Endogenous Technology Choices and the Dynamics of Wage Inequality

Transcription

1 Endogenous Technology Choices and the Dynamics of Wage Inequality Empirical and Theoretical Evidence Jan Department of Economics European University Institute Fondazione Eni Enrico Mattei, February 2013

2 Outline 1 Technology Choice and Skill Premium Dynamics 2 3

3 Dynamics of College Premium and Relative Supply of College Skills in US Figure: College premium and relative supply of college skills (weeks worked by college equivalents divided by weeks worked of noncollege equivalents) in US between 1939 and Data: March CPSs and 1940, 1950 and 1960 censuses

4 Growth of College Premium In 1980 in US college graduates earned 50% more than other employees, in 2005: twice as much as other employees. Similar increase in wage inequality across most OECD countries (Van Reenen (2011)) Overall in 12 OECD countries the college premium has dropped in 80s by 4% and increased in 90s by 16%. Economics and Social implications: Do we all benet from growth? Between 1980 and 2005 in US real GDP grew by 43%, real wages of those with no college degree (3/4 of labour force) grew by 26%

5 Skill-bias of Technology Skill-biased Technology Change explanation: The new global technology, The Information and Communication Technology (ICT) favours skilled workers. Its introduction led to higher demand of skilled workers As a consequence skill premium increased

6 Technology Choice Hypothesis Empirical nding of the paper: Those OECD countries that experienced higher growth of college graduates than other countries experienced higher growth of college wage premium a decade later. 1 Global technology (ICT) is not a sole determiant of how skill-biased is a production process. Firms utilizing ICT inventions still can choose between more and less skill-biased production methods. 2 After increase in supply of skilled workers rms have a possibility to adjust choice of production method (version of the LeChatelier principle) 3 This can explain part of the growth of skill premium across OECD.

7 Technology Choice Example Imagine two production methods, both involving ICT: automatized plant skilled workers programme robots, unskilled workers assist them (clean, guard) high relative demand for skilled workers hand&eye plant unskilled perform key production tasks, skilled workers provide supervision, trainings, logistics, accounting high relative demand for unskilled workers Two sorts of reasons to switch to 'automatized plant' The literature ICT lead to larger improvements in automatized plant than in hand&eye plant there is higher number of skilled workers

8 Outline 1 Technology Choice and Skill Premium Dynamics 2 3

9 The set-up of the model One product, but various available production methods (production functions). Production method i takes a form F i = [ (A is L s ) σ + (A iu L u ) 1 θ σ] σ where L s and L u stand for skilled and unskilled labour inputs Each production method is characterized by a dierent pair (A s, A u ) (productivity parameters) Therefore each production method can be represented in 2D space as a single point. The set of available production methods is determined by the current technology paradigm and is described by: k θ 1 γ Aω is + Aω iu B (1)

10 The set-up of the model (2) Assume (for now) vertical supply of labour (so it does not depend on skill premium) The suppy of skilled relative to supply of unskilled labour follows the process: ( ) ( ) Ls Ls log = log + X t 1 + ξ t (2) L u L t u t 1 where X t 1 is a vector of variables observed by a rm at time t

11 Firms Optimization (1) Firm cannot switch the technology on the spot. This is to reect that rms need a time to redesign a plant and replace the capital. The rm's value function is then: V (A s, A u, L s, L u ) = { [(As = max L s ) σ + (A u L u ) σ] 1 σ w s L s w u L u + A s,a u,ls,lu [ ( +β E V A s,a u, L s,l )]} u subject to 1 γ A ω is + A ω u B. (x denotes the value of variable x next period).

12 Firms Optimization (2) Combining First Order Conditions with respect to labour choices: log ( ws w u ) t ( ) ( ) Ls As = (1 σ) log + σ log L u A t u t (3) Combining FOCs with respect to technology choices and two envelope conditions: [ [(A E s L s) σ + (A ul u) σ] ] σ 1 1 (A s L s) σ 1 L s [ [(A E s L s) σ + (A ul u) σ] ] = A ω 1 s σ 1 1 (A ul u) σ 1 L γa ω 1 u u

13 Firms Optimiyation (3) with E Using First Order Taylor Approximation: log(e [x]) = E [log(x)] ( ) A s log = 1 [ log A u ( L s L u )] = log ω σ log(γ) + ) ( Ls Lu + X t 1 t 1 σ [ ( )] L ω σ E s log L u (4)

14 Equilibrium Since supply of labour is vertical the equations (4) and (3) characterize equilibrium production method choice and skill premium Collecting all the terms: ( ) ( ) ws Ls log = (1 σ) log + σ w u L t u ω σ log(γ) + t t 1 + σ ( ( ) ) 2 Ls log + X t 1 (5) ω σ L u t 1 Equilibrium skill premium is driven by three forces: diminishing returns to skills skill-bias of technological paradigm rm's expectation about today's supply of skills driving technology choice.

15 Dynamics of College Premium and Relative Supply of College Skills in US Figure: College premium and relative supply of college skills (weeks worked by college equivalents divided by weeks worked of noncollege equivalents) in US between 1939 and Data: March CPSs and 1940, 1950 and 1960 censuses

16 Outline 1 Technology Choice and Skill Premium Dynamics 2 3

17 Identication The empirical model can be directly derived from the baseline model: ( ) ( ) ws Ls log = (1 σ) log + σ w u L t u ω σ log(γ) + t Two identication issues: + σ ( ( ) ) 2 Ls log + X t 1 ω σ L u t 1 t 1 Separate the negative eect due to diminishing returns to skills and positive eect of adjusting technology choice. Separate the eect of change in technological possibilities and the eect of adjusting technology choice. (6)

18 Empirical Model The empirical model is: where w it = log w it = α 1 l it + α 2 l it 10 + ˆd t + ε it (7) ( ) ( ) ws Ls and l it = log wu it Lu it

19 Identication - Intuition The model is equivalent to ( w it w t ) = α 1 ( l it l t ) + α 2 ( l it 10 l t 10) +( ε it ε t ) Thus eectively it isolates out the global factors (such as change in skill-biased technology paradigm due to arrival of ICT).

20 Data Technology Choice and Skill Premium Dynamics All data come from EU KLEMS (2008 release) on 23 countries between 1970 and 2005 (unbalanced). The annual data are available, however I use them only in 5 years intervals.

21 Model Estimation

22 Quantifying Technology Choice Between 1990 and 2005 in OECD the skill premium has increased by 28%. The model predicts three sources of change: ( ) ( ) ws Ls log = (1 σ) log + σ w u L t u ω σ log(γ) t + σ ( ( ) ) 2 Ls log + X t 1 ω σ L u t 1 The canonical model (diminishing return to skilled labour) predicts that skill premium should fall by 18% t 1 The technology choice model explains 29% of the dierence between canonical model prediction and observed increase. + (8)

23 Econometric Analysis So far, estimation of the model - a possibility result However, the nding that countries that witness higher growth of college graduates experience higher growth of college premium a decade later is interesting on its own! Lets proceede with the exploratory analysis to explain it.

24 Alternative Explanations ( w it w t ) = α 1 ( l it l t ) + α 2 ( l it 10 l t 10) +( ε it ε t ) Alternative explanations: bias due to inclusion of endogenous variable discussion serial correlation of the error term discussion reverse causality discussion alternative mechanisms: spillover discussion, endogenous adoption discussion

25 Conclusion Countries with faster growth of college graduates witness also faster growth of college premium ten years later. The most likely candidate to explain this pattern is the endogenous technology choice hypothesis: At any point in time a rm has an option to switch between production methods. After increase in supply of college graduates rms decided to switch to more skill-biased production methods. The estimation shows that this switch can account for 1/3 of the growth in skill premium in OECD countries. Future Work.

26 Technology Choice and Climate Change? Existing literature: Jae (2002), Acemoglu, Akcigit, Hanley and Kerr (2012), Kemp (2000). Included in the WITCH model What can we share? evidence for technology switching as a response to change in market conditions the response comes with long delay (10 years) long lags and utilization of cross-section can help to mitigate endogeneity problem in estimation of model.

27 Thank you!

28 Appendix

29 Contribution Contribution to the literature on skill-biased technology change (Acemoglu (2002); Aghion (2002); Goldin and Katz (2008)) Show that switch to skill-biased production method (that lead to increase in skill premium) can happen even without change in technological paradigm Contribution to the literature on Endogenous Technology Choice (Samuelson (1965), Caselli and Coleman (2007), Peri (2009)) Elaborate and estimate a dynamic model (prediction on dynamics of wage inequality; the model is based on Caselli and Coleman static model) Present a piece of empirical evidence to support the hypothesis. Set the microfoundation for the model: How R&D sector can generate variety of production methods.

30 Serial Correlation Suppose that the error term in skill premium regression is the IMA(1,1) process: ε it = η it + βη it 10 (9) back Intuitively β > 0 implies that there is a process whose impact on skill premium is not completed after 10 years. Moreover the process cannot aect all countries equally (e.g. ICT), since this is controlled for. Candidates: Loss of importance of trade unions Globalization

31 Quantitative Results

32 Inclusion of Endogenous Variable w it = α 1 l it + α 2 l it 10 + ˆd t + ε it back Since l it may be correlated with the error term, all estimators are biased What is the sign of the bias for α 2 estimator? The assymptotic bias is E [ˆα 2 α 2 ] = = 1 det E [( l it l t )( l it 10 l t 10)] E [( l it l t )( ε it ε t )] Therefore we would expect the sign of the bias to be negative.

33 Reverse Causality back Skill supply will be correlated with future skill premium if workers predict future changes of skill premium. Workers would need to predict that growth of skill premium in their country will be higher than in other countries. They would need to know it 10 years ahead. If students in 1975 predict a sharp increase in skill premium in 1985 why students in 1970 would not be able to predict it?

34 Spillover Eect back The higher is the number of workers the more productive they are. Some eect should be visible after 5 years - disproved by data

35 Endogenous Adoption back Countries with higher number of skilled workers want to adopt ICT faster than other countries. Growth of Skill premium would need to depend on the level of supply of skills.

36 Quantitative Results

37 Future Work The data shows that in recent year positive correlation between skill supply growth and skill premium growth decade later seems to vanish. More labour mobility across countries? Technological Change greases Technological Choice? Microfoundation and more evidence for the trade-o between more and less skill-biased technologies.