Effects of Intangible Capital on Firm Performance

Transcription

1 Effects of Intangible Capital on Firm Performance Jannine Poletti Lau August 14, 2003 Abstract The main objective of this research is to examine the effect of intangible capital on the market value of the firm via the Tobin s Q ratio. Intangible capital is represented with different accounting variables, which were selected on the basis of agency cost and signalling theories. The empirical model is tested in firm-level panel data for UK and Japan. The model is applied in both static and dynamic specifications. The dynamics of the model were studied to analyse the effect of adjustment costs of Tobin s Q. An attempt was made to determine the effect of technological development of rivals on market value. A comparison of three different estimators: OLS, Within and the GMM, is pursued to analyse the effects of different panel data techniques. The results showed a significant role of intangible capital in influencing the market value of the firm. Key words: intangible capital, firms value, GMM JEL classification: D21; L10; G39 Work in progress, please do not quote or circulate Address for correspondence: The University of York, Department of Economics and Related Studies, Heslington, York, YO10 5DD, United Kingdom, jp132@york.ac.uk 1

2 1 Introduction Intangible assets are an important influence on the performance of the firm. A good management of intangible assets could have a positive impact on the future profits of companies. Several researchers have studied the relevance of intangible capital and more specifically of Research and Development (R&D) at a firm level. In this respect, it is essential to note the difference between R&D, which is registered in both the balance sheet and the profit and loss account. The International Accounting Standards 1 requires that both the definition and the criteria for the recognition of intangible capital as part of the capital in thebalancesheetaremet. Inthecasethatthedefinition is not fulfilled, the transactions on R&D must be recognized as an expense in the profit andloss account. Moreover, the company itself may direct these kind of transactions either to the balance sheet (as capital) or to the profit and loss account (as an expense) as its convenience. The later benefits companies as it diminishes profits and subsequently taxes. Herein, R&D will be referred to as an expenditure recognized in the profit and loss account for future reference. This expenditure is done with the objective of increasing the future profits of the firm, as may be reflected in its market value. To date, researchers have emphasised the relation of R&D and investment (Mulkay et al 2000, Mairesse et al 1999, Bond et al 1999, Toivanen et al 1997). Some other empirical work with intangible capital has been conducted through the productivity of the firm via the Cobb Douglas production function (Goto and Suzuki 1989, Griliches 1994, Wakelin 2000). One of the aims of this research is to analyse the impact of R&D on the market value of the firm. The basis for the implication of this relationship can be observed in figure 1. The upper part of the figure shows the normal financial structure represented in the balance sheet. The invisible equity represents the excess in value of the firm which is recognized by the stock market (below the line). Thus, there should be a counterpart of the invisible equity that can explain the extra value. For this research, it is argued that the counterpart is related to intangible assets, as shown in figure 1. 1 The International Accounting Standards Board, IAS 38: INTANGIBLE ASSETS, Effective for annual financial statements covering periods beginning on or after 1 July

3 Figure 1. The invisible Balance Sheet Market value: Visible + Invisible Equity Intangible assets can be classified in three different categories (see Sveiby 1998): External structure, internal structure and individual competence. External structure represents the popularity of the company with the outside world, or in other words the degree of success (reflected in the increase of profits) that firms would obtain on the basis of being well known to the costumers. Publicity, advertising, brand names and reputation are good examples of external structure. Internal structure represents the positive or negative way that the firm develops from inside. This refers to the administrative structure, technology and computing systems among others. Individual competence is the third category, it reflects employees capacity, skills, relationships, education, experience. An example for this category is R&D, which in the future is expected to bring back the incurred expense plus a premium, so the investment is done with the same objective as any tangible asset. This family of three categorization was developed by a Swedish working group in 1987 and it is known as The Konrad Theory and more recently as Intangible Assets Monitor. The impact that intangible capital has on the stock market valuation has been exploited, among others, by Hirschey 1982, who found that advertising and R&D expenditures have positive and significant market value effects. Some other research involved with market value and R&D has been conducted by Hall 1993, Megna and Klock 1993, Blundell et al 1998, Hall 1999, Klock and Megna 2000, Chen and Steiner For the purposes of this analysis, the measure for market value is represented by Tobin s Q ratio, which may be viewed as a direct measure for expected future 3

4 profits. Tobin s Q is constructed with the market value of debt and equity over the replacement costs. The second aim of the present research is to analyse the effects of different panel data techniques, namely the OLS estimator, the Within estimator, and the General Method of Moments (GMM), with the application of static and dynamic models. The model is estimated from two dataset samples from developed countries: UK and Japan. The samples consist of about 930 and 1577 firms respectively, for twelve years ( ). In addition, the inclusion of spillovers as an extra variable intends to capture the effects of technological innovations from rivals in the model. The proxy chosen to measure the spillovers was the R&D expenditure among firms which belong to the same industrial sector (excluding the firm s own R&D). In the following section, the construction of the empirical model is explained. Section three contains the data description and statistics for both countries. Section four shows the econometric results with the implementation of the different econometric techniques and section five contains the results with the addition of spillovers in the model. 2 The empirical model of corporate value To date, most of the empirical work related with intangible capital has focussed on the US due to the numerous sources of information that are available. In the present analysis of UK and Japan, a similar kind of research has been more limited. A model developed by Klock and Megna (1993, 2000) is used as a starting point. The basic measure is the influence of the intangible capital on Tobin s Q ratio. Tobin s Q ratio represents the value of the firm and can be defined as, q = Mv (1) K 1 + K 2 where K 1 + K 2 are tangible and intangible assets, respectively, as those observed in figure 1. Mv is the market value of debt and equity. The use of q for measuring intangible value is based on the assumption that a company s long-term equilibrium market value must be equal to the replacement value of its assets, giving a q value close to unity. Deviations from this relationship (where q is significantly greater than 1) are interpreted as an unmeasured source of 4

5 value, generally attributed to a company s intangible value. Therefore, in this model, q is expected to be equal 1, where the market value of a firm is equal to the replacement cost of its assets, so the firm would represent its real value in the long term. However, the intangible capital is not observable, so the Q ratio which can be calculated with the observable capital is: q 0 = Mv K 1 (2) As said before, in the long term an equilibrium of q = 1 is expected, so under that condition market value can be expressed as Mv = K 1 +K 2. From equation (2), Mv is substituted to obtain the following relationship with the observable q 0 : q 0 = K 1 + K 2 =1+ K 2 (3) K 1 K 1 K 2 represents all the j intangible characteristics for all the i available observations for each period t, asshowninequation4; K 2 = nx K jit (4) j=1 Equation 3 can be rearrange to represent a linear model for a panel data set as follows: q it = α + X β j K jit K 1it + ε it (5) where α represents the intercept, K 1it is the value of the tangible capital represented by total assets per firm, and K jit is the value of all the unobservable explanatory variables that in some way represent the intangible assets in the firm. β j gives all the parameters of estimation for each of the intangible assets included in the model. ε it is the error term which absorbs the effect of intangibles that were not included. Thus, any premium over the balance-sheet valuation that the stock market places can be a reflection of the company s intangible resources, represented by K jit. One of the aims of this research is to apply a panel data model because of its methodological advantages such as the possibility to analyse the disaggregate dynamic relationship and the availability of a larger number of observations. 5

6 The balance sheet interpretation (described in the introduction) suggests that the excess of market value which is not reflected in the tangible capital is part of the invisible equity. The invisible equity is explained from the firms level by three different structures of intangible assets: external structure, internal structure and individual competence. As said before, individual competence concerns to the employees capacity, skills and knowledge, among others. A good representative for this definition is the Research and Development (RD). Education and experience are other characteristics that describe individual competence. For this study, however, there is no data available that can represent them, so it is expected that the error term will absorb this lack of information. In empirical analysis, internal structure is an important part of intangibles, as a good number of explanatory variables from this category can be observed and measured. Internal structure represents the way that the firm develops from inside. A firm s performance may be observed from some specific accounting information. The influence of the accounting information on market value can be examined via two important theories of capital structure: Agency Cost and Signalling Theories. It is well known that in a market with taxes, debt increases the value of the firm. This impact takes place since debt diminishes profits and subsequently taxes. Managers can manipulate investment projects with the purpose of obtaining loans from bondholders that increases the wealth of the firm. This is called agency cost of debt. In general, funds in a firm come from two main sources: debtholders (bonds) and equityholders (shareholders). Shareholders are looking to maximize their investment so as to pay back to bondholders and pay the respective dividends to shareholders. At the same time, shareholders want to increase their own wealth through the increase in market value of their shares. The way to maximize wealth is by choosing the projects that have a higher rate of return than the cost of capital. These projects will be presented to the bondholders with the objective of obtaining money to finance them. Likewise, to borrow capital implies that the project will pay back at least the same amount of money that was obtained from the bondholders. Managers are able to swap the money obtained from a specific project to a riskier one, which if successful will also increase the wealth of the shareholders. Thus, as part of the internal structure, the effect of leverage (TL) on the value of the firm is included in the model. 6

7 According to signalling theory, the influence of profitability and growth on the firm s value are important. They were studied by Hirschey in 1982, who stated that current profits provide an additional indicator of a firm s profit potential reflected in its value. Profits (P ) are calculated as the sum of net income, interest expense, dividend payments and depreciation and represents the measure of cash flow of the company. It seems reasonable to assume that firmsthatinvestintangibleassetswillreturnprofits that increase the firm s value. Furthermore, dividend payments and changes in dividends policies are regarded as conveying information about permanent earnings (See Brainard et al 1980). Dividends payments (D) can give the firm certain stability which may be reflected in its value. To this end, it is important to be cautious about dividends, which could increase or decrease with the financial policy of the firm. However, it is logical to think that in either case a firm s value is affected as a response. In signalling models the payment of high dividends is used to separate firms with favourable inside information from other firms. Under the endogenous investment hypothesis studied by Hall in 1999, a simple model of capital accumulation is constructed, where corporate securities as a proportion of a productivity index is divided between the payout to the owners of corporations (dividends) and capital accumulation (cash flow per period). From this study a relationship of dividends with firm s value is also implied. For the model in this paper, it is assumed, that all managers anticipate the increase of future payouts via the increase of dividend payments, which is a reflection of good performance therefore increases value. Moreover, the market value of a firm represents the future stream of income discounted to the present value. If the firm s value represents future profits, it is logical to think that the investment of the company, after a time period, will bring back such investment plus a premium, or in the opposite case, when the firm disinvest, a negative effect on value might be expected. As in Hall 2000, the value of the firm is the present value of the future payouts. An amount x of payout is expected depending on the investment in capital held for a productive use, Pr oductive capital t = S t +(1+δ)F t (6) where S is the inventory of the firm (stocks +WIP, datastream X(364)), δ is a rate of obsolescence which, for simplicity, could be equalized to the depreciation rate and F is the value of net fixed assets. Thus, from equation (6) we obtain the investment in productive capital from one period to the other. 7

8 I t = P t P t 1 (7) where I t is the investment for period t. Finally, the value of the firm is measured with only two of the structures mentioned above: internal structure and individual competence. In the internal structure, besides investment, the choice of variables was based on the signalling and agency cost effects. Hence, it is observed that the intangible assets are defined for this research asafunctionofdifferent factors: K jit = f(rd it, TL it,d it,i it,p it, size it, ψ it ), which are defined as: research and development, total leverage, dividends paid, investment in productive capital, profits, respectively. ψ represents those intangibles that were not included in the model which effect is absorbed by the error term. An example of such intangibles is advertising, which is part of the external structure, however, due to the data availability, it was not included. The inclusion of size as a control variable is measured via the log of total assets. This variable has been deflated to 1989 prices to represent a constant measure of size over the whole period of analysis. Finally, for estimation purposes, the following linear specification of the valuation function was adopted; q it = α it + β 1 R&D it TA it TL it D it + β 2 + β TA 3 + β it TA 4 it +β 5 P it TA it + β 6 size it + γ i + λ t + υ it I it TA it (8) where γ i represents the firm effects, λ t represents the period effects and υ it is the error term. TA is total assets, which are also useful to control the effect of firm size and as a deflator, which helps to avoid some possible problems of multicollinearity and heteroskedasticity. 3 Data Description The analysis was applied in two different countries: UK and Japan. The two unbalanced samples where approximately 915 and 1491 firms, respectively. The period under study was from 1989 to 2000 for both samples. The variables were obtained from datastream and included all the non-financial industrial 8

9 sectors 2. The dependent variable is the Tobin s Q ratio, which measures the expected future profits. The construction for this ratio is a proxy which includes the market value of equity and debt over the replacement cost (total assets used as a proxy). The selection of the sample was based on the availability of the information. Both databases were manipulated to obtain a more homogeneous criteria. Some cases that had a negative ratio for Q were deleted. By definition Q cannot be a negative number. The cause of the negative sign came from total assets, which, cannot be negative as the lowest amount for any asset of the firm is zero. Since Tobin s Q ratio (Q) is an important variable and its specification is very susceptible to measurement errors, outliers were controlled. The criteria used was to eliminate observations in a range of 0.2% (including maximum and minimum points). This specification deletes all the outliers where Q>100 in both databases. The last criteria was that the minimum number of years available per firm must be four, so observations that did not follow this rule were deleted. With these modifications the final number of observations (firms times years), were 8,671 for UK and 16,323 for Japan. The variable R&D is represented with the account Research and Development (datastream X119). This figure includes disclosed amounts of expenditure in the year which are not capitalized in the balance sheet. With this variable the stocks of R&D were constructed. Every year starting from 1984, R&D was accumulated until the year 1989, from here, besides the accumulation per year to construct the stock, a depreciation rate of 15% per year (δ) was applied. This methodology has been popularized by Hall (1990) and is based in a standard perpetual inventory equation with declining balance depreciation. The formula applied for the accumulation is as follows: RDstock it =(1 δ)rdstock it 1 + RD it where RDstock it is the end of period stock of R&D and RD it is the expenditure during the year. Table 1 contains the descriptive statistics for both countries. 2 Aerospace + Defence, Automobile and Parts, Beverages Chemicals, Construction + Materials, Diversified Industry, Electronic + Electric, Engineering + Machine, Food + Drug Retailers, Food Producers +proc, Forestry + Paper, Gas Distribution, Health Care, Household Goods + textiles, IT hardware, Leisure and hotels, Media + photography, Mining, Oil and Gas, Packaging, Personal Care + house, Pharmaceuticals, Restaurants and Pubs, Retailers General, Software + Services, Steel + other materials, Support Services, Telecom Services, Tobacco, Transport. 9

10 Table 1: Descriptive Statistics- UK & Japan ( ) UK (915 firms) Median Mean St. Dev Minimum Maximum ratios (all divided by TA) TQ RD TL I P D Japan (1491 firms) Median Mean St. Dev Minimum Maximum ratios (all divided by TA) TQ RD TL I P D Table 1.1 contains the panel data structure, where it was observed that for both countries, the highest number of observations was concentrated in firms with 12 periods (55.7% for UK and 81.1%. for Japan) Table 1.1 Panel data structure number of firms number of years UK Japan Total

11 Table 2. Percentage of R&D analysed over the total population for UK ($millions) year Expenditure on R&D analysed Expenditure on R&D performed in UK % of R&D included in this research ,594 11, % ,697 11, % ,216 12, % ,403 12, % ,499 13, % ,091 14, % ,139 14, % ,508 14, % ,880 14, % ,205 15, % ,918 16, % ,342 na na grand total 71, , % Sources: Datastream and Office for National Statistics and Datastream Table 2 contains the expenditure on R&D performed in UK businesses and compared with the expenditure analysed in this research. The data of R&D performed in UK was obtained from the web-page of UK national statistics. This webpage contains a wide selection of data produced by the Government Statistical Service (GSS) and other statistical bodies in the public sector. R&D related concepts follow internationally agreed standards defined by the Organization for Economic Cooperation and Development (OECD) and published in the Frascati manual. R&D is defined as creative work undertaken in a systematic basis in order to increase the stock of knowledge, including knowledge of man, culture and society and the use of the stock of knowledge to devise new applications. R&D performed in UK businesses excludes R&D funded by UK businesses that is performed overseas or in other sectors of the UK economy (such as higher education; government departments, agencies and non-departmental public bodies; local authorities; and private non-profit organisations). Public corporations are counted as business enterprises. 11

12 Table 3. Distribution of the highest expenses of R&D per industrial sectors UK sector ( millions) year pharmaceutic aerospace IT hardware Food chemicals others Total producers ,156 3, ,424 4, , ,521 5, , ,552 5, , ,499 4, , ,558 4, , ,640 5, , , ,802 6, , , ,026 6, , , ,346 7, ,296 1,047 1, ,461 8, ,398 1, ,012 9,342 Grand Total 18,771 9,283 8,984 6,614 5,845 21,995 71,492 % 26.26% 12.98% 12.57% 9.25% 8.18% 30.77% % Total number of observations % 2.06% 1.57% 1.69% 3.88% 2.27% 88.53% 100% Japan sector ( millions) year IT hardware electronic+elect pharmaceuticals chemicals automobile+parts others Total ric ,776 1, , ,618 8, ,979 1,632 1,029 1, ,758 9, ,773 2,214 1,307 1,264 1,120 2,286 11, ,177 2,797 1,788 1,633 1,369 3,062 15, ,783 3,519 2,348 1,694 1,630 3,889 18, ,768 3,738 2,235 2,193 1,804 4,601 21, ,558 3,896 2,243 2,496 1,847 4,729 18, ,631 2,729 1, ,589 3,607 14, ,882 2,789 2,010 1,920 1,750 3,928 19, ,111 4,234 2,442 2,318 2,263 4,307 22, ,779 4,517 1,431 1,077 2,370 3,423 21, ,921 5,338 2,714 2,536 3,108 5,442 25,059 Grand Total 64,138 39,031 21,666 19,780 20,010 42, ,276 % 30.94% 18.83% 10.45% 9.54% 9.65% 20.58% % Total number of observations % 1.91% 12.46% 2.63% 10.21% 5.59% 67.19% 100% Following this comparison the percentage of firms that report information on R&D expenditures in datastream is increasing every year, so for 2000, a high percentage of R&D analysed is achieved. This strengthens the conclusion that, at least for UK, datastream is a good source of information. The firms on analysis constitutes for 1999 up to 53.5% of the total expenditure of R&D in UK, which is a very representative sample of the population of R&D expenditure in UK. Table 3 shows the highest expenditure on R&D per industrial sector. The currency utilized for both countries is British pounds sterling. A clear inconsistency in IT hardware for UK can be observed from the table above. This three year decrease from was due to a specific firm named Marconi (datastream code: ). Marconi reported zero R&D for three years ( ) after having reported $1,040 million, which represents 98% of 12

13 the total R&D for this sector and 1.5% of the total R&D expenditure in UK. In spite of the change of information on this firm, it is included in the sample. For Japan, IT hardware is the industrial sector with the highest expense, however it only represents 1.91% of the total number of observations. For UK the pharmaceutical sector has the highest R&D expenditure with 26.26%. It has shown a continuous increase from 1989 as the most of the other industrial sectors, the only exception to these continuous increases is the chemical sector which decreased its R&D expenditure by 49.6% from 1989 to Since numbers are not deflated, it is possible to compare the percentages of increase or decrease with the inflation rate for that same period ( ) in UK, which was 47.7%. An important observation is that the concentration of firms in sectors with the highest R&D is small in relation with the whole sample of observations. The number of firms that are included in the top five sectors with the highest R&D expenditures constitute a small percentage of the total number of firms in the sample. The range for both countries is from 1.57% to 12.46%. 4 Regression Analysis The analysis in this section aims to explain the effects of intangible capital on Tobin s Q ratio, which represents the productivity and performance of the firm. In this research, three different panel data techniques were applied to find the best estimator: pooled OLS; Within estimator or Fixed effects estimator, which permits the existence of correlation between the individual effects and the explanatory variables, and; General Method of Moments (GMM) with the first differences transformation. GMM is expect to find the best estimation, as it corrects for the presence of correlated firm-specific effects as well as the biases originated from the endogeneity of explanatory variables with the error term. According to theory, the expected GMM estimators should be within the range suggested by OLS and the Within estimator, which are likely to be biased in opposite directions (see Bond 2002 for a complete discussion). The GMM estimator has been recently studied in empirical applications to the investment theory. Normally it is applied to dynamic models, since the correlation of the lag variable (y i,t 1 ) with the error term is implicit. For this research GMM estimator of the type developed by Arellano and Bond (1991) is applied to both static and dynamic models. An important feature of GMM is the application of an adequate matrix of 13

14 instrumental variables, as they may be subject to large finite sample biases when the instruments available are weak. In this study the variables were obtained from accounting information, where the determination of a variable is closely related with the ratio of Tobin s Q, so simultaneity was taken into account. As Tobin s Q and some of the explanatory variables are simultaneous, they have to be treated as endogenous. Endogeneity is assumed in all the variables to avoid the bias caused from a less flexible model as the one utilized for this research. In tables 4 and 5 some estimates for both countries are presented. The results where estimated with Pcgive for both countries and for all the estimators (OLS, Within and GMM). All regressions include time dummies. Column (1) in tables 4 and 5, present the results for the OLS estimator for UK and Japan respectively. For both countries the AC(2) shows the presence of autocorrelation of the disturbance term. This reflects biases of the coefficients. Column (2) in tables 4 and 5 present the within estimator, which controls for fixed effects with a transformation to eliminate them. The parameters of these two estimators are expected to be the range to find a more efficient estimator with the GMM, or at least an estimator not very distant from either the OLS or within parameters. The methodology to choose the instruments for GMM in columns (3) to (6) follows Blundell et al If υ it in equation (8) is MA(1) (x t =Φ ν t 1 + ν t ), rather than serially uncorrelated, then only the values of untransformed regressors dated t 2 are valid instruments in the transformed equation for period t. If we treat the endogenous variables as predetermined, it would allow us the use of t 1 as an additional instrument. Column (3) presents the results under the first assumption where endogeneity of the explanatory variables (X0s) is considered as well as MA(1) of the disturbances (υ it ). Therefore the instruments available for the endogenous explanatory variables are t 2 and further lags. In practice very remote lags are unlikely to be informative instruments and in addition can result in overfitting bias. Some studies have shown that the loss of relevant information caused by omitting the more distant lags as instruments will often be very modest (see Bond working paper 09/02). To avoid the overfitting bias only the lag t 2 was considered for the model in column (3). Both countries reject the Sargan test, whose null hypothesis is instrument validity. One likely reason for this result is that the endogenous explanatory variables are predetermined with respect to the disturbance (υ it ). Therefore the lag t 1 of the endogenous variables are included as instruments, so they are used in the model instead of t 2. In this case there is the possibility of biases 14

15 due to correlation of the lag t 1andthefirst differenced error-term 4υ it. If the estimate of β for each of the endogenous variables decreases, a downward bias of the coefficient is implicated. The possibility of bias was investigated in two different ways. Each of the endogenous variables were instrumented independently with t 1 while the remainder were instrumented with t 2. The coefficients obtained behave with thesametendencytoincreaseordecreaseforeachofthevariablesaswhenall the endogenous variables were instrumented with t 1 at once. The results of the second approach are presented in column (4). The results show that for UK, there was a downward bias in the coefficients for dividends and profits. For Japan, the downward bias was observed for R&D, and investment. In the presence of measurement error neither t 2 nort 1are available instruments. Column (5) present the results with the exclusion of t 1 and t 2 as instruments for the specific variables where the measurement error was detected. Therefore these variables were instrumented by the lag t 3, and those variables which did not register a downward bias kept instrumented by t 2. For UK, the Sargan test rejected the validity of instruments for column (4) and (5), so it did for Japan. This results can be caused to different factors. The misspecification of the model is a likely cause for two reasons, first, the difficulty in judging for the endogeneity of the explanatory variables; second, the omission of an important variable in the empirical model. Another cause of failure in the GMM estimator is the possibility of weakness in the instrumental variables. In an attempt to treat the misspecification of the model, a further approach was implemented following the Arellano and Bover (1995) suggestion of using lagged differences as instruments. This method is know as the GMM-system. The results do not show any improvement in the validity of the instruments, so this results are not presented in the tables. A final attempt to find valid instruments was completed with the exclusion of all the lags t 2, so considering as valid lags t 3 or earlier (column (6)). A general improvement of the Sargan estimator is registered in both countries, but the Sargan test still suggests the existence of invalid instruments. Since for both countries, the Sargan test is improved in the last specification in column (6), it is considered the most trustworthy specification of the static model, however the interpretation of the results should be cautious due to the sensitivity of the data when the GMM is applied. Due to this conditions, the Within estimator remains to be the most reliable way to pursue an analysis of 15

16 the impact of intangible capital on corporate value in an static model. R&D reflects the amount of knowledge a firm has accumulated, so the expectation for this knowledge is to have a value additive force, as the development of new and successful products might be achieved. From column (2) in tables 4 and 5, it can be observed that the coefficient of R&D has a highly positive and significant influence on determining value for UK, but surprisingly it is not significant for Japan. Furthermore, dividends showed to have a positive and significant influence on companies value, so this result supports the signalling theory, which argues that an stable dividend policy might attract investors. Table 4. Estimators in OLS, Within and GMM for UK UK (930 firms, 8670 observations) (1) (2) (3) (4) (5) (6) OLS estimator Within estimator constant (.000) (.000) (.000) (.000) (.000) RD (.000) (.033) (.604) (.547) (.475) (.210) I (.752) (.580) (.000) (.821) (.124) (.220) D (.000) (.000) (.034) (.032) (.048) (.056) TL (.002) (.001) (.693) (.959) (.991) (.624) P (.067) (.064) (.784) (.157) (.750) (.780) lnassets (.000) (.000) (.404) (.442) (.516) (.015) R square Wald (joint) (.000) (.000) (.000) (.011) (.148) (.000) Sargan (.012) (.000) (.001) (.010) AC(1) (.000) (.472) (.000) (.001) (.000) (.000) AC(2) (.000) (.013) (.425) (.484) (.396) (.411) instruments for GMM levels t - 3 levels t - 2 levels t - 1 levels t - 2 for RD, I, TL, lnassets, t - 3 for P, D. Notes: 1. P values in parenthesis 2. Sargan test for overidentifying instrumental variables 3. All the models include time dummies 4. Results for GMM from the 2- step estimation 5. Wald test for joint significance of all the explanatory variables except for dummies 16

17 Table 5. Estimators in OLS, Within and GMM for Japan Japan (1577 firms, observations) (1) (2) (3) (4) (5) (6) OLS estimator Within estimator constant (.000) (.000) (.000) (.000) (.000) RD (.000) (.197) (.001) (.849) (.012) (.017) I (.864) (.668) (.008) (.003) (.001) (.108) D (.000) (.000) (.380) (.142) (.428) (.141) TL (.000) (.000) (.001) (.000) (.001) (.000) P (.001) (.000) (.222) (.008) (.246) (.929) lnassets (.000) (.000) (.000) (.000) (.000) (.000) R square Wald (joint) (.000) (.000) (.000) (.000) (.000) (.000) Sargan (.000) (.000) (.000) (.000) AC(1) (.000) (.000) (.000) (.000) (.000) (.000) AC(2) (.000) (.065) (.107) (.097) (.105) (.121) instruments for GMM levels t - 2 levels t - 1 levels t - 2 for P, D, TL, lnassets, t - 3 for RD, I Notes: 1. P values in parenthesis 2. Sargan test for overidentifying instrumental variables 3. All the models include time dummies 4. Results for GMM from the 2- step estimation 5. Wald test for joint significance of all the explanatory variables except for dummies levels t - 3. Moreover, the effect of physical investment was not significant for both countries. The behaviour of this variable in the period of analysis is presented in figure 2. Firms tended either to keep constant or decrease the physical investment, contrary to the investment on R&D, which has a general tendency to increase. This opposite behaviour of the investment in knowledge capital via R&D instead of physical investment shows that in the last decade it has been an exchange of resources which designate the money to knowledge rather than tangible capital. 17

18 Figure 2 RD stock / Total Assets Investment / Total Assets Japan UK Japan UK This is not a new finding, however it reinforces the idea that R&D should be considered as capital and not as an expenditure as suggested in accounting systems. Part of the failure of financial statements to fully reflect the real value of the companies is that accounting standards have not been updated with the technological advances of industries, but mainly in changes of operating practices. Dynamic specification Thenextresultsarebasedonthesamemodelbut with the inclusion of adjustment costs via a dynamic specification. Adjusment coststheoryimpliesthattheestimateddynamicsinfluence the dependent variable in the long term. A lag of the dependent variable was introduced to analyse the corresponding influence on the firms value. The same criteria as for the static model was pursued. Tables 6 and 7 present the results of the dynamic model for UK and Japan, respectively. The results are similar to those from the static specification. The null hypothesis for the validity of instruments is not rejected, at the 99% of confidence level, only for UK in the last specification of the model, where the third lag (t 3) was utilized to instrument the variables (column (6)). 18

19 Table 6. Estimators in OLS, Within and GMM for UK (dynamic) UK (1) (2) (3) (4) (5) (6) OLS estimator Within estimator constant (.000) (.000) (.759) (.051) (.058) TQ t (.000) (.382) (.798) (.000) (.766) (.716) RD t (.000) (.008) (.880) (.118) (.094) (.169) I t (.068) (.330) (.000) (.108) (.000) (.542) D t (.000) (.000) (.057) (.051) (.096) (.019) TL t (.000) (.001) (.136) (.539) (.505) (.486) P t E-01 (.076) (.047) (.534) (.866) (.778) (.802) lnassets t (.000) (.001) (.913) (.924) (.891) (.420) R square Wald (joint) (.000) (.000) (.000) (.000) (.000) (.000) Sargan (.000) (.000) (.000) (.036) AC(1) (.033) (.004) (.001) (.976) (.003) (.004) AC(2) (.000) (.549) (.462) (.045) (.292) (.200) instruments levels: t - 2 levels: t - 1 levels: t - 3 for TQ, RD, D, t - 2 for I, TL, P, levels : t - 3 Notes: See notes in table 4 lnassets 19

20 Table 7. Estimators in OLS, Within and GMM for Japan (dynamic) Japan (1) (2) (3) (4) (5) (6) OLS estimator Whitin estimator constant (.000) (.028) (.000) (.005) (.056) TQ t (.000) (.001) (.000) (.000) (.000) (.205) RD t (.000) (.167) (.067) (.068) (.104) (.278) I t (.000) (.625) (.000) (.608) (.002) (.115) D t (.000) (.000) (.770) (.002) (.374) (.387) TL t (.000) (.000) (.083) (.590) (.111) (.403) P t (.130) (.009) (.001) (.227) (.002) (.033) lnassets t (.000) (.000) (.000) (.000) (.000) (.000) R square Wald (joint) (.000) (.000) (.000) (.000) (.000) (.000) Sargan (.000) (.000) (.000) (.000) AC(1) (.017) (.008) (.000) (.012) (.000) (.000) AC(2) (.000) (.010) (.206) (.525) (.153) (.497) instruments t - 2 t - 1 t - 3 D, P, lnassets, TL, t - 2 TQ, RD, I t - 3 Notes: See notes in table 5 As presented in tables 6 and 7, the dynamics of the model show interesting features. The standard errors of the estimators increased considerably in comparison with the OLS and Within estimators. The Within estimator is not valid in this case as the assumption of exogeneity cannot be considered because the inclusion of the lag of Tobin s Q which is an endogenous variable. The GMM estimator shows a very low precision of the estimators and a contradictory effect is seen for most variables. This behaviour suggests that perhaps the GMM estimator is not the most reliable way to analyse this type of model where misspecification can be present. GMM showed to be a very sensitive estimation which could lead to biased estimators. 20

21 Generally, GMM does not present consistent estimators when the set of instrumental variables is exchanged. The imprecision of these parameters is a disappointing feature, since this alternative was expected to improve the estimators for this empirical model. The comparison of the dynamic and static specification with the Within estimator showed that the parameters behaved in the same way, both the standard errors and the coefficients present the same tendencies. By knowing that the Within estimator might be bias in a model with endogenous variables, the similarity of the dynamic (with a definite endogenous variable: Tobin s Q) and the static specification could not be a favourable finding, as it suggests that the estimators from the static model with the Within estimator might be biased. 5 Impact of spillovers on Intangible Capital Spillovers in this section refer to technological change. Specifically, the technological developments created for another firm in the same sector. The belief is that if there is an innovation that will improve the performance of a certain firm, even if that innovation is patented, it could cause certain influence in the future behaviour of the other firms from the same industrial sector. There are different proxies that can be used to measure spillovers, as R&D expenditure, patent information and innovation surveys. For this research, R&D expenditure has been used. The construction of the spillovers variable was determined by the addition of all the R&D expenditure per industrial sector in a specific year minus the R&D expenditure of the firm itself. In this way, the technological level of the sector where the firm belongs is considered. Even though not all the firms from that sector could get advantages from those innovations, this measure reflects the technological opportunity and/or competition. The methodology for the analysis follows Blundell et al 1992 as in the sections before. The impact of R&D spillovers is expected to affect value in future periods. This expectation is considered as it makes sense that any technological development that happen in an industrial sector would take time to spread out among other corporations from the same sector, so the real significant impact would not appear in period t, but in the future. Due to this behaviour, the static specification of the model might not be appropriate. As a consequence, only the dynamic specification is analysed. Similar problems with the extra sensitivity of the GMM are presented. Non of the countries passes the Sargan test for validity of instruments, so again the 21

22 GMM estimators are not very trustworthy in this model. Table 8 presents the results for the dynamic specification for both countries, it includes the OLS, the Within and the GMM estimator under the same specification as in column (5) from the tables above. To analyse the estimators would not be correct as it is a fact that there is endogeneity in both OLS and Within estimators as explained above for the dynamic specification, and GMM shows overfitting bias. Hence the hypothesis of the impact of rivals on corporate value cannot be corroborated with this sample. Generally, the inclusion of spillovers does not present any evidence of affecting the value of the firm, or in any case of improving the precision of the overall model. An explanation for this outcome is that the possible effect of the technological level of the sector on particular firms comes in the long run so the lag for only one period of the variable is not sufficient to show this influence. 22

23 Table 8. Estimation results with spillovers for UK and Japan UK Japan (1) (2) (3) (1) (2) (3) Whitin OLS Whitin estimator estimator estimator OLS estimator constant (.000) (.101) (.001) TQ t (.000) (.784) (.441) (.000) (.001) (.110) RD t (.000) (.008) (.108) (.000) (.161) (.138) I t (.061) (.245) (.000) (.001) (.706) (.871) D t (.000) (.000) (.038) (.000) (.000) (.491) TL t (.000) (.001) (.946) (.000) (.000) (.196) P t E+00 (.175) (.042) (.980) (.210) (.011) (.048) lnassets t (.000) (.002) (.535) (.001) (.000) (.000) R&D spillovers E (.569) (.313) (.761) (.000) (.621) (.051) R&D spillovers t (.814) (.065) (.275) (.000) (.010) (.127) R square Wald (joint) (.000) (.000) (.000) (.000) (.000) (.000) Sargan (.001) (.000) AC(1) (.031) (.004) (.003) (.011) (.006) (.001) AC(2) (.001) (.636) (.191) (.000) (.012) (.371) GMM instruments Notes: See notes in table 5 levels: t - 3 for TQ, RD, TL, R&Dspillovers, D, t - 2 for lnassets,i, P 6 Summary and Conclusions levels: t - 3 for TQ, RD, D, P, TL, R&Dspillovers, D, t - 2 for In this paper an empirical model of market valuation with firm level data was developed. The first objective of the model was to investigate the importance 23

24 of intangible capital in the determination of market value via the use of Tobin s Q ratio (TQ). Adjustment costs of Tobin s Q were included via a dynamic specification. Finally, for further validation a supplementary model specification which included spillovers of R&D was tested. The spillovers were calculated for each industrial sector, this variable attempted to measure the impact of rivals technological level on corporate value. A second objective in this research was to compare different estimators which are relevant in a linear model. The techniques applied were OLS, Within estimator and GMM. The ideal parameters were expected to be found with the GMM in first differences, as they correct for the presence of correlated firm-specific effects as well as the biases originated from the endogeneity of explanatory variables with the error term. According to theory (Bond 2002), the expected GMM estimators should be within the range suggested by OLS and the Within estimator, which are likely to be biased in opposite directions. An important feature of GMM is the application of an adequate matrix of instrumental variables, as they may be subject to large finite sample biases when the instruments available are weak. In this investigation, instrumental variables which followed both econometric and financial theory, were selected. The data employed for the analysis corresponds to two different countries: UK and Japan. The unbalanced panels contain a maximum per year of 930 and 1577 firms for UK and Japan, respectively. The period of analysis was of 12 years ( ). The results demonstrated that GMM was not consistent between countries with the selected instruments. The overfitting bias was a common problem for the empirical model. The standard errors of the estimators with GMM tended to be larger in UK than in Japan. Generally, for the empirical model in this research, the low precision of the GMM directed the analysis to be based on the Within estimator, which showed to be more reliable. However, this application would be valid only for the static model where it may be assumed that variables are not endogenous. The econometric results presented a positive and significant relationship of R&D on TQ in UK,a finding which is consistent with other empirical work in this area (see Hirschey 1982, Hall 1993, Megna and Klock 1993, 2000). However, for Japan, the Within estimator presented high standard errors in R&D. A possible explanation is the difference on approaches from both governments to promote the R&D expenditure. In UK the capital expenditure for R&D qualifies for a 100% first year allowance. In 1996, the government of UK funded around 32% of gross national expenditure on R&D (37.2% of this expenditure was in 24

25 defence). In Japan a tax credit of 20% is given for the excess of the R&D over those from previous years. Moreover, in the period of analysis, physical investment in productive capital (I) showed a tendency to decrease in both countries. However, its effect on market value was not significant in either country. Dividends paid were positive and significant for both countries. Dividend policy appears to have strongly predictable components with firms gradually adjusting dividends to target levels that reflect earnings (see Desai et al 2002). This effect might also be explained with the choice of managers to use dividends to send credible profitability signals to the market. Managers can also manipulate investment projects with the purpose of obtaining loans from bondholders that increases the wealth of the firm, which is known as the agency theory of debt. This behaviour could be suggested from the model, which showed a positive and significant influence of leverage on firms value. In the dynamic specification, to assume consistency of the Within estimator is not adequate since the correlation of the lag variables is implicit, so the application of GMM estimators could be more reasonable. However, in comparison with the static model, coefficients from the Within estimator behaved constant, which could suggest that perhaps there is also some kind of endogeneity in the static model, so to assume the contrary is risky. Generally, results from the GMM estimator should be treated cautiously, as the tests for instrument validity were not very favourable and showed to be very sensitive to small variations in the selection of instrumental variables. The comparison between countries from the static specification was carried out with the results from the Within estimator showed to be more reliable and consistent, but not optimal. Finally, the extension of the model with the integration of the R&D spillovers from rivals was treated only with the dynamic specification, as the expectation about the impact of this variable on corporate value is in future periods. The results for both countries illustrated that a decisive empirical argument can not be established from the data sample used for this research. 25