Productivity Dynamics over the Medium Term

Diego Comin

Dartmouth College

Facts about Productivity and 

Inflation

Slowdown in Productivity Growth, since 2005

until recently

Slowdown in TFP concentrated in

followers

Wide-spread across countries

Recently, TFP grows at pre-GR growth rate

Inflation was higher than expected during

and after GR

But lower than expected once short-run

output converged

-0.06

-0.05

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

0.03

1987 1991 1995 1999 2003 2007 2011 2015

Detrended TFP

‐0.100

‐0.080

‐0.060

‐0.040

‐0.020

0.000

0.020

0.040

0.060

1984.75 1987.75 1990.75 1993.75 1996.75 1999.75 2002.75 2005.75 2008.75 2011.75 2014.75 2017.75

Labor Productivity

‐7

‐6

‐5

‐4

‐3

‐2

‐1

0

1

2

3

0

0.5

1

1.5

2

2.5

3

2000 2002 2004 2006 2008 2010 2012 2014 2016 2018

Inflation and CBO Gap

Core Inflation Output gap (right axis)

Goals

Present model that accounts for these

patterns

Developed to study productivity growth

during the GR

Accounts for productivity growth before

GR, and recent recovery

Also accounts for evolution of inflation

Key mechanism: Cyclical response of

technology adoption

Provide evidence on the mechanism

A historical account of productivity

dynamics and inflation

Model

1. 𝑦 𝑦 𝑦2. 𝑦 𝛼 χ 𝛼 𝑅

3. 𝑅 𝑅 𝑟

4. 𝑅 𝛼 𝜋 +𝛼 𝑦

5. ∆𝜋 𝜅𝑚𝑐 𝜀6. 𝑚𝑐 𝜂 𝑎 𝜂 𝑦

7. ∆𝑦 𝜌∆𝑎

8. ∆𝑎 𝜈 𝜆 𝜈 𝑧 𝑎9. 𝜆 𝛾 𝑦 𝛾 𝑅

10.∆𝑧 χ 𝜃𝑠

0 20 40 60-0.6

-0.4

-0.2

0

0 20 40 60-0.6

-0.4

-0.2

0

0 20 40 60-1.5

-1

-0.5

0

0 20 40 60

-0.2

-0.1

0

0 20 40 603.4

3.6

3.8

0 20 40 600

0.5

1

0 20 40 60

-0.4

-0.2

0

0 20 40 60-2

-1

0

0 20 40 604.7

4.8

4.9

5

EvidenceCyclicality of Adoption

Figure 2: R&D Expenditures by US Corporations, 1983-2013

1984 1987 1990 1993 1995 1998 2001 2004 2006 2009 2012-0.15

-0.1

-0.05

0

0.05

0.1

R&D Expenditure by US Corporations

Log-linearly detrended data. Source: R&D Expenditure by US corporations (National Science Foundation).

Data are deflated by the GDP deflator and divided by the civilian population older than 16 (see Appendix

A.1 for data sources).

6

Cyclical Adoption: A Shred of Evidence

� Survey data: sample of 26 production technologies that di¤used at various timesover the period 1947-2003 in the US (5) and the UK (21).

� mit � fraction of potential adopters that have adopted technology i at time t

�! ratio of adopters to non-adopters rit = mit=(1�mit)�! speed of di¤usion Speedit = 4 ln(rit)

� Econometric speci�cation

Speedit = �i + �1lagit + �2(lagit)2 + � � byt + �it;

lagit time from introduction of technology i

byt �detrended output3

Table 1: Cyclicality of the Speed of Technology Diffusion

I II III IV

yt 3.73 3.7 3.64 4.12(3.59) (2.81) (3.94) (3.17)

yt * US 0.07 -0.74(0.04) (0.53)

lagit -0.057 -0.057(5.22) (4.76)

lag2it 0.001 0.001(2.52) (2.12)

ln(lagit) -0.29 -0.29(6.68) (6.65)

R2 (within) 0.11 0.11 0.13 0.13N technologies 26 26 26 26N observations 327 327 327 327

Notes: (1) dependent variable is the speed of diffusion of 26 technologies, (2) all regressions include technology

specific fixed effects. (3) t-statistics in parenthesis, (4) yt denotes the cycle of GDP per capita in the country

and represents the high and medium term components of output fluctuations, (5)yt*US is the medium term

cycle of GDP per capita times a US dummy, (6) lag represents the years since the technology first started

to diffuse.

speed of diffusion and the cycle. Diffusion speed was lowest in the deep 1981-82 recession;

it recovered during the 80s and declined again after the 1990 recession. It increased notably

during the expansion in the second half of the 90s and declined again with the 2001 recession.

Next, we turn our attention to study the evolution of the speed of technology diffu-

sion during the Great Recession. Due to its recent nature, the evidence we have is more

anecdotal. Eurostat provides information on the diffusion of three relevant internet-related

technologies in the UK.6 Figure 4 plots their average diffusion from 2004 until 2013 with

the business cycle downturns in the UK. The figure confirms the pro-cyclicality of the speed

of diffusion of these technologies. In particular, during the downturn corresponding to the

Great Recession (2008-2009), the average speed of diffusion of our three technologies sharply

declined by 75%. After the Great Recession, the speed of diffusion recovered and converged

6 The measures we consider are the fraction of firms that (i) have access to broadband internet, that(ii) actively purchase online products and services and that (iii) actively sell online products and services(actively is defined as constituting at least 1% of sales/purchases). For each of these three measures weconstruct the speed of technology diffusion using expression (2), and then filter the effect of the lag sincethe introduction of the technology using expression (3) and the estimates from column 3 of Table 1. Theresulting series are demeaned so that they can be interpreted as percent deviations from the average speedof technology diffusion.

8

Figure 3: Speed of Diffusion

1980 1985 1990 1995 2000 2005

−0.3

−0.25

−0.2

−0.15

−0.1

−0.05

0

0.05

0.1

0.15

0.2

Avg. diffusion

Avg. diffusion (3 year MA)

to approximately 8% below average. Beyond its cyclicality, the second observation we want

to stress from the Figure is that fluctuations in the speed of diffusion are very wide, ranging

from 86% above average in 2004 to 74% below the average diffusion speed in 2009.

Finally, Andrews et al. (2015) have recently provided complementary evidence that tech-

nology diffusion in OECD countries may have slowed during the Great Recession. In their

study, they show that the gap in productivity between the most productive firms in a sector

(leaders) and the rest (followers) has increased significantly during the Great Recession.7

Andrews et al. (2015) show that the most productive firms have much greater stocks of

patents which suggests that they engage in more R&D activity. They interpret the increase

in the productivity gap as evidence that followers have slowed down the rate at which they

incorporate frontier technologies developed by the leaders.

These co-movement patterns between the business cycle and measures of investments in

technology development as well as measures of the rate of technology adoption is, in our

view, sufficiently suggestive evidence to motivate the quantitative exploration we conduct

through the lens of our model.

7In manufacturing the productivity gap increased by 12% from 2007 and 2009, and in services by ap-proximately 20%.

9

Figure 4: Diffusion Speed for 3 Internet Technologies in the UK, 2004-2013

2004 2005 2006 2008 2009 2010 2012-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

Diffusion Speed for 3 Internet Technologies in UK

Source: Eurostat; see footnote 6 for details of calculations. Shaded areas are UK recession dates as dated by

UK ONS.

10

0

0.01

0.02

0.03

0.04

0.05

0.06

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

Adoption Costs in Germany

Adoption costs pre& post 08

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

Sales from products introduced or improved

% sales from company improvements in products pre& post 08

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

0.016

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

% Reduction production costs

Unit cost reductions pre& post 08

Historical Account

Q3-87 Q1-93 Q3-98 Q1-04 Q3-09 Q1-15-6

-4

-2

0

2

4

6

8

TFPEndogenous Component of TFPLabor Productivity

Q1-85 Q1-90 Q1-95 Q1-00 Q1-05 Q1-10 Q1-15-15

0

15

2

5

8

AZ

(right axis)

Take Away TFP

Slowdown in TFP largely due to endogenous component

During GR and post-GR: liquidity demand slows down adoption rate

Prior to GR: R&D productivity declines in 2001, and this leads to lower TFP growth from 2005-2008

In recent times: adoption rate stabilized, and TFP grows at pre-GR rate due to catch up

Heterogeneity in adoption between leaders and followers could explain TFP divergence

Q1-10 Q1-12 Q1-14 Q1-16 Q1-18-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

Baseline Model (Data)Exogenous TFP Model

Q1-10 Q1-12 Q1-14 Q1-16 Q1-18-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Baseline Model (Data)

Take Away inflation

Higher than expected inflation during and after GR, due to lower level of endogenous TFP

(Due to TFP convergence) this effect has disappeared and now endogenous TFP leads to lower than expected inflation