Productivity and competition in a global economy* 1 Introduction

Transcription

1 1 Productivy and competion in a global economy* Joseph Plasmans Universies of Antwerp and Tilburg Workshop Rome December 14-15, Introduction This chapter essentially deals wh the interplay of productivy and competion, where aspects of globalisation are discussed. The productivy of a production un is defined as the ratio of a measure of output produced by this production un over a measure of input used during the same time period. If the input measure is comprehensive, then the productivy concept is called Total Factor Productivy (TFP) or Multi Factor Productivy (MFP). Since taking into account all the factors influencing output levels can be unrealistic, MFP may be a more appropriate term to use. Van Ark (2014, p. 3) specifies as follows: The term Total Factor Productivy is mostly used in the academic lerature, even though Multi Factor Productivy correctly recognizes that every productivy measure starts wh assumptions on which factor and non-factor inputs to include. If the input measure is (the number of) labour hours, then the productivy concept is called Labour Productivy (LP). Diewert (2006, p. 1) states that: A problem wh the Total Factor Productivy concept is that depends on the uns of measurement for outputs and inputs. Hence TFP can only be compared across production uns if the production uns are basically in the same line of business so that they are producing the same (or closely similar) outputs and using the same inputs. Therefore, we concentrate in this chapter on firms as production uns operating in the manufacturing market. Usually, MFP Growth (MFPG) can be decomposed into factors such as technical change, technical efficiency (efficient allocation of inputs to outputs), scale effects, input- and output-mix effects (deviations of perfect competion, higher production capabilies), and other components that may be related to productivy changes (uncertainty) (see e.g. Morrison, 1999; Balk, 2008).

2 2 MFPG can be calculated by different approaches, that is, by the growth accounting approach, the index number approach, the distance function approach, and the econometric approach (see e.g., Feng and Serletis, 2008, for an overview). Growth accounting was suggested by Solow (1957) as a method of estimating TFPG. Growth accounting calculation of TFPG requires the explic specification of a neo-classical production function and identifies TFPG as the output change that cannot be accounted for by the growth in those inputs that are explicly included in a specific production function. Under perfect competion and under a Cobb-Douglas production technology, TFPG is commonly called the Solow residual since Bates (2001) explains that the main determinants of output growth are the input growth and TFPG. TFPG (and here also MFPG) is output growth that cannot be directly explained by input growth. The index number approach is an extension of (and a complement to) growth accounting. It involves dividing a (real) output quanty index, or the ratio between two outputs at two different time periods in case of one single output, by an input quanty index to obtain a measure of MFPG. i However, as Feng and Serletis (2008, p. 283) assert: one crical issue regarding this approach is the selection of the appropriate indexes. In fact, statistical indexes are mainly characterized by their statistical properties. These properties were examined in great detail by Fisher (1922) and serve as tests in assessing the qualy of a particular statistical index. and The index that Fisher (1922) found to be the best, in the sense of possessing the largest number of desirable statistical properties, has now become known as the Fisher ideal index. Another index found to possess a very large number of such properties is the discrete time approximation to the continuous Divisia index, usually called the Törnqvist index or just the Divisia index (in discrete time). In fact, the primary advantage of the Fisher ideal index over the Divisia index is that the Fisher ideal index satisfies Fisher s factor reversal test -which requires that the product of the price and quanty indexes for an aggregated good should equal actual expendures on the component goods- while the discrete time approximation of the Divisia index fails that test. However, the magnude of the error is very small -third order in the changes.

3 3 Notice that the index number approach does not require an aggregate production function. The distance function approach to measuring MFPG separates MFPG in two components: changes resulting from a movement toward the production frontier (technical efficiency change) and shifts of this frontier (technical change). The distance function was first introduced separately by Shephard (1953) in the context of production function analysis and by Malmqvist (1953) in the context of consumption function analysis. But was introduced as a theoretical MFPG index by Caves et al. (1982), and then popularized as an empirical MFPG index by Färe et al. (1994). Although the latter empirical MFPG index has several advantages (e.g., neher a specific functional form, nor information on prices are required and does not assume that firms are operating at their efficient level), requires full information about the state of technology at every point in time and all production uns should have identical production functions, which is not realistic. Finally, the econometric approach to MFPG measurement involves estimating the parameters of an aggregator function, being a flexible cost, prof, or production function as the locally flexible generalized Leontief, the translog, and the normalized quadratic specifications. MFPG can then be expressed in terms of the estimated parameters. The advantage of this econometric approach is that we can now identify the various components of MFPG so that this method is more adequate for our purposes (e.g., imperfect competion and scale effects). In this paper we follow the econometric approach to MFPG. We develop a framework whin which we can analyse the link between competive behaviour, economies of scale and MFPG. Such a framework allows us to take into account the posive contribution of scale economies and market power. Furthermore, for relatively small open economies such as Belgium and the Netherlands, scale economies, competion, capal and labour growth are all interrelated. This is important for a society because is only on the basis of a consistent measure of productivy growth and competion that policy-makers can make the right decisions in order to improve or maintain the same level of welfare at all levels of the economy. The analysis and measurement of productivy performance has attracted a great deal of attention ever since Solow (1957) decomposed the growth in output into the growth of inputs and a residual-based productivy term. ii In a series of papers Hall (1986, 1988, and 1990) stresses that the Solow residual is no longer equal to the rate of technical change when there is imperfect competion in product markets, but that the two are related by an equation which includes a

4 4 component involving the markup of a price over marginal cost. Hall s approach has the advantage that does not require to measure the user cost of capal. Hall (1990) even demonstrates that if one relaxes the assumption of constant returns to scale, the above mentioned equation also includes another addional term that can be used to estimate jointly the average markup and the elasticy of scale (see also Klette,1999; Ohinata and Plasmans, 2002). Crépon et al. (1999, 2002, and 2007), Dobbelaere (2004), and Dobbelaere and Mairesse (2013) also introduce imperfect competion in labour markets in Hall s (1990) model. This is argued from the observation that numerous studies have documented large wage differentials across firms or industries for apparently homogeneous types of workers and occupations. Such wage differentials indicate that labour markets are far from being competive. They discuss a generalization of Hall s (1990) approach to allow for the possibily that wages are contractually determined between employers (firms) and employees (workers) according to an efficient bargaining model, where similarly schooled and experienced employees in some high-prof firms can get higher wages than in low-prof firms. We introduce their extension in our model setting. The aim of the strand of lerature up to now was to examine the discrepancies between factor elasticies in the production function and their corresponding shares in revenue due to both imperfections in the product and labour markets. The novelty in our model consists in embedding such a framework to determine changes in productivy growth. Using firm-level data for individual firms in the Netherlands divided over 22 industries for the period , we find that firms set their prices above their marginal costs and workers are bargaining over their salaries. At the manufacturing level, our results, compared wh Dobbelaere and Mairesse s (2013) analysis on French manufacturing enterprises, indicate less imperfect competion in the output market and more imperfect competion in the labour market. Moreover, we find that deviating from the assumption of perfect markets implies heterogeney in production technologies and varying returns to scale whin the 22 industries. We also find that, for many of these industries, firms pricing behaviour tends to be directly associated wh the characteristics of their production technology, including changes in efficiency that translate into productivy growth. Our finding is in line wh well-known models of endogenous growth (e.g., Romer, 1990; Grossman and Helpman,1991; Aghion and Howt, 1992). As a matter of fact, a decrease of the level of product market competion has a posive effect on productivy growth, by augmenting the monopoly rents that reward new innovation. Furthermore, connected to the innovation process

5 5 and labour market imperfections, as also mentioned in Aghion and Howt (1994), we find that labour market imperfection yields a lower productivy growth rate. Firms are subject to idiosyncratic shocks due to product innovations calling for labour force reallocations. As rigidy of the labour market does not allow firms from adjusting their labour factor (quickly), labour market imperfection then causes a lower innovation rate. Wh regard to the globalisation environment in which firms are operating, the most direct link between globalization and productivy growth arises when knowledge acquired in one country can be used to facilate research in another county. This is the issue of international knowledge spillovers. Helpman (2004) reviews a body of empirical research that finds evidence of substantial international knowledge spillovers (see also Lukach and Plasmans, 2000, 2002a and b, and 2005; Plasmans and Lukach, 2001, 2003, and 2010). The above-mentionned findings bring together existing branches of lerature to provide a unique global understanding of the impact of output and labour market behaviour on productivy change. The paper is organised as follows. In section 2, we formulate a MFPG measure that allows for market power and time-varying economies of scale, and use the tradional markup as measure for competion, but corrected for imperfect competion in the labour market. Section 3 describes the data and reports econometric results on market imperfection parameters. In section 4 we report and concisely discuss empirical results for the MFPG measure discussed in section 2. In the final section we conclude wh some policy implications. 2 MFPG based on the markup as a measure of market power We propose a measure for MFPG that allows for both monopoly power and scale economies and is directly related to the markup. The markup is a measure of market power based on price-cost margins that is often applied in the lerature. In fact, research has indicated that the markup is to be preferred over the other often employed measure of market power, the Herfindahl index. We derive the estimating equation for a setting of three inputs (capal, labour and intermediate goods) under imperfect competion in the goods and labour markets. Here, we use gross output, Y, as a measure of output and relate to three specific inputs: Y = A Fi ( K, L, M ) i =1,2, N; t =1,..., T, (1)

6 6 where capal, labour, and intermediate goods, the latter consisting of materials input and energy, are denoted for firm i at period t as K, L, and M, respectively. and Fi () is assumed to be homogeneous of degree A is defined as MFP, so that growth in output can be decomposed into growth in technology and inputs by logarhmic differentiation of the production function (1): dy da K Fi ( ) dk L Fi ( ) dl M Fi ( ) dm =. Y A F( ) K K F ( ) L L F( ) M M i We now relax the conventional assumption of perfect competion in the labour market, allowing both firms and workers unions to have some market power. Many authors have studied the influence of the market power of unions, by introducing wage rigidies through efficiency wages. iii For instance, Hall s model (1990) assumes that the firm wages and level of employment are jointly determined according to an efficient bargaining scheme between the firm and s workers. Interestingly to note now is that, following MacDonald and Solow s (1981) efficient bargaining model, in which both wage and employment are bargained between firms and their workers, can be directly shown that the wage of workers is determined at a level which is higher than the firm s marginal revenue of labour, that is, lny 1 WiktLikt =, where ln L (1 ) Y P ( Y ) ikt t negotiated wage, P ( Y ) is the market price as a function of aggregate output, and t W (2) is the is the Lerner index. Hence, workers in firms wh market power on the output market can earn wages that are much higher than the competive industry wage level. Introducing the nominal input prices, R W and Z as firm i s rental price of capal, wage rate and un price for intermediate goods, respectively, the efficient bargaining model can be summarised as follows. of the wage The workers in the firm bargain wh the firm over both the level of employment L and W. According to MacDonald and Solow (ibid.) the workers objective in their efficient bargaining model can be specified in two alternative ways, that is, eher as the workers (or union membership) aggregate gain from employment, L ( W W ), or, taking account of the unemployment benefs, as LW negotiated wage and W ( N L ), where W is the reservation wage, W the N is the labour supply. MacDonald and Solow (ibid.) judge the first

7 7 specification as the most appropriate one for real life. As a matter of fact, the unemployment benefs may vary in magnude, duration, and eligibily (Bean, 1994); therefore, we advocate MacDonald and Solow s (1981) suggestion. The firm s objective is to maximise s short-run prof given by the difference between the total revenue and the total costs, that is, as P ( Y ) Y W L R K Z M. t The solution to the Nash bargaining problem results in the maximisation of a (multiplicative) weighted average of the workers aggregate gain from employment and the firm s short run prof: where 0,1 W, L, K, M 1, max L ( W W ) P ( Yt ) Y WL RK ZM is the degree of workers (or unions ) bargaining power. Maximising wh respect to employment and to wage, yields the reservation wage (see equation (32) in Appendix B): W P ( Yt ) Y =, L which is the solution to the Nash bargaining model. Hence, the reservation wage is the theoretical wage valid in an imperfectly competive output market and a perfectly competive labour market. Given the equilibrium reservation wage, we can express the elasticy of labour ilt as (for derivations, see Appendix B): = s, (3) ilt ilt where s ikt denotes the share of the cost of input k in the total production value of firm i at period t so that s ilt is the share of the cost of labour of firm i at period t, and is the returns to scale parameter of firm i at period t., ilt Firm i s elasticies of output wh respect to capal, labour, and intermediate goods ( ikt,, respectively) at period t can then be expressed as: imt A Fi () K K Y = = s, (4) ikt ikt A Fi () L L Y = = s,and (5) ilt ilt

8 8 A Fi () M M Y = = s. (6) imt imt Only when the technology is constant returns to scale and the output and labour markets are perfectly competive, the elasticies will be equal to the observed input shares. Due to the imperfect competion on the labour market, establishing the relationship J i J i = k =1 ikt = k =1sikt = s is no longer valid. Indeed, adding the right hand sides of (4)-(6), the correct relationship between and s is found as: = ( s ), (1 ) (7) where the labour market bargaining elasticy,, is now involved. Following the standard convention, we will use the differences of the logarhms to approximate the logarhmic growth rates; logarhms of variables will be denoted as lower case letters. Substuting the output elasticies (4) and (6) into (2) and taking account of production function (1) and the corrected scale elasticy (7), we solve for the MFPG rates resulting output growth equation: ikt ilt imt a from the y = ( s k s l s m ) ( s 1) l a. (8) (1 ) This equation is general in the sense that is derived whout assuming a constant returns to scale technology or perfect competion, neher in the output market nor in the labour market. Defining x sikt k silt l simt m and compact form: where the bargaining elasticy, we can rewre our estimating equation in a (1 ) y = x ( s 1) l a, (9) equals /( ). We assume that the Hicks neutral technological progress is a random variable such that the growth rate of firm i in period growth rate, t t consists of a firm-specific growth rate, a i, and a period specific, which captures the macroeconomic shock that is common across industries in the same period, plus a whe noise, u (which is also called random individual heterogeney ).

9 9 Therefore MFPG a = a u. i t Under the prof-maximising approach, provided that the firm s perceptions of the elasticies of demand remain unchanged, the markup would remain constant over time (μi). Once a firm has discovered a markup of price over marginal costs that serves s purposes, then is que likely that will maintain that markup (Coutts et al., 1978; Basu and Fernald, 1997). The constancy over time does not rule out the possibily for the structural parameters μ and ϕ to vary across firms. In this chapter, we are going to apply model (9) on yearly firm-level data for firms established in the Netherlands. 3 Data and econometric results We extract data from Statistics Netherlands for the years The output and the input variables are defined as follows. As an output measure, we use the value of gross output ( PY ) of each firm i. Labour ( L ) refers to the number of employees in each firm for each year, collected in September of that year. The corresponding wages ( W ) include the total labour costs (gross wages plus salaries and social contributions) before taxes. The costs of intermediate inputs ( ZM ) include costs of energy, intermediate materials and services. The user costs of capal stock ( WK ) are calculated as the sum of the depreciation of fixed assets and the interest rate charges. We use a two-dig NACE deflator of fixed tangible assets calculated by Statistics Netherlands in order to compute the change in the volume index of capal stock ( k ). The nominal gross output and intermediate inputs are deflated wh the appropriate price indexes from the input-output tables available at the NACE Rev. 1 two-digs sector classification. iv The data extracted from the Production Statistics (PS) constutes a highly unbalanced panel (wh a minimum of 1259 firms in 1994 and a maximum of 6277 of enterprises in 1997) of observations spanning over 16 years and over 22 industries. Furthermore, for some firms, we observed a negative correlation between the capal growth rate and the output growth. As a matter of fact, if the firm produces non-tangible goods, even if capal assets are growing, the output may decrease as acquires more technology, which often allows the firm to reduce the output volume. Since our specification of a production function is meant to represent manufacturing firms only, we exclude this type of firms. For the estimates, we only include firms

10 10 for which we have at least two consecutive observations for all variables, ending up wh 7161 firms. The number of time observations per firm (T/firm) varies between two and sixteen years (30660 observations for 7161 firms). Table A1 in the Appendix reports the sectors that were chosen wh a corresponding NACE two-dig code and the corresponding number of firms ( N ). 3.1 Econometric results for the complete sample of Dutch firms Productivy shocks u in MFPG a = a u, such as posive technology shocks, might i t affect the level of factor inputs. It is indeed a plausible assumption that the compose error a includes an unobservable component which is taken into account in the firm s information set before input choices are made. The existence of such components raises the possibily that the input choices are correlated wh u. Hence, we treat all firm specific variables as potentially endogenous. In such a case, Ordinary Least Squares (OLS) estimates would be inconsistent and biased. The estimation of a panel data model wh predetermined variables is typically done by means of Generalized Method of Moments (GMM) estimators applied to the first differences of the variables in the equation of interest, where all the available lags of the predetermined variables are used as instruments. The purpose of this approach is to remove time-invariant unobserved individual heterogeney. Therefore, under the assumption that current random shocks are serially uncorrelated v and defining w ( x, ( s 1) l ) ', the orthogonaly condions can be wrten as E( w u ) = 0, for j = 2,3..., T. The instruments we use are therefore lagged values of j is and ( s 1) l from t 2 and before. The exogeney of the instruments wh respect to the error term is tested by computing Hansen s J test statistic. In addion, we also include time dummies to capture possible unobservable shocks common to all firms. However, this approach yields inaccurate estimates in the case of a panel wh a small number of time periods wh highly persistent data. In this context, as has been stressed in, for instance, Hall and Mairesse (2005), the application of first-differences GMM (FD GMM) estimators wh lagged levels of the series as instruments has produced unsatisfactory results. More specifically, the coefficient of the capal stock is generally low and statistically insignificant, and returns to scale appear to be unreasonably low. Blundell and Bond (1998) suggest that the problem of weak instruments is behind the poor performance of standard GMM x

11 11 estimators in this context. This problem of weak instruments can be overcome by applying an extended GMM estimator proposed by Arellano and Bover (1995). This estimator, labelled as system GMM or SYS GMM, is based on an augmented system which includes level equations wh lagged differences as instruments in addion to the differenced equations wh lagged levels as instruments. First, we focus on the manufacturing industry as a whole over the period whout looking at the potential heterogeney in the markup and the bargaining power parameters among sectors. Estimation results for the entire manufacturing market of our static estimating equation (9) for a range of estimators (level OLS, FD OLS, FD GMM) and of a dynamic specification of this equation, allowing for an autoregressive component in the productivy shocks, for FD GMM and SYS GMM suggest that the derived price-cost markups are not significantly different from 1 and that the corresponding extent of rent sharing is que small (respectively and 0.278). Furthermore, both OLS and FD OLS suggest decreasing returns to scale. The main drawbacks to these estimators are that part of the information in the data is left unused. A fixed-effect estimator uses only the across time variation, which tends to be much lower than the cross-section one for not particularly persistent data. Second, the assumption that the firm s specific attributes are fixed over time may not always be reasonable. The OLS methods tend to underestimate the structural coefficients when the error term of the production function is expected to influence the choice of factor inputs and when the data is not particularly persistent (i.e., the across time variation is much lower than the cross-section one). Despe the signal of non-persistency, we also estimate a dynamic panel data model by considering an AR(1) extension of (9) as indicated above. For this we take as instruments the lagged levels dated at ( t 2 ) and ( 3 t ) in the first-differences equations. As addional instruments in the SYS GMM estimation, we take the lagged differences dated at ( t 1 ). Year dummies have been included in both models. As could be expected from the signal of non-persistency, the AR(1) structure that we assume for the idiosyncratic error term is not needed, as the coefficient of the lagged dependent variable is statistically not significant for the SYS GMM. Accounting for imperfect competion on the labour market yields larger returns to scale than assuming perfect competion. The estimated bargaining power parameter is very high and significant (0.614 wh standard error 0.054, henceforth denoted as 0.614(0.054)) when estimating the static model (9) by FD GMM, reflecting the infuence of labour behaviour on output growth. Nevertheless,

12 12 taking into account the existence of rent sharing translates in a rise in the estimated markup (from 1.04(0.03) to 1.06(0.03)) and in the estimated elasticy of scale (from 0.98(0.03) to 0.99(0.03)). Hence, we find some evidence of imperfect competion on the output market and strong unions power on the labour market in the Dutch manufacturing industries. Markups are significantly and fairly larger than one and returns to scale are constant or moderately decreasing, in the range of 0.9 to Across-Industry Estimates Since firms production behaviour is very likely to vary even across industries, we also investigate across-industry firm heterogeney in estimated markup and rent-sharing parameters. For all 22 manufacturing industries, given in Table 1A in the Appendix, we estimate static equation (9) by DF GMM wh and whout the extension to labour market imperfections, by relying on our trusted FD GMM estimator. As instruments we take an appropriate number of lagged levels. Year dummies are always included. The estimated parameters are reported in Table 1, where is the estimated price-cost mark-up, the corresponding rent sharing (unions bargaining power) elasticy and the estimated scale elasticy; =1 if there is perfect competion on the output market and =0 if there is perfect competion on the labour market.

13 13 Table 1: FD GMM estimates of equation (9) for all 22 industries Industry a (%) ( = 0) ( = 0) N (obs/firm) (0.891) 1.031(0.042) 1.251( (0.042) 0.951(0.039) 1168(3.8) (0.089) 1.107(0.168) 1.019(0.156) (0.205) 1.047(0.192) 16(3.8) (0.185) 1(0.044) 0.887(0.103) (0.108) 0.974(0.101) 286(3.6) (0.092) 0.977(0.035) 0.780(0.067) (0.064) 0.825(0.152) 142(2.8) (0.033) 0.958(0.152) 0.961(0.924) (0.130) 0.948(0.122) 83(3.4) (7.333) 0.982(0.098) 0.933(0.190) (0.089) 0.920(0.083) 262(3.7) (0.135) 1.118(0.162) 1.187(0.204) (0.091) 0.967(0.085) 289(4.0) (0.045) 1.330(0.213) 1.346(0.211) (0.213) 1.003(0.199) 840(3.5) (2.6) (0.068) 1.025(0.094) 1.054(0.100) (0.092) 0.929(0.086) 508(3.9) (0.113) 1.182(0.119) 1.223(0.117) (0.049) 0.956(0.046) 540(3.3) (0.083) 1.169(0.112) 1.294(0.196) (0.088) 1.007(0.082) 428(4.2) (0.741) 0.946(0.086) 0.936(0.077) (0.052) 0.912(0.048) 146(4.6) (0.073) 1.020(0.040) 1.070(0.059) (0.040) 0.947(0.038) 1391(3.6) (0.128) 1.238(0.086) 1.339(0.094) (0.065) 1.148(0.061) 1047(3.1) (0.019) 1.720(0.166) 1.727(0.165) (0.152) 1.254(0.142) 36(2.4) (3.457) 1.009(0.071) 0.967(0.072) (0.072) 0.967(0.072) 265(3.4) (0.929) 1.047(0.189) 1.031(0.173) (0.057) 1.118(0.053) 69(2.4) (0.034) 1.130(0.065) 1.137(0.065) (0.080) 0.967(0.075) 250(3.2) (0.450) 1.324(0.087) 1.344(0.141) (0.096) 1.279(0.090) 210(3.6) (0.146) 1.151(0.047) 1.208(0.092) (0.039) 1.031(0.036) 217(3.2) (0.079) 1.045(0.077) 1.161(0.163) (0.077) 0.978(0.072) 454(3.4) (1) Standard errors in parentheses after the estimates (2) Sample period (3) Dependent variable: output growth y ; a is the average MFPG Results show that the magnude of the estimated markup, elasticy of scale and bargaining power is likely to vary among industries. The price-cost margin is estimated to be lower than 1.01 for the first quartile of industries and higher than 1.16 for the top quartile. However, almost all industries reveal constant or increasing returns to scale (that is likely to be the case for

14 14 manufacturing industries) vi. The sectors that exhib imperfect competion on the output market ( significantly larger than one at 5% level) are the sectors 29, 30, 33, 34, and 35, while whin-industry imperfect competion on the labour market is present in 13 sectors. It also follows from Table 1 that taking into account the existence of rent sharing vii increases the estimated markup bargaining power., where has to be said that in many sectors there is strong evidence of large unions 4 Impact on MFPG Introducing imperfect competion on both output and labour markets, we see that changes in the level of competion vary by industry. This section incorporates these findings to analyse the relationship between imperfect competion, varying returns to scale, and productivy growth. The effect of markups on MFPG, measured in a growth accounting framework, has been addressed in a number of related papers. Azzam et al. (2004) decompose sources of MFPG by economies of scale, markups and demand growth. Using US food industry data for , these authors find that, on average, productivy grew by 0.22 per cent due to markups and 0.10 per cent due to increases in economies of scale. Morrison (1992) finds that the MFPG adjusted for markups in total manufacturing has increased during for Japan, the US, and Canada. It is also noted that variable returns to scale tend to neutralize the implications of the markup adaptation. Kee (2004) finds that the average annual growth rate of the productivy of Singapore s manufacturing sector from 1974 to 1992 is 7 per cent while the MFP index takes imperfect competion and non-constant returns to scale into account, whereas the MFPG is less than 3 per cent by conventional measurement. Note that van Leeuwen and van der Wiel (2003) find evidence of an oppose effect in the (market) service sector when the MFPG is adjusted by a markup ignoring economies of scale. Based on firm-level data between 1994 and 1999, their study finds that the modified MFPG is 0.2 per cent higher than the tradional MFPG.

15 15 Table 2: Averages of MFPG (%) MFPG (0,1] FD GMM residuals for all 22 sectors =0 = 0, = 1 = 0, = =1 Mean Median FD GMM residuals for sectors where >1 Mean Median FD GMM residuals for sectors where 1 Mean Median In Table 2, we summarize our FD GMM estimation results for MFPG. In particular, when adopting the conventional framework of perfect competion and constant returns to scale, the average residually estimated productivy growth (across years) is per cent. When relaxing the assumption of constant returns to scale, this average MFPG for all industries rises to per cent. Strikingly, introducing imperfect competion on the output market dramatically increases the average productivy growth rate (to 1 per cent). This finding is in line wh the majory of models of endogenous growth (e.g., Romer, 1990; Grossman and Helpman,1991; Aghion and Howt, 1992). Indeed, a decrease of the level of product market competion has a posive effect on productivy growth, by augmenting the monopoly rents that reward innovations. Furthermore, connected to the innovation process and labour market imperfections, as in Aghion and Howt (1994), we find that labour market imperfection results in a lower productivy growth rate (MFPG slightly decreases to 0.95 per cent) since rigidy of the labour market does not allow firms adjusting their labour force flexibly. Considering a subsample containing the MFPG estimates of those sectors for which the price-cost markup exceeds 1 and the other subsample wh MFPG estimates of sectors showing evidence of perfect competion on the output market (estimated price-cost markup less than or equal to 1), we find that the average MFPG is remarkably higher for the first subsample of sectors

16 16 (2.3 per cent) than for the second one (0.94 per cent). Again, strong unions bargaining power leads to lower MFPG. Plasmans and Lukach (2010) study backward and forward patent cations in patents granted to firms and instutions in the Netherlands by the Uned States Patent and Trademark Office (USPTO). The study establishes different patterns of patent cation in recent Dutch patents belonging to different industrial classes. They run a model in the set of backward cations made in Dutch applicants patents during and in the set of forward cations to patents issued to firms and organizations in the Netherlands during They compare the patterns of knowledge utilization (represented by backward patent cations) and knowledge dissemination (represented by forward patent cations) and obtain evidence of inter- or intra-firm and inter- or intra-industry knowledge spillovers. In the context of effective competion and innovation policies they advocate for paying special attention to industry specifics when designing policy programs and measures directed at stimulating (international) R&D cooperation and knowledge spillovers. They present evidence that policies for promoting better knowledge exchange among firms should also distinguish between the measures for promoting the inward and the outward knowledge flows for companies in the Netherlands. 5 Conclusion In this paper, we explore both theoretically and empirically a framework where we integrate possible labour market rigidies and imperfect output market behaviour on multifactor productivy growth (MFPG). Embedding Hall s (1990) efficient bargaining model, which introduces a substantial degree of labour market imperfections, we show that rigidies and frictions in the labour market might be crucial for understanding the firms marginal costs and their price setting behaviour. We apply this analysis to 22 industries in Dutch manufacturing for the period By comparing a range of econometric estimators (level OLS, first-differences OLS, first-differences GMM, Arellano and Bover, 1995, and Blundell and Bond, 1998), we estimate a standard output growth equation wh three factors of production and a residual MFPG rate, allowing for an autoregressive component in the productivy shocks. Regarding the imperfectness of markets, we find from a highly unbalanced panel of 7161 firms, dispersed over 22 industries for the years in the Netherlands, some evidence of

17 17 imperfect competion on the output market and strong unions power on the labour market in the Dutch manufacturing industries. Markups are significantly and fairly larger than one and returns to scale are constant or moderately decreasing, in the range of 0.9 to 1.0. On the other hand, the AR(1) structure that we assume for the idiosyncratic error terms is not supported by our data. The variation of imperfect competion across the Dutch industries is large. In particular, we show evidence of imperfect competion on the output market in 5 of the 22 Dutch industries, where the estimated markup is significantly larger than one at the 5% level, and strong unions bargaining power on the labour market in 13 of these 22 industries. To the extent that wages are allocative, we find that labour market imperfections play a main role in eher determining the market behaviour and in assessing the correct MFPG. When adopting the conventional framework of perfect competion and constant returns to scale, the average estimated productivy growth (across years) is -0.7 per cent. When relaxing the assumption of constant returns to scale, this average MFPG for all industries rises to per cent. Moreover, introducing imperfect competion on the output market dramatically increases the average productivy growth rate (to 1 per cent). Hence, a decrease of the level of product market competion has a strong posive effect on productivy growth. Furthermore, we find that labour market imperfections lead to a lower productivy growth rate (MFPG slightly decreases to 0.95 per cent). Wh regard to globalisation has to be stressed that in a static economy, globalisation leads countries to specialise in the activies in which they enjoy a comparative advantage. The same is true in a setting wh endogenous growth wherein one of the activies that each country undertakes is the accumulation of knowledge. On the other hand, and as argued before, knowledge diffusion like the creation of new technologies (wh or whout patenting), can be a source of sustained productivy growth. International integration affects the incentives for investment in activies that foster knowledge diffusion as well as in the productivy of those activies. This provides another link between globalisation and productivy growth. The relationship between international integration and knowledge accumulation ought to vary depending upon the fundamental characteristics of a country, including s factor and resource endowments and s history. It has to be admted that few empirical studies linking productivy growth outcomes to openness or trade policy have allowed for such a dependence. How can we encourage (more) international knowledge spillovers most efficiently? However, previously mentioned empirical

18 18 studies demonstrated the importance of them in determining a country s overall productivy growth performance. These studies also emphasised which economic policies can be used to promote international knowledge spillovers and productivy growth.

19 19 Appendix A Table 1A: NACE 2-dig code and number of firms Code Sector N Food, beverages, tobacco Textiles Wearing apparel Leather Wood Paper products, publishing, printing Publishing, printing and reproduction of recorded media Coke, refined petroleum products and nuclear fuel Chemicals and chemical products Rubber and plastic products Other non-metallic mineral products Basic metals, Fabricated metal products Fabricated metal products, except machinery and equipment Machinery and equipment Electrical and optical equipment Electrical machinery and apparatus Radio, television and communication equipment and apparatus Medical, precision and optical instruments, watches and clocks Transport equipment Motor vehicles and other Other 1106 Total 15976

20 Appendix B The bargaining model 20

21 21

22 22

23 23

24 24 Then, arranging terms, the elasticy of labour can be expressed in terms of the markup, μ, the elasticy of scale, θ, and the degree of workers bargaining power, ϕ: θilt = μ silt - ϕ μ + ϕ θ.

25 25 References Aghion, P. and P. Howt (1992), A Model of Growth through Creative Destruction, Econometrica, vol. 60(2), pp Aghion, P. and P. Howt (1994), Growth and unemployment, Review of Economic Studies, vol. 61(3), pp Arellano, M. and Bover, O. (1995), Another Look at the Instrumental Variable Estimation of Error Component Models, Journal of Econometrics, vol. 68(1), pp Ark, B. van (2014), Total Factor Productivy: Lessons from the Past and Directions for the Future, NBB Working Papers, No. 271, National Bank of Belgium, Brussels. Azzam, A., E. Lopez and R.A. Lopez (2004), Imperfect Competion and Total Factor Productivy Growth, Journal of Productivy Analysis, vol. 22, pp Balk, B.M. (2008), Measuring Productivy Change whout Neoclassical Assumptions: A Conceptual Analysis, Working Paper no. ERS MKT, Erasmus Research Instute of Management (ERIM), Erasmus Universy, Rotterdam. Basu, S. and J.G. Fernald (1997), Returns to Scale in US Production: Estimates and Implications, Journal of Polical Economy, vol. 105(2), pp Bates, W. (2001), How Much Government?: The Effects of High Government Spending on Economic Performance, New Zealand Business Roundtable, Wellington. Bean, C. R. (1994). European unemployment: A survey. Journal of Economic Lerature, vol. 32(2), pp Blundell, R. and S. Bond (1998), Inial Condions and Moment Restrictions in Dynamic Panel Data Models, Journal of Econometrics, vol. 87(1), pp Caves, D.W., L.R. Christensen and W.E. Diewert (1982), The Economic Theory of Index Numbers and the Measurement of Input, Output, and Productivy, Econometrica, vol. 50(6), pp Coutts, K., W. Godley and W. Nordhaus (1978), Industrial Pricing in the Uned Kingdom, Cambridge Universy Press, Cambridge UK. Crépon B., R. Desplatz and J. Mairesse (1999), Estimating Price-Cost Margins, Scale Economies and Workers Bargaining Power at the Firm Level, CREST Working Paper no. G9917, CREST (ENSAE), Paris.

26 26 Crépon B., R. Desplatz and J. Mairesse (2002), Price-Cost Margins and Rent Sharing: Evidence from a Panel of French Manufacturing Firms, Revised version of CREST Working Paper no. G9917, CREST (ENSAE), Paris. Crépon B., R. Desplatz and J. Mairesse (2007), Price-Cost Margins and Rent Sharing: Evidence from a Panel of French Manufacturing Firms, mimeo, CREST (ENSAE), Paris. De Loecker, J. (2011), Producrt Differentiation, Multi-Product Firms and Estimating the Impact of Trade Liberalization on Productivy, Econometrica, vol. 79(5), pp Diewert, W.E. (2006), The Measurement of Productivy, Chapter 6 in W.E. Diewert, Applied Economics, Universy of Brish Columbia, Vancouver. Dobbelaere, S. (2004), Estimation of Price-Cost Margins and Union Bargaining Power for Belgian Manufacturing, International Journal of Industrial Organization, vol. 22(10), pp Dobbelaere, S. and J. Mairesse (2013), Panel data estimates of the production function and product and labour market imperfections, Journal of Applied Econometrics, vol. 28(1), pp Färe, R., S. Grosskopf, M. Norris, and Z. Zhang (1994), Productivy Growth, Technical Progress, and Efficiency Change in Industrialized Countries, American Economic Review, vol. 84(1), pp Feng, G. and A. Serletis (2008), Productivy Trends in U.S. Manufacturing: Evidence from the NQ and AIM Cost Functions, Journal of Econometrics, vol. 142(1), pp Fisher, I. (1922), The Making of Index Numbers, Houghton-Mifflin, Boston. Grossman, G.M. and E. Helpman (1991), Innovation and Growth in the Global Economy, MIT Press, Cambridge MA. Grossman. G.M., E. Helpman, E. Oberfield, and T. Sampson (2017), The Productivy Slowdown and the Decling Labor Share: A Neoclassical Exploration, CESifo Working Papers, No. 6714, CESifo, Munich. Hall, B.H. and J. Mairesse (2005), Testing for Un Roots in Panel Data: An Exploration Using Real and Simulated Data, Chapter 19 in D.W.K. Andrews ens J.H. Stock (eds.), Identification and Inference for Econmetric Models: Essays in Honor of Thomas Rothenberg, Cambridge Universy Press, New York. Hall, R.E. (1986), Market Structure and Macroeconomic Fluctuations, Brookings Papers on

27 27 Economic Activy, vol. 2, pp Hall, R.E. (1988), The Relationship Between Price and Marginal Cost in U.S. Industry, Journal of Polical Economy, vol. 96(5), pp Hall, R.E. (1990), Invariance Properties of Solow s Productivy Residual in P. Diamond (ed.), Growth, Productivy, Unemployment, MIT Press, Cambridge MA, pp Helpman, E. (2004), The Mystery of Economic Growth, Harvard Universy Press, Cambridge MA. Kee, H.L. (2004), Estimating Productivy When Primal and Dual TFP Accounting Fail: An Illustration Using Singapore s Industries, Topics in Economic Analysis & Policy, vol. 4(1), article 26. Klette, T.J. (1999), Market Power, Scale Economies and Productivy: Estimates from a Panel of Establishment Data, The Journal of Industrial Economics, vol. 47(4), pp Leeuwen, G. van and H. van der Wiel (2003), Do ICT spillovers matter: Evidence from Dutch firm-level data, CPB Discussion Paper no. 26, The Hague. Lukach, R. and J. Plasmans. (2000). R&D and Production Behavior of Asymmetric Duopoly Subject to Knowledge Spillovers, CESifo Working Papers, No. 287, CESifo, Munich. Lukach, R. and J. Plasmans (2002a), Measuring Knowledge Spillovers using Patent Cations: Evidence from the Belgian Firm s Data, CESifo Working Papers, No. 754, Cesifo, Munich. Lukach, R. and J. Plasmans (2002b), A Study of Knowledge Spillovers from the Compatible EPO and USPTO Patent Datasets for Belgian Companies in M. Cincera (ed.), Belgian Report on Science, Technology and Innovation, pp , DWTC, Brussels. Lukach, R. and J. Plasmans (2005), International Knowledge Flows from and into a Small Open Economy: Patent Cation Analysis in A. Sphoven and P. Teirlinck (eds.), Beyond Borders: Internationalisation of R&D and Policy Implications for Small Open Economies, pp , Elsevier, Amsterdam. MacDonald, I.M. and R.M. Solow (1981), Wage Bargaining and Employment, American Economic Review, vol. 71(5), pp Malmquist, S. (1953), Index Numbers and Indifference Surfaces, Trabajos de Estadistica, vol. 4(1), pp Manning, A. (2003), Monopsony in Motion: Imperfect Competion in Labor Markets, Princeton

28 28 Universy Press, Princeton NJ. Morrison, C.J. (1992), Unravelling the Productivy Growth Slowdown in the Uned States, Canada and Japan: The Effects of Subequilibrium, Scale Economies and Markups, Review of Economics and Statistics, vol. 74(3), pp Morrison, C.J. (1999), Cost Structure and the Measurement of Economic Performance, Kluwer Academic Press, MA. Ohinata, S. and J. Plasmans (2002), Markups and International Competion wh an Application to the Benelux Countries, Working Paper, Universy of Antwerp. Plasmans, J. and R. Lukach (2001), Measuring Knowledge Spillovers using Belgian EPO and USPTO Patent Data, CESifo Working Papers, No. 430, CESifo, Munich. Plasmans, J. and R. Lukach (2003), Measuring Knowledge Spillovers in the New Economy Firms in Belgium using Patent Cations, Global Business and Economics Review, vol. 5(2), pp Plasmans, J. and R. Lukach (2010), The Patterns of Inter-firm and Inter-industry Knowledge Flows in the Netherlands, CESifo Working Papers, No. 3057, Munich. Romer, P. (1990), Endogenous technological change, Journal of Polical Economy, vol. 98(5), pp Shephard, R.W. (1953), Cost and Production Functions, Princeton Universy Press, Princeton NJ. Solow, R.M. (1957), Technical Change and the Aggregate Production Function, Review of Economics and Statistics, vol. 39(3), pp Törnqvist, L. (1936), The Bank of Finland s Consumption Price Index, Bank of Finland Monthly Bulletin, vol 10, pp Uras, B.R. and P. Wang (2017), Production Flexibily, Misallocation and Total Factor Productivy, NBER Working Paper Series, No , National Bureau of Economic Research, Cambridge MA. To appear as a chapter in Jonathan Michie (ed.), The Handbook of Globalisation, 3 rd Edion, 2018, Edward Elgar, Cheltenham and Northampton. i Under perfect competion on the input markets, the input quanty index is a function of the input prices and quanties for the two periods under consideration (Laspeyres, Paasche, Fisher s (1922) ideal quanty index being the square root of the product of the Laspeyres and Paasche quanty indexes, and Törnqvist s (1936) quanty index). ii Under the assumption of perfect competion on the output and inputs markets, Grossman et al. (2017) develop a neoclassical growth model wh two inputs (capal and labour) in which they explain, also empirically, that the dramatic decline of the labour share in national income since about 2000 in the USA (and Europe) may be caused by

29 29 the slowdown of (measured total factor) productivy growth over roughly the same period. In their paper, they eexplore the possibily that these two seemingly-unrelated phenomena might in fact be connected by a process of neoclassical growth wh endogenous human capal accumulation. iii Dobbelaere and Mairesse (2013) consider three types of competion in the labour market: perfect competion or right-to-manage bargaining, noted as PR, efficient bargaining, noted as EB, and monopsony according to Manning (2003), who abstains from the assumption that the labour supply curve facing an individual employer is perfectly elastic (this type is noted as MO). Considering also perfect and imperfect competion in the product market, denoted as PC and IC, respectively, this results in six regimes in total (PC-PR, IC-PR, PC-EB, PC-MO, IC-EB, and IC-MO), and we have only four (neher PC-MO nor IC-MO). This seems us most plausible since Dobbelaere and Mairesse (2013) themselves conclude that IC-EB is by far the dominant regime (Dobbelaere and Mairesse, 2013, p. 3). Moreover, their underlying production function is assumed to be Cobb-Douglas, while several studies advocate an elasticy of substution different from uny (see, for instance, Uras and Wang,2017, for the case of two inputs, and several other studies for the case of three or more inputs as we have). iv NACE Rev. 1 is a 2-dig activy classification which was drawn up in It is a revision of the General Industrial Classification of Economic Activies whin the European Communies, known by the acronym NACE and originally published by Eurostat in Hence, we deflate firm-level nominal outputs or sales by an industry-wide producer index in the hope to eliminate price effects. This has two major implications. First, will potentially bias the coefficients of the production function if inputs are correlated wh prices, i.e. the omted price variable bias. Secondly, will generate productivy estimates containing price and demand variation. This potentially introduces a relationship between measured productivy and trade liberalisation simply through the liberalisation s impact on prices and demand. De Loecker (2011) provides a partial way to tackle these two issues, but we didn t follow this line in this chapter. v When a variable is predetermined (not completely exogenous), the current period error term u is uncorrelated wh current and lagged values of the predetermined variable but may be correlated wh future values. An unpredictable technology shock will be uncorrelated wh past (and potentially current) production settings, but will surely be correlated wh future ones. vi It has to be noted that globalisation affords innovators the opportuny to explo their new ideas on a larger stage. Firms that develop a new product, improve an old one, or find a better production technique can reap profs not only domestically, but also on sales abroad. This scale effect tends to boost the incentive for knowledge acquision and implies increasing returns to scale. However, in a more global economy, a successful innovator must share the market not only wh other domestic firms, but also wh those that produce abroad. The competion effect of globalisation presents an offsetting disincentive for knowledge acquision and may (sometimes) even lead to decreasing returns to scale. vii Note (again) that rent sharing is found to be evident if (the estimate of) ϕ 0 (no evidence if the estimate of ϕ is not significantly different from zero).