You are on page 1of 15

Applied Economics, 2010, 42, 1941–1953

On the profitability of innovative assets
Dirk Czarnitzkia,c,d,* and Kornelius Kraftb,d
Department of Managerial Economics, Strategy and Innovation, K.U. Leuven, Naamsestraat 69, 3000 Leuven, Belgium b Department of Economics, University of Dortmund, Vogelpothsweg 87, 44227 Dortmund, Germany c Steunpunt O&O Indicatoren, K. U. Leuven, Leumen, Belgium d Centre for European Economic Research (ZEW), Mannheim, Germany
a

Successful innovative activity is a major contribution to the intangible capital of firms. Although its importance is generally acknowledged, the contribution to companies’ profits is a priori unclear. We present the results of an empirical study on the effects of the patent stock on profitability. The database is a representative sample of German manufacturing firms and we use a number of control variables including measures of competition and firm governance. It turns out that the patent stock has a strong and robust effect on profitability.

I. Introduction Innovation is generally considered as a major cause of economic growth and is one important source (along with human capital) of the wealth of the developed countries (see, e.g. the survey by Fagerberg, 1994). Although, a significant part of the investment for science and technology is undertaken by governmental agencies, like universities, research institutions or through subsidy grants, the major part is privately financed. Firms invest into Research and Development (R&D) in order to maximize their individual profits, but not necessarily economic welfare. However, investments in R&D are examples for very high risks of failure of the pursued projects. A large part of all outlays may have no return at all and whether the innovative activity has a positive return on average is a matter that is not clear at the outset. It is a trivial statement that every investment is risky. However, R&D projects have some peculiarities that render such activities particularly uncertain.

R&D is by its definition concerned with a priori unknown outcomes, and therefore the risk of failure is higher than in the case of conventional investment processes (see, e.g. Hall (2002), for a survey on financing difficulties due to uncertain outcomes). Another feature of innovative activity are the long lags between R&D and the introduction of new products and processes (if they are successful at all). Finally, R&D is a classic example of spillover effects to other firms and therefore imitation is frequently an issue, which clearly reduces the profitability of successful R&D projects (see, e.g. Arrow, 1962). Although, imitation is not costless, it is clearly cheaper than original R&D and therefore imitation might well be a profit maximizing alternative to innovative activity.1 At least there seems to be some kind of international division of labour which leads to some countries innovating and others imitating. This reflects, on the one hand, available resources in particular human capital and, on the other hand, production costs namely wages. The risk of

*Corresponding author. E-mail: dirk.czarnitzki@econ.kuleuven.be 1 Very few studies are able to compute the costs of imitation. Mansfield et al. (1981) estimate that imitation amounts to about 65% of the costs of innovation.
Applied Economics ISSN 0003–6846 print/ISSN 1466–4283 online ß 2010 Taylor & Francis http://www.informaworld.com DOI: 10.1080/00036840701749019 1941

as stated at the beginning. if any. Apparently. Only a part of all firms do actually perform R&D.3 Imitation may take place on the national level. Innovation is measured as the patent stock being the depreciated sum of all past patent applications at the individual firm level.g. Section IV discusses the regression results and the final section concludes. 3 This has already been pointed out by Arrow (1962). Hence. 2004). Kraft issue whether the managerial-led firm shows the same behaviour as the owner-led company. The remainder of this article is organized as follows: Section II discusses the literature on profitability and innovation. Using the patent stock as innovation indicator avoids complicated lag structures of past values of patents. they are much smaller than the comparable firm of a typical dataset previously used for profitability studies. In contrast. If this is true.2 The share of R&D performing firms ranges from 14% in Iceland. Section III describes the database employed and presents descriptive statistics of the variables used. which have not had the large expenditures on research and can thus produce at lower costs. It is also questionable. we use some control variables that are usually neglected. it is sometimes argued that the larger Countries that participated in the CIS survey where the 15 EU member states in 2001 plus Iceland and Norway. The median value of employment is 67 and therefore. This is a major difference to nearly all existing studies. It is a long discussed 2 D. we report the results of an empirical study on the effects of innovation on firms’ profits. Moreover. R&D projects may frequently fail and spillovers to competitors may hinder the appropriability of returns. The relationship between innovation and profits is a priori unclear and it is by no means clear that innovative activities really lead to higher returns at the microeconomic level. In this article. it is not clear that innovation has actually a positive impact on a firm’s profits. aside of standard measures. if the projects are unsuccessful. Furthermore. as an imitating firm may choose a variant of the innovative technology.1942 imitation is reflected in many models of the new growth literature (see. In contrast. whether patents effectively prevent imitation. growth of a country’s Gross Domestic Product (GDP) and finally higher income of the population. These firms are clearly not representative for an economy as they are on average quite large. which are based on data from stock corporations. knowledge on innovation may leak out to competitors. Despite the belief that R&D is beneficial for the economy as a whole. The survey by Schleifer and Vishny (1997) discusses this question in detail (cf Gugler et al. In particular. Barro and Sala-i-Martin. R&D is clearly a costly and risky process as several projects might fail and the return. the social and the private return may differ substantially. in fact it is only a minority. it is unclear whether innovative activity is beneficial to the individual firm. the economy usually benefits from innovation as this is one major source for productivity increases. e. it is not clear how reliable this information is. we use a representative sample of firms from which the majority is not required to publish accounting data. This may explain the empirical observation that the majority of firms in industrialized countries chooses not to conduct own internal R&D (Table 1). will be realized with a long delay. On the Profitability of Innovation In the public discussion concerning the competitiveness of an economy figures on innovative activity like patent statistics play a prominent role. Thus. Czarnitzki and K. II. the better it is for the society without taking into account the possibility of wasted resources. Greece and Spain to 41% in Germany and 44% in Belgium.4 However. 2002. Aghion and Howitt. for a recent empirical study). firms do take account of this realistic possibility. Countries not reported in Table 1 did not publish the share of R&D-performing firms.. the other companies think it is not worthwhile to perform it and produce established products by use of a ‘well-known’ technology. 4 Chen and Yang (2005) provide a recent empirical study on the effects of spillovers. . as they are required to publish their balance sheets. 1998. Usually it is assumed that the more innovations are realized. the patent stock reflects previous R&D investment of a firm and approximates its knowledge capital. We use the Mannheim Innovation Panel (MIP) which contains information on a representative sample of German manufacturing firms with more than five employees. Second. Thus. but at least of a potential importance in order to explain profitability. and guides our empirical model specification. This study extends previous research in at least two respects: First. Many authors argue that managers favour growth in comparison to profit maximization. we consider capital ownership of the top management. Table 1 shows the share of R&D-performers in European countries derived from the Third Community Innovation Survey (CIS). profitability would be lower in such firms.

Another example concerning the interaction between innovation. Notes: aAn innovating firm is defined according to the Oslo-Manual (see Eurostat and OECD. There exist not many studies which explicitly consider the lag structure between R&D expenditures and profits or sales at the firm level. It might well be the case that the patent protection works on the national level but to much lesser degree internationally. Grossman and Helpman (1991). however. In brief. Iceland and Norway. imitation and growth is presented by Afonso (2006). Acemoglu et al. 1993). if successful. problems with imitating arise through international imitation. Hence. Hence. An exception is Ravenscraft . (2005) discuss the role of spillovers in the competitive process and their effect on the relation between leaders and followers as well as on the impact of competitive pressure on R&D. Segerstom et al. The other firms conduct R&D which. The value of the R&D input is assured by a system of intellectual property rights that protect a leading firm’s monopoly profits if the innovator is successful. (2006) as well as Dinopoulos and Segerstrom (2006). one problem is the number of lags between R&D expenditures and the effect on profits. The importance of innovations for welfare and also the relevance of spillovers are in particular emphasized in the new growth literature. As said above. The duration in turn depends on the probability that R&D is successful. but also a major concern of policy makers in all industrial countries. The starting point of any innovation is the value of the preceding innovation and this they call intertemporal spillovers. which leads with a Poisson arrival rate to successful innovation. There seems to be some ‘division of labour’ between the highly developed and other countries. consider differences between countries concerning innovation and imitation behaviour. At the same time. International intellectual property rights protection seems to be quite problematic in many countries. 1997. Luxembourg. which determines the return to innovative activity and therewith also the extent of R&D. Aghion et al. b Firm engaged in intramural R&D activity as defined by the Frascati Manual (OECD. He considers production of intermediate and final goods where technical progress is the result of an R&D process. the interest rate and on the duration of the technological leadership of the current monopolist. Innovation and R&D activity in European countries: manufacturing sector in 2000 Belgium (%) Denmark (%) Finland (%) 59 Share of innovators Share of innovators with own 74 internal R&D activityb ¼Share of R&D performers 44 a 1943 France (%) 46 66 30 Germany (%) Greece (%) 67 61 41 27 53 14 53 71 38 49 81 40 Iceland (%) Italy (%) Share of innovatorsa 54 Share of innovators with own 26 internal R&D activityb ¼Share of R&D performers 14 40 37 15 The Netherlands (%) Norway (%) Portugal (%) Spain (%) 55 61 34 39 62 24 45 39 18 38 38 14 Source: European Commission (2004). for the exact definition). some countries are better suited for innovation because of human capital advantages and other specialize in imitation because production costs are lower. This question is certainly not only of academic relevance.On the profitability of innovative assets Table 1. Hence. Innovators benefit from other earlier innovations in that any innovation raises their productivity if they develop a new innovation. the innovations affect the productivity level of the production process. the acquired technological knowledge disseminates to other firms and leads to better innovation capacity there. or has ongoing activities to do so. A prominent contribution is Aghion and Howitt (1992) who consider the invention of a new intermediate good. whose use as input allows more efficient methods to be used in producing the consumption good. the return of an innovation depends on the profits at each time. Hence it is an interactive feed-back process between innovative R&D and imitation. The question of the economic return of innovative activity can only be answered empirically. (1990). a firm is regarded as an innovator if it has introduced at least one new or improved product or new process within the last 3 years. destroys the current leading-edge technology. Innovation in Europe – Results for the EU. the importance of innovation for economic growth is undisputed and therefore it is indispensable for an economy to make sure that sufficient incentives exist for innovation at the firm level. However. Connolly (2003) provides empirical evidence on the impact of trade on imitation (as well as on innovation and growth).

8 Grabowksi and Mueller (1978) as well as Conolly and Hirschey (1984) propose another variable describing imperfect competition. This variable is sometimes called excess return on sales and expresses the following:  S À labour cost À capital cost À material cost ¼ S S with  denoting profits and S being sales. because from a theoretical perspective there is a close relationship between the market share and the price-cost margin. There are numerous studies that follow the same methodology. like Geroski et al. The firms participating in the survey report the typical time lag between the beginning of the development and introduction of the resulting new product. The empirical results of Ravenscraft and Scherer point to a mean lag of 4–6 years. 1997. which will possibly have to leave the market. equal the Lerner index. Efficiency advantages will in the long run also lead to a high concentration ratio as the more efficient firms will grow faster. a positive coefficient of the concentration variable is not sufficient for the conclusion that collusion is at work. a tendency to cartelistic behaviour exists. there are incentives to deviate from such collusive settlements and as R&D is especially difficult to coordinate among firms. but not because of collusion. concentration is high in a particular market.. where the capital costs have been subtracted and need not to be taken into account by capital divided by sales as an explanatory variable as in 5 6 D. if some firms are more efficient than others. our measure is the price-cost margin. . Below we present results with the variable capital intensity defined as total capital/number of employees. The rival hypothesis is that in the presence of few competitors. but the first returns are realized in the next year after starting the project. Then a positive coefficient should be estimated. for a data description). Hence. As it is well known. If firms are in the long-run equilibrium and are operating in the range of their production functions with constant returns to scale. imitation problems are smaller and R&D is more valuable. Czarnitzki and K.1944 and Scherer (1982). They use PIMS5 data on individual business lines of 26 and 42 businesses (the number depends on the time periods available). The PS of firm i in period t is calculated by the perpetual inventory method with a constant depreciation rate as PSit ¼ ðl À ÞPSi.7 The coefficient of the market share estimated simultaneously with the effect of the concentration variable is also interpreted as a measure on firm efficiency.6 The price-cost margin is usually explained by concentration in the industry and the market share of the firm in question. 1990. they will receive a larger market share and at the same time will realize higher profits. We circumvent the problem of long lag structures by using a Patent Stock (PS). all firms should benefit from a high price if concentration implies collusion. The usual way to estimate price-cost margins was introduced by Collins and Preston (1969). The dependent variable is the profit margin. One can therefore write  pq À ACq p À MC ¼ ¼ S pq p with p being the price and q the quantity produced. Hall. See our discussion below. The market share (SHARE) is included here as well. for more detailed descriptions). among others. but is interpreted as a measure concerning barriers to entry. the excess profit return on sales will. Hence. 1993) additionally consider the interaction variable concentration times market share. We use the Herfindahl concentration index (HERF) in order to take account of imperfect competition. Then a negative coefficient of the interaction variable is expected. for example. it may undermine a tendency towards shared monopoly behaviour. The idea is that in highly concentrated industries. tÀ1 þ PAit where PA is the number of Patent Applications in year t and  is the constant depreciation rate that is set to 15% (see Griliches and Mairesse. where RD is the firm-specific R&D PIMS means ‘Profit Impacts of Market Strategy’ (see Schoeffler.g. We use the variable RD*HERF. 8 Cf Uri (1988) for the inclusion of efficiency variables in explaining profitability. 1984. The coefficient of this variable is accordingly not standing for capital costs. With constant returns to sale Marginal costs (MC) are equal to Average Costs (AC). If. For 45% of all companies this is only 1–2 years. Some studies (e. Thus. However. have larger market shares than the others. They suggest the use of a R&D variable interacted with concentration (in addition to concentration and R&D). Kraft other empirical models considering the price-cost margin. a prisoner’s-dilemma situation arises where all firms will deviate from cartelistic behaviour. while 40% express that 2–5 years are needed and 5% reported of a time lag of more than 5 years. 7 Cf Cowling and Waterson (1976). on average across all products produced by the firm.

One important strand of the literature in industrial organization is on the kind of governance of firms.10 Several of the considered variables are potentially endogenously determined. we use lagged values of the exogenous variables whenever possible. Then a negative association with profitability would be estimated.On the profitability of innovative assets intensity (R&D expenditure divided by sales). R&D is an input to the innovation process. Hence. International trade is. We consider this question by the variable OWN. we abstain from this option. Additionally. 1979. This leads to a principal-agent problem with asymmetric information and it can be doubted that incentives are determined in a way that all problems are efficiently solved. More ‘conventional’ control variables are the share of sales exported (EXPORT) at the firm level and imports at the two-digit industry level j [¼ importsj/(importsj þ production of domestic firmsj)] (IMPORT). As we have no data to construct a panel database. The majority of the usually considered large stock companies is not led by the owner but by a manager. If the managerial firm holds also more patents aside of conducting more R&D. Sutton (1998. In order to avoid a possible simultaneous equation bias. as the firm in question is competing not only with domestic companies. In addition. As at least a part of these capital expenditures is sunk. In his model. 11 Another limitation is the cross-sectional character of the dataset. at least at the domestic level. this variable is expected to represent barriers to entry. The dummy variable EAST stands for firms located in Eastern Germany (the former GDR). many authors conclude that they will pursue growth maximization under the restriction of some expected minimum profit. which would bias the coefficient of the patent stock downwards. but in many cases R&D projects fail. who would be able to imitate at lower cost. 2006) hypothesizes and presents empirical evidence for the impact of R&D on concentration. barriers to entry are crucial in order to explain profitability.. Of course there are many other potential influences to profitability that should ideally be used as control variables. On the other hand. this is clearly a limitation of our study. for example. the dummy variable GROUP controls for synergy (dis)advantages. In our crosssectional framework. neglecting the governance question might produce an omitted variable bias. with respect to R&D we have only information from a limited number of firms so that lags would lead to a much smaller sample. we have no information concerning foreign direct investment of our firms. In a dynamic world. These variables are predetermined and can thus be applied in the given context. Other endogenous relations might as well exist. it is also not possible to include the effects of business cycles and uncertainty as it is done by Funk (2006). For example. Some other variables would be of interest as well. The knowledge derived from successful R&D projects might also spillover to competitors. We use the capital intensity (KAPINT) defined as fixed assets divided by the number of employees as a variable that indicates capital requirements.g. Patents are. that is. Ten industry dummies are included as well. If firms are members of a group of companies. Profitability might well be partly determined by management quality or other reasons for unobserved heterogeneity. we include the R&D intensity itself (RD). The R&D intensity must not necessarily have the same impact as patents. increasing competitive pressure. imports and exports supplement the measure of concentration which accounts for domestic competition with international competition. Cf Czarnitzki and Kraft (2004a) for this hypothesis and empirical evidence. size disadvantages have been estimated for Germany (Neumann et al. Therefore. on the 9 10 1945 one hand. STARTUP denotes that the firm in question has been founded during the recent 3 years. 1981) and therefore a negative coefficient is not implausible. Lin and Yeh (2005) point to the interaction between foreign direct investment and R&D. the new growth theory emphasizes the risk of imitation and spillover effects which may affect negatively the return in general and that of innovative activity in particular. This may have two effects: managers may invest more into R&D than owners9 and profitability may be lower at the same time. the percentage of capital ownership held by the top management. Profitability might e. affect the possibility to finance R&D efforts. effective in protecting intellectual property rights and therefore there might well exist a difference between their effect and that of R&D intensity on profits. the dummy variable FOREIGN identifies if the group is led by a foreign parent company. If managers are able to follow their own interests without effective control. However.11 There are also plausible interactions among our regressors. Jensen and Meckling (1976) as well as Jensen (1986) argue that managers choose to reinvest the free cash flow rather than return it to the owners. profitability of the innovative firm would also be lower. Size effects are considered by the number of employees (EMP). However. Thus. In contrast to results from other countries. .

The ownership information of firms is taken from the Creditreform database. One endogenous variable is R&D. the social returns were much larger than the private ones. the bias arising from the initial value of zero in 1980 is vanished over time due to the depreciation rate of the knowledge capital. because it should represent the discounted future profits of a firm. our sample is a cross section of manufacturing firms with five or more employees. (1977). The social returns range from negative values to a maximum of 307% with a median of 56%. the question regarding the return on sales has only been included in the wave concerning the year 2002. They compare this profitability impact with the social rate of return and found quite large figures for both. Creditreform is the largest German credit rating agency and makes its database available to the ZEW for scientific purposes. Descriptive Statistics and Econometric Method In order to receive a database including all the variables mentioned above. Databases. The MIP is an innovation survey conducted by the Centre for European Economic Research (ZEW) on behalf of the German Federal Ministry for Education and Research (BMBF) and is carried out annually since 1992. if it is profitable to do so and hence Sutton’s theory offers another reason for finding a positive impact of innovativeness on profitability. Geroski et al. In line with the market value studies. we had to link several sources. the estimated SEs could be inflated due to multicollinarity. However. He finds that R&D expenditures have a positive impact on profits. the gross rate of return of R&D is 27%. correlations among the explanatory variables should not affect our coefficients. This strategy is followed. (2005) for a recent article on the calculation citation weighted PSs in order to improve the approximation of the value of a firm’s knowledge capital.12 Uri (1988) develops a simultaneous equation model on profitability. Czarnitzki and K. They use data from a sample of firms that agreed to provide private data on the returns from innovation. . Jaffe (1986) estimates a three-equation model with patents.14 The MIP covers the whole manufacturing sector and is a random sample stratified by firm-size classes. industries and regions (Eastern and Western Germany). (1993) consider a sample of British firms and estimate the returns to innovations using a number of control variables. Looking at the correlations among the variables did not point to dramatic interactions. The patents were linked to the MIP by assignee names and addresses using a text field search. Although.13 The major disadvantage of this approach is its limitation to 12 13 D. See Hall (2000) for a survey on market value studies and Hall et al. Czarnitzki and Kraft found that different innovation indicators including the patent stock exhibit a positive impact on ratings. Earlier research One of the earliest studies on this question is Mansfield et al. Conolly and Hirschey (1984) use a sample of 390 ‘Fortune 500’ firms to estimate a simultaneous equation model with R&D intensity (R&D/Sales). Kraft publicly traded stock companies. and is hence negligible. Czarnitzki and Kraft (2004b) suggest using a credit rating as a proxy variable for market value. but would lead to highly selective samples in continental European countries where the vast majority of firms is privately owned. According to his results. though. 14 A few firms are actually smaller than five employees due to differences between the population database used for drawing the sample of the MIP and the firms’ response in the questionnaire. Thus.1946 concentration levels can be explained by exogenous and endogenous factors. An indirect way to evaluate the effect of innovation on profitability is to use the market value as dependent variable. The concentration index and the industry sales Other relevant studies include Pakes (1986) as well as Schankerman and Pakes (1986). R&D and market value as dependent variables. As our data concern the year 2002. They find a positive impact of R&D and a negative one of R&D intensity interacted with concentration. The information to construct the patent stock (PS) stems from the patent database of the German Patent and Trademark Office (DPMA) that includes all patent applications since 1980. market value in excess of book value of assets and concentration as endogenous variables. III. which is used to improve products and to raise the barriers to entry for present outsiders. because credit ratings are available for almost every firm and they should reflect the wealth and thus profitability of rated companies. Most firm level information is taken from the MIP. The initial PS in 1980 was set to zero for all firms. The seminal study in this strand of literature is Griliches (1981). They estimate private returns ranging from negative ones to 214% with a median of 25%. advertising intensity. concentration and advertising. This may still be suitable for the US or the UK. However.

Our initial sample of the MIP wave from 2002 contains 1649 observations on manufacturing firms (after removal of missing or inconsistent values in explanatory variables). we checked for a possible selection bias in the responses by grouping the sample into usable responses and observations with missing values. Some missing values in EXPORT and KAPINT where imputed by mean values defined by industry. First. 267 interviewees did not respond to the question on return on sales properly: 105 indicated category seven ‘do not know’ and 162 did not respond at all.15 The imports are only available at two-digit industry level and stem from the Structural Analysis (STAN) database of the Organization for Economic Development and Cooperation (OECD). IV. The corresponding p-value is 0. As in binary probit models the variance cannot be identified. between 2 and 4%. The return on sales variable was not surveyed as continuous variable because it was expected that the firms are very reluctant to provide information on their exact profit margin voluntarily. Distribution of the return on sales average delay of innovation returns. all other variables are lagged one period. For those who did not respond at all.On the profitability of innovative assets (in order to compute the market share variable) are taken from publications of the German Monopoly Commission and are based on the three-digit industry level. The Wald-statistic on joint significance of the coefficients amounts to 27. EAST.25 which is distributed chi-squared with 23 degrees of freedom. 1. firm size and EAST in order to lose not to many observations. into account. The median return is in the category 2. Usually such estimations include unknown threshold values identifying the cut-off points between the different categories. especially. however. It is remarkable that almost 19% of firms report a negative return on sales for 2002. we used lagged values of our variables whenever possible. those are randomly distributed. FOREIGN) or definitely refer to a single event in the past (STARTUP). 1.16 However.2457 which leads to the conclusion that there are no systematic differences between respondents and nonrespondents and our residual sample with usable information on the return on sales is still random. We use the lagged PS per employee as regressor to reduce collinearity with other variables due to size effects. it was decided to survey the return on sales as categorial variable and classes as shown in Table 2 were defined. Descriptive statistics of all variables used (except 10 industry dummies) are presented in Table 3. on the background that most firms are usually not required to publish an annual report.17 Of course. Despite indicator variables that hardly change over time (industry dummies. GROUP. Then we estimated a probit model on this group variable using our explanatory variables described in Section II. this test only accounts for selection on observables. 17 We also carried out t-tests on mean differences for each explanatory variable that led to the same conclusion. that is. The only exception is RD which is not available in the data for the previous year. For this reason. In order to avoid a simultaneity bias. we consider the interviewees responding ‘do not know’ as neutral. The patent stock is lagged two periods to take the 15 16 1947 Table 2. We use the European standard classification called NACE. that is. Estimation Results We estimate Ordered Probit models to determine the return on sales of the PS along with the other regressors. as shown by Ravenscraft and Scherer (1982). Surveyed categories of the return on sales Return on Return on Return sales (%) Class sales (%) Class on sales 50 (0–2) (2–4) 0 1 2 (4–7) (7–10) (10–15) 3 4 5 Class 415% 6 do not know 7 Fig. The frequencies of our valid responses are displayed in Fig. .

1948 Table 3. hence.013 0.027 0. In addition to the estimations for different subsamples. . Capital intensity has a positive and significant effect.001 0 0. The LR tests rejected homoscedasticity in both samples.334 0. mean of the PS among innovating firms) will ceteris paribus realize a profit margin that is 0.213 0.001 0 0 0 0 0. this result is obtained by an Ordered Probit estimation with known threshold values including a constant term only. only the subsample of Western German firms. The regression results for the full sample are presented in Table 4.237 41.38% for the capital stock and is.403 1 1 1 0. the estimated coefficients directly identify the marginal effects like in a linear regression model and unlike in probit models with unknown thresholds where the estimated parameters are always scaled by the unidentified variance.721 0.783 0.345 96. Hence. Descriptive statistics (1382 observations) Variable RETURN ON SALES PATENT STOCK/EMPa IMPORTa HERFa SHARE (in%)a HERF*SHAREa RD*HERFa EMP/1000a RD EXPORTa EAST STARTUP KAPINT GROUP FOREIGN OWNa.789 515.390 0. Hence.98%. and repeat them for the subsample where OWN is available.065 3.574 0.e. Most firms were newly founded since then and are therefore younger than Western German companies. For example.419 Min 0 0 0. We interpret this result as evidence on barriers to entry. pp.540 143. 192–4) does also present a good example for an Orderd Probit model with known threshold values. This figure amounts to 0. the average profit margin in the sample of Western German firms amounts to 3.243 0. we consider the full sample.247 0.301 SD 1.348 68. the relationship between the return on sales and its covariates appears to be stronger in the Western German sample than in the sample including Eastern German firms.780 1. on average. competition indicators like sellers’ concentration may play a less significant role in Eastern Germany and as many firms in this region of Germany are still struggling to survive in the market economy. like in the case of the PS mentioned above. approximately half the size of the PS variable.307 0. and therefore the interpretation of results focuses on the heteroscedastic models. because we exactly know the threshold parameters from the questionnaire and can hence identify the variance. we can quantify the marginal effect of the explanatory variables exactly.b Mean 2. Eastern German firms are still highly subsidized in order to foster their catching-up 18 19 Verbeek (2000. too.095 0. As the variable OWN is only available for a subsample of 834 firms. even more important. we also calculate the effect of the capital stock per employee at the mean (again among innovating firms) versus the hypothetical situation with capital being equal to zero.423 0.227 0. The most important result is the positive and significant coefficient of the PS.238 2. Moreover.18 We present estimations on two different samples: First.056 0.807 65. Note that there is no need to consider the marginal effects on the probabilities that an observation enters a particular class. Given the threshold values. As expected.267 6.807 0.67 percentage points higher compared to a PS equal to zero. This yields two advantages: we have to estimate less parameters than in the case with unknown parameters and. We are only interested in the marginal effect in the underlying ‘true’ latent model (see the Appendix for an outline of the Ordered Probit model with known thresholds).033 0.367 46.034 0.002 0 0 0 Max 6 0. because the Eastern German economy is still in transition since the German unification in 1990.188 642.19 A Western German firm with an average innovation activity (i.463 3. we tested for heteroscedasticity using Likelihood Ratio (LR) tests. In this case of a categorial variable.000 0. Multiplicative groupwise heteroscedasticity was considered and modelled by a set of 10 industry dummies. b Only 834 observations available. Kraft process.611 16 590.104 0.472 0.488 0.923 1 1 1 D. and second. Czarnitzki and K.106 0. which have an impact on profitability.909 2. there is some effect of innovative activity on profits and an incentive for innovation exists. the relationship between profits and the considered indicators might be less informative than in Western Germany where the industry structure has evolved in a framework of a market economy since the Second World War. In order to compare the impact of capital and patents. A more detailed judgement Note: aLagged values.754 0. Our situation is different in this case.178 0. we first run the estimations with the initial sample of 1382 observations omitting OWN. five size dummies based on the number of employees and the dummy variable EAST.

The innovation effect may already be captured by the PS indicating that this is a good indicator for valuable innovation activities. Those results are partly unexpected.033) (0.596) Heterosced.986 3. RD*HERF has a negative coefficient as estimated by Grabowski and Mueller (1978) as well as Conolly and Hirschey (1984). the profitability decreases with firm size in our sample. for imports and exports. though.067) (3.250** À0.057) (3.412 2.314) (0.532) (0.830) (1.052* EMP/1000b À0.712) Heterosced.038) (0.513) Western German firms Homosced.568) (0. modela 17.808 a (5.575*** 11. Hence if this is (approximately) true.072) (0. the Herfindahl index shows significant coefficients.399** EAST À0.622*** (0.067) (4.745*** À0. 5 and 10% significance levels.559) (0.100 0. The other control variables have no impact.341 À0.998 HERFb 0.554) (0.066* À0.030) (0. As SHARE is insignificant we find no evidence in favour of the efficiency theory of imperfect competition.371*** LR test on joint significance 32.701 1. model (4. ** and * denote 1.001) (0.433*** 1382 920 920 À2573. we need no longer lags and in this case R&D has no significant effect on profits aside of the impact on patents. R&D intensity does usually not change very much between different periods as adjustment costs seem to be considerable.24*** (4.820) (1. Apparently. In contrast to studies from other countries.323 0.847) (0. LR: Likelihood Ratio.406) (0.003) (0.118** 1. but find support for the collusion interpretation.815 0. EXPORT shows a significant positive coefficient in the full sample only.462) (0. Note that in the sample of Western German firms a test on joint significance of the 10 industry dummies does not reject the null hypothesis of all 10 .043 À0.421) (0.011*** 0.047 À0.604) (0.890) (0.001) (0. modela 12.004) (0.553) (0. However.111) (0. Finally. model 17. b Lagged values.024) 1.333) (0.246 À0.006 À1701.000 À0.768 4.24*** of 10 industry dummies ^ ln  1.700) (0.216 0.011*** 0.838) (0. for example.908** GROUP 0.On the profitability of innovative assets Table 4.151 0.001) (0. Among our competition variables.582*** 12. Firm size has a negative effect on the profit margin which is in line with the results by Neuman et al.117) (0.062 À0.723*** IMPORTb À0.315 Constant term 2. We have only data from 1 year on R&D expenditures and as there will be some lag between these and a possible return we cannot make an exact statement about the profitability of R&D.214*** À1. ***.247 STARTUP À1.500) (0.737** À0.657*** 28. 1981). support (although only weakly significant) from another country on the effect of R&D to undermine collusion. 1382 Log-likelihood À2594.139** 4.072 À0.346) (0. these increases in wages are not matched by productivity differences.066) (1.066** À0.412) (0.281 1.024) (0.093) 1.006** 0. Ordered Probit regressions – full sample Dependent variable: return on sales All firms Variable Homosced.79 (5.069 À0.006** SHAREb 0.125* RD 1.295 0.035 KAPINTb 2.100) (1.843) (1.316) (0.000 À0.44 1949 PATENT STOCK/EMPb 11.001) (0.029) 1.905) (0.394 2.716) (1.112) (0.138** 4. our result seems to be interesting. This might be caused by the higher wages that are usually paid in large firms.105) (0.570) (1.000 À0.034) (0. as international trade has frequently a strong impact.426*** (0.429 2.792 EXPORTb 1.058) (4.000 HERF*SHAREb RD*HERFb À0.776) (0. we find an overall positive effect of innovative activity for profitability.592) (0. The R&D intensity (RD) itself has no significant effect.831) (1.384 À1.439*** À0.556 1. on the efficiency of patents versus capital would require information on cost of the PS.363** À0.723 0. The interaction variable market SHARE*HERF is insignificant.003) (0. Heteroscedasticity term includes 10 industry dummies and five size dummies (and EAST in the full sample).616) (0.595 À1714.642*** # of obs.860) (0. As the two studies used American data. But R&D has also no significantly negative impact on profitability and as successful R&D leads to patents.117) Notes: SEs in parentheses. (1979.067 FOREIGN 0.691) (0.003) (0.051 2.487 (0.

180** 6.055) (0.425) (0.089** IMPORTb À0.865*** (0.468*** À0.525 (0.646*** 0.004) (0.002) (0.909) (1.091) (6.288 1.003* À0.039 Constant term 1.027 À0.051) (0. Unfortunately.522) (0.161) (0.916) (0. it is not possible to account for selection within the econometric model as we have no appropriate instruments at hand to model a selection equation.494) (1.048) (0.479* 0.573) (1.126) (0.163) (0.786) Heterosced.161 RD À3.956) (0.004) (0. it turns out that some selectivity is present.567*** 0.039 1.014*** 0.428) (1.432 12. modela 14. a Heteroscedasticity term includes five size dummies. we checked for selectivity in this sample. As OWN is taken from the Creditreform database we have several missing values where we did not find information on the ownership of the corresponding firm in the database.004* À0. The regression for the reduced sample including the firm governance variable OWN are presented in Table 5. Based on these results.757 0.526) (0.424 1. which points to the efficiency interpretation of reduced competition.646) (0.855 À1090.39 (7. It is therefore possible that both forces are at work here. We conclude that our various competition indicators capture the industry differences well and differences in the return on sales are to a large extent driven by innovation.481 EXPORTb 1.740** OWNb 1. and we defined one group for nonmissing values of OWN and the other group where OWN is missing. ***.777) D.031) 1.470 HERFb 0. now the market share as well as the interaction term HERF*SHARE are significant.005) (0.62 (7.101) (6.007* 0.263) (0. b Lagged values. Czarnitzki and K.768* KAPINTb 3.469) (0. The full sample was grouped according to the variable OWN.352) (0. Therefore. model 19.052) (2.616*** (0.386) (0.282 Notes: SEs in parentheses.270) (0.091) (8.249 6. The LR tests yield that homoscedasticity is rejected again.175) (0. but in this case it is sufficient to include the five size dummies in the heteroscedasticity term.057) (0.501) (1.916) PATENT STOCK/EMPb 14. 834 Log-likelihood À1546.005) (0.033 À0.003* RD*HERFb À0.099) (7.453) (0.587) (1. Kraft Western German firms Heterosced. but the strong impact of the PS is still present. We find that no information is available especially for younger firms and Eastern German firms. we omit the variable STARTUP from the regression (because there are only very few cases left in the sample) and point out that the regression including Eastern German firms should be interpreted with some care. The results based on the reduced sample show some differences to the ones from the initial sample.900*** À0.002 EMP/1000b À0. But our results are still different from those reported by Ravenscraft .390** À0. we estimated a Probit model on the group dummy and additionally carried out t-tests on mean differences for all explanatory variables.805*** 1.53*** (6.978) (0.011 2.603) (0.47** of 10 industry dummies ^ ln  1.092) 834 595 595 À1536.454*** À0.850** GROUP 0.112) (0.823*** (0. modela 18.140) (0.754* 3.429) (1.217 À0.696) (0.363 1.051*** 23.286 FOREIGN 0. Aside of the Herfindahl index.638*** # of obs.076) 1.165 À0.001) (0.244 6.155 0.533) (0. 5 and 10% significance levels. coefficients being 0. this result is robust to changes in specification and the sample size.001) (0.075** 1.701*** À0. Again.423 0.180* À3. Ordered Probit-reduced sample including OWN Dependent variable: return on sales All firms Variable Heterosced.663*** 1.180* 5.378 À0.003* À0.799*** LR test on joint significance 21.918) Heterosced.1950 Table 5. It is important to note that the profitability does not differ significantly between both groups and we therefore conclude that the selectivity problem is not too serious in the upcoming regression analysis.642) (0. Hence. Once again.737) (0.014*** 0.896) (0. ** and * denote 1. we find additionally to the impact of concentration also an effect of the market share.445 1.173) (0.031) (2.026) (1.165) (0.661 À1095.243 0.889 0. In this case.076) 1.483 12.036** À0.374 ** SHAREb HERF*SHAREb À0.781) (0.007* 0. but our variables describing the competitive structure have some other effects. model (6.001) (0.314 EAST À0.

Conclusions We present empirical results on the question whether innovative activity has a significant impact on the profits of firms. this 1. The PS has a strong and robust effect on profitability.67 percentage points higher return on sales than a firm not performing innovation activities. one could also test whether patents have a significant effect on V. Because of the positive externalities connected to the spillover effects of R&D. Moreover. as in his case the market share was the dominating variable and the concentration measure had no significant impact. there would be considerable room for political actions on international protection of intellectual property. the results may at least partially spillover to competitors. and at present we have only limited information on those.6 percentage points higher return on sales than a firm that is led by managers who do not hold any capital shares.On the profitability of innovative assets (1983). The Herfindahl concentration index has always a strong positive impact on profits. In contrast.6 percentage points represent a substantial difference in firm performance. we conclude that an incentive for innovation exists in Germany. at the same time. Our additional variable OWN has the expected positive impact. because usually data from large firms are used resulting in a selective sample of companies. capital intensity has a stable positive impact on profits. information concerning the profitability of innovation seems to be of considerable value for the involved firms and public authorities. which points to significant principal-agents problems in manager-led firms in Germany. the positive impact of R&D itself is balanced by the negative ones of wasted R&D and the imitation by rivals. our results point to the conclusion that owner-led firms have a significantly larger profit rate. The insignificance of the R&D coefficient might reflect the positive impact of innovativeness on profitability but at the same time R&D projects to a high-degree failure and. This is supported by our positive effect of the PS which measures successful R&D and. Owner-led firms have a higher profitability rate. Therefore. we have limited evidence in favour of the hypothesis that a high concentration is the result of efficiency advantages of the larger firms. An innovating firm realizes an about 0. The returns have to be compared with the costs. Innovation is specified as the PS and R&D intensity. Imitation is perhaps largely a problem between but not within countries. One would need a long times series on R&D expenditures as the input to the innovative process (and not just the data from 1 year as in the present study). All relevant variables are lagged in order to reduce endogeneity problems (except R&D intensity due to data limitations). As studies on the effect of governance are rare in Germany. The results with respect to the variables representing the competitive structure are mixed. This is a major difference to most other studies conducted in this field. We interpret this result as evidence on barriers to entry that have an impact on profitability. the R&D intensity has no separate. The market share and the interaction variable market share times the Herfindahl index both are only significant in a . This would be the ‘true’ test on the profit effects of innovative activity. Then the effects of the overall outlays for profits in later years have to be calculated. Although. The joint results concerning R&D and patents point to the relevance of the intellectual property rights system for an economy. on average. Provided this statement would be true. On the background of the distribution of profits in our sample. This points to inefficiency of the large firms without competitive pressure. Hence. Thus. A Western German firm that is led by its owners exhibits a 1. Hence. Profitability is in our case defined as profits divided by sales and is equivalent to the price-cost margin. the test on joint significance of the 10 industry dummies does not reject the null hypothesis in the Western German sample. There exist governance inefficiencies in German firms and the managers may require better incentives or closer supervision in order to solve the principal-agent problems. Our result that innovation generates some returns for the firms is just an estimate of the minimal level of the total benefits. We do not find a similar effect for R&D. Most likely this is due to the missing disciplinary effect of competition. we find evidence on the positive impact of innovation on profits. if they are successful. the economy has a considerable advantage from these activities and given the inherent uncertainty of R&D processes in general. 1951 subsample of companies for which information on firm governance is available. the social benefits are usually much larger than the private ones. we think this is a valuable result. The interaction variable is negatively significant. Again. In contrast. In such a study. this is only one part of the story. reflects intellectual property rights that prevent others from imitation to some extent. additional effect. Our representative sample of German manufacturing firms contains many small and medium-sized firms which are not required to publish their balance sheets. Capital intensity remains significant.

G. Barrell. Mansfield. 43. 339–74.. P. 2nd edn. Oxford Review of Economic Policy. Applied Economics. References Acemoglu. corporate finance. (2005) Competition and innovation: an inverted-U relationship. X.1952 profits aside of past R&D activities as. in The Rate and Direction of Inventive Activity: Economic and Social Factors (Ed. Cambridge. Cambridge. (1992) A model of growth through creative destruction. mimeo. on the one hand. Journal of Financial Economics. European Commission (2004) Innovation in Europe – Results for the EU. Mueller. 51. RAND Journal of Economics. RAND Journal of Economics. 1–24. Cambridge. Princeton University Press. 82. and Meckling. O’Mahony. 32. Czarnitzki and K. market structure and profits: a value-based approach. M. 91. Economics Letters. H. C. Economic Journal. 91. B. (2005) The interdependence between FDI and R&D: an application of an endogenous switching model to Taiwan’s electronics industry. (2006) A theory of north–south trade and globalisation. P. and van Reenen.. L. patents are expected to reduce (or even eliminate) imitation possibilities by competitors. Applied Economics. Cambridge. P. and market value. P. Applied Economics. 66. Conolly. J.) R. Chen. 907–18. (2006) Distance to frontier. Econometrica. and Howitt. H. Aghion. (2004a) Management control and innovative activity. Afonso. (2000) Innovation and market value. Jaffee. M. Economics Letters. C. and Howitt. Stockholm School of Economics and CEPR. Jensen. and Hirschey. Mansfield. 76. 24. K. R. (1986) Technological opportunity and spillovers of R&D: evidence from firms’ patents. Innovation. 984–1001. R. 18. and Kraft. Cowling. Journal of the European Economic Association. 9. E. and Yang. E. D. MA.) Z. Nelson. (1994) Technology and international differences in growth rates. 328–43. D. (2004b) Innovation indicators and corporate credit ratings: evidence from German firms. Iceland and Norway. 682–6. J.. R. J. A. Wagner. (2003) The dual nature of trade: measuring its impact on imitation and growth. Gugler. 305–60. (1986) Agency costs of free cash flow. on the other hand. Aghion. . Patents and Productivity (Ed. E. A. and Segerstrom. (2005) Technological knowledge. Bloom. (1977) Social and private return from industrial innovations. Czarnitzki. H. 3. H. Paris. Jurgen Moka for extracting the information on ¨ firm governance from the Creditreform database and the MIP-Team for providing the company data. NBER Working Paper Series No. and Zilibotti. Aghion. R. Connolly. 7. 2361–71. Bell Journal of Economics. G. Eurostat. Jaffe. (1976) Theory of the firm: managerial behavior. Grabowski. American Economic Review. Griffith. M. Blundell. M. 198–211. D. and Economic Performance (Eds) R. 1775–82. (1990) The manufacturing sector master file: 1959–1987. Geroski. K. Mason and M. 609–25. and takeovers. Princeton. S.. J. 323–9. B. 1147–75. (2002) The impact of corporate governance on investment returns in developed countries. pp. A. Cambridge. K. 323–52. Quarterly Journal of Economics. Z. 37. Review of Economics and Statistics. S. G. Czarnitzki. MIT Press.. Griliches. C. 113. B.. 31–55. Economic Journal.. Hall. H. patents are a measure of R&D output or success and. (2006) Skill-biased technological knowledge without scale effects. (2005) Market value and patent citations. 177–98. Acknowledgements The authors thank Thorsten Doherr for providing his search engine to link the patent data with firm-level data. (1981) Imitation costs and patents: an empirical study. M. (1969) Price-cost margins and industry structure. B. H. spillover and productivity: evidence from Taiwanese firm level panel data. 183–7. 72. Kraft Dinopoulos. (1976) Price-cost margins and market structure. (1978) Industrial research and development. Griliches. Journal of Economic Literature. K. 37. MA. 377–84. Review of Industrial Organization. European Commission. 38. Fagerberg. S. (1991) Innovation and Growth in the Global Economy. 76. and economic growth. and Howitt. R&D and patents. E. (1984) R&D. R. intangible capital stocks and firm profit rates. and Sala-i-Martin. and Wagner. and Beardsley. profits. P. and ownership structure. Jensen. 221–40. Quarterly Journal of Economics. 87–92. American Economic Review. R. 36.. G. Griliches. Aghion. Economica. J. M. (1981) Market value.. (2002) The financing of research and development. M. F511–39. in Productivity. 16–38. R. P. and Kraft. (1998) Endogeneous Growth Theory. O. MA. Hall. D. 13–21. Applied Economics. B. selection. Eurostat and OECD (1997) Oslo-Manual – Proposed Guidelines for Collecting and Interpreting Technological innovation Data. (1984) Productivity and R&D at the firm level. 60. agency costs. 701–28. University of Chicago Press. Schwartz. P. and Trajtenberg. B. Luxembourg. F.. H. 4. E. and Waterson. 38. pp. Barro. Machin. and Yurtoglu. (1962) Economic welfare and the allocation of resources for invention. Lin. Cambridge University Press. (2004) Economic Growth. 37–74. MA. Hall. 271–86. 24. 35–51. Funk. P. (2006) Business cylces and research investment. M. Hall. pp. in R&D. (1993) The profitability of innovating firms. and Yeh. 120. and Mairesse. MIT Press. S. J. N. M. Rapoport. P. Arrow. N. B. K. D. Review of Economics and Statistics. and Helpman. J. D. Collins. Z. Journal of Development Economics. NJ. 1789–99. Chicago. and Mueller. and Preston. A. Grossman. MIT Press. W. Romeo.. We are grateful to three referees and the editor for their helpful comments. 3366.

(2000) A Guide to Modern Econometrics. M. 27. Review of Economics and Statistics. Segerstrom. F. (1982) The lag structure of returns to research and development. Let the latent model be yà i ¼ x0i . pp. 755–84. OECD (Organization for Economic Cooperation and Development) (1993) Frascati Manual The measurement of scientific and technical activities standard practice for surveys of research and experimental development. Schoeffler. 1077–92. D. A. structure and performance. T. M. 20. MA. Ravenscraft. in Handbook of Industrial Organization (Eds) M. and Haid. Applied Economics. A. D. (1983) Structure–profit relationships at the line of business and industry level. 227–42. Thorelli. John Wiley & Sons. Pakes. 108–21. (1997) A survey of corporate governance. I. J.On the profitability of innovative assets Neumann. (1979) ¨ Profitability. Sutton. Cambridge. Anant. 14. Paris. Bloomington. Bobel. Neumann. Journal of Industrial Economics. risk and market structure in west German industries. Economic Journal. and Scherer. B. 1052–77. R. C.. A. (1990) A Schumpeterian model of product life cycle. American Economic Review. Journal of’ Economics/Zeitschriftfu¨r Nationalo¨konomie. S. D. N. and Dinopoulos. A. Applied Economics. (1981) ¨ Market structure and the labour market in West German industries – a contribution towards interpreting the structure-perfomance relationship. (2007) Market structure: theory and evidence.. Appendix The ordered probit model with known threshold values The estimations presented in the article are the results of ordered probit models. W. J. I. (1977) Cross-sectional study of strategy. 2301–68. P. and Vishny. 65. Amsterdam. M. and Haid. Uri. Verbeek. Ravenscraft. J. E. S. (1998) Technology and Market Structure. A. North Holland. 603–20. 1953 Schankerman. Porter. Indiana University Press. M. pp. J. 1383–400. 97–109. 54. 41. Schleifer. A. Econometrica. 22–31. MIT Press. Journal of Finance. 737–83. 52.) H. (1986) Patents as options: some estimates of the value of holding European patent stocks.. M. (1986) Estimates of the value of patent rights in European countries during the post-1950 period. 80. Bobel. and Pakes. Chichester. Sutton. in Strategy þ Structure ¼ Performance (Ed. 96. (1988) A re-examination of the relationship between industry structure and economic performance. Armstrong and R.

þ "i . . 5) are usually unknown threshold values which have to be estimated. . Assuming that the errors are normally distributed yields the following probabilities:   0 À x0i . . . N ð1Þ yà is the unobserved dependent variable (the profiti turnover ratio). . . xi the set of regressors and "i the error term. . . . The observed return on sales is 8 à 0 > 0 if yi > > 1 if  5 yà  > > 0 1 i < . > > > 5 if 4 5 yà 5 > i > : 6 if yà 4 5 i k (k ¼ 0. with i ¼ 1. ð2Þ yi ¼ .

Pðyi ¼ 0Þ ¼ È      1 À x0i .

0 À x0i .

    5 À x0i . Pðyi ¼ 1Þ ¼ È ÀÈ   . . . . . .

4 À x0i .

Pðyi ¼ 5Þ ¼ È ÀÈ     5 À x0i .

Recall that the profit variable has been categorized in the survey. Using the true threshold values allows us to identify the variance (and the constant term) and reduces the parameters to be estimated. Firm size is specified as five size classes categorized by ^ the number of employees. Usually the SD  is – as in binary choice Probit models – not identified. which are considered to model the heteroscedasticity. Although. Finally. LR tests do reject the hypothesis of homoscedasticity. the tests on heteroscedasticity allow the variance to vary over industries. firm size and EAST. The homoscedastic SD  is ^ replaced by i with i ¼ expð þ !0i Þ ð4Þ where denotes the vector of additional parameters to be estimated and !i are the variables. In this case. The coefficients can directly be interpreted as marginal effects in the ‘true’ latent model. where we know the threshold values k. All estimated coefficients are scaled by . Pðyi ¼ 6Þ ¼ 1 À È  ð3Þ The joint likelihood function of these probabilities can be estimated with the familiar Maximum Likelihood technique. the results concerning the PS and most control variables remain similar. . and we know the threshold values for each class. however. we are in a situation.

. or email articles for individual use. users may print. download. However.Copyright of Applied Economics is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission.

download. However. or email articles for individual use. .Copyright of Applied Economics is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. users may print.