Institutions and Wage Determination: A Multi-Country Approach

Nunziata, Luca Working Paper Institutions and Wage Determination: A Multi-Country Approach Quaderni - Working Paper DSE, No. 433 Provided in Cooperation with: University of Bologna, Department of Economics Suggested Citation: Nunziata, Luca (2002) : Institutions and Wage Determination: A Multi-Country Approach, Quaderni - Working Paper DSE, No. 433, Alma Mater Studiorum - Università di Bologna, Dipartimento di Scienze Economiche (DSE), Bologna, https://doi.org/10.6092/unibo/amsacta/4865 This Version is available at: https://hdl.handle.net/10419/159274 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc/3.0/ Institutions and Wage Determination: a Multi-Country Approach∗ Luca Nunziata† January 31, 2002 Abstract We present an empirical analysis of the determinants of labour cost in OECD countries, with particular reference to the impact of labour market institutions from 1960 to 1994. The main contribution of the paper is to show that labour market regulations can explain a large part of labour cost rise in OECD countries in the last decades once we control for productivity. These results are consistent with the findings of a companion paper (Nickell et al., 2001) where the effects of institutions on unemployment are examined. The model controls for macroeconomic shocks, and include the possibility of interactions among institutions. We present also a discussion of the potential problems encountered when estimating a macro pooled model like ours. We focus, among other things, on the hypothesis of poolability and on the cointegration properties of the model, suggesting a two way fixed effects specification in GLS form that corrects for heteroskedasticity and serial correlation. The explanatory power of the model is finally tested by means of a series of by country dynamic simulations. Theme: Labour Cost Keywords: Labour Cost, Wage, Labour Market Institutions, Unemployment, MultiCountry Regression JEL-Code: E24, J31, J51. ∗I wish to thank Stephen Nickell for his invaluable comments and insights, John Muellbauer and Stephen Bond for their very helpful remarks. The usual disclaimer applies. †Department of Economics, University of Bologna and Nuffield College, University of Oxford. Email for correspondence: [email protected] . 1 1 Introduction One of the most challenging tasks for contemporary macroeconomics is the construction of robust empirical tests of the explanatory power of alternative theories of unemployment. Recent contributions by Nickell (1997), Elmeskov et al. (1998), Belot and van Ours (2000), Blanchard and Wolfers (2000) and Bertola et al. (2001) have shed some light on the causal relationship that links institutional factors and macroeconomic shocks with the rise and the persistence of high unemployment levels in most of the OECD countries. However we still lack of an empirical account of the way labour market institutions affect unemployment levels through their impact on wage determination. As we will see below, an important part of the explanation of the impact of labour market institutions on unemployment can be derived by the role played by these institutions in the wage determination process. If institutions do matter in the explanation of OECD unemployment patterns, then we should be able to estimate a significant impact on wages as well. Moreover, even if some institutions do not affect unemployment directly, exerting pressure on wages, but through their impact on the matching process of the unemployed to available job vacancies, they should have a significant impact on wages as well through the movements in equilibrium labour demand. The unemployment and the wage models are intimately connected and should yield coherent results. This work and the companion paper Nickell et al. (2001) propose an empirical investigation of the combined effects of wage pressure factors and labour market institutions on unemployment rates and labour cost. Our aim is to produce an empirical investigation of the time series properties of OECD unemployment and to yield some insight on the far from homogeneous experience of each single country. This paper concentrates on labour cost determination in OECD countries, while Nickell et al. present a broader picture, including an account of unemployment patterns and the Beveridge Curve in those countries. The structure of the paper is the following: the next section contains a concise theoretical introduction, section three presents a model of labour cost determination and a brief description of what we intend for labour market institutions. Section four includes a brief overview of the existing multi-country models of labour cost determination. Sections five, six and seven present the empirical analysis, including some comments on the specification of the econometric model. Section eight contains some final remarks1. 2 Unemployment and Labour Cost Determination Our theoretical background is Nickell (1998). According to his simple model, the wage setting framework is described by the equation: w=θ1pe+(1−θ1)p−θ2u−θ3∆u+zw(1) where wages are assumed to be set as a mark-up on averaged actual and expected prices, and zwindicates the vector of all exogenous wage pressure factors. The extent of 1Note that in what follows we use indifferently the term wage and labour cost, intending the real product wage (labour costs per employee normalised on the output price). This includes non wage costs such as employers’ contributions. 2 the mark up is influenced by the state of the labour market, represented by the unemployment rate u(hereinlogs)andthegrowthrateofunemployment∆u. Assuming mark-up price setting behaviour for firms, a standard production function and a mix of anticipated and unanticipated shocks on demand, price, and wages, this stylized model yields the following reduced unemployment equation: u=φ1u−1+φ2(¯zp+¯zw)−φ2(n+λr)+φ3(p+w)(2) where (¯zw,¯zp) represent anticipated autonomous shifts in wages and prices respectively, (w, p) are unanticipated shifts in the same variables, and (n, r) are unanticipated nominal and real demand shocks. In addition, we have that φ1=θ3/θ and φ2=θ−1, with θ=(θ2+θ3). The assumption of rational expectations rules out any effect of anticipated demand shocks on unemployment. A few points can be drawn from this simple setting: •There exists a negative long run relationship between unemployment and real demand factors. This relationship is the obvious consequence of whichever specification we may assume for the production function, since labour is a fundamental input for production. •The relationship between unemployment and wage and price pressure factors (¯zw,¯zp) depends on the impact of unemployment on wages, i.e. on the coefficient θ2. (i) If the unemployment level does not affect wages, i.e. if θ2=0,thenφ1=1and according to equation (2) unemployment will follow a random walk process. In this case its actual level will depend on the history of past shocks, and any unemployment level will be compatible with stable inflation. (ii) If unemployment significantly affects wages, i.e. if θ2>0, then the effect of past shocks falls with time at a rate which increases with θ2. Most importantly, there exists a level of unemployment u∗=1 θ 2 (¯zw+¯zp)suchthatif unemployment is consistently lower than u∗there is a tendency for inflation to grow. The unemployment level u∗is determined by wage and price pressure factors (¯zw,¯zp). The intuition underlying last point is that autonomous changes in unemployment could be offset by labour market forces if higher unemployment triggers a reduction in wages. If this is the case, unemployment will depend on wage pressure institutions. If this is not the case, we expect unemployment to be fully persistent (i.e. it will follow an I(1) process). The analysis of equation (1) is therefore a crucial step in order to understand the sources of OECD unemployment. Consequently, the objectives of the paper are the following: 1. to analyse the impact of unemployment on wages; 2. to investigate the impact of labour market institutions on wages; 3. to investigate the coherence of the wage model with the analogous model of unemployment contained in the companion paper Nickell et al. (2001). 3 This paper concentrates on the estimation of a labour cost equation while Nickell et al. (2001) provides an extensive account of the unemployment model based on the same criteria. The theoretical and empirical relationships between the two analysis have been briefly discussed here. A broader discussion can be found in the companion paper. 3 A Model of Labour Cost Determination Equation (1) may be written as: w−p=−θ2u−θ3∆u+zw−θ1(p−pe)(3) where the term θ1(p−pe) reflects macro shocks. If we want to generalise this simple model we have to allow wages to be ceteribus paribus anchored to productivity in equilibrium. In addition, there are empirical and theoretical reasons to expect some form of persistence in wage dynamics2.Ageneral dynamic reduced form, allowing for a positive productivity trend πand endogenous persistence, is therefore: wt−pt=β0+β1(wt−1−pt−1)−β2ut−β3∆ut+β4πt+γ¯zw,t +ϑst(4) where the same notation used above applies here, with ¯zwbeing the vector of wage pressure factors,and sa vector of nominal and real shocks. According to equation (4) the log real wage level depends respectively on last period’s value, labour market conditions represented by the level and change in unemployment, i.e. utand ∆ut, the productivity trend πt, wage pressure factors ¯zw, and a set of macro shocks st. Given the peculiar nature of our sample period, starting in 1960 when unemployment was virtually zero in some countries, in what follows we do not impose unemployment to be in logs3, since we would introduce spurious variability in the series4. The vector of wage pressure factors ¯zwincludes the set of all labour market institutions that influence the matching process of unemployed individuals to available job vacancies, or directly raise wages in presence of excess labour supply, or both. In any case they will impact wages and unemployment in the same direction, since labour demand, which is negatively related to wages, will move in the opposite direction of unemployment in equilibrium. The institutions can be broadly classified into wage bargaining institutions, institutions related to unemployment benefit provision, employment protection regulations and labour taxation. The empirical counterpart of each institutional dimension is represented by an indicator provided by the Labour Market Institutions Database (LMID), assem2This can depend on various factors, the most obvious being a certain degree of persistence of the impact of macro shocks. Moreover, in the presence of staggered wage adjustment in most countries, only a fraction of total contracts will be renewed every period, with the a resulting strong autocorrelation in the average wage. 3For example, Germany, New Zealand and Switzerland. 4For example, in log form, the rise in unemployment in Switzerland from 1960-64 to 1996-99 (0.2% to 3.7%) has a negative impact on wages which is nearly 300 percent larger than that in Italy where unemployment rose from 3.5% to 10%. This differential seems somewhat implausible. 4 bled by Nickell and Nunziata from different sources, and extensively described in the data appendix of this paper5. The institutions having a direct impact on the wage determination process are the trade union bargaining power and the degree of coordination in wage bargaining. The latter represents the extent to which parties to wage bargaining are able to take account of the macroeconomic consequences of their decisions. Union power consists instead of two major dimensions that are the proportion of employees covered by collective agreements (union coverage) and the union membership rate among active workers (union density). Unfortunately, we can only estimate the wage impact of union density since the data on union coverage is available for a very limited number of years6. As suggested by previous analysis, we expect a positive impact of union bargaining power on wages, although the effect is likely to be offset if bargaining is highly coordinated7. As regards unemployment benefits, a generous system has a direct negative influence on the willingness of the unemployed to undertake a vacant job. Both the amount and the time duration of the monetary transfer to unemployed individuals may matter in this respect. The former is summarised by the benefit replacement rate, i.e. the ratio of the average unemployment benefit on the average salary, aggregated over family types8. The attitude of the unemployed can also be influenced by the returns on non human wealth which may be increasing in the real interest rate9. Employment protection regulations are by definition the set of rules and procedures governing the treatment of dismissals of employed workers. Much of the economic literature insists on the negative impact of these regulations on the employment dynamics10. In addition, they raise the bargaining power of insiders, inducing an upward pressure on bargained wages. However, it is not clear if these regulations exert an effect as significant on the employment level (and therefore on unemployment) in equilibrium. A series of conflicting effects are involved here: inflows into unemployment are reduced by higher protection, while the effect on outflows from unemployment is unclear. Firms may be more cautious about hiring new staff, but the quality of the personnel may be improved by a reduction in the trial-and-error recruitment procedures11. Finally, by labour taxation we mean the tax wedge between the real product wage paid by the employer and the real consumption wage received by the employee12.This consists of the sum of the payroll tax rate, the income tax rate and the consumption tax rate. The question here is how much part of the increase in the tax wedge is born by the 5See also Nickell and Nunziata (2001), the description in Nunziata (2001) and the general discussion in Nickell et al. (2001). A general survey on labour market institutions is provided by Nickell and Layard (1999) while some previous discussions can be found in Layard et al.(1991), Bean (1994) and Nickell (1997). 6See Nickell et al. (2001) for a summary picture of the OECD union coverage patterns. 7See Nickell and Layard (1999) and Boeri et al. (2001). 8No long and comprehensive series on the coverage or on the strictness of the benefit provision system are available for all the countries. Nevertheless, these factors may be of a certain importance in explaining the experience of countries with generous benefits. It is the case of Denmark, where, according to the Ministry of Labour, the implementation of a tighter enforcement system in the 1990s was one of the main causes behind the large reduction in unemployment unaccompanied by any inflationary pressure (see Danish Ministry of Finance, 1999, Chapter 2). 9See Phelps (1994) for a broader discussion. 10See, for example, Bentolila and Bertola (1990), Nickell and Nunziata (2000) and Nunziata (2001). 11See Daniel and Stilgoe (1978) and Evans et al. (1985) for a discussion on the UK. 12The former is normalised on the GDP deflator, while the latter uses the consumer price index. 5 employers or shifted to the employees through a reduction in the after-tax consumption wage. In the first case both wages and unemployment increase, while in the second case labour costs and unemployment are unaffected. The extent to which wages and unemployment respond to changes in the tax wedge tells us which direction prevails. A similar real wage resistance effect can be generated by an adverse shift in the terms of trade or by an unexpected fall in trend productivity growth. 4 Some Previous Multi-Country Models Most of the existing empirical cross country literature on labour cost determination concentrates on the effects of taxation on wages. This is found to be shaped by the institutional configuration of the labour market, however this is defined13. Newell and Symons (1985) use marginal tax data in a simple cross country regression setting, finding a significant negative correlation between marginal tax rates and technical progress, real product wages, investment and labour force participation, and a positive correlation with the unemployment rate. The econometric analysis however is not robust, since the authors do not control for country effects. Alesina and Perotti (1997) propose a more sophisticated panel analysis of 14 OECD countries, for the period 1965 to 199014. They regress unit labour costs on labour tax, a centralization in wage bargaining dummy, total factor productivity and some other controls. The countries are divided into three groups according to their degree of centralization in wage bargaining, and the labour tax variable is interacted with the relative three indicator variables. The empirical evidence produced by the authors supports the view that an increase in labour taxation induces a labour cost increase in countries with an intermediate degree of bargaining centralization15. According to the authors, this result suggests that the relationship between taxation and wages is hump shaped. In other words, highly decentralized countries are characterized by lower distortions while if bargaining is highly centralized then unions internalize the welfare improvements associated with higher tax pressure, demanding lower pay increases16. However, given that all the variables included in the regressions are in log difference the model implies that a 10 percent rise in taxes from the average value in 1960-1990 induces a somewhat implausible 27 percent increase in unit labour costs in the intermediate countries. In addition consumption taxes are omitted from the labour tax variable, which is defined as the sum of direct taxes paid by households, social security taxes paid by employers and employees and payroll taxes as shares of GDP. This omission should be tested, since we should expect that a switch from income taxation to consumption taxation 13The empirical literature on wage determination is immense. A number of other relevant contributions are not mentioned here because of the different emphasis or methodology. See, for example, the macroeconomic papers of Alogoskoufis and Manning (1988) and Bean et al. (1986), as well as the literature on wage curves originated by Blanchflower and Oswald (1990, 1994) that concentrates especially on micro data. 14The countries in the sample are Australia, Belgium, Canada, Denmark, Finland, France, Germany, Italy, Japan, Netherlands, Norway, Sweden, United Kingdom and United States. The base equations include country and year effects. 15The countries belonging to the intermediate bargaining centralization group are Germany, Netherlands and Belgium. 16See Calmfors and Driffill (1988) and Freeman (1988) for evidence. 6 would not change wage earners real income17. Moreover, the authors’ assumption implies, quite implausibly, that by simply replacing payroll and income taxes by consumption taxes, we can reduce labour costs by a huge amount, thereby generating a huge reduction in unemployment. The main concern about the model is however that labour market institutions are taken into account simply dividing the countries into three groups, on the basis of one dimension only, namely bargaining centralization. In fact, several other features of labour market institutions are likely to be crucial in a model of labour cost. This approach also fails to consider the historical evolution of the characteristics of the bargaining process in each country. The changes can be relevant for some countries and could affect the country classification proposed by the authors. As an example, looking at the time series for coordination used in this paper, we see a 25 percent fall in coordination for Sweden from 1980 to 1995 and a 66 percent raise in Italy from 1990 to 1995. Daveri and Tabellini (2000) follow a similar approach in a study of the effects of taxation on unemployment and growth on a group of 14 OECD countries18.Theclaimof the paper is that higher taxation is correlated with greater labour costs and therefore with higher unemployment and a lower rate of growth. Although the theoretical arguments of the authors are in some way different from Alesina and Perotti, the conclusions are comparable. The analysis is constructed using five years averaged data in order to remove cyclical fluctuations, and the tax data is constructed using the methodology suggested by Mendoza, Razin and Tesar (1994). The countries are divided into three groups according to the characteristics of their bargaining system, where the composition of each group is different from the one adopted by Alesina and Perotti, possibly because the authors do not consider only centralization as a subdivision criteria, but other bargaining dimensions too19. The estimation results indicate that taxation has a positive significant impact on unemployment in the countries belonging to the intermediate group, namely continental Europe plus Australia. There is also a positive effect for Canada, US, Japan and Thatcher UK, but weaker. Regarding labour costs, the authors estimate two alternative models, using respectively data on the growth rate of gross real earnings in manufacturing, and the growth rate of net real wages of the average manufacturing worker. The reported 2SLS coefficients in the first more robust case indicate that higher taxation is correlated with higher labour costs in continental Europe and Australia, so that a unit change in the average tax is translated into a 0.49 percent increase of gross real wage. Being the methodology used by Daveri and Tabellini quite similar to the one adopted by Alesina and Perotti, similar comments apply here. In this case too, the tax data does not include consumption taxation, and the unemployment regressions are quite bald, with very few regressors. In addition, the composition of each country group, according to the institutional framework governing wage bargaining, suffers from a certain degree of subjectivity. Indeed, the choice of including one country in one group instead of another ultimately responds to the criteria implied by the authors. Even if we were to rule out subjectivity, there seems to be no reason why we should stick to one rule only. The subdivision criteria can be various and contradictory, with a likely relevant impact on the estimation results. 17See Nickell and Layard (1999). 18With respect to Alesina and Perotti, they include Spain in the sample and exclude Danemark. 19These are union density, union coverage and coordination. 7 Our model intends to respond to these problems and limitations. In what follows we provide an empirical test of the impact of institutions on labour cost determination in OECD countries, together with an investigation of the role played by labour market forces. 5 The Econometric Model We estimate a general version of equation (4) on a sample of 20 OECD countries observed from 1960 to 1994. The baseline specification of the empirical model is the following: wit −pit =β0+β1wit−1−β2Uit −β3∆Uit +β4πit +γ¯zw,it+λhit +ϑsit +φiti+µi+λt+εit (5) where the same notation used above applies here with wit −pit being the real unit labour cost (i.e. real unit wages, salaries and social security contributions paid by resident producers), Uthe unemployment rate (that we do not impose to be in logs), ∆Uthe difference in the unemployment rate, πthe productivity trend, ¯zwa vector of labour market institutions, ha vector of interactions among institutions and other regressors, sa vector of macroeconomic shocks, tia country specific time trend, µia fixed country effect, vtayeardummyandεit the stochastic residual20. More specifically, the vector of labour institutions includes: γ¯zw,it =γ1EPit +γ2BRRit +γ3∆UDit +γ4COit +γ5TWit (6) where EPit is employment protection, BRRit is the unemployment benefit replacement rate, UDit is net union density, COit is bargaining coordination and TWit is the tax wedge, i.e. direct + indirect + payroll tax rate. The vector of interactions in the baseline model has the following form: λhit =λ1URCO+λ2URBRRit +λ3URUDit +λ4TWCOit (7) where each element, denoted with obvious notation, is expressed as interaction between deviations from the world average ¯ ξ, i.e. : VarXVarY it =VarX it −¯ ξVarX·VarY it −¯ ξVarY.(8) Expressing the interactions in this way allows us to read the coefficient of each institution in levels as the coefficient of the ”average” country, i.e. of the hypothetical country characterized by the average level of that specific institutional indicator. The vector of controls for macroeconomic shocks contains the following elements: ϑsit =ϑ1D2TFPit +ϑ2TTSFCit (9) where D2TFP is a total factor productivity shock, calculated as the acceleration in productivity, ∆2TFP,andTTSFC is a terms of trade shock21. 20See the appendix for data definitions and sources. 21This is calculated as imports GDP ∆logP import P GDP where Pimport is the imports deflator and PGDP is the GDP deflator at factor cost. 8 BR EP ∆UD CO TW DWH test F(1,443) =.09 P-value≃0.76 F(1,443) =0.19 P-value≃0.76 F(1,443) =0.07 P-value≃0.79 F(1,443) =0.22 P-value≃0.64 F(1,443) =0.18 P-value≃0.67 Table 7: Durbin - Wu - Hausman tests 7 Estimation Results We present a set of alternative specifications of the labour cost model. These are: 1. the baseline model; 2. including ∆TW; 3. including TWt−1and ∆TWt−1, to check for a lagged effect of the tax wedge; 4. including Union Density, also interacted with unemployment rate and productivity; 5. including Oswald’s Home Ownership variable (Portugal excluded) which represents the proportion of owner occupier households and is a proxy for labour mobility; 6. excluding Portugal for comparison with the previous model; 7. wage equation (i.e. replacing unemployment with its log in the level, delta and interactions effects); 8. including a structural break in the productivity trend; 9. including further shocks; 10. using an alternative measure of bargaining coordination; 11. including coordination types dummies; 12. check 1 of Daveri and Tabellini result, dividing the countries into three groups according to their degree of bargaining coordination; 13. check 2 of Daveri and Tabellini result, dividing the countries into three groups according to their degree of bargaining centralization; 14. estimation on a subsample from 1970; 15. with log unemployment and unemployment growth rate; 16. with log unemployment and unemployment growth rate also interacted with coordination; 17. the static model; 18. the baseline model estimated by OLS. 15 (1) (2) (3) (4) (5) (6) (7) (8) (9) LABC LABC LABC LABC LABC LABC LABC LABC LABC LABC(-1) 0.697 0.696 0.711 0.707 0.687 0.696 0.708 0.694 0.688 [30.29] [29.98] [29.87] [31.71] [29.12] [30.10] [29.65] [29.76] [29.29] U-0.499 -0.476 -0.430 -0.505 -0.474 -0.495 -0.015 -0.488 -0.527 [7.12] [6.70] [6.09] [7.37] [6.77] [7.08] [4.38] [6.98] [7.44] EP 0.023 0.024 0.023 0.020 0.018 0.024 0.026 0.024 0.020 [4.77] [5.02] [4.66] [3.89] [3.40] [4.87] [5.10] [4.85] [3.79] BRR 0.037 0.041 0.039 0.028 0.037 0.035 0.038 0.035 0.044 [3.07] [3.40] [3.05] [2.26] [2.98] [2.84] [2.90] [2.90] [3.55] TW 0.116 0.090 0.071 0.107 0.107 0.108 0.087 0.115 0.101 [3.90] [2.71] [1.90] [3.63] [3.61] [3.59] [2.87] [3.83] [3.42] CO -0.026 -0.025 -0.026 -0.023 -0.024 -0.026 -0.045 -0.027 -0.029 [2.57] [2.45] [2.47] [2.17] [2.32] [2.54] [4.51] [2.63] [2.75] ∆UD 0.202 0.222 0.158 0.225 0.176 0.185 0.182 0.202 0.142 [2.73] [2.97] [2.11] [3.03] [2.33] [2.46] [2.36] [2.72] [1.92] URCO -0.193 -0.199 -0.157 -0.241 -0.197 -0.195 0.004 -0.186 -0.174 [2.68] [2.79] [2.17] [3.21] [2.74] [2.72] [1.29] [2.59] [2.33] URBRR 0.441 0.392 0.271 0.482 0.346 0.414 0.010 0.411 0.337 [2.08] [1.81] [1.25] [2.22] [1.67] [1.96] [1.01] [1.93] [1.61] URUD 0.405 0.374 0.327 0.410 0.466 0.428 0.001 0.390 0.391 [2.08] [1.88] [1.67] [2.02] [2.50] [2.28] [0.05] [2.01] [2.09] TWCO -0.141 -0.148 -0.094 -0.141 -0.127 -0.141 -0.103 -0.133 -0.147 [4.32] [4.60] [2.57] [4.41] [3.91] [4.31] [3.11] [4.10] [4.54] π0.471 0.469 0.444 0.445 0.503 0.467 0.441 0.468 0.505 [12.64] [12.28] [11.57] [12.41] [12.06] [12.53] [11.63] [12.60] [11.62] D2TFP -0.381 -0.401 -0.429 -0.359 -0.434 -0.401 -0.442 -0.399 -0.414 [3.95] [4.17] [4.41] [3.74] [4.47] [4.10] [4.38] [4.14] [4.25] TTS 0.364 0.381 0.311 0.287 0.372 0.368 0.330 0.365 0.307 [6.90] [7.21] [5.91] [6.11] [7.09] [6.99] [5.87] [6.92] [5.69] ∆TW 0.093 -0.067 [1.72] [1.25] π·UD 0.076 [4.64] Constant 2.718 0.948 1.026 -0.608 1.247 -0.105 0.717 -0.438 4.565 [0.44] [0.15] [0.61] [0.37] [0.69] [0.06] [0.41] [0.25] [1.64] D2MS -0.006 [0.73] RIRL 0.130 [3.72] π·d70 -0.002 [1.01] HOUSE 0.140 [1.80] UD 0.003 [0.11] Obs 572 570 569 572 553 553 561 572 559 Countries2020202019192020 20 av. T 29.58 29.55 29.48 29.58 29.95 29.95 29.12 29.58 29.09 Pval Cf=0 Pval f.e.=0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 LABC is log real labour cost per employee, i.e. real unit wages, salaries and social security contributions paid by resident producers. NOTES: Generalised least squares allowing for heteroskedastic errors and country specific first order serial correlation. Each equation contains country dummies, time dummies and country specific trends. (i) When interactions are included, the variables are set as deviations from the mean, so the interactions take the value zero at the sample mean. (ii) The variables unemployment (U), union density (UD), benefit replacement ratio (BRR), employment tax rate (TW), owner occupation (HOUSE) are proportions (range 0-1). Employment protection (EP) and co-ordination (CO) are indices (ranges 0-1.1, 0-2, 1-3 respectively). (iii) π is the productivity trend, D2TFP is the acceleration in TFP, TTS is the terms of trade shock, D2MS is acceleration in money supply, RIRL the real interest rate. z statistics in brackets. Table 8: OECD Labour Cost Models: 1960-1994 16 (10) (11) (12) (13) (14) (15) (16) (17) (18) LABC LABC LABC LABC LABC LABC LABC LABC LABC LABC(-1) 0.726 0.682 0.691 0.700 0.688 0.705 0.700 0.659 [32.79] [29.37] [28.67] [29.64] [23.53] [29.88] [29.16] [19.94] U-0.519 -0.536 -0.352 -0.414 -0.220 -0.455 -0.439 -0.772 -0.534 [7.17] [7.50] [5.09] [5.96] [2.87] [6.59] [6.32] [6.70] [5.00] EP 0.024 0.022 0.018 0.018 -0.004 0.025 0.027 0.037 0.003 [4.80] [4.53] [3.75] [3.74] [0.45] [5.56] [6.57] [4.36] [0.29] BRR 0.040 0.042 0.029 0.036 0.058 0.033 0.034 0.084 0.010 [3.27] [3.43] [2.23] [2.91] [4.04] [2.71] [2.80] [4.57] [0.55] TW 0.059 0.119 0.095 0.103 0.105 0.322 0.065 [2.07] [3.92] [2.98] [3.40] [3.46] [6.78] [1.46] CO -0.004 -0.027 -0.032 -0.026 -0.025 -0.028 -0.034 -0.141 -0.007 [1.01] [2.53] [2.79] [2.42] [2.35] [2.81] [3.38] [8.70] [0.43] ∆UD 0.132 0.243 0.218 0.205 0.271 0.236 0.242 0.341 0.272 [1.80] [3.30] [2.90] [2.73] [3.13] [3.14] [3.24] [2.68] [2.97] URCO -0.207 -0.193 -0.139 -0.132 -0.099 -0.157 -0.105 -0.110 -0.393 [3.34] [2.61] [1.89] [1.79] [1.44] [2.39] [1.71] [0.86] [2.75] URBRR 0.601 0.421 0.087 0.304 -0.337 0.339 0.347 0.750 0.261 [2.68] [1.98] [0.39] [1.39] [1.44] [1.61] [1.62] [2.40] [0.88] URUD 0.400 0.349 0.349 0.281 0.103 0.359 0.322 0.751 0.059 [2.11] [1.82] [1.80] [1.44] [0.48] [1.85] [1.63] [2.91] [0.24] TWCO -0.136 -0.139 -0.089 -0.133 -0.132 -0.152 -0.071 [4.05] [4.13] [2.46] [4.01] [3.93] [2.57] [1.28] π0.423 0.489 0.454 0.438 0.465 0.457 0.468 1.393 0.489 [11.51] [12.95] [12.17] [11.83] [6.44] [12.02] [11.80] [35.40] [8.52] D2TFP -0.420 -0.392 -0.439 -0.404 -0.379 -0.337 -0.345 -0.608 -0.328 [4.33] [4.08] [4.55] [4.22] [3.58] [3.49] [3.57] [3.75] [2.90] TTS 0.359 0.331 0.330 0.318 0.340 0.336 0.332 0.741 0.296 [6.57] [6.22] [6.24] [6.05] [6.17] [6.32] [6.24] [7.80] [4.39] ˆγUR -0.007 -0.000 0.006 [2.09] [0.09] [1.05] Constant -0.934 -0.335 0.492 -0.503 5.529 0.277 0.439 -3.329 -1.893 [0.56] [0.20] [0.30] [0.30] [1.80] [0.17] [0.27] [1.10] [0.92] ˆγUR·CO -0.012 [2.62] TW·Gunc 0.077 0.086 [2.23] [2.52] TW·Gint 0.105 0.094 [3.41] [2.91] TW·Gcoo 0.079 0.051 [2.67] [1.54] COT1 -0.022 [2.87] COT2 0.002 [0.37] COT4 0.009 [1.46] COT5 0.017 [2.33] COT6 0.002 [0.42] Obs 572 572 572 572 470 572 572 572 572 Countries2020202020202020 20 av. T 29.58 29.58 29.58 29.58 24.02 29.58 29.58 29.58 29.58 Pval Cf=0 Pval f.e.=0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 LABC is log real labour cost per employee, i.e. real unit wages, salaries and social security contributions paid by resident producers. NOTES: Generalised least squares allowing for heteroskedastic errors and country specific first order serial correlation, OLS in 18. Each equation contains country dummies, time dummies and country specific trends. (i) When interactions are included, the variables are set as deviations from the mean, so the interactions take the value zero at the sample mean. (ii) The variables unemployment (U), union density (UD), benefit replacement ratio (BRR), employment tax rate (TW) are proportions (range 0-1). Employment protection (EP) and coordination (CO) are indices (ranges 0-1.1, 0-2, 1-3 respectively). (iii) π is the productivity trend, D2TFP is the acceleration in TFP, TTS is the terms of trade shock, ˆγUR is the rate of growth of unemployment, COT1-COT6 are coordination type dummies. TW · Gunc, TW · Gint and TW · Gcoo are interactions between TW and dummies for, respectively, uncoordinated (decentralized), intermediate and coordinated countries in model 12, decentralized, intermediate and centralized countries in model 13. z statistics in brackets. Table 9: OECD Labour Cost Models: 1960-1994 (continued) 17 The estimation output is presented in Tables 8 and 9. All models are estimated using a fixed effect GLS estimator, corrected for by country heteroskedasticity and serial correlation, except for Model 18 that is estimated by OLS. The unemployment coefficient is negative and significant in all specifications. This confirms the theoretical scenario depicted in section 2, showing that the labour market adjustment mechanism is at work. Model 1, our benchmark specification, shows also that the impact of unemployment on wages is shaped by the labour market institutional configuration. The negative wage adjustment triggered by an increase in unemployment is reinforced by higher bargaining coordination, weaker union density and lower unemployment benefits. For what concerns the change in unemployment, we could not find any significance effect. As regards the level impact of wage pressure institutions, we find a positive significant impact of employment protection, benefit replacement rates, the tax wedge and of an increase in union density. The benefit duration indicator is instead never significant. Coordination has a negative effect on wages, both in level and through the interactions with the tax wedge and unemployment. This result is some way in line with the previous findings of Alesina and Perotti (1997) and Daveri and Tabellini (2000) according to which in presence of high bargaining coordination the impact of the tax variable is reduced. However, this specification does not allow to test for the hump shape hypothesis suggested by these authors. Alternative tests of the complementarity effects between the labour tax rate and coordination are therefore provided by Models 12 and 13. In the former model the countries are divided into three groups according to their degree of bargaining coordination, and the resulting indicator variables are interacted with taxation33.Gunc is the dummy for the group of uncoordinated countries, characterized by a coordination level CO < 1.5. Gint is the indicator for the intermediate countries, with 1.5≤CO ≤2, and Gcoo is the indicator for highly coordinated countries with CO > 2.Differently from the analysis of the previous papers, here the countries do not belong necessarily to one group for the whole sample period, but they transfer from one group to another according to the institutional changes occurred from the 1960s to the 1990s. The hump shape hypothesis is only partially confirmed by our analysis. In fact, although the coefficient of intermediate countries is the largest, the tax variable has a positive significant impact on all groups of countries and not on the intermediate group only. In addition, although we cannot reject the hypothesis that the three coefficients differ across the three country groups, the long term impact of taxation on labour cost is very similar. For example, a 10 percentage points rise in tax induces a 2.5 percent increase in labour cost in the not coordinated and coordinated countries and a 3.2 percent increase in the intermediate countries. The test provided by Model 12 is not strictly comparable with the one presented by Alesina and Perotti since they classify the countries according to the degree of centralization in wage bargaining and not of coordination34. In order to allow for comparisons Model 13 performs the same test using a time varying centralization indicator for the con33The results below depend only marginally on the definition of coordinated, intermediate and uncoordinated countries, i.e. on the thresholds we use to construct the group dummies. The criteria reported here are the most sensible in order to have enough observations in each group. However, if we adopt alternative reasonable thresholds’ values we obtain almost identical coefficients and confidence intervals. 34These two variables are positively correlated and have a comparable mean and standard deviation over the sample. Coordination is however more general than centralization, as the latter simply refers to the level at which wages are bargained (plant, firm, industry or economy-wide). 18 struction of the three dummies. The results are again only moderately in line with the hump shape hypothesis. The impact of the tax variable is not significant in centralized countries only while in decentralized and intermediate countries wages are significantly affected by employment taxation. In this case a 10 percentage points rise in tax induces a 3 percent increase in labour cost in the decentralized and intermediate country groups. As expected, the productivity trend has a large positive impact on wages. The long run coefficient is greater than one, suggesting that the long run elasticity of substitution between capital and labour is lower than one35. The effect is reinforced if unions are stronger, as in Model 436. The TFP and the terms of trade shocks have a significant impact on wages, respectively of negative and positive sign, as expected. This means that labour cost fails to accommodate an acceleration of total factor productivity above the trend while an adverse shock in relative import prices induce workers to maintain their real net incomes, increasing wage pressure. The TFP and the real wage resistance effect are robust to the introduction of another set of shocks, namely the real interest rate and the acceleration in money supply, as in Model 9. The coefficient of the latter variable is not significant, while the interest rate coefficient is significant with positive sign. This finding may be interpreted as a corroboration of the hypothesis suggested above of a positive relation between real interest rate and non human wealth of the unemployed. In Model 10 we substitute the coordination variable with the measure provided by Nickell et al. (2001) that tries to capture the various detailed nuances of the variations in coordination. In this specification the coordination variable is not significant in level, but still has a significant effect in reinforcing the unemployment wage adjustment mechanism and in reducing the wage pressure impact of the tax variable. Model 15 includes the unemployment growth rate among the explanatory variables. Its significance with expected sign reinforces the importance of the labour market adjustment mechanism. Moreover, Model 16 shows that the impact of unemployment growth is strengthened when wage bargaining is coordinated, although the change in unemployment becomes insignificant in levels37. Model 11 incorporates a set of dummies that account for changes in types of coordination, based on a taxonomy due to Traxler (1996) and Traxler and Kittel (2000). We include five of the six categorical variables originally set by these authors38.Theseare: CO1=inter associational coordination, i.e. coordination by the major confederations of employers and labour; 35In analytical terms, we can rewrite a general profit maximizing condition AF1AN K,1=W Pκ where Nis labour, Kis capital, Wis the labour cost per employee, Pis the GDP deflator in aggregate and κ=(1−1/η)whereηis the output demand elasticity. Being F1homogeneous of degree zero in its two arguments, after eliminating Kwe obtain N=Y AhW AP κ ,with h<0.Assuming κconstant, this can be written in log linear form as (w−p)=α0+1 σ(y−n)+β 2 σa,whereσis the long run elasticity of substitution. Our estimates imply that σ<1. This means that if wages grow faster then productivity, the labour share increases. 36This result is in line with previous findings in the literature. For example, Booth et al. (1999) identify a positive impact of unions on wages and productivity through greater gains from training. 37We must be however careful in interpreting these results since using unemployment in log form is problematic, as noted at page 4. 38We exclude CO3, pattern setting coordination to avoid multicollinearity. 19 CO2=intra associational coordination, i.e. within the major confederations of employers and labour; CO4=state imposed coordination; CO5=non-coordination; CO6=state sponsored coordination, i.e. with the state joining the bargaining process as an additional party. The model confirms again a significant positive impact of an uncoordinated bargaining framework on wages. In addition, the type of coordination that seems to be most relevant in reducing wage pressure is the inter associational39. Model 2 tests for the significance of changes in taxation and Model 3 tests for lagged effects in taxation. In both cases, the variables are insignificant. Model 5 tests for the impact of the home ownership variable, suggested by Oswald as a proxy for labour mobility40. The idea here is that if a larger share of households are house owners, then workers will be more reluctant to migrate to different regions in order to find a job. This will reduce the labour market adjustment speed in presence of local adverse shocks, preventing one important allocation mechanism, namely labour mobility, to take place. In our model the home ownership variable has a positive and significant (at the 10% level) impact on wages, suggesting that lower mobility may also have a wage pressure effect, through the rigid geographical localization of the labour force41.InModel6wedropthe observations from Portugal, in order to make a comparison with Model 5, since no home ownership data is available for that country. As we can see, the model is quite stable after we drop these observations42. In Model 7 we substitute unemployment with its log in order to compare our findings with the wage equation of Blanchflower and Oswald (1990, 1994). The long run elasticity of wages with respect to unemployment is 5 percent, i.e. half of the one estimated by Blanchflower and Oswald43. This result is not substantially different if we estimate the static version of the model or if we limit our sample to the period 1970-1994. In Model 8 we test the significance of a split trend in productivity after 1970, interacting PROD with a dummy equal to 1 for all years after 1970. This is to check if we are able to capture with a split trend in productivity the rapid growth of the share of labour in most countries in the 1960s. The interaction is insignificant, although it is negative as we would expect. Finally, the labour cost equation is estimated in its static form in Model 17. This is another implicit test of the Pesaran - Smith bias in dynamic heterogeneous panels. The implications of the benchmark model are actually reinforced by the static model. Overall, the empirical results are robust to different specifications and are coherent with the findings of the unemployment model in the companion paper Nickell et al. The institutions and the unemployment rate have a significant impact on OECD wages, once we control for the productivity trend. The next step is then to establish what proportion 39The same results are obtained if we drop the coordination variable, CO, from the model. 40See Oswald (1996). 41The real impact of labour mobility may be stronger than the one estimated in Model 5,as our results may suffer from the high interpolation of the home ownership variable. 42This exercize is also useful in order to control for possible ”Portugal effects” due to the impact of the Portugal revolution in the 1970s. 43This result is however in line with other studies, see for example the analysis of Galiani (1999) for Argentina. 20 of the wage dynamics from 1960 to 1994 can actually be imputed to changes in institutions and unemployment given productivity. Next section provides some dynamic simulations of the baseline model in order to answer to this question. 8 Dynamic Simulations We present a summary of the results of a set of dynamic simulations performed on the baseline version of the labour cost model. The scope of the dynamic simulations is the measurement of the impact on wages of the evolution of labour market institutions and unemployment from 1960 to 1994 in OECD countries. The simulations are produced generating a labour cost series for each country through a recursive procedure that substitute the lagged dependent variable with the previous year prediction of the baseline model. The simulated series are then compared with the one obtained using the same method but constraining one or more regressors at their average 1960s value. In this way we calculate the variation in labour cost that can be imputed to the evolution of the specific regressors over the estimation period. Obviously, the major part of the rise in wages is due to the productivity trend and relative to this institutions are not important. What is important, however, are the impacts of institutions once we control for productivity. Figures 1 and 2 show, respectively, the contribution of the changes in institutions and unemployment (from the 1960s average) to the changes in labour costs in the 1980s and in the 1990s44. From the figures it emerges quite clearly that the role of unemployment is to offset the impact of institutions. In other words, given that wages cannot move too far away from the productivity trend, unemployment has to increase in order to balance the pressure on wages induced by institutions. It follows that the change in unemployment tends to be bigger when the institutional effects are bigger in the opposite direction. Although the role played by each institution varies across countries and time, we see that, on average, the major labour cost changes are generated by the benefit replacement ratio, taxation and employment protection. Indeed, the simulations display a significant institutional impact in countries that were characterized by a noticeable variation in institutions over the sample period. For example, employment protection regulations do not play any role in countries, such as the non-European and Switzerland, where these regulations did not change substantially from the 1960s. On the other hand, stricter employment protection regulations triggered a wage rise in countries like Portugal, Sweden, Germany and France. Higher benefit replacement ratios have increased wages in most of the countries in the sample. On the contrary, the fall in the UK ratios had instead a negative impact in that country. For what concerns coordination, the simulations do not show any effect for the group of countries in which coordination remained stable over the period. However, the increase in coordination in the 1990s was particularly relevant in moderating labour costs in Ireland, France and Italy. 44Note that, in order to make the simulation figures more informative, they show the predicted percentage change in actual real labour cost, and not in the series in logs. 21 Real Labour Cost % change 24 16 8 0 -8 -16 -20 Unemployment Benefits Tax Wedge Coordination Union Density Employment Protection Unemployment Rate AL AU BE CA DK FN FR GE IR IT JA NL NW NZ PG SP SW SZ UK US Figure 1: Dynamic simulations with regressors fixed at 1960s average values: percentage changes in labour cost in the 1980s imputed to specific institutional dimensions Real Labour Cost % change 36 32 24 16 8 0 -8 -16 -24 -32 Unemployment Benefits Tax Wedge Coordination Union Density Employment Protection Unemployment Rate AL AU BE CA DK FN FR GE IR IT JA NL NW NZ PG SP SW SZ UK US Figure 2: Dynamic simulations with regressors fixed at 1960s average values: percentage changes in labour cost in the 1990s imputed to specific institutional dimensions 22 9 Concluding Remarks This paper presents an empirical investigation of the determinants of labour cost in 20 OECD countries observed from 1960 to 1994. The regressions are estimated using a fixed effects estimator in GLS form, correcting for heteroskedasticity and residual serial correlation. We include year dummies in order to control for contemporaneous correlations and we test for panel cointegration, endogeneity and poolability issues among other things. The main findings of the paper are the following: •The unemployment rate has a negative and significant impact on labour cost. Autonomous shocks on unemployment can therefore be offset by labour market forces since higher unemployment triggers a reduction in wages. The estimated long run elasticity of wages with respect to unemployment is 5 percent, i.e. half the one estimated by Blanchflower and Oswald (1990, 1994). •Labour market institutions have a significant impact on wages. The impact can be either direct, inducing wage pressure despite excess supply in the labour market, or indirect, influencing the matching process of unemployed individuals to available job vacancies. •The effect of institutions is coherent with the results of the unemployment model estimated in the companion paper Nickell et al. (2001). These findings confirm the theoretical scenario depicted in the first part of the paper where equilibrium unemployment is determined by autonomous wage pressure institutions. •The productivity trend plays a major role in the evolution of wages in OECD countries in the last decades. The employees’ productivity gains seem to be higher if unions are stronger. The model does not confirm a structural break in the productivity coefficient after the 1970s. •Taxation has a positive impact on labour cost. The model confirms therefore the existence of a shift effect from employees to firms, although smaller than the one estimated by previous papers. The tax effect is moderated by high coordination in wage bargaining. The Daveri and Tabellini hypothesis of major impact of the tax wedge in countries with intermediate degree of coordination is only partially confirmed by our analysis. In fact, the tax wedge is significant in uncoordinated, intermediate as well as coordinated countries, and the impact of taxes seems only to be vaguely hump shaped. •The institutions governing the bargaining process play a major role in shaping wages, both directly as well as through their interaction with unemployment and taxation. Bargaining coordination has a direct negative effect on labour costs, and an indirect effect in reducing the positive impact of taxation and increasing the negative impact of unemployment. •Stronger unions increase wages, although their adverse effect can be offset by high coordination in wage bargaining. 23 •Higher benefit replacement rates have a significant and positive effect on labour cost. They increase workers’ utility in case of the unemployment fall back position raising the average reservation wage. In addition, they reduce the effect of labour market adjustment forces induced by higher unemployment. •The impact of benefit duration is insignificant in all model specifications. •Employment protection has a positive impact on wages, possibly through the increase in the insiders’ rent positions. •Macroeconomic shocks have a significant effect on wages. In particular, the term of trade shock has a positive effect while the acceleration in TFP growth has a negative impact, showing a certain degree of real wage resistance in OECD economies. •The home ownership variable, proposed by Oswald as a proxy for low labour mobility has a positive impact on wages. This suggests that if mobility rates are reduced, wages are likely to rise in order to accommodate local labour demand pressure. 24 Benefit Replacement Ratios data is provided by OECD with one observation every two years for each country in the sample. The data refers to first year of unemployment benefits, averaged over family types of recipients, since in many countries benefits are distributed according to family composition. The benefits are a proportion of average earnings before tax. BD: Benefit Durations An index was constructed by the authors as a weighted average equal to: BD =αBRR 2 BRR 1 +(1−α)BRR 4 BRR 1 where: BRR1is the unemployment benefit replacement rate received during the first year of unemployment, BRR2is the replacement rate received during the second and third year of unemployment and BRR4is the replacement rate received during the fourth and fifth year of unemployment. Note that we give more weight to the first ratio than to the second (α=0.6). If benefit provision stops after one year (min duration), then BRR2=BRR4=0,andBDAV = 0. If however benefit provision is constant over the unemployment years, we have BRR2=BRR4=BRR1,withBDAV =1. TW: Tax Wedge The tax wedge is equal to the sum of the employment tax rate, the direct tax rate and the indirect tax rate: TW =t1+t2+t3. The employment tax rate t1 is calculated as t1= EC (IE −EC) with EC equal to employers’ total contributions and IE equal to wages, salaries and social security contributions. The direct tax rate t2 is equal to t2t=DTt HCRt where DT is the amount of direct taxes and HCR is the amount of households’ current receipts. The indirect tax rate is t3t=TXt−SBt CCt with TX being total indirect taxes, SB subsidies, and CC private final expenditures. All data come from London School of Economics CEP - OECD database, updated by authors using the same criteria. Generally, the data are taken from OECD National Accounts. 31 HO: Housing (proportion owner occupied) The data about owner occupier households, as percentage of total households, is provided by Oswald (1996). This data was recently updated by Oswald, so that the 1990 figure for Spain is 78% not 75%, and the 1960 figure is 51% not 52%, and the 1990 figure for Ireland is 78% not 81%. Oswald data contain information about 19 countries (Portugal excluded) over 4 periods (1960, 1970, 1980 and 1990). This dataset contains an interpolation of Oswald’s data. B.2 Macroeconomic Variables UR: Unemployment Rate Standardized unemployment rate from Layard et al. (1991), updated by authors using the same criteria (see OECD Employment Outlook, 2000, Table A). The missing observations for Portugal are from CEP - OECD database. Data for Italy are based on the US Bureau of Labor Statistics, ”unemployment rates on US concepts”. LABC:LabourCost Labour cost is defined as w−p=log(IE)−log (ET)−log (PGDP ) where IE are compensations of employees by resident producers, namely wages, salaries and social security contributions, ET is total employment and PGDP is GDP deflator at factor cost. All data is from CEP - OECD database, updated by authors using the same criteria, except for PGDP that is calculated from GDP at factor cost, current and constant prices, from OECD Business Sector Database. EPOP: Employment Population Ratio Total civilian employment normalised on the working age population (15-64), from CEP OECD data, updated by authors. RIRL:RealInterestRate Long term real interest rate, constructed using long term nominal interest rate and inflation from OECD Economic Outlook Database. PRODHP: Trend Productivity 32 The labour productivity measure, πin the paper, was calculated as an Hodrick Prescott trend of log real GDP minus log of total employment, i.e. PRODHP = HPtrend {log (YQ)−log (ET)} where ET is total employment, and YQis real GDP at 1990 prices. D2MS: Acceleration in Money Supply This is equal to ∆2ln (MS), where MS is money supply from OECD Economic Outlook Database. LDS: Labour Demand Shock This series consists of the residuals ˆεtof the following 20 by country regressions: log (ETt)=β0+β1log (ETt−1)+β2log (ETt−2)+β3log (ETt−3)+β4log (YQ t)+β5log (WTPt)+εt where ET is total employment and YQand WTP are respectively real GDP and real labour cost at 1990 prices. TTS: Terms of Trade Shock This series is equal to IMP =MC YC∆log Pm PG where MC are imports at current prices, YC is GDP at current prices, Pmis import price deflator and PGis GDP deflator at market prices, both with 1990 as base year. The dataset contains also a factor cost version of the shock (TTSFC), calculated using the series on GDP deflator at factor cost. D2TFP: acceleration in TFP This series is calculated as a three years moving average of ∆2TFP,whereTFPis the series from Nickell and Nunziata (2000). This is based on the Solow residual for each country. TFPHPC: HP cyclical component of TFP This series consists of the Hodrick Prescott cyclical component of the Solow residual, TFPCY = Solow residual - HP trend . 33