Industrial agglomeration effects in Japan: Productive efficiency, market access, and public fiscal transfer




This study examines whether agglomeration economies, market access and public fiscal transfer have a positive or negative influence on the productive efficiency of Japanese regional industries. To attain the research objective, stochastic frontier analysis is applied to a Japanese data set at a prefecture level which consists of estimated spatial and industrial economic activities from 1980 to 2002. An empirical result obtained in this study indicates that both the agglomeration economies and the improvement of market access have a positive influence on the productive efficiency of Japanese manufacturing and non-manufacturing industries. In contrast, the public fiscal transfer has a negative impact on the productive efficiency. These findings indicate that many prefectures (corresponding to States in the United States), which are characterized by weak market access and/or high dependence on public fiscal transfer, are often associated with their low productive efficiency.


Este estudio examina si las economías de aglomeración, acceso a mercados y transferencia fiscal pública tienen una influencia positiva o negativa en la eficacia productiva de las industrias regionales japonesas. Para lograr el objetivo de la investigación se aplica un análisis de frontera estocástica a un conjunto de datos de Japón a escala de prefectura, el cual consiste en actividades económicas espaciales e industriales estimadas entre 1980 y 2002. Un resultado empírico obtenido en este estudio indica que tanto las economías de aglomeración como la mejora en el acceso a mercados tienen una influencia positiva en la eficacia productiva de las industrias japonesas manufactureras y no manufactureras. En contraste, la transferencia fiscal pública tiene un impacto negativo en la eficacia productiva. Estos hallazgos indican que muchas prefecturas (equivalentes a los estados de los Estados Unidos), caracterizadas por un pobre acceso a mercados y/o dependencia de transferencias fiscales públicas, están asociadas a menudo con su baja eficacia productiva.

inline image

1 Introduction

Agglomeration economies are external economies that stem from the location of firms belonging to the same or related industrial sectors. It is widely known that firms can fully take advantage of benefits from industrial agglomerations if they locate closely each other. To empirically investigate the agglomeration economies, many previous studies examined the impact of agglomeration economies on productivity, or so-called ‘technical efficiency’ of firms or industries. They indicated a positive influence of agglomeration economies on productivity of industries. The finding is important from the perspective of regional economy because firms with high efficiency are expected to survive in a competitive market, so contributing to their regional economic growth.

The purpose of this study is to extend the previous studies on agglomeration economies by examining two research hypotheses regarding its influence on industrial productivity at the level of Japanese prefectures (implying a local government unit such as ‘States’ in the United States). One of the two hypotheses, summarized below, measures how market access influences regional agglomeration economies. To examine the hypothesis, this study distinguishes the market access from the other conventional production factors related to agglomeration economies. Then, this study estimates the influence of industrial agglomeration and market access on regional economies from the perspective of industrial productivity.

The effect of market access has been long investigated in previous studies that were interested in a managerial decision-making process on corporate location. For example, Hewings et al. (1998) discussed market orientation becoming a dominant factor in selecting the location of firms. In the theory of agglomeration economies, the best location of firms depends on their market access, which consists of a market size and a transportation cost. For example, Fujita et al. (1999) and Fujita and Thisse (2002) indicated that market access affected the economic performance of firms through their linkages. The recent classification of agglomeration economies provided by Parr (2002) also incorporated such linkage externalities in addition to these related conventional regional and economic factors. Indeed, in the spatial economic analysis, Hewings and Parr (2007) argued that the degree of spatial interaction among regions was more influential and intricate than previously thought. However, such previous studies examined insufficiently the effect of market access from industrial productivity because they could not access data sets on precise economic distance.

To overcome the research difficulty, this study attempts to provide new empirical evidence concerning agglomeration economies by fully utilizing a data set on market access in Japan that contains such an economic distance. The first hypothesis of this study is summarized as follows:

Hypothesis 1:The enhancement of industrial agglomeration and the improvement in market access increase the productive efficiency of Japanese regional industries.

After examining the first hypothesis, this study investigates how governmental spending influences the productivity of regional industries. This analysis is motivated by the recent policy discussion on Japanese fiscal deficit. In the 1990s, after the end of the bubble economy, Japanese policy-makers believed that governmental spending could improve the economic stagnation. Based upon such a belief, the Japanese government spent a large amount of public funds for regional industries via local governments. For example, several highway systems were constructed in Japan during the observed period (1980–2002). It is true that such investment has improved the market access of locations. However, as a negative consequence of the large public spending, Japanese fiscal balance continued to deteriorate over time. As a result of public spending, the total fiscal deficit increased to about 871 trillion yen.1 How to maintain a fiscal balance and how to improve the efficiency of public spending have become a major policy issue of the Japanese government. Under this situation, it is necessary for us to examine the relationship between the productive efficiency of regional industries and the public spending for future economic growth in Japan. In particular, the Japanese government provides local governments with most of the funds for regional public spending. The political system often produces a policy difficulty within local governments. For example, the fiscal transfer from the Japanese central government to local governments negatively affected their performance because the governmental funding reduced their motivation towards an efficient use of the taxpayer's money in supplying public goods.2

Acknowledging the existence of such a political problem, the Japanese government did not change fiscal transfer policy and local government continued to use fiscal resources inefficiently. To provide an empirical perspective on the policy issue in Japan, this study investigates how the fiscal transfer to local governments has affected the productivity of regional industries. The policy concern is summarized by the following second hypothesis:

Hypothesis 2:A fiscal transfer from the central government to local governments negatively affects the productive efficiency of Japanese regional industries.

The remainder of this study is organized as follows. Section 2 briefly reviews literature on agglomeration economies, focusing on the relationship between agglomeration economies and industrial productivity. Section 3 summarizes the methodology used for productivity measurement. Section 4 summarizes the characteristics of Japanese regional economies along with a description on a data set and an empirical model used in this study. The results of our empirical analysis are documented in Section 5. Section 6 concludes this study along with policy implications for the Japanese national land planning reforms.

2 A literature review

2.1 TFP and productive efficiency

During the last two decades, many previous studies measured the impact of industrial agglomerations by total factor productivity (TFP) that is a conventional productivity measure of firms and industries.3 The previous studies indicated that the TFP of a given industry rose by external economies due to the existence of agglomeration economies. In other words, they suggested that an increased level of output was obtained by agglomeration economies along with a reduced level of input. Here, we need to mention that this study uses ‘productive efficiency’ or technical efficiency as a productivity measure and distinguish between the productive efficiency and the conventional TFP. Our rationale concerning the separation is because when we examine the TFP of firms, this study needs to identify two components for enhancing firms' TFP. One of the two components is due to the improvement of individual firm's productive efficiency. The other component is related to the technological progress that advances industry-wide production technology. The technology advancement can produce an upward shift on an industrial production frontier. As a result, even though the first component (i.e. productive efficiency) does not exist in a firm, the firm can be efficient under technological advancement as long as it decreases an amount of inputs under a given level of outputs or increase an amount of outputs under a given level of inputs. To avoid such a methodological problem within the TFP measurement, the distinction between the two components is commonly accepted in production economics. Furthermore, the measure of conventional TFP focuses on production activity of industries and it does not incorporate a level of utility into performance evaluation such as an index to indicate people's living standards. The methodological feature is problematic. Moreover, we need to measure the improvement of productive efficiency by distance shrinkage of each firm from a production frontier over time, whereas the technological progress means an advancement of the production frontier in a time horizon. The conventional research on TFP has a difficulty in distinguishing them by measuring the production frontier.

To overcome several methodological issues concerning the TFP-based measures, Beeson and Husted (1989) measured the productive efficiency of a manufacturing sector at the state level of the United States and found considerable variations in productive efficiency among samples. They explained that these variations were related to several factors such as differences in labour-force characteristics and levels of agglomeration economies. Mitra (1999) estimated the firm-specific efficiency index by applying a stochastic frontier analysis (SFA) to two Indian industries (i.e. electric machinery and cotton and cotton textiles) and verified the relationship between firms' efficiency and agglomeration economies. Moreover, Mitra (2000) measured the TFP growth along with its components (i.e. technological progress and changes in productive efficiency) using regional data on industries in India. The author provided the evidence for an existence of agglomeration economies which made possible for firms to obtain an economic benefit through an improved quality of labour and an enhanced utilization of resources.

2.2 Methodology

Paying attention to research methodology in the previous studies on agglomeration economies, this study finds that Mitra (1999) and some other studies have used a two-stage regression to examine the relationship between technical (or productive) efficiency and agglomeration economies. They estimated a stochastic frontier production function and various efficiencies at the firm level. Then, the estimated efficiencies were regressed by several firm-specific variables in order to identify which factors explained such differences in estimated efficiency measures among firms. Meanwhile, Driffield and Munday (2001) and Tveteras and Battese (2006) and several other studies applied a single-stage regression to the estimation of technical efficiency and its relationship with agglomeration economies. Such a single-stage regression method solved an inconsistency issue that remained in the two-stage approach that was due to an assumption on the independence of inefficiency. For example, Driffield and Munday (2001) investigated both an impact of foreign manufacturing investment on productive efficiency of the UK's industries and an impact on industries' spatial agglomeration. They concluded the importance of industrial agglomeration. In a same research group, Tveteras and Battese (2006) examined the influence of regional agglomeration externalities in the productivity of the Norwegian salmon aquaculture industry. They distinguished between the agglomeration effects on the production possibility frontier and those on the productive inefficiency that is caused by ‘errors’ in the optimization behaviour of firms. Their study supported the existence of externalities for both the production frontier and productive inefficiency. Furthermore, Mitra and Sato (2007) examined the relationship between productive efficiency and agglomeration economies in the Japanese manufacturing sector. Their study indicated that the influence of agglomeration economies depended on the type of industries. Their study also discussed a positive effect of agglomeration economies on productive efficiency.

Finally, the literature survey indicates a positive influence of agglomeration externalities on the productive efficiency of firms or industries. This study explores the policy concern by utilizing a SFA-based production frontier analysis (including the TFP measurement). That is a position of this study.

3 Research methodology

This study applies the production-based SFA to a data set at Japanese prefecture level, as mentioned previously. The SFA was proposed by Aigner et al. (1977); Meeusen and van den Broeck (1977); and Battese and Corra (1977). Various types of SFA models were also proposed by many researchers for the efficiency measurement of production activities. A comprehensive survey, along with a variety of advanced models of SFA, can be found in Kumbhakar and Lovell (2000). Among the proposed SFA models, this study applies the model proposed by Battese and Coelli (1995), because it can examine the mean inefficiency of firms by a single-stage regression, using various explanatory variables for efficiency. The advantage of the SFA model is not limited to a methodological benefit originated from the single-stage estimation. That is, the SFA model can investigate the level of firm-specific efficiency. Furthermore, this study can make a linkage between the SFA-based efficiency measure and regional components to yield a productivity growth. Such a methodological feature is important because this study uses a data set that has a panel structure with a time series aspect regarding each prefecture. It is commonly known that the productivity growth measure, as investigated in this study, needs to be decomposed into several components such as an efficiency change and technological progress.4 Thus, the proposed use of the SFA approach is useful in dealing with such a panel data set.

To describe the methodology proposed in this study, we use a production frontier model which can be specified by the following logarithmic form:


where Yjt is the level of output for the jth (j= 1, . . . , J) prefecture in the tth period and Xijt are the level of the ith (I= 1, . . . , I) input for the jth prefecture (j= 1, . . . , J) in the period t. It is assumed that the technological progress differs among prefectures due to the difference in their regional industrial policies and their idiosyncratic conditions. To capture such a prefecture-specific condition, the dummy variable, dumj, is incorporated in such a manner that the dummy is multiplied by the time t as an independent variable in Equation (1). The symbols (α and β) are parameter estimates. The parameter (α) indicates a constant parameter (α0) and the other parameters represent a level of production technology (αi, i= 1, . . . , I), while β are parameters to capture a technological progress that may vary across prefectures (βj, j= 1, . . . , J). An error term (vjt − ujt) consists of two parts: an observational error term (vjt) and a managerial error associated with productive inefficiency (ujt). The error term (vjt) is assumed to be i.i.d. N (0, inline image) as well as independency on the managerial error term (ujt) and all regressors of the proposed production function. The ujt is a non-negative random variable, which is assumed to be independently distributed as the truncation at zero of the N (µ, inline image) distribution.

Given Equation (1), the productive efficiency, or TEjt, is measured by the ratio of the observed output Yjt to the estimated production frontier output yjt. That is:


Equations (1) and (2) provide the following growth accounting:


where the dot over each variable indicates a percentage change of each variable. The equation measures an output growth that can be decomposed into the three components: (i) a technological progress inline image; (ii) an input growth inline image; and (iii) a change in productive efficiency (inline image).

This study formulates the mean of productive inefficiency (µjt) in the following manner:


where Z are explanatory variables concerning mean inefficiencies. They depend upon regional industries over time. The symbols related to δ are parameters to be estimated. Because of the model structure, the estimated level of productive efficiency (TEjt) is non-negative with an upper bound of unity. That is:


After substituting (4) into (3), the growth accounting is specified as follows:


Thus, the growth rate of an observed output (inline image) is measured by the sum of the three components: (i) the rate of technological progress inline image; (ii) the weighted growth rate of inputs inline image; and (iii) the rate of productive efficiency change inline image. By incorporating an error term (inline image) into the growth rate of TFP, this study can specify it in the following manner:


4 Japanese regional structure, data and empirical model

4.1 Regional structure in Japan

In Japan, the geographical distribution of industries is unique. To explain such a unique feature, Table 1 that shows the regional distribution of economic activity in 2000. For example, the Capital region (listed in the fifth row of the table) comprises Saitama, Chiba, Tokyo and Kanagawa prefectures which accounts for 30.4 percent of economic scale (measured by production share) and 26.1 percent of population scale (measured by population share), respectively. Meanwhile, this region occupies only 7.3 percent of the entire inhabited area in Japan. Such a discrepancy between the economic/population scale and the inhabited area implies a high degree of economic concentration in the Capital region (including Tokyo). The high level of concentration can be confirmed by comparing the economic size (30.4%) of the Capital region with that (16.2%) of the Kansai region. The region is the second largest in terms of economic scale and it comprises Shiga, Kyoto, Osaka, Hyogo, Nara and Wakayama prefectures. Furthermore, the economic size of the Capital region is as large as the sum of the Kansai and the Chubu regions, which is the third largest region in terms of economic scale (14.4%). The Chubu region comprises Gifu, Shizuoka, Aichi and Mie prefectures. The statistics clearly indicates a high degree of economic concentration in the Capital region.

Table 1.  Regional characteristics in Japan (2000)
RegionInhabited area (%)Population share (%)Production share (%)
  1. Note: Hokkaido (Hokkaido), Tohoku (Aomori, Iwate, Miyagi, Akita, Yamagata, Fukushima and Niigata), Kita-Kanto (Ibaraki, Tochigi, Gunma and Yamanashi), Capital region (Saitama, Chiba, Tokyo and Kanagawa), Hokuriku (Toyama, Ishikawa and Fukui), Chubu (Nagano, Gifu, Shizuoka, Aichi and Mie), Kansai (Shiga, Kyoto, Osaka, Hyogo, Nara and Wakayama), Chugoku (Tottori, Shimane, Okayama, Hiroshima and Yamaguchi), Shikoku (Tokushima, Kagawa, Ehime and Kochi), Kyushu (Fukuoka, Saga, Nagasaki, Kumamoto, Oita, Miyazaki and Kagoshima) and Okinawa (Okinawa).

Capital region7.326.130.4

Figure 1 depicts the production share of (i) the manufacturing industries and (ii) the non-manufacturing industries in each prefecture. The two types of industries are widely distributed in the metropolitan regions such as the Capital region, the Kansai region and the Chubu region. The manufacturing industries locate in the non-metropolitan regions such as the Kita-Kanto region and the Tohoku region near the metropolitan region. This unique feature came about because past industrial policy intended promoting the decentralization of manufacturing industries. Meanwhile, the metropolitan areas in the Capital region have a high production share in the non-manufacturing industries that comprise all types of industries, with the exception of the manufacturing industries such as mining and service industries. The service industry has the maximum share in the non-manufacturing industries so that a business trend of the service industry influences the distribution of non-manufacturing industries. An important feature of the service industry is that it depends on the demand scale of each region in a level of population concentration. Consequently, the geographical distribution of non-manufacturing industries is similar to that of population. In other words, a higher share of the non-manufacturing industries is observed in the Capital region where the size of population is very large.

Figure 1.

Production share (%) in 2000

4.2 A data set

The data set used in this study consists of Japan's 47 prefecture-level data for manufacturing and non-manufacturing industries from 1980 to 2002. Since there are many regional statistics, there was need to compile the data set from various sources. The data set is mainly obtained from the Japanese Annual Report on Prefectural Accounts, prepared by the Economic and Social Research Institute, the Cabinet Office.

In the data set, an output (Y) for each prefecture is measured by a nominal value-added that is adjusted by the value-added deflator reported in the System of National Accounts (SNA). Labour input (L) is an estimate of man-hours. Capital input (K) comprises the fixed capital stock adjusted by its rate of utilization. Since the fixed capital stock at a regional level is not publicly available, we use the estimated value of the capital stock for each prefecture, which is obtained from the Central Research Institute of Electric Power Industry (CRIEPI) database.5 The utilization rate in the manufacturing industries is obtained from a set of indexes regarding the operating ratios published by the Ministry of Economy, Trade and Industry (METI). However, there is no publicly available data on the utilization rate in the non-manufacturing industries. Therefore, this study has estimated it in such a manner that we estimate a deviation of the logarithm of inverse of the capital coefficient from its time trend and then uses the deviation as a proxy variable of the utilization rate.6

This study employs population density (DENS) and market access (ACC) as proxy variables of agglomeration economies. They are used as explanatory variables for inefficiency. DENS is defined by the size of population divided by each inhabited area. Meanwhile, this paper defines the following index as a proxy variable of the market accessibility (ACC):


where djkt represents an automobile travel time between regions j and k in the period t7 and Qkt is an amount of gross output by industry that represents the size of the local production market at the kth region (prefecture in this study) in the tth period. As described above, the ACC changes over time because both the automobile travel time and the gross output vary over time. In particular, a large-scale investment in road infrastructure had been implemented over the observation period (1980–2002). The investment has significantly influenced the ACC variable in such a manner that the development of road infrastructure has decreased the travel time between regions over time.

It is also important to note that a production market size (a gross amount of output) does not necessarily equal the demand size or the consumption in each region. The rationale concerning why the gross output is used as a market size is because this research is interested in examining the productive efficiency of industries from a supply side of each regional economy. Thus, the proposed definition is consistent with our research purpose that examines the industrial efficiency from each regional production. Furthermore, the production market in each region consists of the final goods' production market and the intermediate goods' production market. In this study, we use a gross amount of output that is the total sum of productions in the multiple markets.

An important characteristic of ACC is that the degree of market access is measured by a production market size, which is combined with a transportation cost (travel time) of accessing a market. If the market size of the other region (Qkt) increases, the corresponding variable increases. In contrast, if the travel time from a region to another region decreases, the market size of the region is considered to increase so that the value of ACC increases.

In addition to the two variables, this study uses the fiscal transfer ratio (TRANS) as the third variable to explain productive efficiency. TRANS is defined as the ratio of local allocation tax to general finance. Here, local allocation tax implies a public fiscal transfer from the central government to a local government. The amount of the transfer is determined by the differencebetween a fiscal expenditure and a financial income at each municipality. The local allocation tax is included into part of general finances. The general finances imply fiscal resources in Japanese politics, which the local government can use without any restrictions on the purpose of use.8 In other words, if this ratio (i.e. local allocation tax divided by general finance) is high, local governments depend more on public fiscal transfer in arbitrary resources for their intended use, such as road investments.

Japan has invested a large amount of funds in local public infrastructure during the 1980s and the 1990s. Most of such local investments depended upon the funds that the Japanese central government allocated to local governments through the financial transfer mechanism. A policy problem of the fiscal transfer was that an excessive amount of funds was allocated to low productivity regions during the public investment from the 1970s to the 1990s (Yamano and Ohkawara 2000). Unfortunately, this public investment did not produce an expected return from regions by their economic growth. Consequently, the fiscal transfer provided local governments with a systematic distortion for over-investment in their public infrastructures. Moreover, the fiscal transfer was done without any serious evaluation from cost-benefit impacts of each project. As a political byproduct, the local economy heavily depends upon funds from the central government. It is easily imagined that the public investment can save the local economy from temporarily deteriorating even though local governments do not make any serious efforts to stimulate economic growth. Under the policy environment, each local government raises the ratio (the amount of public investment divided by that of economic activity) until it realizes an inappropriate use of public funds.

To visually describe the policy problem related to the Japanese fiscal transfer, Figure 2 depicts the political relationship between the central government and local governments by comparing the share of the public-sector fixed capital formation in the gross regional product (in the horizontal axis) with the fiscal transfer per capita (in the vertical axis). The figure shows an increasing trend (or a positive correlation) between the share of the public-sector fixed capital formation in the gross regional product and the fiscal transfer per capita. In addition to Figure 2, Figure 3 visually implies a similar positive correlation between the share of the construction industry in the gross regional product (in the horizontal axis) and the fiscal transfer per capita (in the vertical axis). This figure shows that the ratio of the construction industry to economic activity is higher in regions where local governments receive a large amount of local allocation tax. As mentioned previously, Japanese local governments depend heavily on the fiscal transfer from the central government. The funds obtained from the governmental fiscal transfer have been long allocated to the construction industry that provides job opportunities, so supporting the regional economy. As a consequence of this fiscal misallocation, their local economies have unsuccessfully reformed their regional industrial structures from the construction industry. They have resulted in inefficient resource allocation and thereby they could not expand their industrial infrastructures through government funding. The remained inefficiency is a major policy problem of the regional economy in Japan.

Figure 2.

Fiscal transfers and public-sector fixed capital formation in prefectures (2000)

Figure 3.

Fiscal transfers and construction industry in prefectures (2000)

4.3 An empirical model and descriptive statistics

Using the variables discussed in Section 4.2., this study uses the following SFA model that comprises a production function and an equation to express the level of inefficiency (due to a managerial error):


where δ are parameters to be estimated. Note that if an explanatory variable improves the level of efficiency, then the parameter (δ) has a negative sign.

Table 2 summarizes the descriptive statistics of the data set on the value-added, capital and labour along with these annual growth rates for every five years (1980, 1985, 1990, 1995 and 2000) as well as that of the entire period. The table indicates two important findings. First, the annual growth rate of the value-added in the two types (manufacturing and non-manufacturing) of industries significantly increased during the 1980s. The annual average growth rate of manufacturing industries was 4.32 percent from 1980 to 1985 and 5.24 percent in the remaining other periods of the 1980s (see the third column of Table 2). The annual average growth rate of non-manufacturing industries was 2.74 percent from 1980 to 1985 and 5.30 percent in the remaining period, respectively (see the fourth column of Table 2). However, the increasing trend was reversed in Table 2. For example, during the 1990s, the annual growth rate of the manufacturing industries decreased to 0.37 percent on average from 1990 to 1995 and 1.55 percent during the remaining period. Similarly, the non-manufacturing industries decreased the annual growth rate to 1.46 percent (1990–1995) and 1.31 percent (1995–2000) on average, respectively. The annual average growth rate of the value-added in the manufacturing industries was 2.41 percent and that of the non-manufacturing industries was 2.50 percent (1980–2002). Such an up-and-down trend in the value-added was due to an economic growth in Japan in the 1980s and an economic recession in the 1990s.

Table 2.  Descriptive statistics
 Value-added (million yen)Capital input (million yen)Labour input (people)
Annual growth rate (%, 1980–1985)4.322.745.797.861.070.82
Annual growth rate (%,1985–1990)5.245.306.067.670.941.16
Annual growth rate (%,1990–1995)0.371.464.465.86−1.531.45
Annual growth rate (%, 1995–2000)1.551.312.884.07−2.04−0.06
Annual growth rate (%, 1980–2002)2.412.504.476.04−0.810.73

Second, the growth rate of employees decreased during the 1990s, resulting in a negative growth rate for the two types of industries. An exception was the non-manufacturing industries from 1990 to 1995. In contrast, the private capital stock of the two industries increased to attain a high growth rate during the 1980s and 1990s. The annual average growth rate of the manufacturing industries was 4.47 percent and that of the non-manufacturing was 6.04 percent from 1980 to 2002. The time trend of employees and capital stocks implies a structural shift of the Japanese industry from labour-intensive to capital intensive.

Paying attention to variables concerning agglomeration economies, Table 3 indicates that the annual average growth rate of the population density decreased from 0.71 percent from 1980 to 1985 to 0.00 percent from 1995 to 2000. The annual average growth rate of the market access variables for the two groups of Japanese industries decreased in the 1990s from the 1980s. However, a relatively high growth rate was maintained in the market access indexes during the observed period. For example, see the bottom of Table 3 that contains 2.37 percent and 2.23 percent for the market access of the manufacturing and non-manufacturing industries, respectively. In contrast, the growth rate of fiscal transfer ratio was negative in the 1980s, reflecting a strong economic growth in Japan. Then, it shifted to an increasing trend in the 1990s, particularly from 1995 to 2000. The increase in fiscal transfer during the 1990s is attributed to the government's policy to increase the growth of regional economy, which had been long stagnating after so-called ‘bubble economy’ at the end of the 1980s.

Table 3.  Descriptive statistics
 Population density (people per area)Market access (million yen)Fiscal transfer ratio (%)
Annual growth rate (%, 1980–1985)0.712.883.28–0.89
Annual growth rate (%, 1985–1990)0.315.325.26–0.16
Annual growth rate (%, 1990–1995)
Annual growth rate (%, 1995–2000)
Annual growth rate (%, 1980–2002)0.312.372.230.49

5 Empirical results

Table 4 summarizes the estimation results of Equation (6). To estimate the proposed model, we use Frontier Version 4.1 which is an econometric software developed by Battese and Coelli (1995) for a specific purpose of estimating SFA models.

Table 4.  Estimation results
 Manufacturing IndustriesNon-manufacturing Industries
  1. Notes: Standard deviation is listed within parentheses. The symbols ** and * indicate the significance at the 1% and 5% level, respectively.

  2. β1: Hokkaido (Hokkaido). β2β8: Tohoku (Aomori, Iwate, Miyagi, Akita, Yamagata, Fukushima, Niigata). β9β12: Kita-Kanto (Ibaragi, Tochigi, Gunma, Yamanashi). β13β16: Capital region (Saitama, Chiba, Tokyo, Kanagawa). β17β19: Hokuriku (Toyama, Ishikawa, Fukui). β20β24: Chubu (Nagano, Gifu, Shizuoka, Aichi, Mie). β25β30: Kansai (Shiga, Kyoto, Osaka, Hyogo, Nara, Wakayama). β31β35: Chugoku (Tottori, Shimane, Okayama, Hiroshima, Yamaguchi). β36β39: Shikoku (Tokushima, Kagawa, Ehime, Kochi). β40β46: Kyushu (Fukuoka, Saga, Nagasaki, Kumamoto, Oita, Miyazaki, Kagoshima). β47: Okinawa (Okinawa).

inline image0.0079(0.0004)**0.0024(0.0001)**
inline image0.6316(0.0485)**0.7259(0.1430)**
Log likelihood1179.99 1758.03 
LR test196.81 484.72 
Number of observations1081 1081 

Table 4 indicates that technology parameters related to production (αK and αL) are statistically significant in each industry. All variance parameters (inline image and inline image) are also statistically significant. Since the degree of change in productive efficiency (TE) is calculated by information that is derived from the estimates of these variance parameters, it is very important for this study to obtain the statistically significant parameter estimates in order to indicate the validity of productive efficiency measures related to the two types of industries. Almost all parameters on the technological progress (βj) are significant and their signs and magnitudes are consistent with our expectation. The result confirms the validity of assumption that the technological progress varies among prefectures. Furthermore, population density (δDENS) and market access (δACC) are the explanatory variables of productive efficiency. As shown in Table 4, the parameter estimates of those variables are negative and statistically significant. The result implies that the two variables (population density and market access) contribute to improving the efficiency of Japanese manufacturing and non-manufacturing industries. Conversely, the estimated parameter of fiscal transfer (δTRANS) is positive and statistically significant for the two groups of industries, so implying that the fiscal transfer has a negative impact on productive efficiency. The result is important because it indicates that a productive efficiency loss is generated by regions where the central government provides a large amount of fiscal transfer. Thus, the two assumptions discussed in Section 1 are valid and the results are consistent with the empirical evidence found by the previous studies on agglomeration economies.

Table 5 summarizes the average of productive efficiency measures and their changes for 11 regions during the period (1980–2002). The value for each region is obtained by calculating the arithmetic average of these efficiency measures at the level of each prefecture. The comparison among regions in Table 5 indicates that the efficiency measures (TE for manufacturing and non-manufacturing) are relatively high in the three metropolitan areas such as the Capital region (0.972, 0.947), Kansai (0.903, 0.842) and Chubu (0.904, 0.866), while the efficiency measures are relatively low in the non-metropolitan areas such as Hokkaido (0.799, 0.764), Tohoku (0.769, 0.751), Hokuriku (0.748, 0.754), Chugoku (0.786, 0.781), Shikoku (0.738, 0.739), Kyusyu (0.762, 0.726) and Okinawa (0.733, 0.755). We can easily imagine the lower productive efficiency in the non-metropolitan areas from the estimated parameters of Table 4. The non-metropolitan areas have a low population density and a relatively low market access. In particular, the areas receive a relatively high level of fiscal transfer to their general finance during the entire observed period (1980–2002), so reducing the efficiency of their resource allocations. The geographical distribution pattern of the efficiency measures corresponds to the regional structure in Japan, as documented in Table 1 and Figure 1.

Table 5.  Productive efficiency
11 regionsManufacturingNon-manufacturing
Capital region0.9720.0340.9470.030
Coefficient of variation0.1091.4830.0901.649

Paying attention to a change in the productive efficiency, this study finds three important policy implications. First, the efficiency change of non-metropolitan areas such as Hokkaido (15.4%, 5.1%) and Tohoku (10.2%, 7.5%) is larger than that of the metropolitan areas like the Capital region (3.4%, 3.0%) and Kansai (–2.9%, –0.7%) even though their average efficiency measures are lower in those of the metropolitan areas. The result implies that an efficiency change from 1980 to 2002 is different from the average efficiency. Such a difference between efficiency and its change was due to the fact that Japan constructed road networks mainly in non-metropolitan areas. Consequently, market access (ACC) could improve in the non-metropolitan areas from 1980 to 2002. Such investments in road construction were done under the government's economic policy that intended to support regional economies by creating employment opportunities in the local construction industry. Thus, the improved traffic access to the Capital region positively influenced the growth of productive efficiency in the non-metropolitan areas, but it produced the dependence on fiscal transfer as a negative influence.

Here, it is important to note that the result on productive efficiency provides a policy implication that is partly consistent with the previous studies that have claimed that highway investment increases the productivity of a region through the enhanced accessibility to a market. For example, Montolio and Solé-Ollé (2009) examined whether the impact of public investment in road infrastructures increased the TFP growth in Spanish provinces and indicated a positive influence of the road investment even though they considered the negative effect of excessive use, or congestion.

Second, the degree of average inefficiency in the non-manufacturing industries is larger than that of the manufacturing industries. For example, an efficiency loss of the latter is approximately 17 percent (= (1 − 0.827) × 100) on average, while that of the former is approximately 20 percent (= (1 − 0.799) × 100) on average. The result indicates that a significant efficiency loss occurred in the two groups of industries. Such a large difference occurs because there is a considerable efficiency difference between the metropolitan and non-metropolitan areas. For instance, the level of efficiency ranges from 0.985 (Tokyo in the metropolitan) to 0.686 (Aomori in the non-metropolitan) in the manufacturing industries. That indicates a difference of 0.299 (= 0.985–0.686) in their efficiency measures. Meanwhile, the efficiency ranges from 0.968 (Kanagawa in the metropolitan) to 0.700 (Miyazaki in the non-metropolitan) in the non-manufacturing industries. The difference between the two prefectures is 0.268 in their efficiency scores. Such an efficiency difference between the metropolitan and non-metropolitan areas is consistent with the geographical distribution of industries depicted in Figure 1.

Finally, to examine the regional differences in productive efficiency further, this study calculates the coefficient of variation (CV) as summarized at the bottom of Table 5. There is no major difference in the CV of efficiency between the two groups of industries. In contrast, there is a difference in the CV of an efficiency change (from 1980 to 2002) between them. For example, the CV of an efficiency change in the non-manufacturing industries is 1.649, while that of the manufacturing industries is 1.483 (see the bottom of Table 5). Thus, the CV of the non-manufacturing industries is larger than that of the manufacturing industries. This indicates that the regional difference in the change of productive efficiency in non-manufacturing industries is larger than that of the manufacturing industry.

Next, we pay attention to growth accounting. Tables 6 and 7 summarize the results on growth accounting. Table 6 summarizes an annual growth rate (%) of the manufacturing industries. Table 7 summarizes that of the non-manufacturing industries. The growth of the TFP and its decomposed factors are described from the fifth to the ninth columns in Tables 6 and 7.

Table 6.  Growth accounting (annual growth rate %) in 1980–2002. Manufacturing industries
11 regionsOutput growthCapital growthLabour growthGrowth of total factor productivity (TFP)
Technological progressChange in productive efficiency (TE)Others
  1. Notes: TFP is calculated by (5). Others. stands for ‘an unexplained efficiency change’. DENS, ACC and TRANS stand for population density, market access and fiscal transfer ratio, respectively.

Capital region1.621.09–1.031.630.030.64–0.07–0.67
Table 7.  Growth accounting (annual growth rate %) in 1980–2002. Non-manufacturing industries
11 regionsOutput growthCapital growthLabour growthGrowth of total factor productivity (TFP)
Technological progressChange in productive efficiency (TE)Others
  1. Notes: TFP is calculated by (5). The others in the last column stand for ‘an unexplained efficiency change’. DENS, ACC and TRANS stand for population density, market access and fiscal transfer ratio, respectively.

Capital region3.211.950.710.460.010.32–0.01–0.23

In examining Tables 6 and 7, we need to mention that the relationship between the TFP growth and the Z-variables is expressed by mean efficiency, as specified by Equation (5). Since the relationship maintains a stochastic feature, the TFP growth incorporates an error term as in Equation (5). Therefore, it may deviate from what is predicted by the explanatory variables, the Z-variables and the dummy variables. When we measure a change between two periods, the change in the stochastic part (an error term) is under a non-zero effect. Hence, we describethe missing element of an unexplained efficiency change as ‘others’ in Table 6 according to Jacob and Los (2007).9

Tables 6 and 7 provides the three important findings. First, there is obvious technological progress from the observed period (1980–2002) and the technological progress accounts for a largest portion of TFP growth. The contribution of the technological progress ranges from 2.56 percent (Shikoku) to 0.99 percent (Hokkaido), indicating a total average of 1.79 percent in the manufacturing industries (see the fifth column of Table 6). Meanwhile, the technological progress ranges from 1.12 percent (Hokuriku) to 0.09 percent (Kita-Kanto), indicating a total average of 0.50 percent in the non-manufacturing industries (see the fifth column of Table 7). The contribution of productive efficiency changes through the three factors (DENS, ACC and TRANS) to TFP growth is smaller than that of technological progress (see the right hand side of Tables 6 and 7). Meanwhile, the enhanced market access (ACC) increases the average TFP growth by 0.52 percent in the manufacturing industries and 0.34 percent in the non-manufacturing industries. Such an explicit contribution of market access (ACC) to the TFP growth is found in the metropolitan areas and other areas.

Second, the contribution of the other (DENS and TRANS) factors is relatively small in the manufacturing industries, being 0.01 percent and –0.04 percent on average, respectively. Their contributions are 0.00 percent (DENS) and –0.01 percent (TRANS) on average in the non-manufacturing industries. In contrast, the impact of the fiscal transfer is not negligible in the manufacturing industries. For instance, it ranges from –0.12 percent for Chubu to 0.01 percent for Okinawa. The impact in the metropolitan areas is larger than that of the non-metropolitan areas. The result implies that the loss of TFP growth caused by the fiscal transfer in the metropolitan areas is larger than the national average because the growth of fiscal transfer in the metropolitan areas is larger than that of the other areas during the observed period 1980–2002 (see the right hand side of Table 8 explained later).

Table 8.  Degree of dependence on local allocation tax (%)
11 regions1980198519901995200020022002/1980
Capital region7.2%5.4%3.5%6.1%11.8%10.5%1.47

Finally, the impact of population density (DENS) on the TFP growth is negligible in this study, namely, being less than 0.03 percent for all the regions. In particular, the impact is almost zero in the non-manufacturing industries10 (see the sixth column of Table 6). Furthermore, we find that the degree of positive impact of market access exceeds that of the negative impact induced by the fiscal transfer so that the productive efficiency change positively influences the TFP growth which is approximately 0.49 percent and 0.33 percent for the manufacturing and non-manufacturing industries, respectively.11

Table 8 summarizes the degree of dependence on local allocation tax. The metropolitan areas have less dependence on public fiscal transfer than the other areas. However, the growth rate of dependence is 1.47 from 1980 to 2002, which is larger than the other regions with the exception of Kansai (1.70) (see the right hand side of Table 8). The level of dependence in most of regions decreases during the period from 1980 to 1990 and increases in the remaining other period, as indicated in Table 8.

6 Conclusion and future extensions

There is an ongoing policy discussion about the national land planning reforms. The Japanese government has been facing a large amount of fiscal deficit. The government needs to implement an effective economic policy to support the regional economic growth at the level of prefectures. To examine the influence of agglomeration economies and fiscal transfer on the productive efficiency of regional industries, this study prepared two assumptions, as summarized in Section 1.

This study found two policy implications from this empirical study. One of the two implications is that the agglomeration economies positively influenced the productive efficiency and the market access contributed to the improvement in the productive efficiency for manufacturing and non-manufacturing industries. The other implication is that public fiscal transfer or local allocation tax negatively influenced the productive efficiency of Japanese industries.

In addition to the policy implications, it is important to mention that the data set used in this study comprises panel data with time series on each prefecture. The effects of regional factors on productive efficiency enhanced the total factor productivity growth over time.

The empirical evidence identified in this study showed that the enhanced market access contributed to the TFP growth through the improved productive efficiency, while the public fiscal transfer had no or negative influence on the TFP growth through the deteriorated productive efficiency. Such effects were particularly evident in the manufacturing industries. The decrease in efficiency linked to the public fiscal transfer was caused by a reduced incentive on managerial efforts in local governments and industries. Furthermore, we found that the negative impact of fiscal transfer on TFP growth in the metropolitan areas was larger than that of the non-metropolitan areas because the degree of dependence on local allocation tax in the metropolitan areas increased faster than that of the non-metropolitan areas, particularly since the late 1990s. These findings were consistent with previous studies on metropolitan areas.

For future extensions of this study, we need to consider four research agendas. First, we will have to investigate the policy influence of agglomeration economies on the economic growth in Japan. The Japanese government is expected to implement the most effective spatial policy based on a cost-benefit analysis of each economic project without any political bias. That is an important future extension of this study. Second, it is important for Japan to support industries in such a manner that they can achieve more efficient spatial patterns of firms' locations. The policy agenda needs to be explored in future. Third, it is necessary for Japan to develop a traffic system that improves inter-regional access, because an economic loss often occurs with limited market access and an inefficient transport network to metropolitan areas. Such limitations become serious policy issues for regional industries. Therefore, we need to investigate the policy issues in future. Fourth, it is necessary for us to design a more efficient system of fiscal transfer. Under the new system, managerial incentives should be given to local governments and industries to attain the growth of regional economy.

Finally, it is hoped that this study makes a contribution in the area of regional science and we look forward to seeing research extensions as discussed in this study.


  • 1

    Nihon Keizai Shinbun (Japan Economic Newspaper on February 11, 2010).

  • 2

    This Japanese problem is often discussed as soft budget constraints (Akai et al. 2003). In other nations, previous studies also demonstrated that the fiscal transfer to local governments significantly affected the technical efficiency of regional economies, particularly in the direction of reducing their level of efficiency. For example, De Borger and Kerstens (1996) examined the efficiency of municipality finance in Belgium, and found that financial dependency on an intergovernmental subsidy reduced the efficiency of municipality finance.

  • 3

    See Fried et al. (1993) for a detailed description on the measurement of productive or technical efficiency, empirical models and their estimation techniques.

  • 4

    See e.g. Färe et al. (1994), who use a nonparametric programming technique (activity analysis) to measure the efficiency, while this paper uses a parametric technique for the same purpose of a decomposition of the productivity growth.

  • 5

    In the CRIEPI database, capital stocks are constructed from the gross investment by using the perpetual inventory method. The CRIEPI database provides the fixed capital stocks classified into manufacturing and non-manufacturing industries for each prefecture. The database can be obtained from CRIEPI at URL:, upon request.

  • 6

    It is expected that variations in the logarithm of the inverse of the capital coefficient, inline image, have a constant slope over time under the production system that employs capital-using technology for a long term. However, the value fluctuates every year in an observed data set. Hence, we assume the fluctuations in inline image can be attributed to a change in a capital utilization as well as to a time trend. Based on this assumption, a proxy of the capital utilization rate can be measured by a residual error term (ε) in the regression below; ln(Y/K) =α+βT+ε, where T is a time trend and β is a time-invariant slope of inline image.

  • 7

    According to Physical Distribution Census (2000), 81.7 percent of the total shipment cost is attributed to automobile transportation, while the shares of marine, air, and rail transportations are merely 13 percent, 4.2 percent, and 1.2 percent, respectively. These statistics suggest that using the travel time of road transportation is preferable for constructing data of market access.

  • 8

    In Japan, the income of local government is divided into fiscal resources which the local government can freely use without any restrictions for the purpose of spending the money, and those constrained by a specific purpose for spending the money.

  • 9

    Jacob and Los (2007) label ‘unexplained assimilation’, which could be translated in the current context as ‘unexplained efficiency change’.

  • 10

    An exception is the Capital region, which shows contributions of 0.03 percent for the manufacturing industries; however, the total average is 0.01 percent. Hence, the effect of the population density is negligible in this study.

  • 11

    Tables 6 and 7 indicate that the ‘Other’ factor is calculated by residuals as unexplained efficiency change. The value of the ‘Other’ factor is not too small to be negligible. Thus, it implies that there are some regional differences that cannot be identified by the proposed model. Since our main research concern is to examine an impact of agglomeration economies and fiscal transfer on TFP, we do not explore the residuals further in this study.