Banque nationale de Belgique SA · EURONEXT:BNB
Banque nationale de Belgique SA — Technical Presentation (May 2013)
54 slides from Banque nationale de Belgique SA's technical presentation, May 10, 2013
Slide 1: Labour Market Policies and Unemployment Dynamics
Labour Market Policies and Unemployment Dynamics A Cross-Country Comparison Fabrice Murtin 1 Jean-Marc Robin 2 1 OECD 2 Sciences-Po and UCL 10th May 2013 Murtin, Robin (OECD, ScPo & UCL) Labour Market Policies 10th May 2013 1 / 40
Slide 2: Introduction
Introduction <O> «f> «<=> «> Dae
Slide 3: Question
Question How much of unemployment dynamics can be accounted for by Labour Market Policies? We shall answer this classic question by constructing a small model of the labour market model such that � Aggregate shocks to GDP fuel the cycle � LMPs induce structural change. The model will be separately e
Slide 4: Question
Question How much of unemployment dynamics can be accounted for by Labour Market Policies? We shall answer this classic question by constructing a small model of the labour market model such that � Aggregate shocks to GDP fuel the cycle � LMPs induce structural change. The model will be separately e
Slide 5: Question
Question How much of unemployment dynamics can be accounted for by Labour Market Policies? We shall answer this classic question by constructing a small model of the labour market model such that � Aggregate shocks to GDP fuel the cycle � LMPs induce structural change. The model will be separately e
Slide 6: The literature: reduced form studies
The literature: reduced form studies Bruno and Sachs (1985), Blanchard and Wolfers (2000), Bertola, Blau and Khan (2007) Log UNR Log LDR Log JFR Log V Log Tightness Initial Replacement Rate 0.130 0.084 -0.062 -0.129 -0.260 ALMP: Placement -0.062 -0.041 0.018 -0.006 0.056 ALMP: Training -0.133 -0.059
Slide 7: The literature: small structural models
The literature: small structural models Small structural model focusing on one policy � Ljungqvist and Sargent (1998) study welfare policies; � Prescott (2004) and Rogerson (2008) highlight the role of labour taxes. Murtin, Robin (OECD, ScPo & UCL) Labour Market Policies 10th May 2013 5 / 40
Slide 8: Our approach
Our approach We develop a hybrid approach that aims at bridging the gap between the two approaches: � Structural shocks to production � Search-matching framework à la Mortensen-Pissarides (RES, 1990) � Rent sharing à la Postel-Vinay-Robin (Ecta, 2002) � Worker heterogeneity as in Robin (Ecta, 2011)
Slide 9: Theory
Theory <O> «f> «<=> «>
Slide 10: Aggregate shocks
Aggregate shocks Time is discrete and indexed by t ∈ N. The global state of the economy is described by a Markov chain yt ∈ {y1 < ... < yN } with transition probability matrix Π = (πij ). Aggregate shocks accrue at the beginning of each period. Murtin, Robin (OECD, ScPo & UCL) Labour Market Policies
Slide 11: Heterogeneity
Heterogeneity There are M types of workers and �m workers of each type (with �M m=1 �m = 1). Each type is characterized by a time-invariant ability xm , m = 1, ..., M, with xm < xm+1 . Firms are identical. Murtin, Robin (OECD, ScPo & UCL) Labour Market Policies 10th May 2013 9 / 40
Slide 12: Matches
Matches Workers are paired with firms to form productive units. The per-period output of a job, if the worker is of ability xm and aggregate productivity is yi , is yi (m) = xm yi . Matches form and break at the beginning of each period, after realisation of the aggregate shock. Murtin, Robin (OECD,
Slide 13: Unemployment and job destruction
Unemployment and job destruction Let ut (m) denote the proportion of unemployed workers of ability xm at the end of period t − 1. �M Let ut = m=1 ut (m)�m denote the aggregate unemployment rate. I denote as Si (m) the surplus of a match (xm , yi ). Only matches with Si (m) > 0 are viable. At the beg
Slide 14: Unemployment and job destruction
Unemployment and job destruction Let ut (m) denote the proportion of unemployed workers of ability xm at the end of period t − 1. �M Let ut = m=1 ut (m)�m denote the aggregate unemployment rate. I denote as Si (m) the surplus of a match (xm , yi ). Only matches with Si (m) > 0 are viable. At the beg
Slide 15: Unemployment and job destruction
Unemployment and job destruction Let ut (m) denote the proportion of unemployed workers of ability xm at the end of period t − 1. �M Let ut = m=1 ut (m)�m denote the aggregate unemployment rate. I denote as Si (m) the surplus of a match (xm , yi ). Only matches with Si (m) > 0 are viable. At the beg
Slide 16: Meetings
Meetings Matching is random and firms with a vacancy can either meet an unemployed worker or an employed worker. Let λt denote the meeting rate. The relative search intensity of employees is set as k = 0.12. At the beginning of period t, � A fraction λt 1{Si (m) > 0}ut (m)�m of unemployed workers me
Slide 17: Meetings
Meetings Matching is random and firms with a vacancy can either meet an unemployed worker or an employed worker. Let λt denote the meeting rate. The relative search intensity of employees is set as k = 0.12. At the beginning of period t, � A fraction λt 1{Si (m) > 0}ut (m)�m of unemployed workers me
Slide 18: Meetings
Meetings Matching is random and firms with a vacancy can either meet an unemployed worker or an employed worker. Let λt denote the meeting rate. The relative search intensity of employees is set as k = 0.12. At the beginning of period t, � A fraction λt 1{Si (m) > 0}ut (m)�m of unemployed workers me
Slide 19: Rent sharing
Rent sharing We assume that employers have full monopsony power with respect to unemployed workers. So they keep all the surplus in this case. However, employed workers search on the job, and Bertrand competition transfers all the surplus to the poached employees. Murtin, Robin (OECD, ScPo & UCL) La
Slide 20: Rent sharing
Rent sharing We assume that employers have full monopsony power with respect to unemployed workers. So they keep all the surplus in this case. However, employed workers search on the job, and Bertrand competition transfers all the surplus to the poached employees. Murtin, Robin (OECD, ScPo & UCL) La
Slide 21: Vacancy creation
Vacancy creation Firms posts vacancies vt until ex ante profits are exhausted (free entry). The total vacancy cost is cvt . For yt = yi , free entry ensures that M � cvt = λt ut (m)�m Si (m)+ m=1 where we denote x + = max(x, 0). A matching function relates the job finding rate λt to tightness θt : v
Slide 22: Vacancy creation
Vacancy creation Firms posts vacancies vt until ex ante profits are exhausted (free entry). The total vacancy cost is cvt . For yt = yi , free entry ensures that M � cvt = λt ut (m)�m Si (m)+ m=1 where we denote x + = max(x, 0). A matching function relates the job finding rate λt to tightness θt : v
Slide 23: Vacancy creation
Vacancy creation Firms posts vacancies vt until ex ante profits are exhausted (free entry). The total vacancy cost is cvt . For yt = yi , free entry ensures that M � cvt = λt ut (m)�m Si (m)+ m=1 where we denote x + = max(x, 0). A matching function relates the job finding rate λt to tightness θt : v
Slide 24: Turnover rates
Turnover rates The following turnover rates can then be computed: Exit rate from unemployment: � 1{St (m) > 0}ut (m)�m ft = λt m ut Job destruction rate: � m 1{St (m) ≤ 0}(1 − ut (m))�m st = δ + (1 − δ) 1 − ut Murtin, Robin (OECD, ScPo & UCL) Labour Market Policies 10th May 2013 15 / 40
Slide 25: Unemployment dynamics
Unemployment dynamics The law of motion of individual-specific unemployment rates is such that, at the end of period t, ut+1 (m) = 1 − [(1 − δ)(1 − ut (m)) + λt ut (m)]1{St (m) > 0} � 1 if St (m) ≤ 0, = ut (m) + δ(1 − ut (m)) − λt ut (m) if St (m) > 0. Note that ut+1 − ut = st (1 − ut ) − ft ut . Mu
Slide 26: Unemployment dynamics
Unemployment dynamics The law of motion of individual-specific unemployment rates is such that, at the end of period t, ut+1 (m) = 1 − [(1 − δ)(1 − ut (m)) + λt ut (m)]1{St (m) > 0} � 1 if St (m) ≤ 0, = ut (m) + δ(1 − ut (m)) − λt ut (m) if St (m) > 0. Note that ut+1 − ut = st (1 − ut ) − ft ut . Mu
Slide 27: The match surplus
The match surplus The surplus of a match (xm , yi ) solves the following (nearly linear) Bellman equation: 1−δ � Si (m) = yi (m) − z + πij Sj (m)+ , 1+r j where we denote x + = max(x, 0) and z is the flow utility of unemployment. � After a productivity shock from i to j the continuation surplus is z
Slide 28: The match surplus
The match surplus The surplus of a match (xm , yi ) solves the following (nearly linear) Bellman equation: 1−δ � Si (m) = yi (m) − z + πij Sj (m)+ , 1+r j where we denote x + = max(x, 0) and z is the flow utility of unemployment. � After a productivity shock from i to j the continuation surplus is z
Slide 29: Empirical analysis
Empirical analysis <O> «f> «<=> «>
Slide 30: Data construction
Data construction Our own calculations from BLS (2011), Eurostat (2011), OECD Employment Outlook (2010). Compare with Murtin, De Serres and Hijzen (2011), Petrongolo and Pissarides (2008), Elsby et al. (2008). Although unemployment data allow for monthly frequency in some countries, the time unit is
Slide 31: Structural parameters
10 structural parameters 1 Markov chain for yt = yi for i = 1, ..., N = 100. Discretization of an AR(1) process for log-wages with autocorrelation ρ and variance σ 2 . 2 Exogenous job destruction rate δ 3 Matching function: f (θ) = φθη 4 Ability: � 500 equidistant nodes in [C , C + 1]. � The distrib
Slide 32: Labour Market Policies (LMPs)
Labour Market Policies (LMPs) 1 UI benefit 2 Initial replacement rate (RR) 3 ALMP expenditure (ALMP expenditure per unemployed wkr/GDP per employee): 1 Employment & Placement Services 2 Training 3 Incentives (e.g. payroll tax reduction for low wage workers) 4 Product market regulation OECD index (PM
Slide 33: How LMP enters the model
How LMP enters the model We assume that LMP changes are not anticipated: Individuals observe the change when it occurs and incorporate it in the surplus calculations with myopic expectations on future changes. Crude intervention mechanisms: Each government intervention shifts structural parameters φ
Slide 34: Estimation
Estimation 1 Estimate parameters by matching moments of detrended series of GDP, unemployment rates, job finding rates, job destruction rates and ‘tightness’ (V /U). 2 Filter productivity shocks by minimising the SSR of GDP. 3 Estimate LMP effects by fitting GDP cycle and actual series (trend and cyc
Slide 35: Estimation
Estimation 1 Estimate parameters by matching moments of detrended series of GDP, unemployment rates, job finding rates, job destruction rates and ‘tightness’ (V /U). 2 Filter productivity shocks by minimising the SSR of GDP. 3 Estimate LMP effects by fitting GDP cycle and actual series (trend and cyc
Slide 36: Estimation
Estimation 1 Estimate parameters by matching moments of detrended series of GDP, unemployment rates, job finding rates, job destruction rates and ‘tightness’ (V /U). 2 Filter productivity shocks by minimising the SSR of GDP. 3 Estimate LMP effects by fitting GDP cycle and actual series (trend and cyc
Slide 37: Remarks on estimation
Remarks on estimation We use GDP instead of GDP per worker because of negative correlation between productivity and employment in AUS and SWE! We arbitrarily set the matching elasticity η in [0, 1]. Surprisingly, for any choice of η there seems to be a choice of the other parameters that fits as wel
Slide 38: Remarks on estimation
Remarks on estimation We use GDP instead of GDP per worker because of negative correlation between productivity and employment in AUS and SWE! We arbitrarily set the matching elasticity η in [0, 1]. Surprisingly, for any choice of η there seems to be a choice of the other parameters that fits as wel
Slide 39: Results: The Business Cycle model
Results The Business Cycle model
Slide 40: Estimated parameters
Estimated parameters C < z ⇒ endogenous job destruction AUS FRA DEU JAP PRT ESP SWE GBR USA Productivity ρ 0.970 0.938 0.933 0.942 0.842 0.972 0.961 0.970 0.959 σ 0.017 0.020 0.024 0.025 0.026 0.026 0.029 0.023 0.015 C 0.701 0.679 0.527 0.514 0.826 0.705 0.700 0.695 0.663 E(α) 0.993 0.990 0.980 0.98
Slide 41: Moments fit
Moments fit AUS DEU ESP FRA true sim. true sim. true sim. true sim. Mean GDP 1.002 1.000 1.001 1.000 1.001 0.993 1.000 0.997 Std log GDP 0.017 0.018 0.021 0.024 0.028 0.029 0.019 0.022 Autocorr. log GDP 0.927 0.980 0.925 0.942 0.963 0.982 0.972 0.954 Mean unemployment 0.074 0.075 0.091 0.092 0.159 0
Slide 42: Moments fit
Moments fit GBR JPN PRT SWE USA true sim. true sim. true sim. true sim. true sim. Mean GDP 1.006 0.995 1.005 0.992 1.007 1.003 1.004 0.995 1.005 1.000 Std log GDP 0.023 0.026 0.027 0.029 0.029 0.027 0.029 0.031 0.017 0.018 Autocorr. log GDP 0.977 0.980 0.946 0.963 0.965 0.859 0.979 0.973 0.960 0.978
Slide 43: Unemployment Cycle
Unemployment Cycle AUS DEU ESP 10 12 unemployment 10 20 Cyclical 8 8 15 6 6 10 1985 1995 2005 1985 1995 2005 1985 1995 2005 FRA GBR JPN 10 unemployment 11 4 8 Cyclical 10 3 9 6 8 4 2 1985 1995 2005 1985 1995 2005 1985 1995 2005 PRT SWE USA 10 8 unemployment 10 8 Cyclical 8 6 6 6 4 2 4 4 1985 1995 20
Slide 44: Correlation Between Actual and Simulated Unfiltered Series
Correlation Between Actual and Simulated Unfiltered Series AUS FRA DEU JAP PRT ESP SWE GBR USA Average Productivity 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Unemployment 0.83 0.68 0.88 0.87 0.73 0.92 0.77 0.87 0.75 0.81 Job Finding Rate 0.70 0.64 0.70 0.79 0.74 0.77 0.31 0.85 0.82 0.70 Job
Slide 45: Conclusions on BC model
Conclusions on BC model Right amplification of productivity shocks In many countries the fit is good beyond unemployment fluctuations. However, some (relative) failure regarding tightness and employment/unemployment flows. This may be an artifact of the HP-filtering procedure as the following analys
Slide 46: Trends
Trends <O> «f> «<=> «>
Slide 47: LMP effects
LMP effects Effect of a one-std increase in policy. We a priori shunt certain channels for identification. φ δ c Initial replacement rate -0.036 ALMP Placement and Employment Services 0.025 -0.114 ALMP Training -0.037 -0.095 ALMP Incentives 0.063 -0.167 Product Market Regulation -0.026 0.112 Employmen
Slide 48: LMP effects on unemployment
LMP effects on unemployment AUS FRA DEU JAP PRT ESP SWE GBR USA Mean IRR (−1 sd) -0.29 -0.40 -0.25 -0.16 -0.24 -0.61 -0.22 -0.32 -0.22 -0.30 Placement (+1 sd) -0.93 -1.25 -0.74 -0.47 -0.69 -1.95 -0.70 -0.99 -0.72 -0.94 Training (+1 sd) -0.35 -0.45 -0.24 -0.16 -0.21 -0.72 -0.25 -0.36 -0.27 -0.33 Incen
Slide 49: Unemployment trends
Unemployment trends AUS DEU ESP 12 Unemployment 10 20 10 8 5 10 6 1990 2000 2010 1990 2000 2010 1990 2000 2010 FRA GBR JPN 14 14 Unemployment 12 6 12 10 10 8 6 4 8 4 2 2 1990 2000 2010 1990 2000 2010 1990 2000 2010 PRT SWE USA 12 8 Unemployment 8 10 8 6 6 6 4 4 4 2 1990 2000 2010 1990 2000 2010 1990
Slide 50: JFR trends
JFR trends AUS DEU ESP Job finding rate 25 60 40 20 40 20 15 1990 2000 2010 1990 2000 2010 1990 2000 2010 FRA GBR JPN Job finding rate 50 60 35 30 40 50 25 20 30 40 15 30 1990 2000 2010 1990 2000 2010 1990 2000 2010 PRT SWE USA Job finding rate 35 90 30 25 60 80 20 15 40 70 1990 2000 2010 1990 2000
Slide 51: LDR trends
LDR trends AUS DEU ESP Job destruction rate 5 4.5 2.2 4 2 4 3.5 1.8 3 1.6 1.4 3 1990 2000 2010 1990 2000 2010 1990 2000 2010 FRA GBR JPN Job destruction rate 4 5 2 3 4 3 1.5 2 2 1 1990 2000 2010 1990 2000 2010 1990 2000 2010 PRT SWE USA Job destruction rate 5 2 6 4 1.5 5 3 2 4 1 1 3 1990 2000 2010 1
Slide 52: Market tightness trends
Market tightness trends AUS DEU ESP 0.4 0.2 Market tightness 0.1 0.15 0.2 0.05 0.1 0.05 0 0 1990 2000 2010 1990 2000 2010 1990 2000 2010 FRA GBR JPN Market tightness 0.15 0.5 0.4 0.4 0.1 0.3 0.2 0.05 0.2 0 0.1 1990 2000 2010 1990 2000 2010 1990 2000 2010 PRT SWE USA 0.08 Market tightness 1.2 0.06 0.
Slide 53: Correlation Between Actual and Simulated Unfiltered Series
Correlation Between Actual and Simulated Unfiltered Series AUS FRA DEU JAP PRT ESP SWE GBR USA Average Productivity 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Unemployment 0.96 0.77 0.68 0.98 0.76 0.82 0.98 0.89 0.49 0.81 Job Finding Rate 0.87 0.89 0.60 0.84 0.41 0.86 0.88 0.92 0.71 0.77 Job
Slide 54: Conclusion
Conclusion One (simple) model of LMP intervention fits them all. Main lessons: � ALMP - Placement and Employment Services: most important (+1sd = -1% UNR). � Other LMPs are less effective (+1 sd = between .2 and .33 pct points). Extension: firm heterogeneity. Work in progess with Jeremy Lise. More ac
Banque Nationale De Belgique — Technical Presentation (May 2013) on 6ix