broadband internet, labor demand, and total factor ... · productivity evolves according to an...

Policy Research Working Paper 8318

Broadband Internet, Labor Demand, and Total Factor Productivity in Colombia

Carlos Ospino

Finance, Competitiveness and Innovation Global Practice GroupJanuary 2018

WPS8318P

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

ed

Produced by the Research Support Team

Abstract

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.

Policy Research Working Paper 8318

This paper is a product of the Finance, Competitiveness and Innovation Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at [email protected].

This paper studies the relationship between information and communication technology, labor demand, and total factor productivity in Colombia. It estimates total factor productivity for the Colombian manufacturing sector using a method that assumes a law of motion where total factor productivity evolves according to an autoregressive pro-cess of order 1 as well as with the past use of broadband technologies. Using fixed effects models, the paper esti-mates the effect of broadband use on the labor demand of different workers, controlling for total factor productivity,

capital, and wages. To address the potential endogene-ity between broadband adoption and labor demand, the analysis uses state-industry level variation in broadband quality (speed) and intensity of use. The results show a pos-itive association of broadband adoption on labor demand, suggesting that adoption of information and communi-cation technology can offset the employment effects of technological growth. Attempts to identify causal effects using an instrumental variable approach were inconclusive.

BroadbandInternet,LaborDemand,

andTotalFactorProductivityinColombia

Carlos Ospino1

JEL codes: D24, L25, L60, O15

Keywords: Total factor productivity, broadband internet, labor demand, production function

estimation.

1This work has benefitted from funding by the World Bank’s Latin America and Caribbean’s Chief Economist Office, under the regional study “Digital technology adoption, skills, productivity and jobs in Latin America”. I appreciate the very helpful comments and suggestions from Joana Silva, Mark Dutz, Rita Almeida, Truman Packard, Giulia Lotti and the participants at the author’s workshop at the World Bank. Daniel Espinosa provided much appreciated research assistance to this project. Please send comments and suggestions to [email protected]

2

IntroductionThis section of the paper shows the international context of internet use in Colombia.

Figure 1 OECD Fixed broadband subscriptions per 100 inhabitants, by technology, June 2015

Source: OECD, Broadband Portal, http://www.oecd.org/sti/broadband/oecdbroadbandportal.htm, June 2015.

Figure 1 shows that as of June 2015, Colombia had a total number of internet subscriptions per 100

inhabitants similar to Mexico (11.2), but lower than Chile (14.6). The latter’s is half of the OECD

average of 28.8 subscriptions per 100 inhabitants. DSL and cable are the dominant technologies in

Colombia and both have the same number of subscriptions (5.2) per 100 inhabitants, despite an

ambitious plan to expand the optical fiber network to almost every municipality ((Ministerio de

Tecnologías de la Información y las Comunicaciones, 2017).

Figure 2 OECD fixed broadband penetration (per 100 inhabitants), percentage increase, June 2014‐June 2015, by country

Source: OECD, Broadband Portal, http://www.oecd.org/sti/broadband/oecdbroadbandportal.htm, June 2016.

00

05

10

15

20

25

30

35

Chile Mexico OECD Colombia

DSL Cable Fibre Satellite Fixed wireless Other

0.00

2.00

4.00

6.00

8.00

10.00

12.00

Chile Mexico OECD Colombia

3

Colombia showed a significant increase in fixed broadband penetration. The growth level was

almost three times the level of the OECD average and near one and a half times the growth level

displayed by Chile. This shows that at the household level, Colombia may still have room to expand

its broadband penetration. At the business level, the story is quite different. In 2003, 86% of small

and medium size enterprises (SME) and 96% of large ones had internet access (CINTEL) (2003). In

2015, 99% of firms in manufacturing and 99.7% in retail and services used internet (CRC) (2015).

This shows that internet access in the last decade has been relatively high in Colombia’s business

sector.

This high coverage of internet access poses the challenge of identifying the gains in productivity due

to the availability of internet and how it relates to labor demand; I therefore focus on a slightly

different question. In the next section, I explore the literature and conclude that the focus, rather

than being on internet access per se, should be on high‐speed broadband internet access for firms.

The second section discusses the relevant literature that is related to this paper. The third describes

the data used in this analysis. The forth section discusses the results of the ICT module in the Annual

Manufacturing Survey for the years 2008‐2014. The fifth section discusses the TFP estimation

procedure. The sixth section presents the main results of the paper in terms of the effect of TFP

gains due to ICT use on labor demand and wages. The sixth section shows robustness and

falsification exercises. The final section concludes with a discussion of the results.

LiteraturereviewThis paper is related to two different branches of the economics literature. The first branch studies

the impact of broadband internet and firm productivity ((Fabling, Grimes, & Grimes, 2016). The

second branch focuses on the estimation of production functions and total factor productivity. The

first contribution of the paper is adding to the scarce empirical evidence about the labor demand

predictions of task‐based models of technological change for heterogeneous firms in developing

countries, with a special focus on broadband internet. The second one is estimating total factor

productivity within the framework of proxy method models of production function estimation

proposed by De Loecker and Warzynski (2012) and De Loecker (2013). The innovation is modeling a

total factor productivity law of motion that explicitly considers the past adoption of high‐speed

broadband internet as a shifter in growth levels. I now discuss each branch of the literature and put

the paper’s contribution in context within each one.

The paper’s theoretical underpinnings follow Brambilla (2018). Her model predicts that because of

ICT adoption, firms will grow, pay higher wages, become specialized in complex tasks and more skill

intensive conditional on skilled labor and ICT being complements. Brambilla (2018) expands

traditional task‐based models of technical progress and labor markets to allow firm heterogeneity

and wage heterogeneity across firms. In this paper, ICT adoption will be modeled as the adoption of

high‐speed broadband internet. Akerman et al. (2015) provide evidence of the skill complementarity

of broadband internet adoption. Related to this point, there is also a growing literature about the

effect of broadband internet on firm productivity ((Bertschek, Cerquera, & Klein, 2013; Colombo,

4

Croce, & Grilli, 2013; Fabling et al., 2016; Grimes, Ren, & Stevens, 2012; Haller & Lyons, 2015). I now

briefly discuss this literature.

Akerman et al. (2015) use Norwegian data and a quasi‐experiment to show that ICTs are innovations

biased towards skilled workers. To identify the intent‐to‐treat effects, the authors exploit the

staggered roll‐out of broadband internet. To estimate the production function and thus

productivity, they use the Levinsohn and Petrin (2003) methodology. The authors find evidence that

the availability of broadband is associated with a substantial increase in the output elasticity of

skilled labor. They find the opposite effect for unskilled labor.

Fabling et al. (2016) study the effect of ultrafast broadband (UFB) internet on labor demand and

productivity in New Zealand between 2010 and 2012. UFB was defined by the government as

download speeds of up to 100Mbps and upload speeds of up to 50Mbps. The authors use the

proximity to schools as an instrument for broadband availability since the government targeted

schools and hospitals in their expansion plans of UFB to reach 80 percent of the population by 2022.

The technology through which UFB can be obtained and the authors focus on is (optic) fiber. These

authors do not find any statistically significant effects and suggest that most gains from UFB would

arise as a result of selection. Their results suggest that gains from UFB would be related to

complementary investments by firms.

Bertschek et al. (2013) do not find a statistically significant effect of broadband on labor productivity,

but find positive effects on firm innovation in Germany. Their analysis covers the period 2001 to

2003 when there was already a high penetration rate of broadband use by firms and focus on DSL

technology. The authors exploit broadband availability at the postal code level to identify the causal

effects.

Colombo et al. (2013) find negligible effects of adopting broadband applications in small and

medium enterprises in Italy during the period 1998‐2004. The authors base their analysis on services

which require broadband speeds and construct their variable of interest using principal component

analysis. Their results point to the complementarity of broadband with other investments at the

firm, those related to management practices.

The paper also relates to the literature about production functions and total factor productivity

(TFP) estimation ((D. A. Ackerberg, Caves, & Frazer, 2015; De Loecker, 2013; De Loecker &

Warzynski, 2012; Gandhi, Navarro, & Rivers, 2016; Levinsohn & Petrin, 2003; Olley & Pakes, 1996).

The paper contributes to this literature by explicitly modelling an endogenous total factor

productivity process which is governed by past productivity as in D. Ackerberg et al. (2006) and the

previous adoption of broadband through broadband technology, namely DSL, cable or optical Fiber.

A more detailed discussion about these methods is postponed to the section where the

methodology is explained.

DatadescriptionThe main data set is composed of an unbalanced panel of 14,479 plants in the manufacturing sector

in Colombia, spanning over 7 years (2008‐2014) of data from the Annual Manufacturing Survey

5

(EAM from its acronym in Spanish.). EAM is a census of manufacturing firms in Colombia that employ

at least 10 workers or have an output value about 73,000 USD.2 It gathers very detailed information

about employment levels and costs, asset levels and investment, output and intermediate materials

use for production. Starting in the year 2008 it also collects an ICT module where information about

internet use and technology can be found. In the sample, 6,999 (48.3%) of the plants have

information for every year and were used as a balanced sample to check the robustness of the

results. I also use broadband speed data at the state level kindly provided by the Communications

Regulation Commission for the years 2008‐2014. The sample used to estimate total factor

productivity was a subsample of these firms and it is described in detail in that section.

ICTuseintheColombianmanufacturingsectorIn this section I describe the results about ICT use in Colombian firms from the information collected

in EAM special modules for the years 2008 through 2014. The pooled sample summary statistics for

the five‐year period are discussed. In this section the analysis is broken down by firm size and I

report information collected in the ICT module about the use of computers at the firm, internet

access, and the type and speed of internet connections.

Table 1 Distribution of firms by total employment size

Source: EAM‐DANE 2008‐2014. Author’s calculations.

The top panel in Table 1 shows that 72% of firms in the sample hired up to 50 workers and are

considered by most definitions small firms; 18% of firms hired between 50 and 200, 8% between

200 and 1,000 and about 2% of firms hired more than 1,000 workers between 2008 and 2014. It is

clear that most firms in the sample are small units of production, and the reader should take this

into account when interpreting the results. The bottom panel restricts the sample to the initial year,

2008, where ICT information started to be collected. The main takeaway is the stability in the

distribution of firm sizes during the period of analysis.

2 137.2 million COP in the year 2013 and exchange rate of 1,867 COP per USD, which is the average exchange rate for that year.

Size Frequency Percent Cum.

[0,50] 91,861 72.07 72.07

(50,200] 22,776 17.87 89.94

(200,1000] 10,382 8.15 98.09

>1000 2,438 1.91 100

Total 127,457 100

Size Frequency Percent Cum.

[0,50] 6,708 69.21 69.21

(50,200] 1,884 19.44 88.65

(200,1000] 862 8.89 97.54

>1000 238 2.46 100

Total 9,692 100

Full sample period

Year 2008

6

Table 2 ICT and firm size

Source: EAM‐DANE 2008‐2014. Author’s calculations. Each column shows the average value of each variable for each bracket of firm

size.

The first column of Table 2 shows the average number of employees in each firm size category, to

put into context the variables described. The table also shows that the average number of

computers at each firm increases with size as does the share of workers that use computers. Column

four shows that the average use of internet is basically universal for firms of all sizes. The final

column shows that having a website tends to increase with firm size. In smaller firms about 1 in 3

workers uses a computer while in large firms this ratio is about 1 in 2. In the Colombian

manufacturing sector, less than a third of workers uses the internet for their work. This percentage

is higher for large firms (those hiring more than 1,000 workers) but is just about 42% of workers. In

small firms, just under 28% of workers use the internet in their work. This suggests that the impact

of internet use or availability on firms’ productivity would be more likely to be observed in larger

firms where internet exposition is greater for workers.

Table 3 Internet and firm size

Source: EAM‐DANE 2008‐2014. Author’s calculations. Each column shows the average value of each variable for each bracket of firm size.

The leading technology in the Colombian manufacturing sector is cable/fiber with just under 46% of

firms using this technology to access the internet. The second most popular technology is ADSL, and

this is particularly true for smaller firms. 25% for small firms use ADSL while only 5% of firms with

more than 1,000 workers use this technology. The most common contracted speed for internet use

is 1‐2 Mbps with just under 36% of firms. However, two‐thirds of large firms report internet speeds

over 2Mbps.

TFPestimationData Cleaning

SizeTotal

employment

Number of

computers

% of

workers

using

computers

Internet

use

% of

workers

use

internet

for work

Has a

website

[0,50] 14.3 3.1 32.2 96.6% 29.7 46.5%

(50,200] 91.1 16.6 35.7 99.9% 32.6 77.3%

(200,1000] 280.0 98.2 42.3 99.8% 36.3 90.2%

>1000 650.1 573.7 55.2 100.0% 47.0 93.7%

Total 63.8 30.2 34.5 97.7% 31.4 58.3%

Size Modem ISDN ADSL Cable/Fiber Wireless Mobile0‐256

Kbps

254‐1024

Kbps

1025‐2048

Kbps

>= 2049

Kbps

[0,50] 6.3% 0.8% 25.3% 39.0% 19.7% 6.5% 7.5% 26.9% 38.7% 24.9%

(50,200] 2.2% 0.5% 27.0% 51.2% 13.4% 5.6% 2.7% 18.6% 33.4% 45.2%

(200,1000] 1.0% 0.3% 13.1% 72.9% 7.7% 5.0% 1.2% 13.7% 24.7% 60.4%

>1000 0.2% 1.1% 5.2% 79.2% 7.0% 7.4% 1.2% 7.7% 16.0% 75.0%

Total 4.7% 0.7% 23.9% 45.8% 16.9% 6.2% 5.8% 23.4% 35.7% 33.8%

7

As discussed above, the sample for the estimation of the production function parameters from

which TFP is calculated was different from the main sample. The selection criteria were determined

using four items: Value added, prices and input consistency, industry size, and survival. I now discuss

each one in detail.

Value added: firms which reported zero or negative value added were deleted.

Wage bill and intermediate inputs: I required that both the value of the wage bill and the

costs of intermediate materials were strictly lower than the value of output. Firms that did

not fulfill this requirement were deleted.

Industry size: I required that that at least 10 firms were part of an industry at any moment

in time to include any firm in that industry.

Survival: For any firm to be included in the production function estimation sample, it had

to report information during four consecutive years.

Input selection

The production function estimation uses capital, labor and intermediate materials as inputs. All

monetary variables were deflated using the Producer Price Index which varies by industry class (two

digits in ISIC Rev. 3) and the base year is 2013. Capital is defined as the value of fixed capital stock

in each year. The stock of capital is constructed recursively using the change in investment levels

and assuming a depreciation value of 5% for all capital. ∗ 1 0.05 . Capital

includes three types of assets: Machinery, office equipment and transportation equipment. I have

decided to leave out buildings and land to make the analysis comparable across businesses and

sectors.3 Labor includes all personnel, regardless of their skill level or contract type. Finally,

intermediate materials include the value of raw inputs, materials and packaging used in the

production process.

Total factor productivity was estimated using the GMM methods proposed by De Loecker and

Warzynski (2012) and De Loecker (2013). I will refer to this methodology as DLW from now on. A

gross output production function was estimated to address the concerns by Gandhi, Navarro, &

Rivers (2016) that value‐added production functions may generate too much dispersion in the

estimated productivity.4 In this paper a firm’s broadband use at each moment was used to identify

the parameters of the Cobb‐Douglas production function in equation (1), which uses labor, capital

and intermediate materials ( , , , respectively). Where is the unobserved total factor

productivity and represents measurement error. The assumption is that the demand for inputs

differs significantly between firms that decide to use broadband internet and those that do not. All

inputs and gross output are expressed in logarithms.

3 Eslava et al. (2004) construct capital stock series which include equipment, machinery, buildings and structures, while excluding vehicles and land. 4 Gandhi et al. (2013) find that in Colombia, under a value‐added production function results imply that the 95th percentile firm would produce more than six times more output whereas under a gross output production function this firm would only produce twice as much.

8

1

As an innovation, the process that governs the evolution of productivity over time can be

determined by the firm’s use of broadband technologies at time t‐1, . The intuition follows De

Loecker's (2013) argument that if productivity evolves exogenously, as in D. A. Ackerberg et al.

(2015) then investment decisions such as exporting or the adoption of high‐speed internet would

have no effect on technological improvements by the firm. We therefore consider an explicit law of

motion which allows past decisions of adopting broadband internet to directly affect productivity

growth.

DLW propose a two‐stage estimation procedure. In the first stage, gross output is regressed on the

production function inputs and on variables that affect input demand. In our case the firm’s

broadband use is assumed to affect input demands differently for firms that have access to

broadband internet and those that do not. The function includes interactions of period t

broadband use, a third‐degree polynomial of all inputs, year and industry fixed effects.

, , , 2

The procedure assumes that (unobserved) total factor productivity, , is a monotonic function of

(observable) inputs and broadband use, , , , . Therefore, unlike other proxy

methods such as Levinsohn and Petrin (2003) none of the parameters is identified in the first stage.

As mentioned, I assume that productivity evolves as a function of the previous period productivity,

broadband use and a productivity shock, similar to De Loecker and Warzynski (2012).

; . Function is approximated using a third‐degree polynomial on lagged

productivity, lagged exporting status and a constant. The vector ≡ , , is used to

construct total factor productivity as:

, , ,

Finally, in the second stage the production function parameters are identified by imposing the

following moment conditions in a GMM estimation.

000

Capital is assumed to be predetermined at time , and therefore uncorrelated with the productivity

shock. Lagged variable inputs and are used in the moment conditions as current values

are likely to be correlated with the productivity shock (e.g. these are freely variable inputs). A single

production function is assumed and estimated for the whole manufacturing sector. In line with this

assumption, Casas and González (2016) showed that aggregating from sector‐specific productivity

provides similar productivity growth patterns to estimating a single production function.

9

This paper aims to assess the effect of TFP growth which can be explained by broadband use on

labor demand and wages. Since the estimation process imposes a structured autoregressive law of

motion on the evolution of TFP which is affected by past broadband use, this exercise can be

performed directly by regressing the variables of interest on the estimated TFP. De Loecker (2013)

points out that methods which do not consider an endogenous TFP process that depend on past

exporting behavior tend to underestimate the relationship between exporting and productivity.

Under the assumption that there are similarities in the decision‐making process between the

adoption of broadband technology and entering exporting markets, I conjecture that not

considering an endogenous TFP process which depends on broadband use is likely to bias the

relationship between ICT use and productivity. To assess the nature of the bias I will conduct

robustness exercises using the productivity estimator of D. A. Ackerberg et al. (2015) and the De

Loecker (2013) procedure which models the productivity law of motion as a function of past

exporting status.

Table 4 shows the estimated production function coefficients for each the methods used in this

paper. My preferred method’s coefficients for labor and capital fall within the range of values of OLS

and fixed effects estimations. Suggesting that both coefficients are biased upward under an OLS

estimation. On the contrary, the materials coefficient is higher than either OLS or fixed effects. Table

5 shows that the correlation of TFP across the three methods is very high.

Table 4 Production function estimated coefficients

Author’s estimations. Standard errors for DLW obtained via bootstrapping with 100 repetitions.

Table 5 Correlation of TFP estimates

Source: EAM‐DANE. Author’s calculations.

DLW estimation is performed under the assumptions that: 1‐Variable input labor demands are a

monotonic function of productivity. 2‐Input demand responds differently for firms that use

broadband internet and those that do not. 3‐TFP evolves as a function of the previous period

productivity and the previous period use of broadband internet. I now provide evidence that these

assumptions hold.

Method Labor Capital Materials

DLW (Broadband Use) 0.466 *** 0.126 *** 0.529 ***

ACF 0.458 *** 0.119 *** 0.539 ***

DLW (Exporting Status) 0.408 *** 0.112 *** 0.588 ***

OLS 0.492 *** 0.199 *** 0.381 ***

Fixed effects 0.373 *** 0.076 *** 0.418 ***

DLW (Broadband Use) ACF DLW (Exporting Status)

DLW (Broadband Use) 1

ACF 0.8257 1

DLW (Exporting Status) 0.9779 0.7849 1

10

Table 6 Input demand and TFP

Table 6 shows that there is a negative correlation, for all inputs, between TFP and input demand in

OLS estimations. This is true for variable inputs that enter the production function as well as for

energy which was not included but has been used by other authors in Colombia (Eslava,

Haltiwanger, Kugler, & Kugler (2004)). This same table shows that input demand is higher for firms

that have broadband internet and those that have broadband internet technologies. In this case

firms that have contracted broadband speeds through broadband technologies (DSL, cable/fiber,

wireless) showed a lower input demand than the rest of firms in the OLS estimation. Once I account

for firm fixed effects, the correlation between capital and productivity becomes positive and firms

that use broadband through broadband technologies have a higher variable input demand. This

interaction was chosen as the indicator for broadband use for the remainder of the paper, as it

captures the use of higher broadband internet speeds. It is an established fact that high‐speed

internet requires technologies such as DSL or optical fiber for its delivery, where regular broadband

tends to be more technology neutral (Calvo, 2012).

Table 7 shows the estimation of the TFP policy function through three different methods: Ordinary

least squares, fixed effects and dynamic panel models. All estimations show that TFP is a function

of past productivity. The coefficient on lagged TFP has the expected positive sign which points to

the persistence of TFP growth within firms. Lagged broadband use is not statistically significant and

lagged broadband technology is only statistically significant in the fixed effects estimation, which is

my preferred specification. In these specifications their interaction is not statistically significant.

(1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES

TFP ‐0.254***‐0.317*** ‐0.052*** 0.026*** ‐1.566*** ‐1.669*** ‐0.156*** ‐0.253***

[0.008] [0.006] [0.013] [0.008] [0.012] [0.009] [0.013] [0.010]

1.broadband 0.847*** ‐0.008 1.372*** 0.005 1.353*** 0.049*** 1.184*** 0.032**

[0.031] [0.010] [0.049] [0.012] [0.047] [0.014] [0.049] [0.016]

1.BBTechnology 0.440*** ‐0.010 0.740*** ‐0.010 0.607*** 0.027*** 0.629*** 0.016

[0.021] [0.007] [0.034] [0.009] [0.032] [0.010] [0.034] [0.012]

1.broadband#1.BBTechnology ‐0.309*** 0.034*** ‐0.601*** 0.004 ‐0.469*** 0.033** ‐0.493*** ‐0.015

[0.032] [0.010] [0.051] [0.013] [0.049] [0.015] [0.051] [0.017]

Constant 4.250*** 4.859*** 12.972*** 12.783*** 21.034*** 21.339*** 12.281*** 12.593***

[0.044] [0.028] [0.069] [0.037] [0.066] [0.041] [0.069] [0.048]

Observations 55,098 55,098 55,098 55,098 55,098 55,098 55,013 55,013

R‐squared 0.103 0.065 0.142 0.030 0.346 0.443 0.170 0.015

Plant FE NO YES NO YES NO YES NO YES

Year FE YES YES YES YES YES YES YES YES

Sector FE YES YES YES YES YES YES YES YES

City FE YES YES YES YES YES YES YES YES

model ols fe ols fe ols fe ols fe

Number of plants 9,917 9,917 9,917 9,910

Standard errors in brackets

*** p<0.01, ** p<0.05, * p<0.1

Total employment Capital Materials Energy

11

Column (4) shows that all terms of a more flexible, third‐degree polynomial of lagged TFP, are

statistically significant. More precisely, that TFP does respond to past broadband use.

Table 7 TFP policy function estimation

ICTuse,productivityandlabordemandIn this section I provide evidence of the relationship between productivity, broadband use and labor

demand. To do so I estimate longitudinal fixed effects linear models as specified by the following

equation:

Β Β Β Β

Where interest lies in coefficient which captures the elasticity of dependent variable to

changes in broadband use. The employment dependent variables are logs of: Total employment,

direct employment, outsourced employment, plant laborers, plant professionals, total plant

workers, managers, permanent, temporary employment, men and women. This rich range of labor

types attempts to identify whether broadband use is related to changes in the labor demand of

different workers. METRO, SECTOR and YEAR, are vectors of indicator variables for each

VARIABLES (1) (2) (3) (4)

Lagged TFP 0.904*** 0.310*** 0.637*** ‐1.352***

[0.002] [0.005] [0.017] [0.156]

Lagged TFP^2 0.257***

[0.023]

Lagged TFP^3 ‐0.012***

[0.001]

Lagged broadband ‐0.003 ‐0.003 0.002 ‐0.002

[0.007] [0.007] [0.010] [0.007]

Lagged BBTech 0.002 0.009* 0.010 0.01**

[0.005] [0.005] [0.006] [0.005]

Lagged broadbandXLagged BBTech ‐0.016** ‐0.004 ‐0.040*** ‐0.003

[0.008] [0.008] [0.010] [0.008]

Constant 0.475*** 3.182*** 1.315*** 6.530***

[0.010] [0.024] [0.221] [0.331]

Observations 44,714 44,714 44,714 44,714

R‐squared 0.831 0.094 0.098

Plant FE NO YES NO NO

Year FE YES YES YES YES

Sector FE YES YES YES YES

model ols fe gmm fe

Number of nordest . 9,363 9,363 9,363

Standard errors in brackets

*** p<0.01, ** p<0.05, * p<0.1

12

metropolitan area, (two‐digit) industry of economic activity and year respectively. Finally, the vector

X captures variables which are theoretically related to labor demand. I have included the lagged

value of capital and the minimum wage (which is interacted with the share of unskilled labor in

industry total employment). All standard errors were two‐way clustered at the state and plant level.

The period of analysis is 2008‐2014 given that information on broadband use is only available for

this time frame. The estimation sample is restricted to firms that report a consistent industry code

during the sample period and therefore the vector of coefficients Β will not be identified separately

form the fixed effect.

The adoption of broadband services may be correlated with unobserved factors, such as the

availability of service providers at the plant’s location or the costs of internet services. Under this

assumption coefficient will be inconsistent to identify the effect of broadband adoption. I

instrument broadband use with an interaction of log broadband speeds available for businesses at

the state level and the industry average broadband use through broadband technology in the

previous year. Therefore, the estimation will pick up the effect of broadband use which can be

explained by the variation in broadband speed availability at the industry‐state level. Broadband

speed is defined as the download internet speeds weighted by the number of subscriptions each

internet service provider reported in each state from 2008 to 2014.

The identification of the effect of broadband use due to broadband service quality use on labor

demand relies on the relevance and exclusion restrictions being satisfied by the instrument. The

instrument was constructed as the product of two variables: 1‐The average use of broadband at the

(three‐digit) industry level in the previous year.5 2‐The (weighted6) average download speeds at the

state level. The first part of the instruments is relevant since it captures the degree of broadband

use in each industry. As Table 15 shows, there are statistically significant differences in labor market

variables depending on whether firms belong to an industry where broadband is more intensively

used. The second part of the instruments is relevant as it captures a precise definition of broadband

internet quality, it is download speeds in the state where the production unit is located.7

The exclusion restriction of the instrument is given by the fact that the average, three‐digit industry

broadband use is predetermined each period. The second part of the instrument exploits the fact

that internet quality to businesses is determined by the service provider infrastructure and should

only affect labor demand through its effects on broadband adoption and use and not directly. Figure

3 suggests that between 2008 and 2014 the average internet speed increased in Colombia. By 2008

only the darkest areas had internet speeds above 2mbs, while by 2014 some of the white areas had

broadband coverage. However, this increase has not radically changed the areas where the highest

broadband speeds were available. The Appendix shows the broadband speed information for every

5 Recall that in the analysis I include industry dummies at the two‐digit level, so this is a finer measure of industry that is still larger than the firm level. 6 The average is weighted by the number of subscriptions each Internet Service Provider has. 7 The state level may still be too broad a measure of broadband availability. However, the data provided to us for this research project by Colombia’s national statistics agency did not include a finer geographical identification variable due to data confidentiality concerns.

13

year. A violation of the exclusion restriction would suggest firms change labor demand as a response

to internet speeds in their state and not because it affects labor demand through its effect on

broadband adoption and use.

Figure 3

Table 8 shows some suggestive statistics for firms across two dimensions. The first is whether they

were classified as being part of an industry where broadband use was above the manufacturing

sector average in the year 2008. The second dimension captures whether the firm currently uses

broadband internet. Current use of broadband is related to higher unskilled labor demand, output

and productivity. However, there is not any additional correlation due to being part of an intensive

broadband use sector. These results suggest a certain degree of complementarity of broadband use

with the demand of unskilled labor in Colombia but not with skilled labor. Even in the presence of a

positive relationship with output and total factor productivity.

(847,2979](582,847](492,582][248,492]

Average Download Speed 2008

(4663,6717](3821,4663](2650,3821][590,2650]


14

Table 8 Broadband use intensity and selected outcomes

I now turn to the main results. Table 9 shows that firms that report using broadband internet appear

to have a higher labor demand than those with slower internet access; about 2.2 log points.

However, there does not appear to be any relationship between broadband use and outsourced,

management and permanent labor demand, which suggests internet use is particularly related to

production workers’ labor demand. Temporary labor displays the highest correlation with

broadband use. Since TFP has been estimated from a structural model, under the set of assumptions

explained before, the results can be interpreted as a causal effect on labor demand.

These results show a negative relationship of productivity gains on labor demand for all types of

workers, holding everything else constant. Columns 2 to 9 compare pairs of possible labor

substitutes in the production process. Taken at face value, increases in productivity from past

adoption of broadband internet would reduce to a higher degree outsourced labor than direct labor,

low skilled than high skilled labor demand, production workers relative to managers, and temporary

workers relative to permanent ones. These results suggest that labor types which may be easier to

adjust, such as outsourced, unskilled and temporary would be the most likely to be substituted by

broadband adoption. Interestingly, the size of the effect is similar for men and women, suggesting

an absence of gender bias in the substitution between labor and productivity growth.

(1) (2) (3) (4) (5)

VARIABLES Laborers Professionals Value of production Output per worker TFP

Intensive & Broadband (1=Yes) ‐0.013 0.017 0.003 0.005 0.008

[0.016] [0.019] [0.013] [0.010] [0.006]

Broadband (1=Yes) 0.017*** 0.007 0.019** 0.003 0.034***

[0.006] [0.011] [0.008] [0.008] [0.004]

Observations 50,466 28,967 52,661 52,616 50,753

R‐squared 0.007 0.003 0.009 0.002 0.005

Number of plants 9,037 5,915 9,378 9,369 9,139

Plant FE YES YES YES YES YES

Year FE YES YES YES YES YES

Sector FE YES YES YES YES YES

City FE YES YES YES YES YES

Std. Errors

Robust standard errors in brackets

*** p<0.01, ** p<0.05, * p<0.1

Cluster Plant/State

15

Table 9 Labor demand and Broadband use

Table 10 shows the results of my instrumental variables approach. As discussed, I used the log

average download speeds at the state level interacted with the lagged average broadband use at

the industry level. While other potential instruments such as the number of broadband

subscriptions and the upload speed were available, the preferred specification uses download speed

given the results of statistical tests which can be seen in Table 15 and Table 16 in the Appendix.8

Except for outsourced and professional workers, the first‐stage statistics are well above the

conventional levels of statistical significance. This is also true about the weak instruments and weak

instrument robust inference statistics.

Correcting for the possible endogeneity in broadband adoption suggests strong positive effects on

labor demand of broadband use. The point estimate on total labor demand is around 22 log points,

but it is not statistically significant. Lack of significance may be due to the use of two‐way clustered

standard errors and the fact that variation of the instrument is at the state and industry level. The

fact that most point estimates are strictly greater than zero provides some evidence of a positive

effect, which nonetheless is not precisely identified.

Alternatively, given the model assumptions under which TFP was estimated, the underlying labor

demand should only be a function of input prices and TFP. Thus, under these assumptions, the effect

of broadband adoption should only be observed through its indirect effect on TFP. Therefore, lack

of statistical significance in the 2SLS estimation is consistent with this theoretical model.9

8 See Table 15 which shows the first‐stage regression and Table 16 which shows relevance, and weak inference statistics. 9 I thank the participants at the applied microeconomics seminar at Banco de la República in Cali, especially Juan Esteban Carranza, Camila Casas and Salvador Navarro for pointing out this alternative explanation.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)VARIABLES Total Direct Outsourced Laborer Professional Production Management Permanent Temporary Women Men

Broadband (1=Yes) 0.022*** 0.023*** ‐0.007 0.021*** 0.015** 0.023*** 0.006 0.008 0.037** 0.018*** 0.021***

[0.005] [0.006] [0.018] [0.005] [0.007] [0.005] [0.006] [0.007] [0.017] [0.005] [0.006]

TFP ‐0.332*** ‐0.291*** ‐0.745** ‐0.341*** ‐0.099** ‐0.304*** ‐0.195*** ‐0.204*** ‐0.328*** ‐0.253*** ‐0.277***

[0.056] [0.049] [0.324] [0.078] [0.049] [0.055] [0.037] [0.039] [0.083] [0.041] [0.051]

Observations 50,753 50,255 13,479 48,805 27,910 49,823 48,020 44,098 33,711 49,050 50,186

R‐squared 0.071 0.042 0.027 0.046 0.004 0.047 0.021 0.015 0.013 0.029 0.040

Number of plants 9,139 9,082 2,716 8,837 5,759 8,989 8,809 8,196 6,593 8,920 9,064

Plant FE YES YES YES YES YES YES YES YES YES YES YES

Year FE YES YES YES YES YES YES YES YES YES YES YES

Sector FE YES YES YES YES YES YES YES YES YES YES YES

City FE YES YES YES YES YES YES YES YES YES YES YES

Std. Errors

model


*** p<0.01, ** p<0.05, * p<0.1

Cluster Plant/State

fe

16

Table 10 2SLS estimation Labor demand and Broadband use

RobustnessexerciseIn this section I use a balanced panel sample to assess the degree to which sample selection could

change the previous results. The balanced panel has a slightly higher share of large firms than the

unbalanced panel in the main exercise, but the size distributions are not too different from one

another.

Table 11 Size distribution in the balanced panel.

Using a balanced panel does not change the main results, so there should be little concern about

the effect of sample selection in the OLS regression. The fact that the balanced sample is comprised

of about 70% of plants in the period is also reassuring. When I instrumented broadband on the

balanced panel, the instrument loses its relevance, making the comparison between the two

samples difficult. This suggests, however, that entry and exit dynamics could be related to

broadband quality in a way that is systematically different for firms that remain in the sample since

2008 and those that enter or leave. Therefore, the 2SLS results in the main section should be

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

VARIABLES Total Direct Outsourced Laborer Professional Production Management Permanent Temporary Women Men

Broadband (1=Yes) 0.217 0.017 1.672* 0.182 ‐1.217 0.166 ‐0.100 ‐0.011 0.607 ‐0.047 0.361

[0.293] [0.396] [1.001] [0.393] [2.076] [0.393] [0.409] [0.415] [0.812] [0.285] [0.339]

TFP ‐0.304*** ‐0.250*** ‐0.698* ‐0.316*** ‐0.001 ‐0.280*** ‐0.157*** ‐0.174*** ‐0.316*** ‐0.213*** ‐0.267***

[0.060] [0.060] [0.360] [0.093] [0.133] [0.068] [0.050] [0.053] [0.080] [0.048] [0.056]

Observations 44,816 44,396 11,597 43,030 24,517 43,970 42,347 38,880 29,570 43,266 44,298

R‐squared 0.009 0.035 ‐0.668 0.012 ‐0.599 0.017 0.009 0.012 ‐0.077 0.021 ‐0.082

Number of plants 8,853 8,802 2,574 8,549 5,503 8,703 8,522 7,909 6,354 8,629 8,777

Plant FE YES YES YES YES YES YES YES YES YES YES YES




Errors

Instrument

Weak identification test (Kleibergen‐

Paap rk Wald F statistic) 12.73 12.01 4.126 13.42 2.841 12.39 12.77 11.02 9.504 13.96 13.25


*** p<0.01, ** p<0.05, * p<0.1

Cluster Plant/State

Download speed at state level

Size FrequencyPercent Cum.

[0,50] 42,241 70.19 70.19

(50,200] 10,868 18.06 88.25

(200,1000] 5,509 9.15 97.41

>1000 1,561 2.59 100

Total 60,179 100

Size FrequencyPercent Cum.

[0,50] 5,863 68.2 68.2

(50,200] 1,699 19.76 87.96

(200,1000] 802 9.33 97.29

>1000 233 2.71 100

Total 8,597 100

Full sample period

Year 2008

17

interpreted as being driven mainly by entry and exit dynamics in the manufacturing sector. Further

work to account for these dynamics would be an interesting extension to the current results.

Table 12 Labor demand and Broadband use

Table 13 14 2SLS estimation Labor demand and Broadband use

DiscussionColombia is a country where businesses have a relatively high level of internet access. However,

there is a great amount of heterogeneity regarding internet speed. Considering that the

identification strategy proposed for the paper relies on the geographical nature of internet rollout

to identify the causal effect of ICT on labor demand which is mediated through productivity gains, a

finer measure of internet access was considered. The measure combines broadband speeds (greater

than 1Mbps) and broadband technology use (DSL, cable/fiber, wireless). Figure 4 and Figure 5

showed that this measure displayed both time and industry variation, making it a good candidate to

model productivity gains due to broadband use. Since broadband download speed might be a key

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)VARIABLES Total Direct Outsourced Laborer Professional Production Management Permanent Temporary Women Men

Broadband (1=Yes) 0.019*** 0.021*** ‐0.012 0.017** 0.018* 0.022*** 0.004 0.011 0.020 0.015*** 0.018***

[0.006] [0.007] [0.022] [0.007] [0.009] [0.007] [0.007] [0.008] [0.019] [0.005] [0.006]

TFP ‐0.276*** ‐0.238*** ‐0.580** ‐0.305*** ‐0.076* ‐0.269*** ‐0.176*** ‐0.176*** ‐0.289*** ‐0.220*** ‐0.230***

[0.052] [0.045] [0.272] [0.084] [0.040] [0.060] [0.031] [0.039] [0.076] [0.035] [0.046]

Observations 34,602 34,232 10,511 33,236 19,944 33,944 32,839 30,384 24,095 33,488 34,282

R‐squared 0.060 0.033 0.019 0.040 0.003 0.042 0.019 0.014 0.010 0.025 0.032

Number of plants 6,363 6,322 2,248 6,150 4,180 6,256 6,160 5,751 4,830 6,225 6,323

Plant FE NO NO NO NO NO NO NO NO NO NO NO




Std. Errors

model


*** p<0.01, ** p<0.05, * p<0.1

Cluster Plant/State

fe

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

VARIABLES Total Direct Outsourced Laborer Professional Production Management Permanent Temporary Women Men

Broadband (1=Yes) ‐0.121 ‐0.399 1.296 ‐0.086 ‐2.233 ‐0.095 ‐0.232 ‐0.572 0.601 ‐0.319 ‐0.045

[0.247] [0.417] [1.082] [0.365] [1.926] [0.371] [0.382] [0.530] [0.816] [0.356] [0.320]

TFP ‐0.263*** ‐0.199*** ‐0.637** ‐0.293*** 0.060 ‐0.257*** ‐0.155*** ‐0.127** ‐0.322*** ‐0.190*** ‐0.224***

[0.060] [0.063] [0.316] [0.102] [0.128] [0.075] [0.054] [0.065] [0.091] [0.056] [0.052]

Observations 34,602 34,232 10,511 33,236 19,944 33,944 32,839 30,384 24,095 33,488 34,282

R‐squared 0.033 ‐0.143 ‐0.407 0.029 ‐2.067 0.026 ‐0.027 ‐0.205 ‐0.082 ‐0.069 0.028

Number of plants 6,363 6,322 2,248 6,150 4,180 6,256 6,160 5,751 4,830 6,225 6,323





Errors

Instrument

Weak identification test (Kleibergen‐

Paap rk Wald F statistic) 6.114 5.409 2.662 6.852 4.456 6.803 5.672 6.131 5.684 5.201 7.062


*** p<0.01, ** p<0.05, * p<0.1

Cluster Plant/State


18

factor explaining the adoption of broadband internet, it was considered a suitable candidate for an

instrument.

In this paper, I proposed and estimated total factor productivity under the assumption that it

evolves according to a process which depends on past broadband adoption. I estimated the effect

of broadband adoption on labor demand for different worker types. I find positive correlations

between broadband adoption and labor demand of most worker types of about 2%. The

instrumental variable estimation showed a positive but statistically insignificant effect of broadband

use on total labor demand. The fact that I used two‐way clustered standard errors could have

affected the statistical significance of the test. Alternatively, lack of statistical significance could

come from the fact that broadband should only affect labor demand through its direct effect on TFP

growth. In Table 8, I showed a positive relationship between broadband use and total factor

productivity, which suggests a positive complementarity between broadband use and productivity.

To further asses the possible skill complementarity of broadband, I interacted total factor

productivity and broadband adoption. Table 17 in the Appendix shows that labor demand for

unskilled labor is negatively correlated to productivity increases even further for firms that use

broadband, regardless of the method used to estimate TFP. Table 18 shows that for exporting firms

although unskilled labor demand is negatively related to TFP growth, the value of production and

output per worker are positively related to TFP growth.

The paper contributes by providing evidence about the effect of ICT adoption which is associated

with productivity growth on labor demand. The lack of significance in the instrumental variable

estimations suggests that other identification strategies are needed to capture the causal effect of

broadband use on labor demand. Particularly, since the instrument appears to be highly sensitive

to entry and exit dynamics. Although several possible instruments were available, download speed

appears to be a strong predictor of broadband adoption to a greater extent than the number of

subscriptions or upload speed, but more work should be conducted in determining the appropriate

instruments to pin down the causal effects of broadband adoption on labor demand and

productivity.

References(CINTEL), C. de I. de las T. (2003). Análisis del Mercado Servicios de Banda Ancha en Colombia.

Bogotá.

(CRC), C. de R. de C. (2015). IV Reporte de industria del sector TIC. Bogotá.

Ackerberg, D. A., Caves, K., & Frazer, G. (2015). Identification Properties of Recent Production Function Estimators. Econometrica, 83(6), 2411–2451. https://doi.org/10.3982/ECTA13408

Ackerberg, D., Caves, K., & Frazer, G. (2006). Structural Identification of Production Functions. MPRA Paper No. 38349, 12(38349), 1–35. Retrieved from http://folk.uio.no/rnymoen/Ackerberg_Caves_Frazer.pdf

Akerman, A., Gaarder, I., & Mogstad, M. (2015). The skill complementarity of broadband internet. Quarterly Journal of Economics, 130(4), 1781–1824. https://doi.org/10.1093/qje/qjv028

19

Bertschek, I., Cerquera, D., & Klein, G. J. (2013). More bits – more bucks? Measuring the impact of broadband internet on firm performance. Information Economics and Policy, 25(3), 190–203. https://doi.org/10.1016/j.infoecopol.2012.11.002

Brambilla, I. (2018). Digital Technology Adoption and Jobs: A Model of Firm Heterogeneity.

Calvo, A. G. (2012). Universal Service Policies in the Context of National Broadband Plans. OECD Digital Economy Papers, (203). https://doi.org/10.1787/20716826

Casas, C., & González, A. (2016). Productivity Measures for the Colombian Manufacturing Industry * (Borradores de economía No. 947). Cali.

Colombo, M. G., Croce, A., & Grilli, L. (2013). ICT services and small businesses’ productivity gains: An analysis of the adoption of broadband Internet technology. Information Economics and Policy, 25(3), 171–189. https://doi.org/10.1016/j.infoecopol.2012.11.001

De Loecker, J. (2013). Detecting learning by exporting. American Economic Journal: Microeconomics, 5(3), 1–21. https://doi.org/10.1257/mic.5.3.1

De Loecker, J., & Warzynski, F. (2012). Markups and firm‐level export status. American Economic Review, 102(6), 2437–2471. https://doi.org/10.1257/aer.102.6.2437

Eslava, M., Haltiwanger, J., Kugler, A., & Kugler, M. (2004). The effects of structural reforms on productivity and profitability enhancing reallocation: Evidence from Colombia. Journal of Development Economics, 75(2 SPEC. ISS.), 333–371. https://doi.org/10.1016/j.jdeveco.2004.06.002

Fabling, R., Grimes, A., & Grimes, A. (2016). Picking up speed : Does ultrafast broadband increase firm productivity ?, (November).

Gandhi, A., Navarro, S., & Rivers, D. (2016). On the Identification of Production Functions: How Heterogeneous is Productivity? Mimeo, (May 2006).

Grimes, A., Ren, C., & Stevens, P. (2012). The need for speed: impacts of internet connectivity on firm productivity. Journal of Productivity Analysis, 37(2), 187–201. https://doi.org/10.1007/s11123‐011‐0237‐z

Haller, S. A., & Lyons, S. (2015). Broadband adoption and firm productivity: Evidence from Irish manufacturing firms. Telecommunications Policy, 39(1), 1–13. https://doi.org/10.1016/j.telpol.2014.10.003

Levinsohn, J., & Petrin, A. (2003). Estimating Production Functions Using Inputs to Control for Unobservables. Review of Economic Studies, 70(2), 317–341. https://doi.org/10.1111/1467‐937X.00246

Ministerio de Tecnologías de la Información y las Comunicaciones. (2017). Proyecto Nacional de Fibra Óptica. Retrieved February 12, 2017, from http://www.mintic.gov.co/portal/vivedigital/612/w3‐propertyvalue‐647.html

Olley, G. S., & Pakes, A. (1996). The Dynamics of Productivity in the Telecommunications Equipment Industry. Econometrica, 64(6), 1263–1297.

20

AppendixTable 15 First Stage Regression for 2SLS Estimation

Note: Each column shows the sample over which the 2SLS estimation took place. The dependent variable is broadband use.

Table 16 First Stage test statistics

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

Sample Total Direct Outsourced Laborer Professional Production Management Permanent Temporary Women Men

TFP 0.090*** 0.090*** 0.051 0.109*** 0.060*** 0.095*** 0.085*** 0.079*** 0.064*** 0.088*** 0.091***

[0.019] [0.018] [0.044] [0.032] [0.016] [0.022] [0.018] [0.018] [0.019] [0.019] [0.019]

Capital (Lagged) ‐0.001 ‐0.001 ‐0.022** ‐0.001 ‐0.012* ‐0.002 ‐0.002 0.002 ‐0.010 ‐0.001 ‐0.002

[0.008] [0.008] [0.009] [0.008] [0.007] [0.007] [0.007] [0.008] [0.010] [0.008] [0.008]

Minimum Wage X Share of Laborers in Industry 0.000* 0.000* ‐0.000 0.000* 0.000 0.000* 0.000 0.000 0.000** 0.000* 0.000*

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

Excluded instrument from structural equation 0.060*** 0.057*** 0.042** 0.061*** 0.030* 0.061*** 0.062*** 0.050*** 0.054*** 0.061*** 0.061***

[0.017] [0.016] [0.021] [0.017] [0.018] [0.017] [0.017] [0.015] [0.018] [0.016] [0.017]

Observations 44,816 44,396 11,597 43,030 24,517 43,970 42,347 38,880 29,570 43,266 44,298

Number of plants 8,853 8,802 2,574 8,549 5,503 8,703 8,522 7,909 6,354 8,629 8,777





Errors

Instrument


*** p<0.01, ** p<0.05, * p<0.1


Cluster Plant/State

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

Statistic Total Direct Outsourced Laborer Professional Production Management Permanent Temporary Women Men

1St Stage F Test 12.73 12.01 4.13 13.42 2.84 12.39 12.77 11.02 9.5 13.96 13.25

Prob> F 0.0014 0.0018 0.0545 0.0011 0.1034 0.0016 0.0014 0.002 0.0048 0.0009 0.0011

Weak identification test (Cragg‐Donald

Wald F statistic) 19.27 17.22 3.03 18.99 3.08 19.65 19.33 11.98 11.27 19.18 20.05

Weak‐instrument‐robust inference

(Anderson‐Rubin Wald test) 0.6 0 5.62 0.24 0.58 0.19 0.06 0 0.75 0.03 1.38

P‐Value 0.4452 0.9672 0.0269 0.6303 0.4541 0.6646 0.8126 0.9801 0.394 0.8727 0.2505

Note: The Stock‐Yogo weak ID test critical values: 10% maximal IV size is 16.38.

21

HeterogeneouseffectsofbroadbanduseTable 17

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

VARIABLES

TFP#1.broadband ‐0.087** 0.001 ‐0.012 0.015

[0.042] [0.023] [0.025] [0.013]

Broadband (1=Yes) 0.421** 0.035 ‐0.304*** 0.012 ‐0.008 ‐0.190 0.080 ‐0.082 ‐0.266* ‐0.068 ‐0.032 0.041

[0.194] [0.061] [0.112] [0.106] [0.129] [0.209] [0.109] [0.106] [0.149] [0.064] [0.054] [0.329]

TFP ‐0.293*** ‐0.099** ‐0.253*** 0.061***

[0.068] [0.040] [0.051] [0.017]

Lagged Capital 0.081*** 0.075*** 0.066***0.074***0.072***0.069*** 0.091*** 0.084*** 0.097*** 0.003 0.004 0.029***

[0.006] [0.007] [0.007] [0.007] [0.007] [0.007] [0.007] [0.009] [0.010] [0.007] [0.007] [0.008]

Minimum Wage 0.000 0.000 0.000 ‐0.000* ‐0.000** ‐0.000** 0.000*** 0.000*** 0.000***0.000***0.000***0.000***

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

TFP#1.broadband ‐0.007 0.006 0.025 0.010

[0.016] [0.034] [0.029] [0.014]

TFP(DLW) ‐0.232*** ‐0.051 ‐0.236*** ‐0.004

[0.052] [0.038] [0.043] [0.018]

TFP#1.broadband 0.055*** 0.035 0.048* ‐0.007

[0.020] [0.036] [0.025] [0.056]

TFP(ACF) ‐0.566*** ‐0.186*** 0.307*** 0.939***

[0.100] [0.041] [0.070] [0.143]

Observations 48,805 48,573 48,573 27,910 27,834 27,834 50,747 50,442 50,442 50,747 50,442 50,442

R‐squared 0.047 0.021 0.055 0.004 0.003 0.005 0.032 0.021 0.030 0.003 0.001 0.183

Number of plants 8,837 8,737 8,737 5,759 5,704 5,704 9,138 9,030 9,030 9,138 9,030 9,030

Plant FE NO NO NO NO NO NO NO NO NO NO NO NO

Year FE YES YES YES YES YES YES YES YES YES YES YES YES

Sector FE YES YES YES YES YES YES YES YES YES YES YES YES

City FE YES YES YES YES YES YES YES YES YES YES YES YES

Std. Errors


*** p<0.01, ** p<0.05, * p<0.1

Laborers Professionals Value of production Output per worker

Cluster Plant/Depto

22

Table 18

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

VARIABLES

TFP#.1exports ‐0.335** ‐0.008 0.059*** 0.094*

[0.137] [0.032] [0.022] [0.056]

Exports (1=Yes) 1.550** 0.936* 0.307 0.051 0.130 0.033 ‐0.227** ‐0.418*** ‐0.184 ‐0.406 ‐0.428* 0.003

[0.620] [0.553] [0.259] [0.132] [0.202] [0.231] [0.097] [0.091] [0.338] [0.263] [0.244] [0.696]

TFP ‐0.321*** ‐0.096** ‐0.266*** 0.058***

[0.074] [0.044] [0.047] [0.014]

Lagged Capital 0.081*** 0.086*** 0.077*** 0.074*** 0.097*** 0.094*** 0.091*** 0.117*** 0.136*** 0.003 0.019*** 0.050***

[0.006] [0.011] [0.010] [0.007] [0.009] [0.009] [0.007] [0.011] [0.009] [0.007] [0.005] [0.007]

Minimum Wage 0.000* ‐0.000 ‐0.000 ‐0.000* ‐0.000*** ‐0.000** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000***

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

TFP#.1exports ‐0.240 ‐0.024 0.131*** 0.122*

[0.148] [0.054] [0.024] [0.064]

TFP(DLW) ‐0.345*** ‐0.090* ‐0.367*** ‐0.035**

[0.077] [0.046] [0.064] [0.016]

TFP#.1exports ‐0.046 0.001 0.044 0.004

[0.045] [0.040] [0.058] [0.120]

TFP(ACF) ‐0.664*** ‐0.206*** 0.365*** 1.101***

[0.093] [0.052] [0.062] [0.131]

Observations 48,805 72,460 72,460 27,910 41,671 41,671 50,747 75,077 75,077 50,747 75,077 75,077

R‐squared 0.050 0.038 0.073 0.004 0.008 0.010 0.032 0.043 0.040 0.004 0.006 0.216

Number of Plants 8,837 10,033 10,033 5,759 6,643 6,643 9,138 10,334 10,334 9,138 10,334 10,334

Plant FE YES YES YES YES YES YES YES YES YES YES YES YES

Year FE YES YES YES YES YES YES YES YES YES YES YES YES

Sector FE YES YES YES YES YES YES YES YES YES YES YES YES

City FE YES YES YES YES YES YES YES YES YES YES YES YES

Std. Errors


*** p<0.01, ** p<0.05, * p<0.1

Laborers Professionals Value of production Output per worker

Cluster Plant/State

23

Figure 4 Broadband use through broadband technology by industry 2008‐2014.

Figure 5 Broadband use through broadband technology by industry 2008‐2014.

.5.6

.7.8

.9.5

.6.7

.8.9

.5.6

.7.8

.9

2008 2010 2012 2014 2008 2010 2012 2014 2008 2010 2012 2014 2008 2010 2012 2014

15 16 17 18

19 20 21 22

23 24 25 26

95% CI lpoly smooth: BBTechXbroadband

lpoly smoothing grid

Graphs by ciiu2_r3

.5.6

.7.8

.9.5

.6.7

.8.9

.5.6

.7.8

.9

2008 2010 2012 2014 2008 2010 2012 2014

2008 2010 2012 2014 2008 2010 2012 2014

27 28 29 30

31 32 33 34

35 36

95% CI lpoly smooth: BBTechXbroadband

lpoly smoothing grid

Graphs by ciiu2_r3

24

Figure 6 Average download and upload broadband speed by state 2008‐2014. Source: Comisión de Regulación de Comunicaciones.

(847,2979](582,847](492,582][248,492]


(439,1501](308,439](253,308][176,253]

Average Upload Speed 2008

(1227,2766](960,1227](765,960][211,765]


(571,924](469,571](396,469][134,396]


25

(1822,3704](1702,1822](1435,1702][270,1435]


(897,1148](825,897](720,825][188,720]


(2442,4198](2197,2442](1773,2197][337,1773]


(1128,1376](1042,1128](878,1042][198,878]


26

(3068,5091](2643,3068](1879,2643][593,1879]


(1280,1699](1200,1280](936,1200][416,936]


(3699,5643](3240,3699](2376,3240][559,2376]


(1422,2075](1300,1422](1056,1300][443,1056]


27

(4663,6717](3821,4663](2650,3821][590,2650]


(1368,2151](1245,1368](1050,1245][517,1050]


broadband internet, labor demand, and total factor ... · productivity evolves according to an...

Documents