Labor demand in a macro-econometric model: Neo-classical vs. Keynesian specifications Dr. Heike Joebges IMK Summer School August 7th, 2009 www.boeckler.de
Why do we care about the labor demand? Policy implications! What is the best response to the current crisis for Germany? Wage moderation and further labor market reforms or wage increases coupled with minimum wage introduction etc.? What are the implications for the euro area? 2
Do high wages/wage increases cause unemployment? YES! The prevailing view in Germany Up to the current crisis: praise of past wage moderation & past labor market reforms (e.g. Deutsche Bundesbank (2007), SVR (2007), European Commission (2007), OECD (2008)) Since the current crisis: growing calls for more wage moderation in order to save employment (e.g. organization of employers, research institutes, ) 3
Do high wages/wage increases cause unemployment? NO! Looking at case studies for big countries Relatively higher wage increases go in line with higher employment growth higher consumption & domestic demand growth higher GDP growth yet: less export growth Only for small countries, the reverse holds. 4
Outline Motivation Country comparisons Macro-econometric analysis Neo-classical employment equation Keynesian employment equation Simulation results Summary 5
Outline Motivation Country comparisons Macro-econometric analysis Neo-classical employment equation Keynesian employment equation Simulation results Summary 6
Country comparisons:* Labor market Total economy, 1999=100 Compensation of employees per hour Employment (in hours) 150 106 AT 140 130 NL UK 1 FR 104 102 NL UK 1 FR 120 AT 100 110 DE 98 DE 100 1999 2001 2003 2005 2007 96 1999 2001 2003 2005 2007 AT = Austria, DE = Germany, FI = Finland, FR = France, NL = Netherlands, UK = Great Britain 1 national currency Quelle: Reuters EcoWin (Eurostat-national accounts); IMK-calculations. 7 *Selected countries with similar labor costs (Joebges et al. 2008)
Country comparisons:* Consumption & Exports Total economy, 1999=100 Exports (real) Private Consumption (real) 220 200 NL DE 180 160 140 AT UK 1 120 FR 100 1999 2001 2003 2005 2007 130 UK 1 120 FR AT 110 NL DE 100 1999 2001 2003 2005 2007 AT = Austria, DE = Germany, FI = Finland, FR = France, NL = Netherlands, UK = Great Britain 1 national currency Quelle: Reuters EcoWin (Eurostat-national accounts); IMK-calculations. 8 *Selected countries with similar labor costs (Joebges et al. 2008)
Country comparisons:* GDP & domestic demand Total economy, 1999=100 Gross domestic product (real) Domestic demand (real) 130 125 UK 1 AT 120 115 FR 110 DE 105 NL 100 1999 2001 2003 2005 2007 130 UK 1 125 FR 120 NL 115 110 AT 105 100 DE 95 1999 2001 2003 2005 2007 AT = Austria, DE = Germany, FI = Finland, FR = France, NL = Netherlands, UK = Great Britain 1 national currency Quelle: Reuters EcoWin (Eurostat-national accounts); IMK-calculations. 9 *Selected countries with similar labor costs (Joebges et al. 2008)
Outline Motivation Country comparisons Macro-econometric analysis Neo-classical employment equation Keynesian employment equation Simulation results Summary 11
Econometric analysis of labor demand How to specify the employment equation? How does the employment equation fit in the macroeconometric model? Macro-model sets the frame for the equation! 12
Overview: Macro-econometric Models Theory based models for policy simulations (Quest II, Multimod): sound theoretical foundation; partly calibrated coefficients; Data based models for short-term forecasting (OFCE, Fair): try to fit data as well as possible, theoretical foundation is secondary; Forecasting and policy simulations (NIGEM, OEF): different equations for different purposes, i.e. estimated and calibrated coefficients. 13
IMK-Model for Germany Focus Short- to medium-term macroeconomic forecasts Analysis of different macroeconomic policies Structural model (47 stochastic equations) Equations guided by economic theory, but good datafit is necessary No calibration Same equations for forecasting and for policy simulations National Accounts Statistics raw data 14
IMK-Model for Germany Philosophy Based on Keynesian/New-Keynesian elements Crucial difference between short- and long-term Short-term: prices and wages only partly flexible; Long-term: adjustment mechanism towards steady state (adjustment of prices, wages, ) Real effects of economic policy Existence of unemployment in the long run Existence of nominal rigidities Market spillovers 15
IMK-Model for Germany Estimation approach Analysis of time series properties Single error correction equations Tests for serial correlation Stability tests Evaluation of the forecasting quality of the stochastic equation (dynamic in-sample and out-of-sample forecast) Evaluation of the behavior of the stochastic equation inside the model (ex post simulation) 16
IMK-Model for Germany Data Data All data in logs (excluding: rates, ratios, dummies) Quarterly data Raw data, not seasonally adjusted All data from national accounts, starting 1980 Q1 Almost all variables are I(1) Structural breaks! (Reunification, European Monetary Union, ) 17
Outline Motivation Country comparisons Macro-econometric analysis Neo-classical employment equation Keynesian employment equation Simulation results Summary 18
I) Neo-classical equation: Theory In the short run, employment is driven by demand factors, but in the long run, only by supply factors: Real output Real wage costs Real user costs of capital or: relative labor to capital costs (the ratio of real wage costs and real user costs of capital) Wage moderation increases employment 19
I) Neo-classical equation: Variables Employment: Persons employed [Hinz/Logeay (2006): hours worked] Real output: Real GDP Real wage costs: Compensation of employees per hours worked* Proxy for producer prices: GDP-deflator Real user costs of capital [Barrel et al. 1996]: Deflator of non-residential private investment Real interest rates (short-term: 3m; long-term: 10y) 20 *including: income tax & social security taxes of both employers & employees
I) Neo-classical equation: Results for the co-integration relation: Real GDP elasticity can be restricted to 1 Real wage elasticity is significantly negative, point estimate about -0,3 for persons employed [-0,6 for hours worked]* No substitution effect** (relative factor price elasticity is not significant for reunified Germany) System approach (VECM) confirms the elasticiy estimates and the single equation approach (weak exogeneity of real wages and real output) 21 *Hinz/Logeay 2006; **in constrast to Barrell et al. 1996
Outline Motivation Country comparisons Macro-econometric analysis Neo-classical employment equation Keynesian employment equation Simulation results Summary 22
II) Keynesian equation: Theory Aggregate demand for goods and services determines supply and thereby employment In the short-run, labor is the only mobile production factor; in the medium-run, capital stock adjustment Employment demand depends on total demand and the capital stock Unemployment is the result of insufficient demand, which could be raised by economic policy Expansionary economic policy increases employment 23
II) Keynesian equation: Variables Employment: Persons employed Real output: Real GDP Real capital stock: Real capital stock (last period) (Indicator construction: start value (DESTATIS) plus real investment excluding construction investment & real depreciation) 24
II) Keynesian equation: Results Real GDP-elasticity can be restricted to 1 Capital stock elasticity is significantly negative point estimate: -0,5 25
Comparing the employment equations: Co-integration relation Neo-classical equation: Employment = real GDP -0,3*real wage + trend + c Keynesian equation: Employment = real GDP -0,5*real capital stock +trend+c 26
Comparing the employment equations: Which equation performs better? Hard to discriminate between Keynesian/ Neo-classical employment equations Both equations perform well with regards to test statistics, robustness, & out-of-sample performance! Slightly better out-of-sample performance for the Keynesian equation, esp. starting 2001 Yet: implications for employment differ enormously! 27
Outline Motivation Country comparisons Macro-econometric analysis Neo-classical employment equation Keynesian employment equation Simulation results Summary 28
Simulation: Negative wage* shock** 2 0-2 % -4-6 Neo-classical employment equation Neoklassische Beschäftigungsgleichung -8-10 Keynessche Keynesian Beschäftigungsgleichung employment equation 1 2 3 4 5 6 7 8 9 10 29 *compensation per employee; **differences to baseline scenario in percent
Simulation: effect on employment* 2,0 1,5 1,0 0,5 Neoklassische Beschäftigungsreaktion Neo-classical employment equation % 0,0-0,5-1,0 Keynesian employment equation Keynesianische Beschäftigungsreaktion -1,5 1 2 3 4 5 6 7 8 9 10 30 *differences to baseline scenario in percent
Simulation: effect on real GDP* 0,5 0,0 % -0,5 Neo-classical Neoklassische employment Beschäftigungsgleichung equation -1,0 Keynessche Keynesian Beschäftigungsgleichung employment equation -1,5 1 2 3 4 5 6 7 8 9 10 31 *differences to baseline scenario in percent
Simulation: effect on real private consumption* 0,0-0,5-1,0 % -1,5-2,0-2,5 Neo-classical Neoklassische employment Beschäftigungsgleichung equation -3,0-3,5-4,0 Keynesian Keynessche Beschäftigungsgleichung employment equation -4,5 1 2 3 4 5 6 7 8 9 10 32 *differences to baseline scenario in percent
Simulation: effect on real exports* 3,5 3,0 2,5 Keynesian employment equation Keynessche Beschäftigungsgleichung % 2,0 1,5 Neoklassische Beschäftigungsgleichung Neo-classical employment equation 1,0 0,5 0,0 1 2 3 4 5 6 7 8 9 10-0,5 33 *differences to baseline scenario in percent
Simulation: effect on consumption deflator* 0,5 0,0 1 2 3 4 5 6 7 8 9 10-0,5-1,0 % -1,5-2,0 Neo-classical employment equation Neoklassische Beschäftigungsgleichung -2,5-3,0-3,5 Keynesian employment equation Keynessche Beschäftigungsgleichung 34 *differences to baseline scenario in percent
Outline Motivation Country comparisons Macro-econometric analysis Neo-classical employment equation Keynesian employment equation Simulation results Summary 35
Summary Neo-classical vs. Keynesian employment equation: Both equations perform equally well with regards to test statistics, robustness, & out-of-sample performance! Hard to discriminate between the two equations Out-of-sample performance is slightly better for Keynesian equation, esp. from 2001 onwards Reaction to shocks is similar, if oil price or demand shocks are modeled; only wage shocks make a huge difference (and only for employment) 36
Summary (continued) Arguments in favor of the Keynesian equation: Stagnating employment after years of wage moderation cannot be explained with the neoclassical employment equation* Model is better fitting the data Future research: Better discrimination, if equation is specified for hours worked instead of persons employed? 37 *...neither with labor market institutions (Bassanini et al. 2006)
Thank you! 38
Neoclassical employment equation Dependent Variable: DLOG(DE_EE-DE_RES_EE) Method: Least Squares Date: 11/14/06 Time: 13:17 Sample (adjusted): 1981Q2 2005Q4 Included observations: 99 after adjustments Variable Coefficient Std. Error t-statistic Prob. LOG(DE_EE(-1)-DE_RES_EE(-1))-LOG(DE_GDP00(-1)) -0.220436 0.021937-10.04837 0.0000 LOG(DE_COEE(-1))-LOG(DE_PGDP00(-1)) -0.064728 0.013277-4.875031 0.0000 C 1.246657 0.132436 9.413243 0.0000 @TREND -0.000601 7.52E-05-7.995935 0.0000 S91Q1 0.015363 0.002674 5.745648 0.0000 I91Q1 0.282305 0.005508 51.25275 0.0000 DLOG(DE_EE(-3)-DE_RES_EE(-3)) -0.098304 0.029583-3.322965 0.0013 DLOG(DE_EE(-4)-DE_RES_EE(-4)) 0.515623 0.057450 8.975184 0.0000 DLOG(DE_GDP00(-1))+DLOG(DE_GDP00(-2)) -0.050927 0.009478-5.372937 0.0000 DLOG(DE_GDP00) 0.095358 0.018839 5.061848 0.0000 I91Q1(-3) 0.031576 0.009713 3.250836 0.0016 I91Q1(-4) -0.164672 0.018881-8.721483 0.0000 R-squared 0.994099 Mean dependent var 0.003879 Adjusted R-squared 0.993353 S.D. dependent var 0.033119 S.E. of regression 0.002700 Akaike info criterion -8.877855 Sum squared resid 0.000634 Schwarz criterion -8.563295 Log likelihood 451.4538 F-statistic 1332.462 Durbin-Watson stat 1.815085 Prob(F-statistic) 0.000000 39
Neoclassical employment equation Prognoseguete (dynamic, in-sample, Sample (adjusted): 1981Q2 2005Q4) Root Mean Squared Error 166.7584 Mean Absolute Error 123.8024 Mean Absolute Percentage Error 0.391491 Theil Inequality Coefficient 0.002712 Bias Proportion 0.001645 Variance Proportion 0.006768 Covariance Proportion 0.991587 Prognoseguete (dynamic, out-of-sample, Sample (adjusted): 1981Q2 2001Q4) Root Mean Squared Error 170.4378 Mean Absolute Error 148.4559 Mean Absolute Percentage Error 0.428285 Theil Inequality Coefficient 0.002451 Bias Proportion 0.532123 Variance Proportion 0.002521 Covariance Proportion 0.465356 Prognoseguete (dynamic, out-of-sample, Sample (adjusted): 1981Q2 2000Q4) Root Mean Squared Error 433.1084 Mean Absolute Error 409.7640 Mean Absolute Percentage Error 1.178743 Theil Inequality Coefficient 0.006181 Bias Proportion 0.895106 Variance Proportion 0.008998 Covariance Proportion 0.095896 Prognoseguete (dynamic, out-of-sample, Sample (adjusted): 1981Q2 1999Q4) Root Mean Squared Error 231.9727 Mean Absolute Error 191.9230 Mean Absolute Percentage Error 0.551352 Theil Inequality Coefficient 0.003317 Bias Proportion 0.382732 Variance Proportion 0.028472 Covariance Proportion 0.588795 40
Neoclassical employment equation 30 1.2 20 10 0-10 -20-30 84 86 88 90 92 94 96 98 00 02 04 1.0 0.8 0.6 0.4 0.2 0.0-0.2 84 86 88 90 92 94 96 98 00 02 04 36000 35600 35200 34800 34400 34000 33600 33200 32800 CUSUM 5% Significance In-Sam ple Prognosen "eq_de_ee_5" 95 96 97 98 99 00 01 02 03 04 05 CUSUM of Squares 5% Significance 41 YH AT YH AT+1.96*YH ATS E YH AT-1.9 6*YH ATSE de_ee Abh. Bes chaeftigte, Inl.
Neoclassical employment equation Out-Of-Sample Prognosen "eq_de_ee_5" Out-Of-Sample Prognosen (2) "eq_de_ee_5" 36400 36000 36000 35600 35200 35500 35000 34800 34500 34400 34000 33600 34000 33500 33200 95 96 97 98 99 00 01 02 03 04 05 33000 95 96 97 98 99 00 01 02 03 04 05 Out-Of-Sam ple Prognosen (3) "eq_de_ee_5" Out-Of-Sample Prognosen (4) "eq_de_ee_5" 36400 36000 36000 35600 35200 35500 35000 34800 34500 34400 34000 33600 34000 33500 42 33200 95 96 97 98 99 00 01 02 03 04 05 33000 95 96 97 98 99 00 01 02 03 04 05
Keynesian employment equation Dependent Variable: DLOG(DE_EE-DE_RES_EE) Method: Least Squares Date: 11/20/06 Time: 15:21 Sample (adjusted): 1981Q2 2005Q4 Included observations: 99 after adjustments Variable Coefficient Std. Error t-statistic Prob. LOG(DE_EE(-1)-DE_RES_EE(-1))-LOG(DE_GDP00(-1)) -0.212295 0.022403-9.476184 0.0000 C 1.872430 0.207688 9.015596 0.0000 LOG(DE_CSTOCK00(-1)) -0.109144 0.012811-8.519472 0.0000 S91Q1 0.031735 0.004130 7.684436 0.0000 I91Q1 0.274198 0.006119 44.80760 0.0000 DLOG(DE_EE(-4)-DE_RES_EE(-4)) 0.362250 0.068722 5.271250 0.0000 DLOG(DE_GDP00(-1))+DLOG(DE_GDP00(-2)) -0.029472 0.011703-2.518239 0.0137 DLOG(DE_GDP00) 0.116759 0.027073 4.312782 0.0000 I91Q1(-4) -0.113091 0.022755-4.969905 0.0000 Z1-0.009537 0.003208-2.973324 0.0039 Z2 0.005531 0.001631 3.391565 0.0011 Z3 0.001819 0.001777 1.023465 0.3091 S91Q1*Z1-0.002478 0.002278-1.087990 0.2797 S91Q1*Z2-0.004771 0.001644-2.902363 0.0047 S91Q1*Z3-0.003815 0.001968-1.938080 0.0560 DLOG(DE_CSTOCK00)+DLOG(DE_CSTOCK00(-1))+DLOG(DE_CSTOCK00(-2))+DLOG(DE_CSTOCK00(-3)) 0.096677 0.016744 5.773985 0.0000 R-squared 0.995102 Mean dependent var 0.003879 Adjusted R-squared 0.994217 S.D. dependent var 0.033119 S.E. of regression 0.002519 Akaike info criterion -8.983306 Sum squared resid 0.000526 Schwarz criterion -8.563893 Log likelihood 460.6736 F-statistic 1124.202 Durbin-Watson stat 1.727360 Prob(F-statistic) 0.000000 43
44 Keynesian employment equation Prognoseguete (dynamic, in-sample, Sample (adjusted): 1981Q2 2005Q4) Root Mean Squared Error 154.4936 Mean Absolute Error 116.9241 Mean Absolute Percentage Error 0.380390 Theil Inequality Coefficient 0.002513 Bias Proportion 0.000206 Variance Proportion 0.000027 Covariance Proportion 0.999767 Prognoseguete (dynamic, out-of-sample, Sample (adjusted): 1981Q2 2001Q4) Root Mean Squared Error 122.8920 Mean Absolute Error 95.22245 Mean Absolute Percentage Error 0.273853 Theil Inequality Coefficient 0.001770 Bias Proportion 0.000991 Variance Proportion 0.030811 Covariance Proportion 0.968198 Prognoseguete (dynamic, out-of-sample, Sample (adjusted): 1981Q2 2000Q4) Root Mean Squared Error 297.7411 Mean Absolute Error 269.7591 Mean Absolute Percentage Error 0.774579 Theil Inequality Coefficient 0.004258 Bias Proportion 0.809648 Variance Proportion 0.005017 Covariance Proportion 0.185335 Prognoseguete (dynamic, out-of-sample, Sample (adjusted): 1981Q2 1999Q4) Root Mean Squared Error 214.4531 Mean Absolute Error 176.8086 Mean Absolute Percentage Error 0.507052 Theil Inequality Coefficient 0.003067 Bias Proportion 0.404633 Variance Proportion 0.000054 Covariance Proportion 0.595313
Keynesian employment equation 30 1.2 20 1.0 10 0.8 0 0.6 0.4-10 0.2-20 0.0-30 86 88 90 92 94 96 98 00 02 04-0.2 86 88 90 92 94 96 98 00 02 04 36500 CUSUM 5% Significance In-Sam ple Prognosen "eq_de_ee_10" CU SUM of Squ ares 5% Significance 36000 35500 35000 34500 34000 33500 33000 32500 95 96 97 98 99 00 01 02 03 04 05 45 YH AT YH AT+1.96*YH ATS E YH AT-1.9 6*YH ATSE de_ee Abh. Bes chaeftigte, Inl.
Keynesian employment equation Out-Of-Sample Prognosen "eq_de_ee_10" Out-Of-Sam ple Prognosen (2)"eq_de_ee_10" 36400 36000 36000 35600 35200 35500 35000 34800 34500 34400 34000 33600 34000 33500 33200 95 96 97 98 99 00 01 02 03 04 05 33000 95 96 97 98 99 00 01 02 03 04 05 Out-Of-Sample Prognosen (3) "eq_de_ee_10" Out-Of-Sam ple Prognosen (4) "eq_de_ee_10" 36400 36000 36000 35600 35200 35500 35000 34800 34500 34400 34000 33600 34000 33500 46 33200 95 96 97 98 99 00 01 02 03 04 05 33000 95 96 97 98 99 00 01 02 03 04 05
Literature Barrel, R./Pain, N./Young, G. (1996): A cross-country comparison of the demand for labour in Europe, Weltwirtschaftliches Archiv 132 (4): 638-650. Bassanini, A./Duval, R. (2006): Employment Patterns in OECD countries: Reassessing the Role of Policies and Institutions, OECD Economics Department Working Paper no. 486. Deutsche Bundesbank (2007): Monthly Bulletin, August, S. 47-48. Hinz, D./Logeay, C. (2006): Forecasting Employment for Germany, IMK Working Paper No. 1. Joebges, H./Logeay, C./Peters, D./Stephan, S./Zwiener, R. (2008): Deutsche Arbeitskosten steigen im europäischen Vergleich nur gering, IMK Report Nr. 34, November. Sachverständigenrat (SVR 2007): Jahresgutachten. European Commission (2007): Raising Germany s Growth Potential, DG ECFIN Occasional Paper Nr. 28, February. OECD 2008: Economic Surveys, Vol. 7, April. 47