Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

47
Training Subsidies and the Wage Returns to Continuing Vocational Training: Evidence from Italian Regions Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

description

Training Subsidies and the Wage Returns to Continuing Vocational Training: Evidence from Italian Regions. Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale). Training in this paper is. Formal - PowerPoint PPT Presentation

Transcript of Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Page 1: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Training Subsidies and the Wage Returns to Continuing

Vocational Training: Evidence from Italian Regions

Giorgio Brunello (Padova)Simona Comi (Milano Bicocca)Daniela Sonedda (Piemonte

Orientale)

Page 2: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Training in this paper is Formal Continuing vocational rather than

initial vocational (after full time education has ended)

Mainly workplace training initiated by firms (but not exclusively)

Page 3: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Training matters Broad consensus among policy

makers that training matters for employment, productivity and individual well being

Yet applied economists do not have a consensus view on the wage returns to training

Page 4: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Two extreme cases Lynch, 1992, finds that one week of

training raises hourly wages by 0.2% Bartel, 1995, finds that one day of

training raises wages by 2 percent The literature often finds returns of at

least 3 percent for a week of private sector training – large relative to returns to 1 year of schooling (10 percent)

Page 5: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Estimating these returns is difficult

itititit xw 'ln

Participation in training is not random

Training correlated with individual un-observables (ability)

Page 6: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Methods used in the literature Fixed effects estimates:

If un-observables are time invariant the within estimator is appropriate

required assumptions are: 1. earnings growth is the same for

participants and non-participants 2. temporary shocks that affect

wages do not affect training

Page 7: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

IV estimates We need an exclusion restriction

A variable which affects training but does not affect directly wages or the probability of receiving a positive wage

Difficult

Page 8: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

The Acemoglu and Pischke model Imperfect product and labour

markets General skills Firms are willing to train even when

the imparted skills are general Frictions and imperfections reduce

the transferability of skills

Page 9: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Sketch of the model Two periods First period: training takes place and the

employer pays the training costs At the start of second period the match

may end because of an exogenous shock If the match survives, bargaining over

wages Production occurs

Page 10: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Notation

)()(1)()()(

vfGqwvf

output

Worker outside option

wages

Probability of employment

Page 11: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Wage setting Nash bargaining

The firm has outside option equal to zero

Outcome of the bargain )()()()( vfvw

Training costs are bygones

Page 12: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

The training decision

scvfq )()()()1)(1()(

s= training subsidy

0)()()()1)(1( ''' scvfgq

Training increases with the subsidy

The subsidy affect training directly and wages and employment indirectly via training

Page 13: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Implication Training policies such as training subsidies

are a good candidate to instrument training in an earnings regression

However: national training policies that affect all individuals equally cannot work

We need that training policies affect only some groups (ex:

training policies only in some areas) the intensity of policies differs among groups

Page 14: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

In this paper

We use regional training policies (training grants) as instrument

They differ across regions AND over time

Page 15: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Italian institutional setup Training policies are regional

policies Regional governments have

substantial autonomy in Allocation of training expenditures to

their budget Timing of their invitations to tender Ability to pay quickly

Page 16: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

CVT policies in Italy Levy / grant type (funded by social security

contributions with a grant mechanism to award funds) European Social Fund (largest; Objective 4

during 1994-99; Directives 1 and 2 during 2000-06– lifelong learning)

Laws 236/94 and 53/00 Industry based training funds (from late 2004)

Tax deductions (Tremonti 2001 and 2002) – time dummies

Page 17: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

ESF, laws 236/93 and 53/00

Funded at least in part by a compulsory levy of 0.30% on payroll

Regional implementation, especially from 2001

Regional and time variation in expenditure plans and invitations to tender (impegni)

Page 18: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Our data We collect from regional

publications the regional invitations to tender associated to Laws 236 and 53

Data on ESF expenditure plans and invitations to tender partly from ISFOL and partly from the National Audit Court (Corte dei Conti)

Page 19: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

ABR

BAS

CAL

CAM

EMI FVG

LAZ

LIGLOMMAR

PIE

PUG

SAR

SIC

TAA

TOS

UMB

VEN

010

2030

4023

6/53

real

trai

ning

ince

ntiv

es p

er h

ead;

sum

199

7-20

04

0 20 40 60 80 100FSE real training incentives per head; sum 2000-2004

Cumulated Real Training Incentives, by Region; real euros per head

Page 20: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Resources allocated to training subsidies from the 0.30% compulsory levy

Source: ISFOL, 2006

Page 21: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Table 1. Regional Planned Training Expenditures. Cumulated stock 1994-2005. Real Euros per head. ESF 236 and 53 Piemonte

100.74

23.10

Lombardia 64.98 15.58 Trentino Alto Adige 251.36 35.43 Veneto 94.77 21.06 Friuli Venezia Giulia 144.33 32.88 Emilia Romagna 153.32 31.47 Liguria 89.51 19.96 Toscana 71.60 15.75 Marche 62.66 16.82 Umbria 104.14 23.52 Lazio 61.48 16.63 Abruzzi 68.96 19.86 Campania 26.00 12.97 Puglia Basilicata

10.69 42.27

11.91 30.68

Calabria 18.12 7.36 Sicilia 34.86 5.10 Sardegna 69.27 15.74

Page 22: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

11.

52

2.5

3st

ock

of tr

aini

ng in

cent

ives

1998 2000 2002 2004 2006year

PIE LOMTAA

12

34

1998 2000 2002 2004 2006year

FVG EROLIG VEN

12

34

1998 2000 2002 2004 2006year

TOS UMBMAR LAZ

05

1015

stoc

k of

trai

ning

ince

ntiv

es

1998 2000 2002 2004 2006year

ABR CAMPUG

010

2030

40

1998 2000 2002 2004 2006year

BAS CALSIC SAR

Page 23: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Empirical model

irtirtWrtWirtrtWirt TQXw 31''ln

irtrtTrtTirtrtTirt TSQXT ''

T=training stock

TS: stock of training incentives per head at constant prices

Page 24: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

The specification Contextual effects

Regional and time dummies (wage bargaining is national in Italy)

Changes in regional labour markets Regional unemployment rate

Changes in R&D expenditure Regional share of high tech industries

Reverse causality – negative shocks reduce wages and induce regions to spend more on incentives

First lag of training incentives

Page 25: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

The data

Match regional data on training incentives with micro data (ILFI)

ILFI: indagine longitudinale sulle famiglie italiane

Collects current and retrospective information

Page 26: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Why ILFI? Has info on wages and covers relevant

period (1999, 2001, 2003, 2005) Pluses:

Has good info on training – not only training incidence but also training episodes – plus retrospective info: can be used to compute training stock as discounted number of episodes

Minuses: info on training duration has many missing

values – we decide not to use it Tends to omit shorter episodes (usually the

case in household surveys)

Page 27: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Table 2. Sum of Training Episodes and Percentage of Workers receiving any Training. By region. Year: 2005 Sum of

training episodes

% of trained

workers

Sum of training episodes for

trained workers Piemonte

1.071

0.442

2.419

Lombardia 0.725 0.355 2.042 Trentino Alto Adige and Veneto 1.362 0.514 2.647 Friuli Venezia Giulia 2.111 0.722 2.923 Emilia Romagna 0.704 0.422 1.666 Liguria 0.653 0.307 2.125 Toscana 1.209 0.493 2.450 Marche and Umbria 0.958 0.375 2.555 Lazio and Abruzzi 0.982 0.377 2.604 Campania 0.271 0.171 1.583 Puglia Basilicata and Calabria

0.788 1.057

0.384 0.228

2.050 4.625

Sicilia and Sardegna 0.557 0.285 1.950 Source: see the Appendix

Page 28: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Training stock

1)1( ititit TIT

Page 29: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Presentation of results

First stage estimates 2SLS (LATE) Variations on the main theme

Page 30: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Table 4. First stage estimates – full sample and subsample with positive earnings. Private sector employees only. Dependent variable: cubic root of the training stock and training stock T. 13 regions. (1) (2) (3) (4) Full sample

Cube root of T 1998-2005

Full sample Linear T

1998-2005

Subsample with wage>0

Cube root of T 1999,2001,2003,

2005

Subsample with wage>0 Linear T

1999,2001,2003, 2005

Lagged incentives stock 0.002*** 0.003*** 0.002*** 0.003** [0.000] [0.001] [0.001] [0.001] F-test Elasticity

29.93 0.408

14.40 0.310

10.10 0.350

6.16 0.300

Observations 11495 11495 4850 4850 R-squared 0.140 0.118 0.153 0.123 Note: Clustered robust standard errors in brackets; *** p<0.01, ** p<0.05, * p<0.1; each regression includes a constant, regional, year and industry dummies. For additional controls see paper.

Page 31: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Implications (ceteris paribus)

One additional real euro per head spent in training subsidies from time t-x to time t-1 increases the discounted training stock at time t by 0.6 percent, a small amount.

To increase the average individual training stock by 10%, regions would have to spend an additional 13.47 euro per head (40 million euro in Lombardia)

Page 32: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Table 5. Ordinary least squares and IV estimates. Dependent variable: log monthly real earnings. 13 regions. Years 1999, 2001, 2003 and 2005 (1) (2) (3) (4) RE RE RE-IV RE-IV Training Stock (Cube Root) 0.047*** 0.273** [0.012] [0.123] Training Stock 0.021*** 0.149** [0.006] [0.060] Marginal effect of current episode Marginal effect of a week of training Elasticity

0.032 0.007 0.010

0.021 0.005 0.007

0.186 0.044 0.063

0.149 0.035 0.050

Observations 4850 4850 4850 4850 Note: Robust standard errors in brackets: *** p<0.01, ** p<0.05, * p<0.1. Each regression includes a constant, industry, regional and time dummies.

Page 33: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Comments No evidence of weak instruments with

cube root specification 2SLS estimates: one additional training

episode - that raises T by one unit -raises monthly earnings by 18.6 percent

A week of training raises earnings by 4.4 percent (average duration: 21 days)

LATE, not ATE or ATT

Page 34: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

0.0

1.0

2.0

3.0

4m

argi

nal r

etur

n

0 10 20 30time since investment

5 percent depr rate 15 percent depr rate

Figure 3. Marginal returns to a week of training

Marginal returns decline over time, especially in our baseline

Page 35: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Average marginal effect

Over a 20 years period, the average marginal effect of a training episode is 1.35% in the preferred specification

Page 36: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Exploring heterogeneous effects Interact both the training stock

and the instrument with gender, age and firm size

In the case of firm size there are significant differences

Page 37: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Table 6. Effect of incentives on training and IV estimates of the effect of training on log wages. With interactions with firm size. 13 regions. (1) (2) Lagged incentives stock x 100 0.168*** [0.017] Lagged incentive stock x Firm size>100 dummy x 100

0.033*** [0.012]

Lagged Training Stock Lagged Training Stock x Firm Size>100 Dummy Elasticity firms with less than 100 employees Elasticity firms with more than 100 employees

0.401

0.410

0287*** [0.125]

-0.114* [0.066]

Marginal effect firms with less than 100 employees Marginal effect firms with 100 or more employees

0.219

0.109

Observations 11495 4474

Page 38: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

0.0

2.0

4.0

6m

argi

nal r

etur

n

0 10 20 30time since investment

firms with < 100 employees firms with at least 100 employees

Figure 4. Marginal returns to a week of training; depr. rate:0.15

Page 39: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Interpretation of results: I Small firms with less than 100

employees often don’t have the resources and the facilities to train

Small firms train much less than large firms

Marginal benefits of training are decreasing in the quantity of training

When policies induce smaller firms to train, the benefits are much larger

Page 40: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

MB

MCS

MCL

Page 41: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Interpretation II Small firms have lower bargaining

power In order to retain their trained

employees, they need to pay higher wage premia

Page 42: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Potential biases I Informal training

Additional subsidies raise formal training and reduce informal training: we over-estimate effects

Nothing we can do as informal training is not measured

Page 43: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Potential biases II Additional incentives induce firms to

choose longer training course and reduce shorter courses: we over-estimate effects

More incentives affect training quality as less productive courses are added in: we over-estimate effects

We regress average duration on training incentives and find no significant effect. If quality is related to duration this suggests that these biases may be small

Page 44: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Back of the envelope If T increases by 1 today annual earnings of

compliers increase by 2645 euro (from 14222) – this is not the average treatment effect

1 euro spent in subsidies increases the training stock today by 0.002; hence earnings increase by 5.29 (2645*0.002) for compliers

After 10 years these increases are only 20 percent of current increases

Page 45: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Yet

Since the average effect on the treated is different from LATE, we cannot go further than this – we would need to know the wage return of a broader group in the population of interest

Page 46: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Conclusions We find evidence that

The wage returns to training for those affected by training policies are relatively high

These large effects are mostly limited to small firms; trained workers in large firms who comply with the training policies have much lower returns

Page 47: Giorgio Brunello (Padova) Simona Comi (Milano Bicocca) Daniela Sonedda (Piemonte Orientale)

Conclusions Training incentives work but

moderately so: one euro per head spent in an average region (3 million euro) increases the stock of training by 0.6 percent