Advertisement

Asia-Pacific Journal of Regional Science

, Volume 2, Issue 2, pp 453–475 | Cite as

Local economic structure, productivity growth, and industry life cycle: evidence from Indonesia

  • Khoirunurrofik
Article
  • 743 Downloads

Abstract

This paper examines the effects of dynamic agglomeration economies on the productivity growth of the industries in Indonesia’s regions. The study introduces employment market potential into the city-industry growth estimation for controlling local size and preventing overestimation of the agglomeration effects. The results suggest that both specialization and diversity are important for city-industry growth and that some externalities are stronger in different time periods. The effects of specialization and diversity on medium-term growth (2000–2010) are stronger than on long-term growth (1990–2010), in addition to new positive effects of competition. A detailed analysis across industries reveals a strong relationship between local industrial structure and performance—productivity and employment growth—which is associated with industry maturity within its lifecycle stages.

Keywords

Employment market potential Local industrial structure Industry life cycle 

JEL Classification

R11 R12 O47 L20 

1 Introduction

The manufacturing industry’s role as an engine of sustained regional growth became an important topic in Indonesian development after the fiscal decentralization of 2001. However, a 2012 World Bank report asserted that the sector has been trapped in a “growth recession” and suffered “slow” or “weak” growth (p. 2); these drawbacks prevented the industry from returning to the level of performance it exhibited prior to the Asian financial crisis—that is, when it was significantly contributing to economic growth.

For that reason, empirical study, therefore, becomes very important to provide evidence as to what kinds of factors can increase the sector’s productivity growth from the point of view of the local industrial structure. This local view is important, because the decentralization policy introduced since 2001 should open the door for local government to promote policies inducing industry growth. As a result, the policy changed the population distribution of cities across the country. This implies that city size is not stable and would change due to the interaction between centripetal and centrifugal agglomeration forces (Abdel-Rahman and Anas 2004). In Indonesia, the number of cities in respect to city size classification has changed over time.1 It shows a declining trend in the number of small–medium cities and an increasing trend in the number of metro-megapolitan cities. For instance, there were 154 small–medium cities in 1990, 77 in 2000, and 53 in 2010.

Since the seminal paper by Glaeser et al. (1992), many empirical works attempted to explain the relationship between local industrial structure—namely, specialization, competition, and diversity—and growth patterns in cities. The studies usually specify these variables as representative of dynamic agglomeration externalities in employment growth regression and, later on, in  total-factor productivity (TFP) growth regression. One important variable in such a kind of growth regression is city size. Given that city size significantly influences local economic growth,2 a variable controlling city size becomes an important factor in determining the magnitudes of externalities.

The objectives of this paper are to estimate the effects of dynamic externalities of agglomeration economies on TFP and employment growth in both the long run (1990–2010) and the medium run (2000–2010) and to introduce employment market potential to control city size within the relationship between local industrial structure and city growth. We expect to contribute to the empirical literature in two ways. First, we provide evidence of the importance of the employment market potential for controlling a local size and subsequently affecting the source type and the magnitude of dynamic agglomeration externalities. This corrects the overestimation of regional employment by controlling for local size growth instead of using regional employment (for example, Cingano and Schivardi 2004; Almeida 2007). Second, we provide evidence of a changing local industrial structure, identified in both the long-term and medium-term analyses, toward stronger diversity and the new role of competition in the medium term.

This study explores a unique long-panel plant-level data set for Indonesian manufacturing from 1990 to 2010. We measure local economic performance in TFP and employment growth. This paper calculates TFP using a control function approach to account carefully for input endogeneity. While Cingano and Schivardi (2004) employed an Olley-Pakes (OP) estimator, we prefer to use the Levinsohn-Petrin (LP) method similar to Almeida and Fernandes (2013), with respect to data availability, for estimating the firm’s production function. The aggregate TFP at the industry-city level is weighted by plant output. Knowing the potential of reversed causality between the employment potential of a market and city-industry growth, we apply the ordinary least square (OLS) and instrumental variables (IV) estimation methods. In more detail analyses, we also run regressions for each period: long term (1990–2010) and medium term (2000–2010). Furthermore, we conduct an empirical investigation across industries to examine whether the industry lifecycle theory can explain the impact of industrial structure on city growth.

The result indicated that the employment market potential has strong effects on city size and subsequently affects the source type and the magnitude of dynamic agglomeration externalities on both productivity and employment growth. The overestimation of regional employment was corrected by controlling local size. The instrumental variable estimation further improved the estimation by solving the potential of reversed causality. The empirical evidence also showed that specialization and diversity positively impact TFP growth in long-term periods. It also showed that only diversity contributed to employment growth in the similar period. The results generally confirmed the importance of specialization and diversity for city-industry growth, as suggested by Duranton and Puga (2000) and empirically found by De Lucio et al. (2002) and Henderson et al. (1995).

This paper is organized as follows. The first section provides an overview of the importance of the research and its novelty. The second part offers an analysis of the theoretical background and empirical studies related on the subject literature. The third section describes the data and the construction of variables, and the empirical modeling and related estimation issues are reported in the fourth and fifth sections, respectively. Finally, the results and analysis are described in the sixth section. The seventh section provides our conclusions.

2 Literature review

Rosenthal and Strange (2004) highlight the importance of geographic scope in studying agglomeration economies. To account for neighboring agglomeration effects, we introduce employment market potential for controlling a local size, by summing local employment and the employment of neighboring cities, weighted by distance. Melo et al. (2009) argue that market potential can absorb spatial spillover or regional externalities from neighboring regions over space and outside geographic boundaries. Combes et al. (2010) and Holl (2012), in France and Spain, respectively, are among the studies that explore the role of market potential in firm-level productivity. After instrumenting market potential with long-lag variables and local geographic characteristics, they found a positive impact of market potential on plant productivity levels. However, the current paper differs from those works, since we focus on long-run TFP growth and city-industry level, rather than on yearly changes in the TFP plant level.

One of the first works to address the role of knowledge spillover on local economic growth, Glaeser et al. (1992) explains how urban areas and local economies develop over time through the contributions of three types of externalities: intra-industry knowledge spillover, interindustry knowledge spillover, and local competition. By virtue of spatial proximity, firms and workers within a particular industry located near each other can enjoy knowledge spillover from similar or different technologies, access a pooled market of labor and employment skill, and benefit from intermediate input sharing, all of which enhances firm productivity (Gill and Goh 2010). In a dynamic context, these external scale economies, or intra-industry knowledge spillover effects, are known as Marshal, Arrow, and Romer (MAR) externalities (Glaeser et al. 1992). On the other hand, interindustry exchanges of ideas and technology among different kinds of industries could create more variety in business services, enlarge market size on the supply and demand sides, and facilitate more product innovation and firm growth (Gill and Goh 2010). In a dynamic context, these effects are known as Jacobs externalities (Glaeser et al. 1992). The third type of externality known as Porter externalities stems from the recognition that local competition also plays a role in firms’ development. Local competition is a main source of pressure on firms to create innovative products and adopt new technologies (Glaeser et al. 1992).

The empirical literature on dynamic externalities emerged to offer contradictory findings as the result of different approaches to measuring local growth. When growth is measured by employment, the results tend to support the existence of Jacobs externalities (Glaeser et al. 1992; Combes 2000). However, Henderson et al. (1995) provide some evidence that both specialization and diversity can contribute to employment growth, depending on the maturity of the industry. Conversely, some authors prefer to measure local growth using TFP. They find that MAR externalities, and to some extent, Porter externalities are the important externalities leading to growth (Dekle 2002; Cingano and Schivardi 2004; Almeida 2007). These authors argue that there is possibly an identification problem in the employment growth regression and pointed out that the subsequent interpretation of employment growth overlooked the positive link between productivity growth and employment growth.

The employment growth regression may suffer from some limitations, as noted by Dekle (2002), Cingano and Schivardi (2004), and Combes et al. (2004). These authors argue that the connection between employment growth and productivity growth is not necessarily, nor always, positive; therefore, it remains a problem of interpretation. Duranton and Puga (2014) argue that the results from employment growth regression might be valid in a sector with constant markup and an elastic price of demand. In such a sector, the increased productivity results in higher output, larger revenue, and increased employment. However, the results do not hold in a sector with an inelastic price of demand, such as the traditional manufacturing industry, in which increased productivity may lead to declining employment. To deal with this problem, Dekle (2002) and Cingano and Schivardi (2004) use TFP growth instead of employment growth as a proxy for local economic performance. Their results indicate that specialization effects often positively affect TFP growth in Japanese prefectures, while diversity does not significantly affect TFP in the Italian local labor system. Similar evidence in Almeida (2007) supports the existence of MAR externalities on aggregate productivity growth in most sectors in Portuguese regions.

3 Data

This study employed data from the Statistik Industri, an unpublished electronic data set on the annual survey of large and medium firms conducted by Indonesia’s Central Bureau of Statistics (BPS), from 1990 to 2010. In this paper, large firm is a firm which employ more than 100 labors, while medium firm employ 20–99 labors. All values in this research were expressed in 2000 real values. We used the wholesale price index (WPI) published monthly in BPS’s bulletin, Statistik Bulanan Indikator Ekonomi. We gathered data on road length from BPS, while land area data were collected from the Ministry of Home Affairs.3 Furthermore, we used data from the Village Potential Survey (PODES) of BPS to generate variable on the share of households connected to electricity, the share of coastal area, and the land used by the non-agricultural sector. We used the geographic information system (GIS) Euclidean distance to calculate market potential. Since the number of districts in Indonesia has changed over time, particularly since 2001 (after the implementation of regional autonomy), we regroup newly created districts back into their parent districts, keeping the 1990 configuration of 284 districts at 26 provinces. This regrouping will allow us to compare across districts over years from 1990 to 2010.

4 Model specification: TFP and employment growth model

This study applied a two-step empirical approach to agglomeration economies modeling: (1) plant-level production function estimation and (2) productivity and employment growth estimation. To address a possible bias due to input endogeneity in the production function, in the first step, we used a semiparametric estimation of TFP introduced in Levinsohn and Petrin (2003) for each three-digit standard industrial classification (SIC). Following their method, we used capital and electricity consumption as a proxy for unobserved productivity shock.4 From the estimated TFP and employment at the plant level, we calculated a weighted average of the industry-city TFP growth using plant output as the weight. Accordingly, we also calculated employment growth from plant-level data.

In the second step, we applied OLS and IV estimations to examine how dynamic agglomeration externalities affected TFP and employment growth, after controlling for the average age of the local industry and regional characteristics such as land area and share of non-agricultural land. We applied IV estimation to deal with the possible simultaneity bias between employment market potential and productivity growth, which is due to firm selectivity. That is, a plant might choose a location in the most productive and agglomerate regions and introduce reversed causality into the model.

We specified the econometric model for testing the effects of dynamic agglomeration externalities on city growth into two specification models: the TFP growth model and the employment growth model. All specifications were estimated using OLS and IV estimations at the industry-city level and include industry dummies at the three-digit SIC level. We examined the estimations of local economic performance using TFP growth as a dependent variable compared to using employment growth. Detailed information on variable definitions and data sources is given in the Appendix (Table 7).

The general framework for modeling the relationship between dynamic agglomeration externalities and cityindustry growth was specified according to the framework of de Groot et al. (2009) following the seminal work of Glaeser et al. (1992) our TFP model extends the TFP growth regression formulated in Cingano and Schivardi (2004) by replacing initial city employment with employment market potential. We calculated TFP growth from 1990 to 2010 and set other variables to the conditions of the initial year, 1990. Furthermore, we ran OLS and then IV regressions to account for endogeneity stemming from the fact that market potential determines productivity growth, but the productivity growth might also determine market potential (via an influence on the location decision of firms and employees). We estimated the model using OLS and IV with a two-stage least-squares (TSLS) estimator and specified the TFP growth model as
$$\begin{aligned} {\text{TFPgrowth}}_{ir90 - 10} & = \alpha_{0} + \upbeta_{1} {\text{lnTFP}}_{ir90} + \upbeta_{2} {\text{lnMpemp}}_{r90} + \upbeta_{3} {\text{lnArea}}_{r90} \\ & \quad + \upbeta_{4} {\text{lnAge}}_{ir90} + \upbeta_{5} {\text{Nonagriland}}_{r90} + \upbeta_{6} {\text{lnSpe}}_{ir90} + \upbeta_{7} {\text{lnComp}}_{ir90} \\ & \quad + \upbeta_{8} {\text{lnDiv}}_{ir90} + \gamma_{i} + e_{ir} , \\ \end{aligned}$$
(1)
where \({\text{TFPgrowth}}\) is the TFP growth of industry i in region r, and \({\text{TFP}}\) is the TFP level of industry i in region r. \({\text{Mpemp}}\) is the variable for employment market potential, Area is the land area, Age is the average age of plants in industry i in region r, and \({\text{Nonagriland}}\) is the share of non-agricultural land in region r. The main interest variables are the three types of dynamic externalities of industry i in region r noted as Spe, Comp, and Div for specialization, competition, and diversity, respectively. Finally, we added industry dummies \(\gamma_{i}\) for industry i to account for unobserved variables at the industry level, and e is the error component.
Since TFP as measurement of productivity is highly influenced by capital intensity, which is more immobile, compared to employment. Therefore, this research will use another measurement as an alternative which have more mobile capability that is employment. In the employment growth model, we used employment growth as a dependent variable. However, we substituted the initial TFP with the initial wage, and the model was formulated as follows:
$$\begin{aligned} {\text{Empgrowth}}_{ir90 - 10} & = \alpha_{0} + \upbeta_{1} {\text{lnWage}}_{ir90} + \upbeta_{2} {\text{lnMpemp}}_{r90} \upbeta_{3} {\text{lnArea}}_{r90} \\ & \quad + \upbeta_{5} {\text{Nonagriland}}_{r90} + \upbeta_{6} {\text{lnSpe}}_{ir90} + \upbeta_{7} {\text{lnComp}}_{ir90} \\ & \quad + \upbeta_{8} {\text{lnDiv}}_{ir90} + \gamma_{i} + e_{ir} . \\ \end{aligned}$$
(2)

We measured dynamic externalities based on the employment number. The variable \({\text{emp}}_{i,j,r}\) denotes the plant-level employment of plant i in industry j within region r. Variable \({\text{emp}}_{j,r}\) represents the industry-level employment of industry j in region r, while \({\text{emp}}_{j',r}\) stands for the industry-level employment of industries other than industry j in region r. Furthermore, \({\text{emp}}_{r}\) stands for the region-level employment of region r, while \({\text{emp}}\) indicates the national total employment. These notations were applied to measure the specialization, competition, and diversity. To get a better identification between MAR externalities and Jacobs’ externalities, we calculated dynamic agglomeration externalities variables based on the three-digit industrial classification suggested in Beaudry and Schiffauerova (2009).

To choose variables representing agglomeration economies, we measured them using a relative measurement index, where the numbers were derived from comparisons among city-industry levels and national-industry levels. Following Combes (2000), scale (MAR) externalities using employment specialization (Spe) in industry j in region r at time t were calculated as the ratio of the employment share of industry j in region r to the employment share of industry j in the national industry. That is, we specified specialization as follows:
$${\text{Spe}}_{j,r} = \frac{{{\raise0.7ex\hbox{${{\text{emp}}_{j,r} }$} \!\mathord{\left/ {\vphantom {{{\text{emp}}_{j,r} } {{\text{emp}}_{r} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${{\text{emp}}_{r} }$}}}}{{{\raise0.7ex\hbox{${{\text{emp}}_{j} }$} \!\mathord{\left/ {\vphantom {{{\text{emp}}_{j} } {\text{emp}}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\text{emp}}$}}}}.$$
(3)

A value greater than 1 indicates that the industry in a district is locally more specialized than elsewhere in Indonesia. We expect that industrial specialization will increase productivity growth, because knowledge flows are more important within industries.

Furthermore, we derived a variable representing Porter externalities, also following Combes (2000), as the ratio of the inversion of the local Herfindahl index using plant-level data to the inversion of the national Herfindahl index using industry-city data. Thus, industry competition (Comp) faced by a plant that belongs to industry j in region r was measured as follows:
$${\text{Comp}}_{j,r} = \frac{{{\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 {\mathop \sum \nolimits_{i\varepsilon j,r} ({{\text{emp}}_{i,j,r} /{\text{emp}}_{j,r} } )^{2} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\mathop \sum \nolimits_{i\varepsilon j,r} ( {{\text{emp}}_{i,j,r} /{\text{emp}}_{j,r} } )^{2} }$}}}}{{{\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 {\mathop \sum \nolimits_{j} \left( {{\text{emp}}_{j,r} /{\text{emp}}_{j} } \right)^{2} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\mathop \sum \nolimits_{j} ( {{\text{emp}}_{j,r} /{\text{emp}}_{j} } )^{2} }$}}}}.$$
(4)

According to Porter (1990), local competition could pressure firms to create innovative products, adopt new technology, and increase productivity growth. A value greater than 1 is interpreted that industry j in region r that is locally more competitive than elsewhere in Indonesia.

Finally, we measured diversity (Div) to represent the Jacobs externalities, following Marrocu et al. (2013) who modified the diversity index computed by Combes (2000). This index is more focused on the employment level of the rest of the industry in a given region; it directly measures the diversity level faced by a plant in a specific industry, so that it simultaneously captures industrial and regional dimensions. Having already calculated the values of the Herfindahl index based on the employment numbers from the rest of the economy in the given region, this index provides a better measurement of Jacobs externalities. Moreover, the estimated coefficient has a straightforward interpretation, as suggested in Marrocu et al. (2013). The diversity of industry j in region r is calculated as follows:
$${\text{Div}}_{j,r} = \frac{1}{{\mathop \sum \nolimits_{{\begin{array}{*{20}c} {j^{\prime}} \\ {j \ne j^{\prime}} \\ \end{array} }} \left( {\frac{{{\text{emp}}_{{j^{\prime} ,r}} }}{{({\text{emp}}_{r} - {\text{emp}}_{j} }}} \right)}}.$$
(5)

A high value of diversity means a region is more diversified; therefore, productivity growth will increase if cross-industrial knowledge flows are more important than the other externalities.

The novelty of this paper is the introduction of employment market potential as a proxy for local size to control the relationship between dynamic agglomeration externalities and productivity growth. We followed Holl (2012), measuring the employment market potential as the sum of the own regional employment and the regional employments of neighboring areas weighted by the inverse of the GIS distance within a threshold. This variable assumes that the firms’ or workers’ decisions include geographical advantages and spatial environment considerations to enhance firm productivity and maximize profits. The employment market potential is formulated as
$${\text{Mpemp}}_{rt} = {\text{emp}}_{rt} + \mathop \sum \limits_{s \in R284} \frac{{{\text{emp}}_{st} }}{{d_{rs} }} ,$$
(6)
where Mpempr is the employment market potential in region r, empr and emps are the regional employments in regions r and s, respectively, and d is the distance from the district capital r to the district capital s. The threshold distance of d is 25 kilo-meters, and R284 is defined as the total number of districts or cities referring to year 1990. In addition to that, we also comparing long-term impact and medium-term impact as there were no.

5 Estimation issues and instrumental variables

Since the employment market potential in our empirical model is considered an endogenous variable, it is assumed to be correlated with the error term in the OLS regression and potentially results in biased estimates. Therefore, we employ the IV technique to correct this potential bias, following Combes et al. (2010) and Holl (2012). The biggest challenge in an IV analysis is finding a credible instrument. The two conditions of relevance and exogeneity must be satisfied to achieve unbiased estimates (Combes et al. 2010). Combes et al. (2010) and Holl (2012) demonstrated sets of valid instrumental variables to deal with endogeneity between market potential and productivity growth in the cases of France and Spain. Following them, we use long-lagged variables, such as the market potential of population in 1983, which were determined previously and may relate to market potential but which no longer plausibly influence current productivity growth. Moreover, we use geographic characteristics that may be sources of various influences on market potential like ruggedness, types of rocks, and type of physiography. Ruggedness might not only determine population and employment growth in certain areas, but it might also affect firms’ or peoples’ decisions in that area in construction of buildings, roads, and other infrastructure. Likewise, the geology and physiography variables describe the presence of various characteristics of the soil that may affect settlement patterns and direct human activity in a particular area (Combes et al. 2010; Holl 2012).

We calculate population in 1983 from the Village Potential Survey (PODES) of BPS. Following Combes et al. (2010), we measure ruggedness as the difference between the highest and lowest altitude within a city; it is also constructed from PODES. Furthermore, we identify 12 types of rocks from the geological map of 2010 published by the Geology Agency of Indonesia’s Ministry of Energy. We simplify the 12 types into four—sedimentary, volcanic, cretaceous sedimentary, and other—each city is accorded the type that dominates its landscape. Similarly, from the Geology Agency, we also gather physiography details from a map that shows 12 earth morphology types that also can be aggregated into four types of physiography: low plain, low hills, high plain, and mountain areas.

We check the validity of our instruments by calculating the partial correlation between the log of the employment market potential and the instruments, as suggested in Holl (2012). We find strong correlations between instruments and market potential, as presented in Table 1. There is a consistent result between partial correlation and OLS estimation in both significance level and sign. Specifically, we identify the positive effects of the long-lagged market potential of the population in 1983 and the physiography on the market potential employment. On the other hand, we find negative effects of ruggedness and geology on the employment market potential.
Table 1

Partial correlation of instruments and employment market potential

 

Partial correlation coefficient with Mppemp (Ln)

OLS estimation dependent variable Mppemp (Ln)

Mppop83 (Ln)

0.6046***

1.4997***

Ruggedness (Ln)

− 0.1588***

− 0.1070***

Geology

− 0.1753***

− 0.0152

Physiography

0.1107*

0.05814***

OLS estimations include dummies of industry and use robust standard errors

Significance levels: *p < 0.10, **p < 0.05, ***p < 0.01

6 Results and discussion

We presented summary statistics of the variables used in our empirical model in Table 2. At a glance, we can see larger heterogeneity in employment growth compared to TFP growth. It also shows a higher variation of regional employment than market potential employment, indicating that it is more viable to use market employment to proxy local city size while considering large quantities of regional employment. The table also demonstrates that specialization and competition measurements are incredibly more dispersed than that of diversity.
Table 2

Descriptive statistics of variables

Variable

Label

Mean

SD

CV

Annual growth 1990–2010 (# of observations = 1869)

 Productivity growth

TFP growth

0.054

0.077

1.43

 Employment growth

Employ growth

0.014

0.069

4.85

Industry-region initial level 1990 (# = 1869)

   

 Initial TFP level

Initial TFP

817

2426

2.97

 Initial wage rate level

Avg. Wage

2.14

6.52

3.05

Regional industry average age

Avg. Plant age

12.67

9.67

0.76

Regional characteristics in 1990 (# = 232)

   

 Regional employment

Regemp

10,964

24,002

2.18

 Market potential

Mpemp

22,381

27,914

1.25

 Regional area

Area

5861

11,508

4.69

 Non-agriland

Non-agriland

0.39

0.22

0.56

Agglomeration economies in 1990 (# = 1869)

 Specialization

Spe

4.22

14.98

3.55

 Industry competition

Comp

0.09

0.30

3.20

 Industry diversity

Div

5.90

4.35

0.74

SD standard deviation, CV coefficient of variance

6.1 Analysis of the TFP growth models

Before we discuss the results from the estimation, we first scrutinize the validity of the instrumental variables to ensure the accuracy of our empirical approach. As we have several optional instruments (depicted in Tables 3, 4), choosing the one with higher accuracy to instrument, the market potential requires a large value of the first-stage F statistic and a high p value of the Hansen J test. The first-stage F statistic on the instruments is always significantly very large for our attempted instruments. According to Stock and Yogo (2005) critical values for weak instrument testing, our variables’ first-stage F statistics pass the test of weak instruments, giving us confidence that we have strong instruments. Likewise, the large p values of the Hansen J test (testing for overidentification of restrictions) confirm the first-stage F statistic, suggesting that we do not have weak instrument problems.
Table 3

City-industry productivity growth: TFP growth model

Dependent variable

TFP growth

Estimation methods

OLS

IV

 

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Initial TFP

− 0.041***

− 0.041***

− 0.041***

− 0.039***

− 0.039***

− 0.039***

− 0.039***

 

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

Regemp

0.012***

0.010***

     
 

[0.001]

[0.002]

     

WRegemp

 

0.010***

     
  

[0.003]

     

Mpemp

  

0.020***

0.011***

0.011***

0.012***

0.011***

   

[0.002]

[0.004]

[0.004]

[0.004]

[0.004]

Area

0.005***

0.006***

0.006***

0.006***

0.006***

0.006***

0.006***

 

[0.001]

[0.001]

[0.001]

[0.001]

[0.001]

[0.001]

[0.001]

Avg. plant age

− 0.005***

− 0.005***

− 0.004***

− 0.005***

− 0.005***

− 0.005***

− 0.005***

 

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

Non-agriland

0.017***

0.018***

0.017***

0.018***

0.018***

0.018***

0.018***

 

[0.006]

[0.006]

[0.006]

[0.005]

[0.005]

[0.005]

[0.005]

Spe

0.006***

0.006***

0.005***

0.003**

0.003**

0.003**

0.003**

 

[0.001]

[0.001]

[0.001]

[0.001]

[0.001]

[0.001]

[0.001]

Comp

− 0.001

− 0.002

− 0.001

0.003

0.002

0.002

0.002

 

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

Div

0.006***

0.006***

0.006***

0.009***

0.009***

0.009***

0.009***

 

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

[0.002]

_cons

0.158***

0.071**

0.051*

    
 

[0.019]

[0.031]

[0.027]

    

Instruments

       

 Mppop83 (Ln)

   

Y

Y

Y

Y

 Ruggedness

   

N

Y

N

N

 Geology

   

N

N

Y

N

 Physiography

   

N

N

N

Y

 Weak IV test (first stage F)a

   

1040.095

561.028

542.882

526.875

 Wu–Hausman test (p value)

   

0.0011 

 0.0008

 0.0026

0.0013 

 Overidentification (J test)

   

0.000

0.049

1.734

0.379

 (p value)

    

0.824

0.188

0.538

 N

1869

1869

1869

1869

1869

1869

1869

 R2

0.418

0.422

0.421

0.362

0.362

0.363

0.362

Estimations include dummies of industry. Instrumented variable: employment market potential (Mpemp)

White standard errors are reported in brackets

Significance levels: *p < 0.10, **p < 0.05, ***p < 0.01

aEstimated using STATA commands ivreg2, see Baum et al. (2007)

Table 4

Long- and medium-term city-industry productivity and employment growth

Periods

IV

Dependent variable

TFP growth

Emp growth

Periods

90–10

00–10

90–10

00–10

 

(1)

(2)

(3)

(4)

Initial TFP

− 0.039***

− 0.065***

  
 

[0.002]

[0.003]

  

Initial wage

  

− 0.006**

− 0.007*

   

[0.003]

[0.004]

Mpemp

0.011***

0.008

− 0.003

0.001

 

[0.004]

[0.006]

[0.007]

[0.010]

Spe

0.003**

0.003

− 0.010***

− 0.011**

 

[0.001]

[0.002]

[0.003]

[0.005]

Comp

0.002

0.009***

− 0.001

− 0.002

 

[0.002]

[0.003]

[0.002]

[0.003]

Div

0.009***

0.014***

0.010***

0.008**

 

[0.002]

[0.004]

[0.002]

[0.003]

_cons

    

Weak IV test (first stage F)a

561.028

366.192

495.408

354.681

Wu–Hausman test (p value)

0.001

0.0000

0.680

0.7228

Overidentification (J test)

0.049

0.405

0.506

0.908

(p value)

0.824

0.524

0.477

0.341

N

1869

2513

1869

2513

R 2

0.362

0.33

0.188

0.102

Estimation in Table 8 indicates strong correlation between instruments and market potential. It also shows positive effects of long-lagged market potential of the population in 1983 and physiography on the market potential employment. On the other hand, we find the negative effects of ruggedness and geology on the employment market potential. This result is supported by statistical tests that indicate that our instruments are valid for better estimation in the second-stage regression, as shown in Table 3.

Table 3 shows that the estimated coefficients on the instrumental variables in columns (4)–(7) are smaller than OLS estimates in column (3). In addition, we test for the endogeneity of the regressors using the Durbin–Wu–Hausman test. Rejecting the null hypothesis, we find statistical evidence of endogeneity in the TFP growth regression, and therefore, the IV and OLS estimates are significantly different. Thus, we focus the discussion on the IV results, although we also report the OLS results (particularly for the analysis by period and industry). Ultimately, we prefer Mppop83 and ruggedness as the instrumental variables, with magnitudes presented in column (5) of Table 3.

After using the IV estimation approach to address the possible interactions between higher productivity growth and greater employment potential of a market, we find that market potential has a strong positive impact on productivity, supporting the result of Combes et al. (2010) and Holl (2012). The result indicates that employment market potential has strong effects on city size and, subsequently, affects the source type and magnitude of dynamic agglomeration externalities on productivity growth. The approach corrects the overestimation of the regional employment’s influence on local size. We observe that instrumenting for market employment always results in a lower estimation of the corresponding point estimates. This indicates that OLS estimates are biased upwards due to simultaneity problems. For further analysis, we take up the estimates of column (5) as our benchmark estimates based on their results in the first-stage F test and Hansen J test. We use this benchmark to investigate the robustness of our results further, to analyze different periods of growth, and to examine the impact across industry groups.

Table 3 shows the positive effects of specialization and diversity on city-industry TFP growth, supporting the MAR and Jacob externalities. In this respect, the result seems to be consistent with the result in Henderson et al. (1995), finding evidence of MAR externalities in the traditional industries and of both Jacobs and MAR externalities in the new high-technology industries in the United States. Furthermore, our findings also parallel the work of De Lucio et al. (2002), which finds significant effects of specialization and diversity on TFP growth in the case of Spain.

In the case of Indonesia, the present paper partly supports the findings in Sjöholm (1999)—finding strong evidence of diversity on productivity growth—and the findings in Widodo et al. (2014)—identifying specialization as having positive effects on city growth and diversity having negative effects. However, our method is different from the previous literature. We carefully applied methods ignored by the previous authors to address input endogeneity of the firm production function and to set a strong approximation of local size to control local industrial structure. While previously the amount of labor force or region’s population is used as the representation of local size, our research will use market potential as replacement for representation. The control variables seem to have the expected signs. Larger land area and non-agricultural land lead to faster growth of a city industry, indicating comparative advantages of the city. Those factors can facilitate firms’ accumulation of more resources in producing goods and finally supporting growth. However, we identify that as industry grows older, productivity growth decreases.

To confirm the robustness of our results, we performed robustness checks, as reported in Appendix Table 8. The table presents the different specifications. The estimates from the benchmark model are presented in column (1) for comparison; excluding the high-technology industries and other manufacturing, the industries are presented in columns (2) and (3), respectively. Columns (4) and (5) provide results from alternative measures of productivity using different weights of TFP aggregation to calculate TFP growth and labor productivity growth. The initial-related variable is also changed accordingly. Our results are consistent in both signs and significance levels, indicating that our empirical models are robust to a variety of specifications and alternative measures of productivity growth.

6.2 Productivity growth by time periods: long term and medium term

As was discussed in the “Introduction”, to avoid the effect of the 1997 Asian financial crisis, and at the same time focus on the period after the decentralization policy in Indonesia, we provided an alternative, shorter, medium-term time-period analysis (2000–2010). Table 4 shows the IV estimates in columns (1)–(4). The Wu–Hausman test for endogeneity in columns (1) and (2) indicates that the OLS and IV estimates of TFP growth regression are significantly different. However, this is not the case for the employment growth regression, in which the Wu–Hausman test in columns (3) and (4) suggest that the results of both methods are relatively similar. It also should be noted that the effects of the employment market potential on TFP growth become insignificant when analyzed over a shorter period. We suspected that the growing number of large cities within 2000–2010 might have reduced and eliminated the role of the employment market potential in controlling local size, though the employment market potential still influenced the coefficients of the agglomeration variables.

The table shows that specialization and diversity positively affect TFP growth in the long-term period 1990–2010 [column (1)]. However, a contrasting result was shown for employment growth, as we found that specialization had negative effects on employment growth, although diversity still had positive effects. The results generally confirm and incorporate the findings of both Glaeser et al. (1992) with employment growth and Cingano and Schivardi (2004) with TFP growth. More precisely, our results were consistent with the findings of Henderson et al. (1995) that specialization and diversity played important roles in employment growth, conditional on the industry type.

We obtained a different identification of larger effects of diversity with additional positive effects of competition instead of specialization in the medium-term period 2000–2010 [column (2)]. As far as the effect’s magnitude is concerned, the role of externalities was stronger and broader on city growth, showing that the local industry may need to be adjusted accordingly. In that case, we need to take into account that the effects are obviously different between the long term and medium term. We believe that regional competition after the decentralization policy increased and enhanced local productivity growth.

Comparing the different growth measurements (see Table 4), we confirmed that the relationship between employment growth and productivity growth need not be positive, as noted by Combes et al. (2004) and Cingano and Schivardi (2004). Furthermore, we also found consistent negative effects of specialization on employment growth (consistent with the findings of Glaeser et al. 1992) in both terms, as is shown in columns (3), and (4). Duranton and Puga (2014) argued that price elasticity of demand is the main factor in determining the relationship between productivity and employment. They explained that for mature industries, which usually have inelastic demand, increased productivity growth was associated with lower employment growth. Therefore, using a TFP growth estimation does not necessary result in a positive effect of specialization on productivity growth if the manufacturing sectors are comprised of more small-sized firms or new-entry firms. Typically, these firms are more likely in favor of Jacobs externalities due to their dependence on the external environment provided by diversity.

6.3 Productivity growth by industry

We classified the 23 industries of the two-digit SIC into six groups: (a) traditional, (b) heavy, (c) transportation equipment, (d) machinery and electronics, (e) high technology, and (f) other industries, following Henderson et al. (2001). We only reported the IV estimates by industry that showed large values from the first-stage F test and high p values of the Hansen J test. Thus, if those values were small due to a small number of observations, we did not include them in Tables 5 and 6. We, therefore, reported only the estimation results by industry for the traditional, heavy, and machinery and electronics industries.
Table 5

Long-term city-industry productivity and employment growth by industry

1990–2010

IV

Industry group

Traditional

Heavy

Mach&Elect

Dependent variable

TFP

EMP

TFP

EMP

TFP

EMP

 

(1)

(2)

(3)

(4)

(5)

(6)

Initial TFP

− 0.039***

 

− 0.041***

 

− 0.038***

 
 

[0.002]

 

[0.004]

 

[0.005]

 

Initial wage

 

− 0.010***

 

− 0.006

 

0.041**

  

[0.003]

 

[0.004]

 

[0.017]

Mpemp

0.014***

0.007

0.009

0.006

− 0.007

− 0.134***

 

[0.005]

[0.010]

[0.006]

[0.011]

[0.013]

[0.040]

Spe

0.005***

− 0.006

0.001

− 0.007

− 0.001

− 0.073***

 

[0.002]

[0.004]

[0.002]

[0.005]

[0.004]

[0.023]

Comp

0.000

− 0.002

0.007

− 0.006

0.009

0.031**

 

[0.002]

[0.003]

[0.004]

[0.005]

[0.008]

[0.015]

Div

0.008**

0.006*

0.011***

0.016***

0.023**

0.019

 

[0.003]

[0.003]

[0.004]

[0.004]

[0.009]

[0.014]

Weak IV test (first stage F)a

308.18

242.03

141.15

190.19

14.42

7.47

Wu–Hausman test (p value)

0.0081

0.6308

0.0785

0.6109

0.918

0.0123

Overidentification (J test)

0.527

4.635

0.120

2.566

0.217

0.695

(p value)

0.468

0.031

0.729

0.109

0.641

0.405

N

1020

1020

542

542

114

114

R 2

0.368

0.186

0.387

0.228

0.468

0.153

Estimations include dummies of industry. Instrumented variable: employment market potential (Mpemp). Each regression includes control for industry characteristic of age average and for regional characteristics of area and non-agriland. Instrumental variables: mppop83 and ruggedness. White standard errors are reported in brackets.

Significance levels: *p < 0.10, **p < 0.05, ***p < 0.01

aEstimated using STATA commands ivreg2, see Baum et al. (2007)

Table 6

Medium-term city-industry productivity and employment growth by industry

2000–2010

IV

Industry group

Traditional

Heavy

Mach&Elect

Dependent variable

TFP

EMP

TFP

EMP

TFP

EMP

 

(1)

(2)

(3)

(4)

(5)

(6)

Initial TFP

− 0.064***

 

− 0.073***

 

− 0.070***

 
 

[0.003]

 

[0.005]

 

[0.008]

 

Initial wage

 

− 0.012***

 

− 0.010*

 

0.066**

  

[0.004]

 

[0.006]

 

[0.031]

Mpemp

0.009

0.013

0.021*

0.025

− 0.008

− 0.208***

 

[0.008]

[0.013]

[0.012]

[0.016]

[0.039]

[0.077]

Spe

0.005*

− 0.005

0.009**

− 0.004

0.001

− 0.105***

 

[0.003]

[0.006]

[0.004]

[0.007]

[0.008]

[0.039]

Comp

0.007

0.001

0.000

− 0.017***

0.029*

0.024*

 

[0.004]

[0.004]

[0.007]

[0.006]

[0.017]

[0.014]

Div

0.016***

0.003

0.011

0.012**

0.001

0.000

 

[0.005]

[0.005]

[0.007]

[0.005]

[0.013]

[0.015]

Weak IV test (first stage F)a

225.70

197.49

96.87

93.06

5.92

13.72

Wu–Hausman test (p value)

0.000

0.792

0.048

0.496

0.461

0.063

Overidentification (J test)

1.357

0.812

2.248

11.54

0.961

0.269

(p value)

0.244

0.367

0.134

0.001

0.327

0.604

N

1359

1359

718

718

157

157

R 2

0.322

0.106

0.386

0.133

0.475

− 0.031

Estimations include dummies of industry. Instrumented variable: employment market potential (Mpemp). Each regression includes control for industry characteristic of age average and for regional characteristics of area and non-agriland. Instrumental variables: mppop83 and ruggedness. White standard errors are reported in brackets.

Significance levels: *p < 0.10, **p < 0.05, ***p < 0.01

aEstimated using STATA commands ivreg2, see Baum et al. (2007)

The disaggregated analysis by industry is consistent with the aggregate analysis in attributing specialization and diversity as the major factors of city-industry growth. The impact of both externalities varies substantially across industries. We also observed that, between the long-term and medium-term analyses, the effect of specialization and diversity changed, seen in larger effects of diversity in the traditional industries and specialization in the heavy industries. Furthermore, we found that the productivity growth of the machinery and electronics industries strongly depended on diversity in the long-term analysis, but it then changed to depend on competition in the shorter analysis. The analysis by industry using employment growth shows that diversity has the strongest effects on the machinery and electronics in the long term. We also identified the positive effects of diversity in the traditional industries. Interestingly, we also observed positive effects of competition in the machinery and electronics industry in the medium term.

Considered as mature industries, traditional and heavy industries usually depend on specialization. However, our results showed that those industries were also affected by diversity. Looking at the data, about 55.83% of the traditional industries and 50.83% of the heavy industries are small firms (during 1990–2010). These statistics support the fact that mature industries also need diversified environments, since the number of small firms is dominant. It is also important to note based on our result that initial wage has a negative effect on employment growth. This is due to the fact that industries in Indonesia are dominated by traditional industry; thus, an increase in wage will likely to decrease firm’s production, especially mature firms, and firms might be moving into machinery tools to produce output. According to the theory of the “nursery city” by Duranton and Puga (2001), the authors argued that a diversified environment is suitable for new plants or small firms, whereas specialized cities are important for mature industries. Therefore, diversity is still important for productivity growth even in mature industries. Small firms usually depend on external environments to acquire knowledge and learn about innovation in large cities. These results are consistent with the product lifecycle theory provided by Duranton and Puga (2001). In this theory, diversity is more important for the firm in the initial development of a product to learn from a cross-industrial environment. Once the new product is established and the firm is ready to start mass production, the firm may relocate to specialized areas, benefiting from the surrounding mature industries.

Furthermore, the effect of competition on productivity growth is revealed in the medium term for the machinery and electronics industries. This finding supports the MAR externalities theory that suggests that firms in similar industries, or in a cluster, grow more rapidly due to their competition. However, at the same time, competition also validates the Jacobs externalities theory, since diversified environments create pressure for firms to innovate for survival. In general, our results partly fit the prediction of Duranton and Puga (2000) in that mature industries are more productive in specialized cities, while younger industries grow faster in diversified cities.

7 Conclusions

This study was designed to determine the effects of dynamic agglomeration externalities on productivity growth in Indonesia. This paper pointed to a key finding that the employment market potential controlling local size has strong effects on determining the magnitudes of externalities and correcting overestimation for each city size classification. We found vigorous evidence of the positive impact of market potential on TFP growth in the long term but not on employment growth. We also provided strong evidence of the importance of specialization and diversity on city-industry growth and that, as suggested by Duranton and Puga (2000), specialized and diverse cities can coexist. It explains that MAR externalities and Jacobs externalities are important for Indonesian manufacturing growth, but the former appears stronger than the latter even though the latter captures a wider range of industries. In addition, new evidence of Porter externalities appears in the machinery and electronics industries.

The evidence of changing local industrial structure identified in both long-term and medium-term analyses, toward stronger specialization, and new evidence of the role of competition in the medium term were discovered. The positive effects of specialization on TFP growth and diversity on TFP growth are related to the industrial composition of the manufacturing sector in Indonesia. It suggests that as an industry becomes more mature, cities tend to become more specialized. Moreover, this evidence is supported by the fact that competition is important in stimulating innovation and increased productivity. However, because of the dominating small firms in traditional and heavy industries, the role of diversity is still important. This result supports the idea of the industry lifecycle theory by Duranton and Puga (2001).

Footnotes

  1. 1.

    City size is classified into three categories: “small–medium” (population under 500,000), “large” (population between 500,000 and 1,000,000), and “metro-megapolitan” (population over 1,000,000).

  2. 2.

    For instance, Rosenthal and Strange (2004) found that the impact of city size on productivity is between 3 and 8%.

  3. 3.
  4. 4.

    Stata command “levpet” created by Petrin et al. (2004) was used to estimate the plant-level production function.

Notes

Acknowledgement

I would like to thank to Muhammad Irfan Shaleh for sharing the GIS Euclidean-Distance and geological data.

References

  1. Abdel-Rahman HM, Anas A (2004) Theories of systems of cities. Handb Reg Urban Econ 4:2293–2339CrossRefGoogle Scholar
  2. Almeida R (2007) Local economic structure and growth. Spat Econ Anal 2(1):65–90CrossRefGoogle Scholar
  3. Almeida R, Fernandes AM (2013) Explaining local manufacturing growth in Chile: the advantages of sectoral diversity. Appl Econ 45(16):2201–2213CrossRefGoogle Scholar
  4. Baum CF, Schaffer ME, Stillman S (2007) Enhanced routines for instrumental variables/GMM estimation and testing. Stata J 7:465–506Google Scholar
  5. Beaudry C, Schiffauerova A (2009) Who’s right, Marshall or Jacobs? The localization versus urbanization debate. Res Policy 38(2):318–337CrossRefGoogle Scholar
  6. Cingano F, Schivardi F (2004) Identifying the sources of local productivity growth. J Eur Econ Assoc 2(4):720–744CrossRefGoogle Scholar
  7. Combes P-P (2000) Economic structure and local growth: France, 1984–1993. J Urban Econ 47:329–355CrossRefGoogle Scholar
  8. Combes P-P, Magnac T, Robin J-M (2004) The dynamics of local employment in France. J Urban Econ 56:217–243CrossRefGoogle Scholar
  9. Combes P-P, Duranton G, Gobillon L, Roux S (2010) Estimating agglomeration economies with history, geology, and worker effects. Agglomeration economics. University of Chicago Press, Chicago, pp 15–66CrossRefGoogle Scholar
  10. De Groot HLF, Poot J, Smith MJ (2009) Agglomeration externalities, innovation and regional growth: theoretical perspectives and meta-analysis. In: Capello R, Nijkamp P (eds) Handbook of regional growth and development theories. Edwar Elgar, Cheltenham, UK, pp 256–281Google Scholar
  11. Dekle R (2002) Industrial concentration and regional growth: evidence from the prefectures. Rev Econ Stat 84(2):310–315CrossRefGoogle Scholar
  12. Duranton G, Puga D (2000) Diversity and specialisation in cities: why, where and when does it matter? Urban Stud 37(3):533–555CrossRefGoogle Scholar
  13. Duranton G, Puga D (2001) Nursery cities: urban diversity, process innovation, and the life cycle of products. Am Econ Rev 91(5):1454–1477CrossRefGoogle Scholar
  14. Duranton G, Puga D (2014) The growth of cities. In: Handbook of economic growth, vol 2, pp 781–853. http://dx.doi.org/10.1016/B978-0-444-53540-5.00005-7
  15. Gill IS, Goh C-C (2010) Scale economies and cities. World Bank Res Obs 25:235–262CrossRefGoogle Scholar
  16. Glaeser EL, Kallal HD, Scheinkman JA, Shleifer A (1992) Growth in cities. J Polit Econ 100(6):1126–1152CrossRefGoogle Scholar
  17. Henderson J Vernon, Kuncoro A, Turner M (1995) Industrial development in cities. J Polit Econ 103(5):1067–1090CrossRefGoogle Scholar
  18. Henderson JV, Lee T, Lee YJ (2001) Scale externalities in Korea. J Urban Econ 49(3):479–504CrossRefGoogle Scholar
  19. Holl A (2012) Market potential and firm-level productivity in Spain. J Econ Geogr 12:1191–1215CrossRefGoogle Scholar
  20. Levinsohn J, Petrin A (2003) Estimating production functions using inputs to control for unobservables. Rev Econ Stud 70:317–341CrossRefGoogle Scholar
  21. Lucio D, Juan J, Herce JA, Goicolea A (2002) The effects of externalities on productivity growth in spanish industry. Reg Sci Urban Econ 32(2):241–258CrossRefGoogle Scholar
  22. Marrocu E, Paci R, Usai S (2013) Productivity growth in the old and new Europe: the role of agglomeration externalities. J Reg Sci 53:418–442CrossRefGoogle Scholar
  23. Melo PC, Graham DJ, Noland RB (2009) A meta-analysis of estimates of urban agglomeration economies. Reg Sci Urban Econ 39(3):332–342CrossRefGoogle Scholar
  24. Petrin A, Poi BP, Levinsohn J (2004) Production function estimation in stata using inputs to control for unobservables. Stata J 4:113–123Google Scholar
  25. Porter M (1990) The competitive advantage of nations. MacMillan, LondonGoogle Scholar
  26. Rosenthal SS, Strange WC (2003) Geography, industrial organization, and agglomeration. Rev Econ Stat 85(2):377–393CrossRefGoogle Scholar
  27. Rosenthal SS, Strange WC (2004) Evidence on the nature and sources of agglomeration economies. In: Henderson JV, Thisse JF (eds) Handbook of regional and urban economics, vol 4. Elsevier, Amsterdam, pp 2119–2171Google Scholar
  28. Sjöholm F (1999) Productivity growth in indonesia: the role of regional characteristics and direct foreign investment. Econ Dev Cult Change 47(3):559–584CrossRefGoogle Scholar
  29. Stock JH, Yogo M (2005) Testing for weak instruments in linear IV regression. In: Andrews DWK, Stock JH (eds) Identification and inference in econometric models: essays in honor of Thomas J. Rothenberg. Cambridge University Press, Cambridge, pp 80–190CrossRefGoogle Scholar
  30. Widodo W, Salim R, Bloch H (2014) Agglomeration economies and productivity growth in manufacturing industry: empirical evidence from indonesia. Econ Record 90(s1):41–58CrossRefGoogle Scholar

Copyright information

© The Japan Section of the Regional Science Association International 2018

Authors and Affiliations

  1. 1.Department of Economics, Faculty of Economics and Business, Institute for Economic and Social Research (LPEM)Universitas IndonesiaJakartaIndonesia

Personalised recommendations