Issuu

Bottlenecks: Sectoral Imbalances and the US Productivity Slowdown Daron Acemoglu David Autor daron@mit.edu dautor@mit.edu MIT MIT Christina Patterson christina.patterson@chicagobooth.edu Chicago Booth

OECD, July 9, 2021 1 / 28

Surge in innovation in ICT and electronics

Suggests we are on the verge of ”singularity” (Kurzweil 2005, Bostrom 2014, Diamandis and Kotler 2012) 2 / 28

But slowdown in productivity growth in recent decades 1.3

1.2

1.1

1 1980

1990

2000

2010

2020

year Private Business Sector Manufacturing

Suggests we are in a new age of slower growth (Cowen 2011, Gordon 2017) 3 / 28

Slowdown is not just a US phenomenon 1.3

1.2

1.1

.9 1985

1990

1995

2000

2005

2010

year Spain France USA

Italy UK

Germany Japan

Consistent with Andrews, Criscuolo and Gal (2016), showing that slow-down is pervasive across industries and countries, and driven by technological followers. 3 / 28

The bottleneck hypothesis reconciles these two facts

Our hypothesis: Uneven technological advances create endogenous bottlenecks that hold back aggregate productivity Simple Theoretical Framework: New technologies require simultaneous improvements in several inputs when advancements are complements Empirical Evidence: Greater dispersion in an industry’s suppliers’ TFP growth has large negative effect on its own TFP growth

4 / 28

Illustrative Case Study: GPS

Historical navigation: sight-lines, optical instruments and tables 1970s: radio-positioning with limited geographic coverage 1978: GPS satellites launched by military, providing geolocation and date-and-time information (atomic-level time-keeping) 1983: GPS opened for civilian use Subsequent innovations that relied on GPS: I Precision agriculture I Synchronization of power transmission systems I Weather prediction I Innumerable consumer-facing technologies (ride hailing, targeted advertising)

5 / 28

Outline for Presentation

1. Motivating theoretical framework 2. Data Sources 3. Empirical evidence for productivity bottlenecks I Evidence for the importance of innovation linkages I Importance of innovation in driving these patterns I International evidence

6 / 28

Outline for Presentation

6 / 28

Motivating Theoretical Framework: Setup Goal: Derive estimating equations linking an industry’s productivity to the distribution of supplier productivity Production: N sectors producing Yit 1−∑j ∈Si αij

Yit = Bi Ait Lit

∏ Xijt

αij

(1)

j ∈ Si

I I I I I

Lit : labor Ait : productivity of sector i Si : time-invariant set of inputs in production αij : Input share of industry j in sector i Bi : Normalizing constant

7 / 28

Motivating Theoretical Framework: Quality Improvements

Quality-ladder structure: Ajt = λnjt Critical assumption: arrival rate of innovation (φit ) depends on how technologically advanced inputs are ! φit = H

∑ αij h(Ajt )

(2)

j ∈ Si

I h and H are monotone functions determining relationship between inputs and innovation I Convex h: most advanced inputs determine innovation I Concave h: greater dispersion hinders innovation

8 / 28

Endogenizing Innovation Effort Modify arrival rate of innovations to be a function of research effort zit 1 φit = H γ

∑ αij h(Ajt )

!1− γ γ

zit

j ∈ Si

Firms maximize payoff given per-unit research cost (κit ) and reward to innovation (πit ) ! 1 1 πit 1−γ ∗ H ∑ αij h (Ajt ) φit = γ κi j ∈ Si Arrival rate of innovations proportional to exogenously-specified process: I Endogenous innovation not central for empirical work I Suggests caution in drawing normative conclusions 8 / 28

Motivating Theoretical Framework: Deriving Estimating Equations Taking a second-order Taylor expansion of Eq (1) around the mean and a first-order expansion of H around 0 gives: i i φit ≈ ηmean Āit + ηvariance var ({αij Ajt }j ∈Si ),

(3)

I Āit ≡ ∑j ∈S αij Ajt : (cost-share weighted) mean of productivities of the inputs to sector i i I var ({αij Ajt }j ∈Si ): (weighted) variance of the productivity as of the inputs to sector i i I ηmean ≡ H 0 (0)h0 (Āit ) i I ηvariance ≡ H 0 (0)h00 (Āit ).

In concave case: lagging inputs hold back innovation, meaning lower innovation when the variance is high for the same mean

9 / 28

Main Empirical Spec: Link TFP growth to dispersion of TFP growth among suppliers ∆TFPit = β mean ∑ αijt −1 ∆TFPjt + β variance varjt∆TFP + δi + δt + ε it j

where varjt∆TFP

≡ ∑ αijt −1 ∆TFPjt − ∑ αijt −1 ∆TFPjt j

Differences from theoretical equation: 1. TFP as the measure of innovation 2. Linking 5-year TFP growth, not levels

Our hypothesis: H0 : β mean > 0

β variance < 0

10 / 28

Outline for Presentation

10 / 28

Data Sources 1. Input-Output Linkages: Detailed Benchmark Input-Output Tables in 1977, 1982, 1987, 1992, 1997 and 2002 2. U.S. Manufacturing Productivity: NBER-CES Manufacturing Industry Database I 462 1997 NAICS Industries from 1977-2007 3. U.S. Non-Manufacturing Productivity: BLS Multifactor Productivity Statistics I 42 1997 NAICS Industries from 1987-2007 4. International Productivity: KLEMS I 30 1997 NAICS Industries from 1987-2007 5. Direct Measures of Innovation: I Industry-level patents from 1976-2006 I Network of patent citations (Acemoglu et al. 2016) 6. USPTO patents for patent-level analysis: I To come. . . 11 / 28

Outline for Presentation

11 / 28

Higher average supplier TFP growth predicts higher TFP growth TFP growth vs. Average Supplier TFP Growth All Manufacturing Industries

Customer TFP Growth (Residualized)

0 β = 0.81 s.e. = .13

-.1 -.1

-.05

.05

.15

Average Supplier TFP Growth (Residualized)

12 / 28

Higher variance of supplier TFP growth correlated with lower TFP growth TFP growth vs. Variance of Supplier TFP Growth All Manufacturing Industries

Customer TFP Growth (Residualized)

.05 β = -.744 s.e. = .121

-.05

-.1

-.15 -.05

.05

.15

Variance of Supplier TFP Growth (Residualized)

13 / 28

Negative relationship with variance not driven by outliers TFP growth vs. Variance of Supplier TFP Growth All Manufacturing Industries

Customer TFP Growth (Residualized)

.05

β = -.487 s.e. = .157

-.05 -.1

-.05

.05

Variance of Supplier TFP Growth (Residualized)

Back

13 / 28

Bottleneck patterns present across all sectors

Input Average

(1)

(2) Manufacturing

(3)

(4)

(5) All Industries

(6)

0.425 (0.139)

0.810 (0.130) -0.744 (0.121)

0.676 (0.170)

0.343 (0.178)

0.915 (0.161) -0.905 (0.158)

0.636 (0.183)

Input Variance Input Bottom Decile

0.059 (0.113) -0.110 (0.033)

Input Top Decile

Observations R-Squared

2772 0.108

2772 0.133

2772 0.118

0.164 (0.099) -0.117 (0.034) 2016 0.079

2016 0.102

2016 0.090

Notes: All regressions include year fixed effects. Columns 1-3 include stacked 5-year changes from 1977-2007 and columns 4-6 include include stacked 5-year changes from 1987-2007. Standard errors are clustered at the industry level. Industries are unweighted. 14 / 28

Bottleneck patterns present across all sectors

Input Average

(1)

(2) Manufacturing

(3)

(4)

(5) All Industries

(6)

0.425 (0.139)

0.810 (0.130) -0.744 (0.121)

0.676 (0.170)

0.343 (0.178)

0.915 (0.161) -0.905 (0.158)

0.636 (0.183)

Input Variance Input Bottom Decile

0.059 (0.113) -0.110 (0.033)

Input Top Decile

Observations R-Squared

2772 0.108

2772 0.133

2772 0.118

0.164 (0.099) -0.117 (0.034) 2016 0.079

2016 0.102

2016 0.090

Bottleneck patterns present across all sectors

(1)

Input Average

0.425 (0.139)

Input Variance

(2) (3) Manufacturing 0.810 (0.130) -0.744 (0.121)

Input Bottom Decile

(5) All Industries

(6)

0.343 (0.178)

0.915 (0.161) -0.905 (0.158)

0.636 (0.183)

0.059 (0.113) -0.110 (0.033)

Input Top Decile

Observations R-Squared

0.676 (0.170)

(4)

2772 0.108

2772 0.133

2772 0.118

0.164 (0.099) -0.117 (0.034) 2016 0.079

2016 0.102

2016 0.090

Variance of Supplier TFP has grown over time All Manufacturing Industries

Variance of Supplier TFP Growth

.04

.03

.02

.01

No Input Weighting

No Computers

1977-1987

1987-1997

1997-2007

All Industries 17 / 28

Rising variance of supplier TFP growth explains significant fraction of TFP slowdown ∆TFP ∆TFPtCF = ∆TFPt − βbvariance (vart∆TFP − var1977 −1987 )

Manufacturing Industries .06

.04

.02

-.02 1977-1987

1987-1997

1997-2007

Actual TFP growth Counterfactual TFP growth with variance at 1977-1987 level

Weights TFP growth by 1987 real value-added share All Industries 18 / 28

Examples of Bottleneck Industries Endogenous bottleneck sectors: created by explosive growth in other sectors I Bottleneck Industries: biggest negative effect on variance when increasing TFP by 10% I Outlier Industries: biggest positive effect on variance when increasing TFP by 10% Outlier Industries

Bottleneck Industries

Semiconductor and Related Devices Electronic Computers Iron and Steel Mills Computer Storage Devices

Petroleum Refineries Pharmaceutical Preparation Turbine and Turbine Generator Set Units Printed Circuit Assembly

Redistributing 20pp of TFP growth from 10 fastest-growing to bottom 50% of industries increases aggregate TFP by over 1pp Limited Industries

19 / 28

Bottleneck patterns are robust across many specifications Coefficient on Upstream Variance Baseline Industry Trends VA Weights Covariance 10-year Changes Lagged Dep. Var. China Shock Dropping Computers Fixed IO Table All Inputs -2.5

Average Term

-2

-1.5

-1

-.5

All Industries 20 / 28

Bottlenecks present when using TFP in other advanced economies as instrument

Baseline

Simultaneity concern: Common productivity shocks cause mechanical correlation International IV approach: isolate productivity changes common across several advanced economies I Instruments: mean and variance of supplier TFP growth in France, Germany, and the UK OLS

Industry FEs

2SLS

OLS

2SLS

-2

-1.5

-1

-.5

Coefficient on Variance

Alternate Formulation 21 / 28

Innovation linkages are a mechanism that accounts for bottleneck patterns

Are these patterns capturing innovation?

22 / 28

Innovation linkages are a mechanism that accounts for bottleneck patterns

Are these patterns capturing innovation? 1. Patterns driven by TFP, not prices or quantities

Table

22 / 28

Innovation linkages are a mechanism that accounts for bottleneck patterns

Are these patterns capturing innovation? 1. Patterns driven by TFP, not prices or quantities

Table

2. Bottleneck patterns appear using innovation networks

Table

22 / 28

Innovation linkages are a mechanism that accounts for bottleneck patterns

Are these patterns capturing innovation? 1. Patterns driven by TFP, not prices or quantities

Table

2. Bottleneck patterns appear using innovation networks

Table

3. Bottleneck patterns appear using patents as measure of innovation (next two slides)

22 / 28

Bottleneck patterns appear using patents as measure of innovation

Patenting growth vs. Average Supplier Patent Growth

Patent growth vs. Variance of Supplier Patent Growth

All Manufacturing Industries

All Manufacturing Industries .05

Customer Patent Growth (Residualized)

β = 0.81 s.e. = .044

-.2

-.4

β = -1.56 s.e. = .228

-.05

-.1

-.15 -.3

-.2

-.1

Average Supplier Patent Growth (Residualized)

-.02

.02

.04

.06

Variance of Supplier Patent Growth (Residualized)

23 / 28

Bottleneck Patterns with Patenting

Input Average Input Variance

Ind. Fixed Effects Upstream Network Observations R-Squared

(1) # Patents

(2) Top 20% Patens

(3) International IV

(4) Citation Network

1.095 (0.049) -0.750 (0.495)

1.190 (0.049) -0.687 (0.092)

0.924 (0.073) -1.951 (0.582)

1.890 (0.038) -0.617 (0.191)

no IO 1804 0.438

no IO 1804 0.897

no IO 1804 0.711

no Cites 1350 0.882

Standard errors are clustered at the industry level. Time fixed effects are included in all specifications and industry fixed effects are included where indicated. Industries are defined using SIC codes.

23 / 28

Outline for Presentation

23 / 28

Bottleneck patterns confirmed using cross-country data ∆TFP ∆TFPi,c,t = β mean ∑ αi,j,c,2000 ∆TFPj,c,t + β variance varj,c,t + δi,t + δc,t + ε i,c,t j

Upstream Average

(1)

(2)

(3)

(4)

0.258 (0.075)

0.270 (0.080) -0.824 (0.212)

0.108 (0.080) -0.535 (0.143)

-0.229 (0.113) -0.444 (0.157)

X X

982 0.076

X 982 0.363

Upstream Variance

Year FEs Country FEs Year*Country FEs Year*Industry FEs Observations R-Squared

982 0.065

X X 982 0.401

Notes: Standard errors are clustered at the industry level. All regressions include stacked 5-year changes from 1987-2007 for 30 industries and 9 countries: Spain, France, the US, Austria, Finland, the Netherlands, Italy, Germany and the UK 24 / 28

Conclusion

25 / 28

Examples of Limited Industries

Limited Industries: large contributors to value added but are inhibited by high TFP growth dispersion among their suppliers. Select Limited Industries Surgical and Medical Instruments Relay and Industrial Controls Gasoline Engine and Engine Parts Guided Missile and Space Vehicles Industrial Valves Back

26 / 28

Rising TFP Variance in Manufacturing: Unweighted All Manufacturing Industries

Variance of TFP Growth

.006

.004

.002

1958-1973

1974-1989

1990-2004

2005+

Back 26 / 28

Rising TFP Variance without Computers and Electronics

Excluding the Computer/Electronics Sector

Excluding the Computer/Electronics Sector .015

Variance of Supplier TFP Growth

.005

Varaince of TFP Growth

.004

.003

.002

.001

.01

.005

0 1958-1973

1974-1989

1990-2004

2005+

1977-1987

1987-1997

1997-2007

Back

26 / 28

Rising TFP Variance Across All Industries All Industries

Variance of Supplier TFP Growth

.04

.03

.02

.01

1987-1992

1992-1997

1997-2002

2002-2007

Back 26 / 28

Negative relationship with variance not driven by outliers TFP growth vs. Variance of Supplier TFP Growth All Manufacturing Industries

Customer TFP Growth (Residualized)

.05

β = -.487 s.e. = .157

-.05 -.1

-.05

.05

Variance of Supplier TFP Growth (Residualized)

Back

26 / 28

Counteractual Aggregate TFP Growth ∆TFP ∆TFPtCF = ∆TFPt − βbvariance (vart∆TFP − var1987 −1992 )

All industries .04

.02

-.02

-.04 1987-1992

1992-1997

1997-2002

2002-2007

Actual TFP growth Counterfactual TFP growth with variance at 1987-1992 level Variance contribution for given period

Back

26 / 28

Robustness of Bottleneck Patterns: Average Supplier TFP Coefficient on Upstream Average Baseline Industry Trends VA Weights Covariance 10-year Changes Lagged Dep. Var. China Shock Dropping Computers Fixed IO Table All Inputs .5

1.5

2.5

Back 26 / 28

Robustness of Bottleneck Patterns: All Industries

Coefficient on Upstream Variance

Coefficient on Upstream Average

Baseline

Industry Trends

VA Weights

Covariance

10-year Changes

Lagged Dep. Var.

China Shock

Dropping Computers

Fixed IO Table

All Inputs

All Inputs -4

-3

-2

-1

1.5

Back

26 / 28

Prices, Quantities and Productivity

TFP Average TFP Variance

(1)

(2)

(3)

(4)

Baseline 0.810 (0.130) -0.744 (0.121)

Prices 0.815 (0.113) -0.686 (0.232) 0.006 (0.085) -0.051 (0.204)

Employment 0.720 (0.134) -0.703 (0.118)

0.224 (0.045) 0.166 (0.219)

Combined 0.602 (0.107) -0.655 (0.233) -0.141 (0.091) -0.123 (0.201) 0.264 (0.051) 0.117 (0.218)

no 2772 0.149

no 2772 0.152

Input Price Average Input Price Variance Input Employment Average Input Employment Variance

Ind. Fixed Effects Observations R-Squared

no 2772 0.133

Notes: Standard errors are clustered at the industry level. All regressions include year fixed effects and industry fixed effects are included where indicated. Sample includes stacked 5-year changes for manufacturing industries from 1977-2007.

Back

27 / 28

Robustness: Instrumental-Variables Estimates Using Ranked TFP Growth

Upstream Average Upstream Variance

Ind. Fixed Effects Observations R-Squared First-Stage F-Stat

2SLS Estimates 0.928 1.093 (0.242) (0.236) -0.667 -1.480 (0.326) (0.393)

LIML Estimates 0.928 1.094 (0.245) (0.237) -0.664 -1.482 (0.330) (0.395)

no 2478 0.117 9.39

yes 2478 0.374 9.47

Notes: Standard errors are clustered at the industry level. Time fixed effects are included in all specifications.Time fixed effects are included in all specifications. Back

27 / 28

Bottleneck patterns persist using citation networks Consider dispersion in TFP growth among “idea suppliers” I Fraction of a sector i’s patent citations that cite sector j (Acemoglu et al. 2016)

Input Average (Citations) Input Variance (Citations)

(1)

(2)

(3)

(4)

1.421 (0.227) -1.201 (0.388)

1.143 (0.274) -1.542 (0.706)

1.137 (0.227) -0.639 (0.663) 0.295 (0.113) -0.437 (0.173)

0.853 (0.294) -0.909 (0.831) 0.296 (0.100) -0.481 (0.164)

no None 1844 0.125

yes None 1844 0.408

no None 1844 0.133

yes None 1844 0.415

Input Average (IO) Input Variance (IO)

Ind. Fixed Effects Industry Weighting Observations R-Squared

Notes: Standard errors are clustered at the industry level. The regressions include stacked 5-year changes from 1987-2007 and include 462 manufacturing industries. Time fixed effects are included in all specifications and industry fixed effects are included where indicated.

Back

28 / 28