The IT Boom and Other Unintended Consequences of Chasing the American Dream
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Introduction Empirics Model Estimation Counterfactuals
The IT Boom and Other Unintended
Consequences of Chasing the American Dream
Gaurav Khanna Nicolas Morales
UC San Diego Richmond Fed
09.17.21
The views expressed are those of the authors and do not necessarily reflect
those of the Federal Reserve Bank of Richmond or the Board of Governors.
Khanna & Morales And Other Unintended Consequences 1 / 33
Introduction Empirics Model Estimation Counterfactuals
Innovation Boom in US IT
US Computer Scientists as a Share of College Grads / STEM workers
CS fastest growing occupation in 1990s (and expected to
stay fastest growing) (BLS 1996)
Khanna & Morales And Other Unintended Consequences 1 / 33
Introduction Empirics Model Estimation Counterfactuals
Immigrants mostly Computer Scientists
Immigrants as Fraction of Workers by Occupation
Khanna & Morales And Other Unintended Consequences 2 / 33
Introduction Empirics Model Estimation Counterfactuals
Most of this Driven By India
Foreign-born CS workers in the US by Country
By 2014, more than 70% of H1Bs went to Indians
And 86% of all CS H-1Bs went to Indians (5% to China)
Khanna & Morales And Other Unintended Consequences 3 / 33
Introduction Empirics Model Estimation Counterfactuals
US is a significant employer of Indian IT workers
Ratio of Indian college graduates in US relative to India by sector.
By the end of the period, for every 10 Indians in Indian IT,
there was 1 in the US.
Khanna & Morales And Other Unintended Consequences 4 / 33
Introduction Empirics Model Estimation Counterfactuals
Raises Expected Wage Premium for CS in India
Figure: Relative Wages CS to non-CS for Indians
Khanna & Morales And Other Unintended Consequences 5 / 33
Introduction Empirics Model Estimation Counterfactuals
Raises Expected Wage Premium for CS in India
Figure: Relative Wages CS to non-CS for Indians
In 2010 CS wage is 4 times higher in the US
and 15% of Indian CS work in the US
Khanna & Morales And Other Unintended Consequences 5 / 33Introduction Empirics Model Estimation Counterfactuals
Indian students enrolled in Engineering/CS
Led to increase in CS enrollment in India
Khanna & Morales And Other Unintended Consequences 6 / 33Introduction Empirics Model Estimation Counterfactuals
Indian students enrolled in Engineering/CS
“growth (in training and degrees) was driven by larger
salaries in the IT industry abroad” (Bhatnagar 2005)
(since few domestic IT jobs in 90s)
Khanna & Morales And Other Unintended Consequences 7 / 33Introduction Empirics Model Estimation Counterfactuals
Boom in Indian IT firms
1 But H-1Bs capped – not all workers can go abroad.
2 Visas expire after 3-6 years – returns bring acquired skills
3 Firms in India tap into skilled workforce.
Khanna & Morales And Other Unintended Consequences 8 / 33Introduction Empirics Model Estimation Counterfactuals
Boom in Indian IT firms
1 But H-1Bs capped – not all workers can go abroad.
2 Visas expire after 3-6 years – returns bring acquired skills
3 Firms in India tap into skilled workforce.
IT Sector Output in India
Khanna & Morales And Other Unintended Consequences 8 / 33Introduction Empirics Model Estimation Counterfactuals
India Becomes Major Exporter of IT
Share of IT Exports: US, India and Rest of the World
India overtakes the US in 2005 as an exporter of IT
Khanna & Morales And Other Unintended Consequences 9 / 33Introduction Empirics Model Estimation Counterfactuals
What role did Immigration play in the spread of
this tech boom to the other side of the world?
Khanna & Morales And Other Unintended Consequences 10 / 33Introduction Empirics Model Estimation Counterfactuals
This Paper
Evidence of ‘brain-gain’ driven by migration prospects
Change in Prob(migration) affect occupation-choice in India
IV: variation in H-1B cap + baseline occupation shares
Khanna & Morales And Other Unintended Consequences 11 / 33Introduction Empirics Model Estimation Counterfactuals
This Paper
Evidence of ‘brain-gain’ driven by migration prospects
Change in Prob(migration) affect occupation-choice in India
IV: variation in H-1B cap + baseline occupation shares
Build 3 country, 2 sector, quantitative GE model
(di Giovanni et. al. 2015, Desmet et. al. 2018, Llull 2018, Burstein et. al.
2019, Colas 2019, Caliendo et. al. 2020, Morales 2020)
Dynamic occupational choice for college grads: CS vs Other.
Indians: uncertain on migration when choosing occupations
Migrants can return once visas expire (‘brain circulation’)
Trade in IT good and final good.
Innovation in CS improve technology
Khanna & Morales And Other Unintended Consequences 11 / 33Introduction Empirics Model Estimation Counterfactuals
This Paper
Estimate key model parameters:
India labor supply response to US demand shocks
Dynamic labor supply: short-run vs long-run elasticity.
Counterfactual: decrease H-1B cap by 50% from 1995-2010
Khanna & Morales And Other Unintended Consequences 12 / 33Introduction Empirics Model Estimation Counterfactuals
This Paper
Estimate key model parameters:
India labor supply response to US demand shocks
Dynamic labor supply: short-run vs long-run elasticity.
Counterfactual: decrease H-1B cap by 50% from 1995-2010
Immigration raised welfare in US (0.007%) & India (0.13%)
US IT sector 0.11% larger & India’s is 35% larger
Khanna & Morales And Other Unintended Consequences 12 / 33Introduction Empirics Model Estimation Counterfactuals
This Paper
Estimate key model parameters:
India labor supply response to US demand shocks
Dynamic labor supply: short-run vs long-run elasticity.
Counterfactual: decrease H-1B cap by 50% from 1995-2010
Immigration raised welfare in US (0.007%) & India (0.13%)
US IT sector 0.11% larger & India’s is 35% larger
Importance of: Trade, endogenous college decisions,
imperfect substitution, heterogeneous abilities, remittances
Endogenous labor supply is key to quantify welfare:
Welfare gains of immigration for US (0.007% vs 0.04%) and
India (0.13% vs -0.05%).
Khanna & Morales And Other Unintended Consequences 12 / 33Introduction Empirics Model Estimation Counterfactuals
India Labor Response to Migration Opportunities
Did migration to US impact occupational choice in India?
Leverage variation across occupations o & regions r in India
Ln(Nort ) = γ1 Ln (Migration probort ) +δor + δrt + ϵort
| {z }
Indians in US past 5 yrsort
Tot. US Mig. Applicantst
Khanna & Morales And Other Unintended Consequences 13 / 33Introduction Empirics Model Estimation Counterfactuals
India Labor Response to Migration Opportunities
Did migration to US impact occupational choice in India?
Leverage variation across occupations o & regions r in India
Ln(Nort ) = γ1 Ln (Migration probort ) +δor + δrt + ϵort
| {z }
Indians in US past 5 yrsort
Tot. US Mig. Applicantst
Two issues:
1 Demand shocks between India and US might be correlated
2 Unobserved preference shocks for occupations may coincide
Khanna & Morales And Other Unintended Consequences 13 / 33Introduction Empirics Model Estimation Counterfactuals
India Labor Response to Migration Opportunities
Did migration to US impact occupational choice in India?
Leverage variation across occupations o & regions r in India
Ln(Nort ) = γ1 Ln (Migration probort ) +δor + δrt + ϵort
| {z }
Indians in US past 5 yrsort
Tot. US Mig. Applicantst
Two issues:
1 Demand shocks between India and US might be correlated
2 Unobserved preference shocks for occupations may coincide
1 Construct local demand control: o-r-industry (k) variation
k
!
X Nor,1994
Localort = Ln P × Total employment in k and t
k k Nor,1994
2 But OLS still biased γ1 . ϵort correlated with preferences
and demand shocks
Khanna & Morales And Other Unintended Consequences 13 / 33Introduction Empirics Model Estimation Counterfactuals
India Labor Response to Migration Opportunities
Ln(Nort ) = γ1 Ln(Indians in US past 5 yrsot ) + γ2 Localort + δor + δrt + ϵort
Instrument I: for Indians in US past 5 yrsot :
Isolate US demand for migrants (and changes in US policy)
Derive predicted migrant stock from H-1B caps H-1B cap
Khanna & Morales And Other Unintended Consequences 14 / 33Introduction Empirics Model Estimation Counterfactuals
India Labor Response to Migration Opportunities
Ln(Nort ) = γ1 Ln(Indians in US past 5 yrsot ) + γ2 Localort + δor + δrt + ϵort
Instrument I: for Indians in US past 5 yrsot :
Isolate US demand for migrants (and changes in US policy)
Derive predicted migrant stock from H-1B caps H-1B cap
(N Indians in US)o,1990
H-1B Instrumentot = × (Predicted H-1B stock)t
(N Indians in US)1990 | {z }
| {z } time shifter
Initial share
Changes to the H-1B cap plausibly uncorrelated with
Indian preference and demand shocks.
Alt Instruments Test correlated shocks
Khanna & Morales And Other Unintended Consequences 14 / 33Introduction Empirics Model Estimation Counterfactuals
India Labor Response to Migration Opportunities
OLS 2SLS
Log Employment All Log Employment Young Log Employment All Log Employment Young
Log Prob(Migrate to US) 0.204*** 0.169*** 0.740*** 1.065***
(0.0455) (0.0495) (0.262) (0.311)
Demand Control 0.654*** 0.716*** 0.724*** 0.833***
(0.214) (0.239) (0.225) (0.259)
N 3,114 3,114 3,114 3,114
1st stage F-stat 67.64 67.64
Regressions include occupation-region and region-time fixed effects. SE clustered at the occupation-region level. Columns 2
and 4 count only college graduates who are between 25 and 40. Main explanatory variable is the log number of Indian college
graduates, who migrated to the US in the past 5 years to work in occupation o. Sample includes 38 occupations, 30 Indian
states and 5 periods. Data from ACS (USA) and NSS (India), for 5 periods: 1994, 2000, 2006, 2010, and 2012.
Khanna & Morales And Other Unintended Consequences 15 / 33Introduction Empirics Model Estimation Counterfactuals
India Labor Response to Migration Opportunities
OLS 2SLS
Log Employment All Log Employment Young Log Employment All Log Employment Young
Log Prob(Migrate to US) 0.204*** 0.169*** 0.740*** 1.065***
(0.0455) (0.0495) (0.262) (0.311)
Demand Control 0.654*** 0.716*** 0.724*** 0.833***
(0.214) (0.239) (0.225) (0.259)
N 3,114 3,114 3,114 3,114
1st stage F-stat 67.64 67.64
Regressions include occupation-region and region-time fixed effects. SE clustered at the occupation-region level. Columns 2
and 4 count only college graduates who are between 25 and 40. Main explanatory variable is the log number of Indian college
graduates, who migrated to the US in the past 5 years to work in occupation o. Sample includes 38 occupations, 30 Indian
states and 5 periods. Data from ACS (USA) and NSS (India), for 5 periods: 1994, 2000, 2006, 2010, and 2012.
Alternative instruments: Demand for migrants Results
Non-Indian migrants in US by year.
Non-Indian migrants in US by year and occupation.
Khanna & Morales And Other Unintended Consequences 15 / 33Introduction Empirics Model Estimation Counterfactuals
India Labor Response to Migration Opportunities
OLS 2SLS
Log Employment All Log Employment Young Log Employment All Log Employment Young
Log Prob(Migrate to US) 0.204*** 0.169*** 0.740*** 1.065***
(0.0455) (0.0495) (0.262) (0.311)
Demand Control 0.654*** 0.716*** 0.724*** 0.833***
(0.214) (0.239) (0.225) (0.259)
N 3,114 3,114 3,114 3,114
1st stage F-stat 67.64 67.64
Regressions include occupation-region and region-time fixed effects. SE clustered at the occupation-region level. Columns 2
and 4 count only college graduates who are between 25 and 40. Main explanatory variable is the log number of Indian college
graduates, who migrated to the US in the past 5 years to work in occupation o. Sample includes 38 occupations, 30 Indian
states and 5 periods. Data from ACS (USA) and NSS (India), for 5 periods: 1994, 2000, 2006, 2010, and 2012.
Alternative instruments: Demand for migrants Results
Non-Indian migrants in US by year.
Non-Indian migrants in US by year and occupation.
Dimensions of the data Results
Re-weighting explanatory variable at region level
Occupation-by-time only
Khanna & Morales And Other Unintended Consequences 15 / 33Introduction Empirics Model Estimation Counterfactuals
Additional Analysis
No Pre-Trends Pre Trends
Shift-share tests Baseline
Khanna & Morales And Other Unintended Consequences 16 / 33Introduction Empirics Model Estimation Counterfactuals
Additional Analysis
No Pre-Trends Pre Trends
Shift-share tests Baseline
Testing for correlated shocks Test correlated shocks
Adding trade demand controls
No demand from other countries (e.g. UK, Canada)
Khanna & Morales And Other Unintended Consequences 16 / 33Introduction Empirics Model Estimation Counterfactuals
Additional Analysis
No Pre-Trends Pre Trends
Shift-share tests Baseline
Testing for correlated shocks Test correlated shocks
Adding trade demand controls
No demand from other countries (e.g. UK, Canada)
(a) Indians in US vs UK (b) Indians in US vs Canada
Khanna & Morales And Other Unintended Consequences 16 / 33Introduction Empirics Model Estimation Counterfactuals
Model Overview
Dynamic Labor Supply for college graduates
College-major and Occupation choice: CS vs. Other
Uncertainty on migration when choosing occupations →
drives “brain gain”.
Product market and Labor Demand:
3 Countries: (1) US, (2) India, (3) Rest of World
2 Tradeable Sectors: (1) Final Goods, and (2) IT Good
CS generate spillover on technology when working in IT
Khanna & Morales And Other Unintended Consequences 17 / 33Introduction Empirics Model Estimation Counterfactuals
Dynamic Labor Supply
3 types: Non-college, College-CS (cs), College-Other (g)
1 Before joining labor market, college students choose major:
g g
max{βEt Vtcs
+1 + F̄ + σηi,t , βEt Vt+1 + σηi,t }
cs
Rational expectations over wages and migration cap.
o iid
Heterogeneous preferences: ηi,t ∼ Type I Extreme Value.
Khanna & Morales And Other Unintended Consequences 18 / 33Introduction Empirics Model Estimation Counterfactuals
Dynamic Labor Supply
3 types: Non-college, College-CS (cs), College-Other (g)
1 Before joining labor market, college students choose major:
g g
max{βEt Vtcs
+1 + F̄ + σηi,t , βEt Vt+1 + σηi,t }
cs
Rational expectations over wages and migration cap.
o iid
Heterogeneous preferences: ηi,t ∼ Type I Extreme Value.
2 Dynamic, occupation-choice thereafter
(
o
Vt,a = max wo
t + χ(a) × 1(ot ̸= ot−1 ) + ζo +
o |{z} | {z } |{z}
current wage switching cost Occ. Avg taste
)
βEt [Vto+1,a+1 ] + o
σηi,t
| {z } | {z }
future payoffs preferences
Occu-switching: mitigates wage impacts of immigration
Khanna & Morales And Other Unintended Consequences 18 / 33Introduction Empirics Model Estimation Counterfactuals
Migration Expectations in India
Timing: Choose occupation → Work/Study in India in t →
If CS, draw immigration probability pt+1 → If selected,
migrate and work in t + 1
Et [Vtcs
+1,a+1 ] = Et [pt+1 × Vt+1,a+1 + (1 − pt+1 ) × Vt+1,a+1 ] ,
us in
| {z } | {z }
Payoff migrate Payoff stay
Prob(Migration):
H1B capt+1
pt + 1 =
CSt,in
Khanna & Morales And Other Unintended Consequences 19 / 33Introduction Empirics Model Estimation Counterfactuals
Migration Expectations in India
Timing: Choose occupation → Work/Study in India in t →
If CS, draw immigration probability pt+1 → If selected,
migrate and work in t + 1
Et [Vtcs
+1,a+1 ] = Et [pt+1 × Vt+1,a+1 + (1 − pt+1 ) × Vt+1,a+1 ] ,
us in
| {z } | {z }
Payoff migrate Payoff stay
Prob(Migration):
H1B capt+1
pt + 1 =
CSt,in
Exogenous probability of returning to India
in,R
+1,a+1 = κwus,t+1 + βEt+1 [(1 − ϱ)Vt+2,a+2 + ϱVt+2,a+2 ] ,
Vtus cs us
κ migration sensitivity, ϱ return probability.
Khanna & Morales And Other Unintended Consequences 19 / 33Introduction Empirics Model Estimation Counterfactuals
Consumers and Production
CES production in final goods s = f , and IT goods s = c
σy
σy −1
!
Z σy −1
γ
where yf i = zf i yci Nf1−γ , s = {f , c}
σy
Ys = ysi di i
i∈Ω
IT good yci is intermediate good in Final output
Growth and innovation in IT affects downstream sectors
Khanna & Morales And Other Unintended Consequences 20 / 33Introduction Empirics Model Estimation Counterfactuals
Consumers and Production
CES production in final goods s = f , and IT goods s = c
σy
σy −1
!
Z σy −1
γ
where yf i = zf i yci Nf1−γ , s = {f , c}
σy
Ys = ysi di i
i∈Ω
IT good yci is intermediate good in Final output
Growth and innovation in IT affects downstream sectors
Ni nested CES composite of 3 types of workers:
(1) non-grads, (2) non-CS grads, (3) CS
(4) Return CS can ‘bring back knowledge’
Greater intensity of CS in IT sector
As CS workforce increases, demand for complements rise
Sector- and Skill-biased technical change over time
Khanna & Morales And Other Unintended Consequences 20 / 33Introduction Empirics Model Estimation Counterfactuals
Technology and International Trade
Rest of the World W produces both goods.
3 countries export and import both goods.
US and India compete for world market
Production can shift to other countries
Trade mitigates impacts: workers/production shift sectors
Khanna & Morales And Other Unintended Consequences 21 / 33Introduction Empirics Model Estimation Counterfactuals
Technology and International Trade
Rest of the World W produces both goods.
3 countries export and import both goods.
US and India compete for world market
Production can shift to other countries
Trade mitigates impacts: workers/production shift sectors
In each country and sector, productivities z drawn from
−θ
Frechet F (z ) = e−Tk z . (Eaton-Kortum 2002)
Tc country-specific technology parameter for each sector
Khanna & Morales And Other Unintended Consequences 21 / 33Introduction Empirics Model Estimation Counterfactuals
Technology and International Trade
Rest of the World W produces both goods.
3 countries export and import both goods.
US and India compete for world market
Production can shift to other countries
Trade mitigates impacts: workers/production shift sectors
In each country and sector, productivities z drawn from
−θ
Frechet F (z ) = e−Tk z . (Eaton-Kortum 2002)
Tc country-specific technology parameter for each sector
CS innovators: (spillovers) increase Tc,k = T (CSc,k )
Benefit all workers
Substantial consumer surplus
Equilibrium
Khanna & Morales And Other Unintended Consequences 21 / 33Introduction Empirics Model Estimation Counterfactuals
First Estimate Fundamental Elasticities
1 India’s labor supply response to migration opportunity
Estimated using IV ✓
2 Computer Science innovation elasticity: Tk = T (CSk )
Estimate impact on patents
Khanna & Morales And Other Unintended Consequences 22 / 33Introduction Empirics Model Estimation Counterfactuals
First Estimate Fundamental Elasticities
1 India’s labor supply response to migration opportunity
Estimated using IV ✓
2 Computer Science innovation elasticity: Tk = T (CSk )
Estimate impact on patents
Log (P atents)j,t = δj + δt + βtech CS
[ j,t + εj,t
IV: (1) Immigrants concentrated in CS, (2) H1B cap/policy changes
3 Labor supply elasticity (wrt wages)
Dynamic: Long run v Short run elasticity
SMM Identification Details
Khanna & Morales And Other Unintended Consequences 22 / 33Introduction Empirics Model Estimation Counterfactuals
First Estimate Fundamental Elasticities
1 India’s labor supply response to migration opportunity
Estimated using shift-share ✓
2 Computer Science innovation elasticity: Tk = T (CSk )
Estimate impact on patents ✓
3 Labor supply elasticity (wrt wages)
Dynamic: Long run v Short run elasticity
SMM Identification Details
Khanna & Morales And Other Unintended Consequences 23 / 33Introduction Empirics Model Estimation Counterfactuals
Labor supply elasticity (wrt wages)
“Innovation Shocks” shift out labor demand
‘Trace out’ labor supply curve and identify parameters
Khanna & Morales And Other Unintended Consequences 24 / 33Introduction Empirics Model Estimation Counterfactuals
Labor supply elasticity (wrt wages)
“Innovation Shocks” shift out labor demand
‘Trace out’ labor supply curve and identify parameters
Change in relative labor supply (CS vs. Other) after 1%
permanent increase in relative wage:
Long vs. Short run elasticity.
Khanna & Morales And Other Unintended Consequences 24 / 33Introduction Empirics Model Estimation Counterfactuals
Model Fit: Out-of-Sample Test
(a) Relative Wage for Computer (b) Relative Wage for non-CS
Scientists: US to India College Grads: US to India
(c) Price Index IT US (d) IT Output US
Khanna & Morales And Other Unintended Consequences 25 / 33Introduction Empirics Model Estimation Counterfactuals
What If We Restricted Migration?
Reduce migration cap by 50%
Khanna & Morales And Other Unintended Consequences 26 / 33Introduction Empirics Model Estimation Counterfactuals
Ambiguous Predictions
Effects on the sending country: India
Brain-drain vs Brain-gain (skill acquisition) and
Brain-circulation: return migrants
(Stark 2009, Easterly & Nyarko 2009, Abarcar & Theoharides 2020)
Khanna & Morales And Other Unintended Consequences 27 / 33Introduction Empirics Model Estimation Counterfactuals
Ambiguous Predictions
Effects on the sending country: India
Brain-drain vs Brain-gain (skill acquisition) and
Brain-circulation: return migrants
(Stark 2009, Easterly & Nyarko 2009, Abarcar & Theoharides 2020)
Effects on the receiving country: US
Immigrants expand IT sector vs Production shifting abroad
Deteriorate terms of trade: IT exports cheaper (Krugman
1979, Davis & Weinstein 2002, Acemoglu, Gancia & Zilibotti 2015)
Khanna & Morales And Other Unintended Consequences 27 / 33Introduction Empirics Model Estimation Counterfactuals
Ambiguous Predictions
Effects on the sending country: India
Brain-drain vs Brain-gain (skill acquisition) and
Brain-circulation: return migrants
(Stark 2009, Easterly & Nyarko 2009, Abarcar & Theoharides 2020)
Effects on the receiving country: US
Immigrants expand IT sector vs Production shifting abroad
Deteriorate terms of trade: IT exports cheaper (Krugman
1979, Davis & Weinstein 2002, Acemoglu, Gancia & Zilibotti 2015)
Effects on workers in both countries
More CS (lower wages) vs more innovation (higher wages)
For non-CS: more demand (higher wages) vs more switching
into non-CS (lower wages)
(Borjas 1999, Kerr & Lincoln 2010, Hunt & Gauthier-Loiselle 2010,
Peri et. al. 2015, Doran et. al. 2017, Bound Khanna & Morales 2017)
Consumers: better off from lower prices, more productivity
Khanna & Morales And Other Unintended Consequences 27 / 33Introduction Empirics Model Estimation Counterfactuals
Impact of Migration in 2010
No occupational choice
Baseline In both countries In India only
Wages
US CS workers -0.97%
India CS workers -11.5%
Occupational Choice
US CS (native plus immigrant) 1.65%
US CS workers -5.87%
India CS workers 56.6%
IT production
US IT output 0.11%
India IT output 35.55%
Welfare
Welfare of US natives 0.007%
Welfare in India 0.13%
Khanna & Morales And Other Unintended Consequences 28 / 33Introduction Empirics Model Estimation Counterfactuals
Impact of Migration in 2010
No occupational choice
Baseline In both countries In India only
Wages
US CS workers -0.97% -3.28% -0.80%
India CS workers -11.5% 0.60% 0.9%
Occupational Choice
US CS (native plus immigrant) 1.65% 7.50% 1.95%
US CS workers -5.87% - -5.56%
India CS workers 56.6% - -
IT production
US IT output 0.11% 4.20% 1.30%
India IT output 35.55% -10.2% -9.80%
Welfare
Welfare of US natives 0.007% 0.04% 0.009%
Welfare in India 0.13% -0.05% -0.05%
Brain Gain > Brain Drain: If can’t switch occupations
India CS / IT can’t grow; US IT output will rise
Khanna & Morales And Other Unintended Consequences 29 / 33Introduction Empirics Model Estimation Counterfactuals
Distributional Welfare
(a) Welfare Change: US native (b) Welfare Change: Workers in
Workers (USD mn) India (USD mn)
Khanna & Morales And Other Unintended Consequences 30 / 33Introduction Empirics Model Estimation Counterfactuals
Mechanisms
Return Migration: Results
1 Bring back tech knowhow and enlarge Indian IT sector.
2 No return migration almost doubles US welfare gains.
Innovation Spillover: Results
1 Key for brain-gain result. Robust to changes in magnitude.
Khanna & Morales And Other Unintended Consequences 31 / 33Introduction Empirics Model Estimation Counterfactuals
Mechanisms
Return Migration: Results
1 Bring back tech knowhow and enlarge Indian IT sector.
2 No return migration almost doubles US welfare gains.
Innovation Spillover: Results
1 Key for brain-gain result. Robust to changes in magnitude.
Trade and Remittances: Results
1 Trade shifts resources to other sectors in US
2 Lower trade costs: more shift in IT production to India
3 Remittances shift where income is spent
Alternative counterfactuals:
1 Vary cap-size – results are non-linear but monotonic Results
2 Restrictions in later years: different consequences Results
Khanna & Morales And Other Unintended Consequences 31 / 33Introduction Empirics Model Estimation Counterfactuals
Alternative modeling specifications
Endogenous college decision. Results
US: More high-school grads. India: More college grads.
Slightly higher welfare for both.
Imperfect substitution: Natives vs. Immigrants in US.
Higher welfare in US, IT sector larger. Results
India captures lower market share.
Heterogeneity in abilities & migration selection heterogeneity
US has larger welfare gains: more skilled migrants.
India has lower response, marginal workers switching to CS,
more able workers leaving.
Khanna & Morales And Other Unintended Consequences 32 / 33Introduction Empirics Model Estimation Counterfactuals
US Immigration Policy Partly Affected Structural
Change in India
Halving H-1B migration reduces welfare by 0.14%
$6.6 bill or $50K per migrant ($45K goes to migrant)
Distributional consequences of migration:
In US and India, native computer scientists wages lower
Non-CS better off from immigration.
Important to model occupational switching, trade,
innovation, price changes, wage expectations....
Endogenous skill acquisition key to quantify gains
Khanna & Morales And Other Unintended Consequences 33 / 33Extra
Thank You!
Khanna & Morales And Other Unintended Consequences 34 / 33Extra
H-1B cap changes
Back
Khanna & Morales And Other Unintended Consequences 35 / 33Extra
Alternative Instruments
(N Indians in US)o,1990
Alt. IVot = × (Migrants from Other Countries)ot
(N Indians in US)1990 | {z }
| {z } time shifter
Initial share
Table: Alternative Instrumental Variables
Non Indian Migrants IV Non Indian Migrants by Occupation
Panel A: Alternative Instruments Log Employment All Log Employment Young Log Employment All Log Employment Young
Log Indians in US 0.629*** 0.770*** 0.582** 0.915***
migrated in past 5 years (0.118) (0.138) (0.240) (0.291)
Demand Control 0.710*** 0.795*** 0.703*** 0.814***
(0.222) (0.250) (0.221) (0.254)
N 3,114 3,114 3,114 3,114
1st stage F-stat 178.1 178.1 66.71 66.71
Notes: ∗ ∗ ∗ = p < 0.01, ∗∗ = p < 0.05, ∗ = p < 0.1 All regressions include occupation-region
and region-time fixed effects. SE clustered at the occupation-region level. Back
Khanna & Morales And Other Unintended Consequences 36 / 33Extra
Correlated Shocks: (1) Trade Controls
ind
!
X Nor,1994
Trade Controlort = Ln × Total value of US imports from INind,t
ind
Nor,1994
Table: Control for Trade Shocks
Baseline IV With return migrants
Log Employment All Log Employment Young Log Employment All Log Employment Young
Log Indians in US 0.732*** 1.059*** 0.698*** 1.034***
migrated in past 5 years (0.260) (0.309) (0.257) (0.308)
Demand Control 0.638*** 0.773*** 0.634*** 0.770***
(0.224) (0.261) (0.223) (0.260)
US Imports Control 0.0348*** 0.0262* 0.0346*** 0.0260*
(0.0125) (0.0151) (0.0125) (0.0150)
N 3,114 3,114 3,114 3,114
1st stage F-stat 67.44 67.44 46.24 46.24
Notes: ∗ ∗ ∗ = p < 0.01, ∗∗ = p < 0.05, ∗ = p < 0.1 All regressions include occupation-region
and region-time fixed effects. SE clustered at the occupation-region level. Back
Khanna & Morales And Other Unintended Consequences 37 / 33Extra
Correlated Shocks: (2) Exclude Demand Controls
Table: No Change When Exclude Local Demand Control
Baseline IV With return migrants
Log Employment All Log Employment Young Log Employment All Log Employment Young
Log Indians in US 0.875*** 1.178*** 0.835*** 1.150***
migrated in past 5 years (0.261) (0.309) (0.258) (0.308)
N 3,380 3,380 3,380 3,380
1st stage F-stat 68.44 68.44 68.62 68.62
Notes: ∗ ∗ ∗ = p < 0.01, ∗∗ = p < 0.05, ∗ = p < 0.1 All regressions include occupation-region
and region-time fixed effects. SE clustered at the occupation-region level. Back
Khanna & Morales And Other Unintended Consequences 38 / 33Extra
Correlated (3) Demand from Other Countries
(a) Indians in US vs UK (2012) (b) Indians in US vs Canada
(c) Indians in US vs UK (2000) (d) Indians in US vs UK
Khanna & Morales And Other Unintended Consequences 39 / 33Extra
Pre Trends Tests
Baseline Without CS Occupations No Demand Controls
Log Emp All Log Emp Young Log Emp All Log Emp Young Log Emp All Log Emp Young
Log Indians in US (t+1) -0.149 0.468 -0.291 0.464 -0.316 0.258
migrated in past 5 years (0.322) (0.434) (0.388) (0.536) (0.330) (0.427)
Demand Control 1.093*** 1.379*** 1.185*** 1.387***
(0.360) (0.415) (0.357) (0.407)
Observations 1,833 1,833 1,811 1,811 1,833 1,833
F stat 37.72 37.72 31.45 31.45 41.92 41.92
Back
Khanna & Morales And Other Unintended Consequences 40 / 33Extra
Shift-Share: Correlations with Base Shares
Education Levels at Baseline Labor Characteristics
Primary Secondary Upper Secondary Unemployed Family Enterprise Domestic Work
Log Indians in US -0.0157 0.0535 0.00366 -0.00760 -0.00943 -0.0113
migrated in past 5 years (0.0273) (0.0465) (0.0628) (0.00463) (0.00991) (0.0161)
Observations 1,154 1,154 1,154 3,114 3,114 3,114
F stat 33.43 33.43 33.43 67.64 67.64 67.64
Back
Khanna & Morales And Other Unintended Consequences 41 / 33Extra
Shift-Share: Baseline Correlations
Share of Indians in US (1994) x 100
P(Male) in 1994 -0.412 -0.45 -1.12 -1.212
(0.209) (0.214) (0.311) (0.292)
Age in 1994 0.0328 0.0366 0.00716 -0.00549
(0.00984) (0.0102) (0.0169) (0.0153)
Highest Grade: Primary Schooling in 1994 -1.852 -2.108 -2.183 0.128
(0.251) (0.265) (0.499) (0.532)
Highest Grade: Middle Schooling in 1994 -0.521 -0.616 -1.694 -0.180
(0.220) (0.227) (0.354) (0.369)
Highest Grade: High School in 1994 1.139 1.203 -0.369 0.725
(0.307) (0.314) (0.454) (0.452)
Unemployed and Looking for Work 1994 -12.04 -9.097 -142.8 -87.91
(26.98) (27.20) (193.4) (183.4)
Self Employed 1994 -0.229 -0.185 -0.936 -0.408
(0.177) (0.181) (0.364) (0.347)
Family Worker 1994 -0.870 -0.830 3.661 6.731
(0.633) (0.646) (2.032) (1.937)
Domestic Worker 1994 0.279 -0.846 -3.546 -3.317
(6.843) (6.947) (7.956) (7.380)
College Graduates 1994 2.323
(0.306)
Constant 0.642 0.580 3.511 2.106
(0.472) (0.490) (0.809) (0.772)
Region Fixed Effects No Yes Yes Yes
Occupations All All Skilled Skilled
Observations 2,432 2,432 954 954
Adjusted R-squared 0.0386 0.0331 0.0278 0.105
Khanna & Morales And Other Unintended Consequences 42 / 33Extra
Sensitivity to Dropping Occupations-Regions
Figure: Sensitivity to Occupations and States
Dropping Occupations One at a Dropping Each State One at a
Time Time
Back
Khanna & Morales And Other Unintended Consequences 43 / 33Extra
Correlations Between Residualized Growth and
Baseline Shares
Figure: Correlations Between Growth and Baseline Shares
Employment vs Baseline Share Change in Employment vs
Weight Baseline Share Weight
Back
Khanna & Morales And Other Unintended Consequences 44 / 33Extra
Dimensions of the Data
(N in India)o,r,1991 (Indians in US)o,1990
IVort = × × (Predicted H-1B stock)t
(N in India)o,1991 (Indians in US)1990 | {z }
| {z } time shifter
Initial share
Table: Regional Reweighting and Occ-Time Regressions
Regional-reweighted Log Employment All Log Employment Young Log Employment All Log Employment Young
Log Indians in US 0.925** 1.245** 0.810*** 0.923***
migrated in past 5 years (0.457) (0.533) (0.281) (0.346)
Demand Control 0.725** 0.822** 0.728*** 0.828***
(0.298) (0.370) (0.280) (0.316)
N 3,087 3,087 3,087 3,087
1st stage F-stat 9.429 9.429 12.10 12.10
Occ-Time Level Main Specification With Occupation trends
Log Employment All Log Employment Young Log Employment All Log Employment Young
Log Indians in US 0.882*** 0.974*** 0.900*** 0.995***
migrated in past 5 years (0.293) (0.344) (0.306) (0.359)
N 187 187 177 177
1st stage F-stat 16.50 16.50 15.39 15.39
Back
Khanna & Morales And Other Unintended Consequences 45 / 33Extra
Equilibrium
Set of prices, wages, and level of technology such that:
Consumers maximize utility taking prices as given.
College grads choose major and occupation, taking wages as
given, and forming expectations.
Firms maximize profits taking wages and prices as given.
Trade between the three regions is balanced.
Output and labor markets clear.
Trade share by country/sector:
T s (ds ξt,b s )−θ
s
πt,k,b = P t,b s t,k,b
s s −θ ,
b Tt,b (dt,k,b ξt,b )
Khanna & Morales And Other Unintended Consequences 46 / 33Extra
Equilibrium
Price index by sector/country:
!− 1
θ
−θ
,
X
s s s s
Pt,k = γ̄ Tt,k ′ (dt,k,k ′ ξt,k ′ )
k′
The flow between occupations for college graduates:
o,o ′
o,o′ exp( σ1k V̄t,k,a )
πt,k,a = o,o ′ o,o
,
exp( σ1k V̄t,k,a ) + exp( σ1k V̄t,k,a )
Back
Khanna & Morales And Other Unintended Consequences 47 / 33Extra
Solution Process
1) Set fundamental elasticities from the literature: Details
Elasticities of substitution (estimate for India)
Trade elasticity.
2) Initially guess labor supply parameters
Occu-switching costs, mean occupation preference,
preference dispersions, education costs
Migration sensitivity (κ): (match ‘brain-gain’ elasticity)
Khanna & Morales And Other Unintended Consequences 48 / 33Extra
Solution Process
1) Set fundamental elasticities from the literature: Details
Elasticities of substitution (estimate for India)
Trade elasticity.
2) Initially guess labor supply parameters
Occu-switching costs, mean occupation preference,
preference dispersions, education costs
Migration sensitivity (κ): (match ‘brain-gain’ elasticity)
3) For those parameters calculate equilibrium Details
Time varying parameters to match data trends
Technology parameters and trade costs
4) Compute simulated moments and compare to data
If different, go back to 2) and calculate equilibrium with
new set of parameters
Intuition, Identification and Moments Identification
Khanna & Morales And Other Unintended Consequences 48 / 33Extra
Elasticities and Time-varying parameters
Product market elasticities Back
τ = 1.7 (Katz and Murphy 1992)
λ = 2 (Ryoo and Rosen 2004, Burstein et. al 2020)
ϵ = 30 (match return wage premium in 1994).
θ = 8.28 (Eaton and Kortum 2002)
Parameters calibrated every year to match trends:
γk : share of income from the final goods sector spent on IT
αt,k : share of expenditures from the final goods sector in
non-college graduate
δt,k : within-country relative wages between CS and non-CS
graduates
s : rel. wages of non-grads between India and the US.
Tt,k
dt,k,b : observed trade flows.
s
Return rate: 0.23 (Lee 2016).
Khanna & Morales And Other Unintended Consequences 49 / 33Extra
Identification
All labor supply parameters jointly estimated. Back
Intuition of identification based on specific moments:
Share of CS in 1995 and 2010:
Identifies the mean taste for other occupation
The change in the CS share over time identifies the
responsiveness of supply to wages → σ
Relative gross flows between those who switch and those
who stay in previous occupation 1995-2000:
Identifies baseline switching costs.
Ratioof CS Share (45-60yrs old) to CS Share (31-60yrs
old), 2010:
Identifies age-dependent switching costs.
Khanna & Morales And Other Unintended Consequences 50 / 33Extra
Identification
Ratio of CS Share (25-30yrs old) to CS Share (31-60yrs
old), 2010:
Identifies the education cost
More young people means the education cost is lower than
switching cost.
Response in CS labor supply to a 10% decrease of the
migration cap.
Identifies migration sensitivity, how much migration
opportunities affect CS choices.
5 US moments for 5 parameters and 6 India moments for 6
India parameters. Back
Khanna & Morales And Other Unintended Consequences 51 / 33Extra
Identification
Table: Empirical vs. Simulated moments Back
US India
Data Moments Simulated Moments Data Moments Simulated Moments
Share CS 1995 2.94% 2.92% 0.12% 0.11%
[2.71%-3.17%] [0.06%-0.19%]
Share CS 2010 3.90% 3.68% 2.76% 2.82%
[3.66%-4.15%] [2.28%-3.24%]
Transition rate 95-00 1.39% 1.20% 0.19% 0.70%
[0.10%-0.28%] [0.1%-0.28%]
Ratio CS Share [25-30]/[31-60] 2010 1.00 1.16 3.96 3.27
[0.83-1.17] [2.45-5.47]
Ratio CS share [45-60]/[31-60] 2010 0.87 0.81 0.19 0.15
[0.80-0.94] [0.05-0.33]
Supply response to migration 0.74 0.76
[0.25-1.28]
Simulated method of moments results comparing empirical moments to data moments. ‘Supply
response to migration’ is based on response estimated in empirical section. ‘Transition rate’ is
defined as the net occupational switching rate between CS and non-CS occupations. ‘Ratio CS
Share’ is the share of CS workers in age group 25-30 and age group 31-60. ‘Share CS’ is the
share of the college graduate workforce that is in CS. 95% confidence intervals for empirical
moments are in parenthesis.
Khanna & Morales And Other Unintended Consequences 52 / 33Extra
Identification
Table: Labor supply parameter estimates Back
US India
Mean taste for non-CS occupations 0.47 0.45
[0.42,0.51] [0.33,0.58]
Preference dispersion parameter 0.46 0.26
[0.40,0.52] [0.18,0.32]
Education cost for CS degrees -0.09 -0.02
[-0.10,-0.08] [-0.02,-0.01]
Occupation switching costs -0.65 -1.09
[-0.83,-0.48] [-1.33,-0.85]
Age-specific switching cost -0.29 -0.01
[-0.34,-0.24] [-0.01,-0.005]
Utility weight on US wage 2.72
[1.85,3.58]
Estimated labor supply parameters based on the simulated method of moments exercise. These
parameters jointly determine the short-run and long-run labor supply elasticities. ‘Mean taste
for non-CS’ is the average preference for non-CS occupations. ‘Preference dispersion’ is the
variance in tastes for occupations. ‘Switching costs’ are occupation switching costs between CS
and non-CS occupations. ‘Education cost for CS’ is the average cost / non-preference for CS
degrees. All costs and tastes in units of the numeraire (consumption basket). ‘Utility weight
on US wage’ is the non-pecuniary premium on US wages (relative to India) for Indian workers.
Confidence intervals for parameters are the 5th and the 95th percentiles of the bootstrapped
(with replacement) sample.
Khanna & Morales And Other Unintended Consequences 53 / 33Extra
Technology Spillovers
Technology level depends on CS workers: (T (CSk,t )).
Khanna & Morales And Other Unintended Consequences 54 / 33Extra
Technology Spillovers
Technology level depends on CS workers: (T (CSk,t )).
Technology in sector j year t:
Log (P atents)j,t = δj + δt + βtech CS
[ j,t + εj,t
Khanna & Morales And Other Unintended Consequences 54 / 33Extra
Technology Spillovers
Technology level depends on CS workers: (T (CSk,t )).
Technology in sector j year t:
Log (P atents)j,t = δj + δt + βtech CS
[ j,t + εj,t
We Leverage:
(1) Immigration concentrated in CS, and
(2) H1B cap / policy changes Back
!
Imm CSj,0
CSj,t = δj + δt + γ Imm CSt + ϵj,t
Empj,0
Khanna & Morales And Other Unintended Consequences 54 / 33Extra
Technology Parameter between 0.22 and 0.24
CS workers Log(Patents) Log(Patents)
Shift Share 1.258***
(0.309)
CS workers 5.45e-06** 5.79e-06**
(2.59e-06) (2.56e-06)
Observations 275 275 275
R-squared 0.877 0.967 0.967
Number of Industries 25 25 25
Additional Controls No. of Firms No. of Firms None
F stat 16.67 16.31
Elasticity 0.226 0.240
SE (0.107) (0.126)
Peri et al (2014) = 0.27; Kerr and Lincoln (2010) = 0.1 to 0.4;
Bound, Khanna, Morales (2017)=0.23; Khanna & Lee (2018)=
0.2 Back
Khanna & Morales And Other Unintended Consequences 55 / 33Extra
With No Return Migration
Table: Brain Drain vs Brain Gain: Main Outcomes
No occupational No migration No return
Baseline
choice in India but reallocation migration
Occupational Choice
US CS (native plus immigrant) 1.65% 1.95% -0.26% 2.51%
US CS native -5.87% -5.56% -0.27% -16.80%
India CS 56.64% -10.21% 64.14% 40.97%
IT production
US IT output 0.11% 1.30% -0.79% 0.62%
India IT output 35.55% -9.80% 36.97% 28.46%
Total Welfare
Welfare of US natives 0.007% 0.009% -0.002% 0.012%
Welfare in India 0.130% -0.047% 0.149% 0.106%
Back
Khanna & Morales And Other Unintended Consequences 56 / 33Extra
Technology Spillovers
Table: Varying the Elasticity of Endogenous Technological Spillovers
US spillover=0.23 Spillover = 0.1
Baseline No spillovers
No spillover India in both countries
IT production
US IT output 0.11% 0.22% 0.28% 0.18%
India IT output 35.55% 21.81% 22.38% 27.56%
Welfare
Welfare of US natives 0.007% 0.003% 0.006% 0.005%
Welfare in India 0.130% -0.050% -0.052% 0.026%
Total (with migrants) 0.142% 0.085% 0.087% 0.111%
Back
Khanna & Morales And Other Unintended Consequences 57 / 33Extra
Robustness to Parameters
Table: Effect of Migration: Robustness to Parameters
Baseline τ =3 θ=4 λ = 2.5 ϵ = 10
Occupational choice
US CS (native plus immigrant) 1.6% 1.3% 0.6% 3.5% 1.5%
US CS native -5.9% -6.0% -6.2% -3.9% -5.9%
India CS 56.6% 55.8% 50.1% 62.2% 56.3%
Welfare
Welfare of US natives 0.007% 0.021% -0.002% 0.021% 0.006%
Welfare in India 0.13% 0.15% 0.09% 0.15% 0.14%
Combined welfare 0.13% 0.11% 0.13% 0.11% 0.14%
Back
Khanna & Morales And Other Unintended Consequences 58 / 33Extra
Trade and Remittances
Table: Increased trade, remittances and age-specific switching cost
10% lower No age-specific
Baseline Remittances
trade costs switching cost
IT production
US IT output 0.11% -0.07% 0.11% -0.06%
India IT output 35.55% 20.28% 35.61% 35.45%
Welfare
Welfare of US natives 0.007% 0.005% 0.006% 0.005%
Welfare in India 0.130% 0.123% 0.132% 0.132%
Combined (with migrants) 0.142% 0.136% 0.142% 0.141%
Back
Khanna & Morales And Other Unintended Consequences 59 / 33Extra
Vary the Cap Sizes
Figure: Welfare Loss by Different Cap Sizes (USD mn)
Back
Khanna & Morales And Other Unintended Consequences 60 / 33Extra
Change the Period the Restriction Begins
(a) Welfare Change: US natives (b) Welfare Change: India
(USDmn) non-migrants (USDmn)
Back
Khanna & Morales And Other Unintended Consequences 61 / 33Extra
Alternative modeling specifications
Table: Imperfect substitution between native and migrants, and
endogenous college
Imperfect Endogenous
Baseline
substitution college
IT production
US IT output 0.11% 0.93% 0.20%
India IT output 35.55% 33.27% 31.57%
Occupational choice
US CS (native plus immigrant) 1.65% 1.79% 1.69%
India CS 56.64% 52.29% 48.13%
US college non-CS 0.24% 0.24% 0.17%
India college non-CS -1.04% -0.91% 1.81%
US high-school - - 0.03%
India high-school - - -0.18%
Welfare
Welfare US natives 0.01% 0.04% 0.01%
Welfare in India 0.13% 0.12% 0.14%
Combined (with migrants) 0.14% 0.18% 0.14%
Back
Khanna & Morales And Other Unintended Consequences 62 / 33Extra
Heterogeneity in Abilities
Table: The Impact of More Immigration
Heterogeneity in
Baseline
Abilities
US India US India
IT output 0.24% 36.22% 1.83% 13.79%
Welfare
Always CS -1.14% -12.93% -2.34% -3.40%
Switchers -0.58% -11.73% -0.01% -5.57%
Always non-CS graduates -0.02% 0.93% 0.06% 0.46%
Non college 0.12% 0.02% 0.14% -0.04%
Total Natives 0.005% 0.13% 0.08% 0.06%
Migrants 41.43% 50.14%
Back
Khanna & Morales And Other Unintended Consequences 63 / 33You can also read
