist.ac.uk ] 1

1. Introduction
While the present paper has a number of important novel aspects, it follows in the footsteps of a well-established literature on the market valuation of intangible assets that dates back to Griliches (1981). This literature has been given renewed emphasis by the growing interest in intellectual capital, which, broadly speaking, focuses on the leverage of knowledge into value added for the firm (Hall, 1993a; Lynn, 1998; Ruggles, 1998). Thus, the role of R&D and patents, which pervade the earlier valuation of intangible assets literature can be reinterpreted in the light of the stages of the various activities companies go through in accessing and managing knowledge to produce competitive advantages in an increasingly dynamic capitalist economy. In this paper the question is examined using a financial market value based measure of firm performance and several measures of intangible assets, which can also be thought of as proxies for the various stages in the process of accessing and leveraging knowledge into intellectual capital. The pros and cons of the use of financial market valuations as a measure of firm performance have been widely discussed in the literature. In particular, it has the merit of being forward-looking (i.e. reflecting the markets views of the companies potential future returns), but it is dependent on the accuracy of the market in making such assessments. This measure of firm performance has the additional merit of exploiting the accounting tautology to infer the value of intangible assets from the gap between the market value of companies and the book value of their tangible assets a gap highlighted during the merger and acquisition of knowledge-intensive companies. The choice of proxies for intangible assets is based upon the common theoretical proposition that R&D, patents, and trade marks reflect the accumulation and exploitation of intellectual capital. R&D is seen as a process by which the firm accesses and produces new knowledge, while patenting and trade marking are processes the company uses to appropriate the value-enhancing benefits of such knowledge. Thus, the companies stocks of R&D-knowledge, patents and trade marks are potentially valuable intangible assets. In this paper, we explore the relationship between intangible assets and the dynamic performance of UK companies using the specific, econometric framework laid down in the groundbreaking study by Griliches (1981) and explored by a number of other authors, such as Hall (1993b). Accordingly, firm performance is measured by an approximation of the Tobin q ratio, the ratio of the firms marginal financial market value to marginal replacement cost. Under the assumption of efficient capital markets, q is used to infer the value of intangible assets. The aim is to estimate the actual contribution of measured intangibles to the variation in q across 146 UK production companies over the period 1990 to 1994 inclusive.1 The framework assumes that the
1

This data set was drawn from a larger data set constructed as part of the ESRC/DTI IP Initiative (Firm Performance and the Valuation of Intellectual Property, ESRC No. L325253005) and a subsequent ESRC project on trademarks (The Economic Role and Value of Trademarks, ESRC No. L325253036). A great deal of work has been undertaken to provide an accurate matched data set of financial accounts and intellectual property data. We would particularly like to thank Mark Longland for all his help in constructing and guiding our use of this data.

various factors that determine a firms performance are linearly and additively separable.2 To date the Griliches framework has most often been applied to US data, largely because of the relative abundance of R&D and patent data for the USA. Most of these applications have involved the use of panel data on production companies, the notable exception being Cockburn and Griliches (1988) who used cross-sectional data. The empirical specifications vary according to the dependent variable, the combination of patent and R&D variables, the measurement of the variables, and the choice of control variables. The dependent variable is generally either a measure of the firms Tobin q or its market value, with the replacement cost variable transferred to the right hand side of the equation (e.g. Ben-Zion, 1984). The measures of intangible assets used include the flow of R&D expenditure, the stock of R&D knowledge, unexpected R&D, patent or innovation variables, and patent stock variables. Control variables variously account for the effects of the quality of patents (e.g. Cockburn and Griliches, 1988, and Hall et al., 1999), financial and accounting variables, the spillover of intellectual capital, time-specific, firm-specific and industry-specific variables. Despite these variations in specification, a fairly robust set of results emerges, covering the period from about 1968 to the present day. First, R&D makes a significant and positive contribution to firm performance. Second, the contribution is both direct and indirect, involving the effects of own-R&D and, where tested, the spillover effects of other firms R&D. Third, the contribution of R&D to firm performance fell significantly during the 1980s a result emphasised by Hall (1993b). Fourth, when the R&D and patent variables are both included, the R&D tends to be more influential (e.g. Cockburn and Griliches, 1988), although there is some disagreement about the extent to which R&D and patents do or do not measure the same thing. Fifth, where both anticipated and surprise variables of the intangibles are included, the surprise variables generally play the more important role.3 Finally, there is some evidence that differences in firm performance are not fully explained by differences in measured intangible capital, in other words, there are significant differences in the effects of unobserved firm- (and time-) specific factors. Very few attempts have as yet been made to apply the Griliches framework to UK data. Those that have include Stoneman and Bosworth (1994), Green, et al. (1996), Toivanen, et al. (1998) and Bosworth and Mahdian (1999). The results are reasonably consistent with those produced by the US studies. Stoneman and Bosworth find evidence that R&D plays a more dominant role in the explanation of market values than do patents, especially when they are both included in the equation. Bosworth and Mahdian provide evidence that R&D and patents perform broadly the same role in the explanation of the market value of UK pharmaceutical companies. Toivanen, et al.
2 3

For alternative solutions to the aggregation problem, see Hall et al. (1999). There is a largely separate literature on the financial valuation effects of various types of news, based upon events-studies: see, for example, Austin (1995), Chaney, et al. (1991), Shapiro and Switzer (1993) and Zantout and Tsetsekos (1994).

find that ongoing R&D contributes to firm performance, but that the first declaration of R&D expenditure in the company accounts has a significant impact on market value. There appears to have been no other testing of the anticipated versus surprise (new-news) model for the UK. This paper: (i) adds further evidence for the UK; and (ii) provides the first UK estimates from a new-news model. In addition, it more generally extends the Griliches tradition by: (i) adding a trade mark variable; and (ii) examining and interpreting the effect of unobserved firm-specific factors. The present paper is one of the first to study the effect of trade mark activity within the Griliches framework. There are several reasons why trade mark activity might be expected to promote the dynamic performance of firms. First, it is associated with the proliferation of brands, reflecting the modification of old products and the launch of new products. Second, it is associated with brand-name advertising, reflecting the reputation and market position of the firm. In short, trade mark activity, linked to the branding of products, represents a potentially important source of competitive advantage. The first study to account for the effect of trade marks within this framework found that they have had a positive influence on company performance in the UK pharmaceuticals sector (Bosworth and Mahdian, 1999). With regards to the second extension to the Griliches framework, the present paper is also one of the first to more rigorously investigate and interpret the role of unobserved firm-specific factors, which have been found to explain significant and systematic differences in the performance of firms beyond that explained by observed variables (again see Bosworth and Mahdian, 1999). The results reported below suggest that firm-specific effects tend to absorb the effects of intangible assets. If, for example, the levels of each companys R&D, patent and trade activity are reasonably constant over time, there are strong grounds for believing that their impact might be assimilated within firm-specific effects. This might explain why the unanticipated changes in intangibles tend to dominate the anticipated, when these have been analysed using US data. Thus the new results of the new-news model are useful here, to see if this is the case. Insofar as firm-specific factors are empirically important, they will tend to encapsulate the markets view of the value of intangible assets. Thus, we explore two avenues of investigation. First, we look at the distribution of the effects of these firm-specific factors, as measures of differential firm performance. Evidence from previous work on both the value of patents and the returns to venture capital investments suggests that the distribution may have a Paretian upper tail, although there is also evidence of concavity in the Pareto-Levy space (Scherer, 1996; Bosworth and Jobome, 1999). Comparisons are made with these earlier results, which argue that the distribution reflects the inherent riskiness of investments in intangible assets. Second, we regress the firm fixed effects on average R&D, patents and trade marks, to find out whether there is any evidence that the impact of the intangibles has been assimilated within the firm fixed effects. If R&D, patenting and trade mark activity are inherently risky, we do not expect a perfect match. Nevertheless, if the above hypothesis is correct, on 4

average, companies that have higher R&D expenditures and greater numbers of patents and trade marks are expected to exhibit higher market values. This is particularly the case if the market itself uses these variables as proxies for the generation of future profits. The rest of the paper is divided into four further sections. The next section describes the theoretical and conceptual framework that underpins the approach adopted in this paper. In particular, it addresses the question of differences in dynamic performance across companies. Section (3) outlines the choice of variables and the empirical specification of the model. Section (4) provides a detailed description of the data and results. The final section discusses the implications and draws the principal conclusions.

2. Methodology: Theoretical and Conceptual Issues

2.1 Tobin q Framework
A number of different approaches have been used to explore the relationship between intangible assets and firm performance, including case studies and regression techniques, each with their own merits and problems.4 The present paper adopts the econometric framework used by Griliches (1981), his co-workers and others.5 The basic assumption is that the various factors influencing firm performance are additively separable. Formally, its solution to the so-called disaggregation problem the problem of valuing the individual assets of a heterogeneous firm amounts to estimating the following equation (Hall, 1998): (1) Vit ( A, K ) = qit ( Ait + K it ) where Vit represents the market value of firm i in period t, A denotes the firms entire stock of tangible assets, and K is a vector of its intangible assets (or intellectual capital). The scalar, , describes the overall scale effect (equal to unity under constant returns to scale), the vector gives the shadow value of each intangible asset relative to tangible assets (marginal contribution of intangible assets to the market valuation averaged across companies in the sample), q is the absolute value (when 1) and the coefficient q is equivalent to Tobins q (and identical to Tobins q if it is assumed that there is only one type of asset, a tangible, A1). On log transformation, Equation (1) becomes

Hall (1998) discusses the merits and problems of using total factor productivity, profit growth, and market value to measure the performance of firms within in an econometric framework. Griliches (1979) provides a brief description of the merits and problems of the case study framework. See, for example, Ben Zion (1984), Hall (1993), Griliches (1981), and Cockburn and Griliches (1988).

(2) log Vit = log qit + log Ait + log(1 + K it / Ait ) Where K it / Ait denotes the normalised intangible assets, or intensity of the assets. Most users of this framework approximate log(1 + K it / Ait ) the bracketed term by K it / Ait , resulting in inaccuracies for K/A greater than 0.15.6

2.2 Distribution of Firm-Specific Fixed Effects and Dynamic Performance

Although the estimation of Equation (2) using company-level panel data has often accounted for the effects of unobserved firm-specific factors, these effects have never been explored beyond the usual tests of the random versus fixed effects hypotheses. In practice, these effects appear to contain important information about the distribution of firm-performance. Insofar as market values reflect a companys dynamic performance, the distribution of these effects should reflect the perceived success of the firm in generating future profits from given values of tangible and intangible assets. This is a particularly important within the present framework, since the data include the input values of R&D (i.e. expenditure) and count values of patents and trade marks, even though the success of R&D and the commercial value of patents and trade marks is known to vary widely. The present paper therefore examines the effects (y) both visually and by further regression analysis. The distribution is examined visually for similarities with earlier findings in the literature: (i) linearity of the Paretian upper-tail (Scherer, 1996; Pareto, 1897; Levy, 1925 and 1937); (ii) evidence of concavity in Pareto-Levy (P-L) space (Bosworth and Jobome, 1999).7 Two types of diagnostic plot are inspected, in particular, the relative frequency distribution and the Pareto-Levy diagram. In both cases, the absolute firm-specific effects data produced by the semi-logarithmic market value equation are transformed into relative values using a device suggested by Halvorsen and Palmquist (1980). This device takes the anti-log to the base e of the absolute firm-specific effect and subtracts one. The frequency distribution is examined for evidence of a long upper-tail. An accompanying cluster close to the lower end would indicate that the
6

In other words, the applied equation involves approximating ln(1 + x) with x, where x is the weighted sum of the normalised independent variables. This approximation is better the smaller is the ratio of intangible assets to tangible assets. Thus: x = 0.2 implies that ln(1+x) = 0.18, resulting in a 10% error, but higher values of x result in a larger error (e.g. x = 0.32 results in a 15% error). Therefore the model can be estimated with more than 90% accuracy only if x < 0.2. However, as x increases the equation is likely to be more revealing. This implies a trade-off between revelation and reliability. The problem of reliability probably exists in all or most of the previous work. An obvious exception is Hall & Kim (1997) who use the non-linear form of the market value equation explicitly. Some studies adopt alternative approaches in an attempt to avoid this problem, for example, Hirschey (1982), Connolly & Hirschey (1990), Sougiannis (1994) and Green, Hirschey (1990), Sougiannis 1994) and Green,et al. (1996). However, in a number of cases this involves other restrictions, such as assuming constant returns to scale ( it

it (i.e. with this approach is that if q 1 then ( qit 1) Ait will end up in the disturbance and potentially bias the estimate of either the relative or absolute shadow value of K). Concavity appears to be the result of the interaction between the annual distribution of returns and the length of period over which these returns are obtained.

=1

) and market equilibrium ( qit = 1 ). The difficulty

returns are very risky. Following Paretos original work on income distribution (Pareto, 1897), the P-L curve plots the proportion of the population (P) with a value of y or more against y itself on a double-logarithmic diagram (Cramer, 1969; Cowell, 1995). If the graph is exactly a straight line throughout its length, then the underlying distribution is known as the Pareto (or the Pareto-Levy) distribution, having a very close fit to a power function of the form: (3)

f ( y ) = ky

where k and are the parameters. The absolute slope of the Pareto curve gives Paretos . Given the parameters, the proportion of the population that has y or less is given by the formula

y (4) F ( y ) = 1 y Where y is the lower income limit for which the Pareto distribution is defined.
All of the corresponding inequality statistics that satisfy the principle of transfers are easy to calculate. The Gini coefficient, for example, is given by the formula (5)

G=

For these measures, a Pareto distribution with relatively high exhibits relatively less inequality. The Pareto distribution has been shown to characterise the upper tail of the distribution of income and various other variables.8 This may not be the case with the distribution of firm performances, as the precise result may depend upon the area of study. One common finding of most areas of study, however, is that, for one reason or another, the associated distributions are not normal. For example, the general conclusion vis a vis financial assets is that the distribution of price changes exhibits high kurtosis (more peaked with fatter tails than the normal distribution), but the evidence that the distribution is skewed is more contentious (Peiro, 1999, p. 860). There are a multitude of other seemingly unconnected areas where there is evidence of skewness, for example, where Pareto-type power law distributions occur (Bak, 1997). These include geological layers of turbidite, volcanic rumbling, earthquakes and city sizes, but also economic variables, such as cotton prices and patent values. There are a number of studies of some relevance to the present paper that emphasise the skewed distribution of the value of individual intangible assets (i.e. that some patents are enormously valuable and most are worth little or nothing). They include
8

1 2 1

Pareto himself did not discover the limitation to the upper tail, as his income distribution statistics were left-hand censored.

the work by Harhoff, et al. (1997) on the value of patents in the USA and Germany; Scherers (1996) study of mainly US patent values; and the brief review of the distribution of returns to patents by Lanjouw, et al. (1996), partly based upon Scherers work, but including some additional information. These initial results suggest that the distribution of returns to R&D and innovation is complex, and concave in P-L space, with some evidence of a linear Paretian upper tail (Bosworth and Jobome, 1999). At the time of writing, the only study to date to use the distributional results from panel data set estimates is Bosworth and Mahdians (1998) study of UK pharmaceuticals companies. This also reports a concave distribution in the P-L space.

3. Empirical Specification
The present paper estimates two versions of Equation (2), a traditional stocks and flows version and a news-augmented version, which uses anticipated and unanticipated variables. In both cases, intangible assets are proxied by R&D, patents, and trade marks. Since the stocks of the different kinds of intangible capital cannot be measured directly, they are approximated using current and past flows. The following equation is used to approximate the stock of R&D: (6) KRDt =

RD
k =0

t k

(1 ) k = 1 RDt ,

where t denotes the current period, t - k denotes past periods, d denotes the constant rate of depreciation, and k denotes the number of compounding periods. The stock of R&D is thus approximated as a weighted sum of the current and past expenditures on R&D, and ultimately by the current level of R&D, on the assumption of a constant growth rate (g) in R&D expenditure, such that b = 1/[1 (g d)].9 In a similar way, the stocks of patents and trade marks are assumed to be proportional to the annual number of patent grants and the annual number of trade mark approvals respectively. These approximations may be substituted into the market-value equation to show that the estimated reflects the ratio of a stock to a flow and therefore relates to the unmeasured stocks. Although there are various stages in both patenting and trade marking procedures that suggest the use of slightly different measures (i.e. applications, publications, registrations, etc.), there are a priori grounds for preferring publications in the case of patents and applications in the case of trade marks as these are the first time the relevant information is released publicly. In addition, experiments with a variety of lag structures suggest that the contemporaneous values of all of the intangible assets generally outperform other alternatives. In all cases, the measures of intangible assets
9

This approach is consistent with many previous studies in the USA and UK, notably Ben Zion (1984) (for a discussion, see Greene, et al. 1996, pp. 193-4). Where stocks and flows have both been tested, there is generally evidence that they do broadly the same job (Stoneman and Bosworth, 1994).

are normalised by the firms fixed tangible assets. The news-augmented version of Equation (2) is constructed by dividing the normalised R&D, patents, and trade mark variables into their predicted and residual elements, on the basis of a simple regression of each of the variables against time. In the second version of Equation (2), these anticipated and unanticipated variables are substituted for the actual R&D, patents, and trade marks in the stocks and flows equation. Both of the empirical equations include control variables. The choice of control variables is based on both statistical relevance and theoretical consistency. To the extent that they are valuation-relevant, the inclusion of control variables allows tighter inferences to be made about the contributions that intangible assets make to market value. There are a number of financial and accounting variables about which the markets might be expected to take an interest, and which prove to be statistically relevant to firm performance. Information on these is easily found in the firms annual statements on income, cash flow, and the balance of assets and liabilities. They are profit before tax (PBT), cash flows for investment in the firms assets (INVEST), ordinary dividends (ORDIV), long-term investments and receivables (FINA), current assets (CURA), and shareholders funds (EQUITY).10 Aside from the impact of financial and accounting variables, both of the empirical versions of Equation (2) employed here also control for the influence of unobserved factors, by the use of firm-specific and time-specific dummies. The time-specific dummies are used to control for inflation, since the data are given in nominal terms and are not readily convertible into real terms. The inclusion of both firm and time dummies also helps to whiten the residuals (Baltagi, 1995). The specifications actually used assume that the slope coefficients are constant across time and firms; that the differences in spatio-temporal firm performance would otherwise show up as differences in the constant; that the differentials are fixed and estimable, rather than randomly distributed; and that the residuals are white.11 The idea developed in Section (2.2) above is that the fixed, firm-specific effects capture useful and, to date, unexplored information on the performance of firms (Lindenberg and Ross, 1981). This data is examined graphically as discussed in Section (2.2) and more rigorously by estimating an equation of the form:

10

The rationale for the inclusion of these variables can be gleaned from the corporate finance literature (e.g., Ross, Westerfield and Jaffe, 1996). According to the classic dividend discount model, a firms historic dividends should be an important signal to shareholders of its future performance. In this context, profits and investment should provide an indirect indication of the shareholders future cash flows. Other models of corporate finance demonstrate that equity, as an element of a firms capital structure, may embody useful valuation information on its relative tax breaks, relative agency costs (of equity and debt), risk of financial distress and growth prospects. Finally, current assets (including financial assets) may offer an indication of a firms short-term management of cash flow and thus such valuation factors as its risk of financial distress, and its trade-off between liquidity and returns. See Greene (1993, pp. 466-480) for a detailed account of the fixed-effects model, and the decision whether to use fixed or random effects.

11

(7) yi = 1 ln Ai + 2

RD PAT TM X + 3 + 4 + , A i A i A i A i

Here yi is the estimated firm-specific effect for firm i for the entire period of the

RD PAT TM , and are the within-firm period averages of the A i A i X normalised R&D, patents, and trade marks variables respectively, and A
study, A , i is a vector of within-firm period averages for normalised financial and accounting control variables. Here also i is the relative effect (i.e., shadow value) of each of the normalised intangibles, and is a vector of the relative effects of the normalised financial variables.

4. Data and Results

4.1 Data
The data consist of observations on the financial market value, book value, R&D expenditures, patent publications, trade mark applications, and other financial and accounting performance indicators of 146 publicly traded production firms in the U.K. between 1990 and 1994.12 The study is limited to this period because of the absence prior to 1990 of comprehensive requirements for UK firms to disclose R&D expenditures and the consequent paucity of data on R&D expenditure before 1990. All data are valued on an end of calendar year basis, with adjustments being made to the length and timing of the reporting period where necessary. Firm performance is measured by the Tobin q ratio. In theory this should be the ratio of a firms marginal value to its marginal replacement cost, and in equilibrium would be equal to one (Tobin, 1969). In practice the lack of suitable data and the difficulty in computing a theoretically correct value means that, in this study, as in most other studies, Tobins q has to be approximated. In this paper, each firms Tobin q is calculated by dividing the product of its common stocks market value and long-term debts book value (Vi) by the book value of its fixed tangible assets (Ai). As with most approximations, including several theoretically-correct qs, such as the Lindenberg and Rosss q, this ratio is based on average rather than marginal values, because the latter are extremely difficult to calculate. The approximation used here assumes that both the value of debt and the replacement cost of fixed assets are equal to their book
12

The immediate source of the data is a database held at the Oxford Intellectual Property Research Centre (OIPRC). Its sources include Extels Firm Analysis data set, the UK Patent Office and the European Patent Office. A detailed description of the OIPRC data can be found in Longland, et al. (1997). Following Extels classification system, our data has been drawn from four production sectors, and therefore excludes data on service sector firms.

10

values. This is a necessary assumption due to the difficulty of collecting data both on the market value of debt and the replacement cost of fixed assets (see Chung and Pruitt, 1994). It also omits information on such variables as the value of preferred stock, the value of short-term debt, and the cost of assets other than fixed tangibles. The omission of the latter is not serious, since it does not interfere with the process of inferring the value of intangible assets. Indeed, the omission of intangibles is deliberate for this purpose, and otherwise unavoidable due to the omission of many such assets, such as brand names and goodwill, from firms accounts. R&D expenditure (RD) is used either as a flow variable or as a proxy for the stock of R&D. The rationale for this was presented in Section (3) above. Likewise, the number of UK published patents acquired by the firm and its subsidiaries (PU), and the number of patents published by the European Patent Office (PE) are also used both as flow variables and as proxies for the corresponding stocks.13 Unusually, the intellectual property data also includes the number of trade mark applications, TM. This count covers all British trade mark applications by firms, including their subsidiaries. The trade mark and patent data are scaled up by one million to avoid loss of accuracy when fitting the market-value equations to the data. All of the explanatory variables are normalised by the book value of the tangible fixed assets. Table (1) shows the mean and variance of these and the other variables used in the present study.
Table 1. Summary Statistics

Variable LnQ LnA TM/A RD/A PU/A PE/A PBT/A ORDIV/A INVEST/A FINA/A CURA/A EQUITY/A

Mean 1.0552 18.9252 0.0726 0.1687 0.0488 0.0230 0.2509 0.0949 0.2084 0.0860 1.8414 1.2935

Standard Deviations 0.6989 2.0727 0.2527 2.1906 0.1970 0.0817 0.3122 0.0738 0.3072 0.1691 1.2559 0.7910

13

Again, a detailed discussion of all the data, their strengths and limitations, can be found in Longland, et al. (1997).

11

4.2 Results: Basic Stocks and Flows Equation

The empirical specifications of the Griliches model used in this study were established via a process of eliminating theoretically inconsistent and statistically irrelevant financial and accounting control variables. Ordinary Least Squares (OLS) estimation of the resultant stocks and flows equation yields the results shown in Table (2), Column (iii).14 The specification explains roughly 99 per cent of the variation in the dependent variable, lnMV. The parameter estimates are used here primarily to infer the quality of the relationship between the various measures of intangible assets and the Tobin-q measure of firm performance. In this respect, the results are rather surprising in the light of previous research, since they indicate that none of the intangible asset variables (i.e. R&D, patents and trade marks) or any combination of them play a significant role in the explanation of the variation in market values. However, a test of the joint significance of the unobserved, firm-specific factors the so-called fixed effects indicates that they capture systematic differences in firm performance. This suggests two lines of enquiry: first, to examine the distribution of returns for evidence that the markets have assimilated the firms intangible assets with unobserved factors; and second, to control for the effects of unanticipated versus anticipated changes in the measured stocks of intangible assets. These lines of enquiry are supported both by Stoneman and Bosworth (1994), and by the results shown in Table (2), Column (ii), produced by estimating the market-value equation without firm-specific factors. The latter results indicate that both trade mark applications and European patent applications have a significant and positive impact on firm performance.15

14

Given the exploratory nature of the present research, no attempt has been made here to test the coefficients of the market-value equations for the presence of either heteroskedastic or autocorrelated residuals. There are several reasons to assume that the inferential consequences are not grave (see Baltagi, 1995, pp. 3-5). Furthermore, the incorporation of firm dummies and time dummies helps us to assume that the residuals are random. Both TM and EPO are also sensitive to the inclusion of the financial control variables; they have a positive effect on market value in the absence of financial control variables, although TM are only significant positive at the 7 per cent level in the presence of financial control variables. However, there are a priori grounds for including the financial control variables (as well as the firm-specific factors) and the estimates that accompany their inclusion are therefore preferred.

15

12

Table 2. The Market Value Equation

Independent Variable (i) LnA RD/A PRD/A SRD/A PU/A PPU/A SPU/A PE/A PPE/A SPE/A TM/A PTM/A STM/A PBT/A ORDIV/A INVEST/A FINA/A CURA/A EQUITY/A Constant Firm dummies Year dummies No. Obs 0.2953 (0.0747) 4.1136 (0.3389) 0.3713 (0.0531) 0.3119 (0.0985) 0.1499 (0.0203) -0.0741 (0.0305) 1.1651 (0.2130) No Yes 585 0.9611 0.4014(0.0583) 1.9459 (0.4071) 0.1073 (0.0366) 0.2216 (0.1009) 0.2073 (0.0306) -0.1022 (0.0504) 4.0319 (1.1696) Yes Yes 585 0.9906 0.1828 (0.0742) -0.1020 (0.0561) 0.5439 (0.0859) -0.2542 (0.1106) 0.3877 (0.6970) 3.7039 (0.3135) 0.3025 (0.0488) 0.2168 (0.0915) 0.0713 (0.0205) -0.0456 (0.0281) 1.2238 (0.2037) No Yes 584 0.9678 -0.1157 (0.1464) -0.0590 (0.0627) 0.4096 (0.0589) 1.7592 (0.4138) 0.1137 (0.0375) 0.2499 (0.1031) 0.2108 (0.0311) -0.1032 (0.0503) 4.2662 (1.2064) Yes Yes 584 0.9907 0.7318 (0.2342) 0.0090 (0.1485) 0.9725 (0.3412) -0.0820 (0.2932) -0.2013 (0.4640) 0.1273 (0.1740) -0.1350 (0.1017) -0.0149 (0.0777) -0.1629 (0.1272) -0.2404 (0.1511) -0.3126 (0.2329) 0.0337 (0.0860) (ii) 0.9396 (0.0099) -0.0028 (0.0071) (iii) 0.7501 (0.0653) -0.0003 (0.0039) 0.0935 (0.0129) 0.0162 (0.0465) 0.0317 (0.0468) 0.0144 (0.0257) Regression (iv) 0.9595 (0.0092) (v) 0.7435 (0.0689)

Notes: The dependent variable is ln MV. The data set comprises annual observations on 146 firms for the period 1990 to 1994. Emboldened figures indicate that the coefficient is significant at the 5% level. Standard errors are shown in parentheses. The firm dummies are collectively significant at the 5% level for all regressions. PRD/A is the predicted RD/A based on a regression of RD/A on time. SRD/A is the residual RD/A. PPU/A is the predicted PU/A based on a regression of PU/A on time. SPU/A is the residual PU/A PPE/A is the predicted PE/A based on a regression of PE/A on time. SPE/A is the residual PE/A PTM/A is the predicted TM/A based on a regression of TM/A on time. STM/A is the residual TM/A

13

The relative frequency distribution of the returns to unobserved firm-specific factors which are likely to reflect the effects of all intangibles (as none were separately or jointly significant in the earlier regression) are shown in Figure (1). The distribution is highly skewed, with the bulk of the observations being concentrated at the lower end of the distribution. The corresponding P-L relationship, shown in Figure (2), offers an alternative perspective on the distribution. It indicates that the distribution is concave in the P-L space and has a long Paretian upper tail (covering roughly 40% of the sample population), with a Pareto constant of approximately 2.2 and a corresponding Gini coefficient of 0.3.16 The results are consistent with those reported in Scherer (1996), both for patent values and for returns to venture capital investments. They are also consistent with the analysis of income distributions, which report values of the Pareto constant in the range 1.5-2.5 for the upper tail.17 Considered together, the figures suggest that the distribution of company performance is highly skewed, consistent with the highly skewed returns to R&D and values of patents and trade marks. This, in turn, is consistent with the extremely high risk of R&D, new product launch and other innovation activities (Allen, 1970; Crawford, 1987; Sanders, et al. 1958; Stevens and Burley, 1987).
Figure 1. Relative Frequency Distribution of the Firm specific Effects

16

The Pareto constant is estimated by an OLS regression of log(P) on log(y) over a suitable domain of y, established using line-of-eye judgement. According to Mandelbrot (1963) the value of the Pareto constant should not be trusted if it is larger than 3.

17

14

Figure 2. Pareto Levy Diagram for Firm specific Effects (y)

OLS estimation of the relationship between the firm-specific fixed effects and the within-firm averages of tangible fixed assets, intangible assets, and the financial control variables over the period 1990-1994 yield the results shown in Table (3), Column (ii).18 The equation fits the data fairly well, explaining roughly 67% of the cross-sectional variation in the dependent variable, the firm-specific effect. The coefficients for both the trade mark and European patent variables are significant and positive. The number of European patent applications appears to be a very good, if indirect, indicator of a firms ability to generate value. However, the results suggest that the R&D and UK patent variables have neither a direct or indirect impact on firm performance. There are two likely interpretations of this result: first, that while the investment in intangibles is inherently risky, firms with a larger number of patents and/or trade marks are more likely to own at least one that is commercially successful; and, second, that the market uses the numbers of patents and trade marks as an imperfect proxy for the firms capacity to generate future profits. The truth is likely to be some combination of the two.

18

The corresponding regression of the relative firm-specific fixed effect on the same explanatory variables yields results (available on request) that exhibit heteroskedasticity. The implied transformation of the relative into the absolute effect both produces a model with a better fit and solves the problem of heteroskedasticity.

15

Table 3. The Firm Specific Effects Equation

Variables (i) LnA TM/A RD/A PU/A PE/A PBT/A ORDIV/A INVEST/A FINA/A CURA/A EQUITY/A Constant No. Obs. (ii) 0.2062 (0.0182) 0.6181 (0.1781) -0.0215 (0.0316) -0.4497 (0.2924) 2.6196 (0.7005) -0.5257 (0.1713) 3.4417 (0.6817) 0.6171 (0.1280) 0.0158 (0.2074) -0.0824 (0.0386) 0.0024 (0.0543) -3.2670 (0.3778) 146 0.6734 Regression (iii) 0.2150 (0.0180) 0.6269 (0.1768) -0.0241 (0.0314) 0.0344 (0.2902) 2.5461 (0.6954) -0.5723 (0.1701) 3.5757 (0.6768) 0.5980 (0.1270) -0.0098 (0.2059) -0.1120 (0.0383) 0.0285 (0.0539) -3.3990 (0.3751) 146 0.6930

Notes: The dependent variable is the absolute firm-specific effect, y. Emboldened figures indicate that the coefficients are significant at the 5% level. Standard errors are shown in parentheses. The Cook-Weisburg test for heteroscedasticity using the fitted values of the dependent variable indicates that the residuals are homoscedastic.

4.3 Results: News-Augmented Model

We noted at the very outset that, if the intangible of assets of the company did not vary greatly over time, their role might be taken up by the FFEs. This might be an explanation for the US finding that it is the unanticipated, rather than anticipated values of intangibles that are important in the explanation of market value. Thus, we test this for the UK. OLS estimation of the news-augmented market-value equation yields the results shown in Table (2), Column (v). The equation again explains about 99 per cent of the variation in lnMV. The results indicate that R&D, patents, and trade marks do not influence the variation in firm performance, even after allowance is made for news. As above, the firm fixed effects are jointly significant. It is nevertheless instructive to look at the results for the news-augmented equation when the firm dummies are omitted (see Column (iv)). This reveals that the predicted variables for R&D, European patents, and trade marks and also the surprise element of trade mark data all have a significant influence on firm performance.

16

Graphical examination of the relative frequency distribution of the fixed, firm-specific effects (FFEs) associated with this specification are shown in Figure (3). The results indicate that the distribution of FFEs is again skewed to the right, as in the earlier results. Likewise, the P-L relationship, shown in Figure (4), reveals a long Paretian upper tail, with a Pareto constant whose value is approximately 2.1, corresponding to a Gini coefficient of 0.3. Thus, these results also correspond with those reported in Scherer (1996) and the income distribution literature (see Cowell, 1995).
Figure 3. Relative Frequency Distribution of the Firmspecific Effects

17

Figure 4. ParetoLevy Diagram of the Firmspecific Effects (y)

The results of estimating the firm-specific effects equation are shown in Table (3), Column (iv).19 As in the earlier results, both the trade mark and European patent variables have a significant influence on firm performance, while the UK patent and R&D variables appear to be irrelevant to firm performance. Thus, our earlier conclusions about this still apply.

5. Conclusions
This paper used econometric techniques to explore the contribution of intangible assets to the performance of 146 UK firms between 1990 and 1994. The econometric framework adopted was almost identical to that set out in the seminal work of Griliches (1981), but the present study had a number of innovative features, including (i) the application of the model to UK data; (ii) the inclusion of a wide range of measures of intangible assets (i.e., R&D expenditures, patent publications and trade mark applications); and (iii) the exploration of the firm-fixed effects as measures of dynamic performance. Two approaches reported in the literature were taken: a basic stocks and flows approach (typified by Hall, 1993b); and a news-augmented approach (typified by Griliches, 1981). Both were extended to involve not only the estimation of a fixed-effects, market-value equation, but also a graphical and regression-based examination of the distribution of firm-specific effects.

19

As above, the results of the corresponding relative effects regression exhibit heteroskedasticity, which implies that the present transformed model both has a better fit and solves the problem of heteroskedasticity.

18

Estimation of the stocks-and-flows market-value equation produced some rather surprising results when viewed in the context of previous applications of the Griliches framework: neither R&D expenditure, patent publications, trade mark applications, or any combination of these variables appeared to have a significant influence on firm performance. However, other results indicated several lines of enquiry. First, the estimated equation indicated systematic differences in the performance of firms that the observed variables failed to capture, but which were picked up by firm-fixed effects. Second, estimation of the equation in the absence of firm dummies revealed that both the European patent and trade mark variables had a positive influence on firm performance. Our reaction to these results was twofold. First, it seemed possible that insofar as the levels of R&D, patents and trade mark activity for each firm were relatively stable, FFEs would dominate, leaving little for these intangibles to explain. One way around this was to attempt to construct surprise variables, particularly as the studies that had used both anticipated and unanticipated changes in intangibles had suggested that the latter tended to dominate. Second, whatever the outcome of this, the significant role played by FFEs suggested that these were an important summary measure of the stocks of intangible assets (intellectual capital) and of variation in dynamic firm performance. This suggested the need to explore the distribution of FFEs and to see if they could be explained in terms of stocks of intangible assets. In this context, the results of estimating the news-augmented equation were particularly revealing. The results were similar to those produced in Part 1 of the analysis for the fixed-effects, stocks and flows equation. In other words, when the FFEs were included in the equation, we could find no significant role for the intangible assets. However, when the model was estimated without firm dummies, the results indicated that European patent applications, trade mark applications, and R&D expenditures do influence firm performance when allowance is made for surprise changes in those variables. This suggests that the FFEs and the set of intangible asset measures are doing broadly the same job in the equations. If this is the case, then it should show up in terms of a relationship between the two. Hence, this paper reported the results of both a graphical and regression-based examination of the firm fixed effects (i.e. the coefficient on the firm-specific dummy variables). The distributions of the firm-specific effects for both the stocks-and-flows and the news equations were highly skewed rightwards. In other words, the majority of the companies reported low values, with only a small proportion reporting high values. There was evidence of concavity in the Pareto-Levy space, as predicted by Bosworth and Jobome (1999) and Bosworth and Mahdian (1999). In addition, there was evidence of linearity in the right hand side of the distribution (i.e. evidence of a Paretian upper-tail) as suggested by Scherer (1996). This indicated a wide range of firm performance, consistent with the underlying riskiness of investments in intangibles. 19

Estimation of a simple cross-sectional equation indicated, in both cases, that there was a significant and positive relationship between, on the one hand, trade marks and, especially, patent applications and, on the other hand, the firm fixed effects, and thus indirectly firm performance. Both UK patents and R&D expenditures failed to explain variations in the firm-specific effects, and thus indirectly firm performance. The UK patents have often been found to be less significant than their European counterparts in our previous work (op cit.). Similarly, we have previously often found that R&D and patents do broadly the same job, although it has generally been the R&D rather than the patents that have dominated. One interpretation of the relative importance of R&D and patents in the present study is that the former is risky in both a technological (i.e. does not lead to a patentable invention) and a commercial sense (i.e. even if it leads to a patent, the associated invention may not be commercially viable), while the latter only carry the commercial element of risk. Overall the results of the analysis suggest that the market considered only a relatively few firms capable of generating substantial if indirect returns to their intangible assets. This is entirely consistent with previous findings relating to the value of patent protection (Pakes, 1986; Pakes and Schankerman, 1984; Harhoff, et al. etc.). It is also consistent with the results of previous empirical research on the value of venture capital investments (Scherer, 1996). The results also suggest that market sentiment was based crucially on the specific characteristics of those firms. Previous research suggests that these characteristics include the appropriability environment in which the firms operate, and the corresponding competitive advantages for which measured intangibles were intended to be a proxy. For example, Lindenberg and Ross (1981) show that firms with high q ratios tend to be those with very strong brand images, know-how, and so forth, whereas those with the lowest q ratios have generally been operating in highly competitive, highly regulated and/or shrinking industries. If this is the case, it gives some support to the contention that the stocks of measured intangibles have been assimilated in the valuation process with unobserved firm-specific factors. It is perhaps too early to speculate on the nature of the underlying processes that give rise to this complex, skewed distribution. However, it is well known that the process that underlies the Pareto distribution is related to that which underlies the lognormal distribution (see Cowell, 1995). The basic idea is that if proportionate changes in income (y) are normally distributed and there is a lower bar below which y cannot fall, then the overall distribution of y will eventually be Paretian. In this case, one could argue that the distribution of differential firm performance might be related to the chances that innovations are a commercial success and/or a technological success. These chances of success might in turn be related to the dynamism and structure of the overall economy within which innovation takes place (Dasgupta and Stoneman, 1987). Cowell (1995), for instance, notes that the Pareto distribution can be shown to result from simple hypotheses about the formation of returns within bureaucratic systems. In other words, given a hierarchical structure of payment and a fairly stable relationship between the remuneration of strong and weak firms, the resulting distribution could 20

be Paretian. This, one could argue, is not inconsistent with the directional and cumulative character of technological history described by Gould (1996), and the attendant and potentially Lamarkian inheritance of favourable innovations which could be said to characterise industrial clustering and concentration in a dynamic, capitalist economy.

21

References
Allen, J.M. (1970). Project Estimates and Outcomes in Electricity Generation Research. Journal of Management Studies, 7(3), 271-287. Austin, D.H. (1995). An Event-Study Approach to Measuring Innovative Output: Case of Biotechnology. American Economic Review: Papers & Proceedings, 83(2), 253-258. Bak, P. (1997). How Nature Works: the Science of Self-organised Criticality. Oxford: Oxford University Press. Baltagi, B.H. (1995). Econometric Analysis of Panel Data. Chichester: Wiley. Ben Zion, U. (1984). The R&D and Investment Decision and its Relationship to the Firms Market Value: Some Preliminary Results. In Z. Griliches (ed.), R&D, Patents and Productivity. Chicago: Chicago University Press. Bosworth, D. and G. Jobome (1999). The Measurement and Management of Risk in R&D. International Journal of Technology Management, (forthcoming). Bosworth, D. and H. Mahdian (1999). Returns to Intellectual Property in the Pharmaceutical Sector. Economie Appliquee, 52(2), 69-93. Chaney, P.K., T.M. Devinney and R.S. Winter (1991). The Impact of New Product Introductions on the Market Value of Firms. Journal of Business, 64(4), 573-610. Chung, K.H. and S.W. Pruitt (1994). A Simple Approximation of Tobins q. Financial Management, 23(3), 70-74. Cockburn, I. and Z. Griliches (1988). Industry Effects and Appropriability Measures in the Stock Markets Valuation of R&D Investment. American Economic Review: Papers and Proceedings, 78(2), 419-423. Connolly, R.A. and M. Hirschey (1990). Firm Size and R&D Effectiveness: A Value-Based Test. Economic Letters, 37, 277-281. Cowell, F.A. (1995). Measuring Inequality (2nd edition). London: Prentice Hall / Harvester Wheatsheaf. Cramer, J.S. (1969). Empirical Econometrics. Amsterdam: North Holland. Crawford, M.C. (1987). New Product Failure Rates: A Reprise. Research Management, 30(4), 20-24.

22

Dasgupta, P. and P. Stoneman (eds.) (1987). Economic Policy and Technological Performance. Cambridge: Cambridge University Press Gould, S.J. (1996). Lifes Grandeur: The Spread of Excellence from Plato to Darwin. London: Vintage. Green, J.P., A.W. Stark and H.M. Thomas (1996). UK Evidence of the Market Valuation of Research and Development Expenditures. Journal of Business Finance and Accounting, 23(2), 191-216. Greene, W.H. (1993). Econometric Analysis. Oxford: MacMillan. Griliches, Z. (1979). Issues in Assessing the Contribution of Research and Development to Productivity Growth. Bell Journal of Economics, 10, 92-116. Griliches, Z. (1981). Market Value, R&D, and Patents. Economic Letters, 7, 183-187. Griliches, Z. (ed.) (1984). R&D, Patents, and Productivity. Chicago: Chicago University Press. Hall, R. (1993a). A Framework Linking Intangible Resources and Capabilities to Sustainable Competitive Advantage. Strategic Management Journal, 14, 607-618. Hall, B.H. (1993b). The Stock Markets Valuation of RD Investment During the 1980s. American Economic Review: Papers and Proceedings, 83(2), 259-264. Hall, B.H. (1998). Innovation and Market Value. Prepared for the NIESR Conference on Productivity and Competitiveness, London, February 5-6, 1998; forthcoming in R. Barrell, G. Mason and M. OMahoney (eds.) Productivity, Innovation and Economic Performance, NIESR/Cambridge University Press. Hall, B.H., A. Jaffe, & M.Trajtenberg (1999). Market Value and Patent Citations: A First Look. Prepared for the Conference in Commemoration of Zvi Griliches 20 Years as Director of the NBER Program on Productivity and Technological Progress, Cambridge, Massachusetts, March 5 and 6, 1999. Hall, B.H. and D. Kim (1997). Valuing Intangible Assets: The Stock Market Value of R&D Revisited. Nuffield College, Harvard University, and NBER: Work in Progress. Halvorsen and Palmquist (1980). Interpretation of Dummy Variables in Semi-log Equations. American Economic Review, 70(3), 474-475. Harhoff, D., F. Narin, F.M. Scherer, and K. Vopel. (1997). Citation Frequency and the Value of Patented Inventions. Mannheim: ZEW Discussion Paper No. 97-27. 23

Hirschey, M. (1982). Intangible Capital Aspects of Advertising and R&D Expenditures. Journal of Industrial Economics, 30(4), 375-390. Lanjouw, J.O., A. Pakes and J. Putnam (1996). How to Count Patents and Value Intellectual Property. Conference on New Developments in Intellectual Property: Law and Economics, St Peters College, Oxford. Levy, P. (1925). Calcul des Probabilits. Paris: Gauthier-Villars. Levy, P. (1937). Thorie de lAddition des Variables Alatoires. Paris: Gauthier-Villars. Lindenberg, E.B. and S.A. Ross (1981). Tobins q Ratio and Industrial Organisation. Journal of Business, 54(1), 1-32. Longland, M., C. Greenhalgh, P. Hayward and D. Bosworth (1997). Matching the Intellectual Property and Accounts Data of UK Firms. Working Paper, Intellectual Property Research Institute, St Peters College, Oxford. Lynn, B.E. (1998). The Management of Intellectual Capital: the Issues and the Practice. Hamilton (Ontario): Society of Management Accountants of Canada. Mandelbrot, B. (1963). New Methods in Statistical Economics. Journal of Political Economy, 71, 421-440. Pakes, A. (1986). Patents as Options: Some Estimates of the Value of Holding European Patent Stocks. Econometrica, 54, 755-784. Pakes, A. and M. Schankerman (1984). The Rate of Obsolescence of Patents, Research Gestation Lags, and the Private Rate of Return to Research Resources. In Z. Griliches (ed.), R&D, Patents and Productivity. NBER Conference series. Chicago: Chicago University Press. Pareto, V. (1897). La Courbe de la Rpartition de la Richesse. In Recueil Publi par la Facult de Droit lOccasion de lExposition Nationale Suisse, pp. 373-387. Lausanne: Universit de Lausanne. Peiro, A. (1999). Skewness in Financial Returns. Journal of Banking and Finance, 23, 847-862. Ross, S.A., R.W. Westerfield and J.F. Jaffe (1996). Corporate Finance (4th edition). London: Richard D. Irwin. Ruggles, R. (1998). The State of the Notion: Knowledge Management in Practice. California Management Review, 40(3), 80-89.

24

Sanders, B., J. Rossman and L.J. Harris (1958). The Economic Impact of Patents. Patent, Trademark and Copyright Journal, 2, 340-363. Scherer, F.M. (1996). The Size Distribution of Profits from Innovation. Harvard University Working Paper. (Patents: What We Know; What Must We Learn? Keynote Address, Applied Econometrics Association: The Econometrics of Innovation (Patents), Strasbourg.) Shapiro, D.M. and L.N. Switzer (1993). The Stock Market Response to Changing Drug Patent Legislation: the Case of Compulsory Licensing in Canada. Managerial and Decision Economics, 14, 247-259. Sougiannis, T. (1994). The Accounting Based Valuation of Corporate R&D. The Accounting Review, 69(1), 44-68. Stevens, G.A. and J. Burley (1997). 3000 Raw Ideas = One Commercial Success. Research Technology Management. Stoneman, P. and D. Bosworth (1994). Feasibility Study for the Development of an Innovation Scoreboard. Final Report to the DTI. London: Stoy Hayward Consulting. Tobin, J. (1969). A General Equilibrium Approach to Monetary Theory. Journal of Money, Credit, and Banking, 1(1), 15-19. Toivanen, O.P., P. Stoneman and D. Bosworth (1998). Innovation and the Market Value of UK Firms, 1989-1995. Working Paper. Warwick Business School. Warwick University. Zantout, Z.Z. and G.P. Tsetsekos (1994). The Wealth Effects of Announcements of R&D Expenditure Increases. Journal of Financial Research, 17(2), 205-216.

25