A single Honda automobile is made of 20,000 to 30,000 parts produced by hundreds of different plants and firms. The maverick vision of Henry Ford, whose dream of total self-sufficiency in the production of automobiles was embodied in the massive River Rouge plant but proved to be out of step with the course of economic history. Instead, the immense productivity gains of the past several centuries have relied on an extensive division of labour across plants which trade specialised inputs with one another in convoluted networks. Some key unanswered questions are how and why these networks of plants and flows of intermediates vary across countries, and how they are related to economic development.
An early literature, e.g., Hirschman (1958), reasoned these industry linkages were essential for economic development and focused on how to promote the formation of robust input markets in poor countries and target investment to the industries with the strongest linkages. However, before the data and methods to test these ideas became available, one-sector models that abstracted from intermediate goods altogether became the standard framework for studying growth. Recent work by Ciccone (2002), Acemoglu et al. (2007), Jones (2011) and others has shown that distortions in input markets can in principle explain a large fraction of productivity differences between countries, but this literature has remained largely theoretical. In Bartelme and Gorodnichenko (2105), we build on these recent studies and analyse the empirical relationship between linkages and aggregate productivity.
Linkages at work
In the first step, we develop a simple neoclassical framework in the spirit of Jones (2011) to link the observed input-output structure of the economy to technological constraints as well as various distortions in input and output markets. These distortions diminish the gains from using intermediate inputs, make linkages weaker, and reduce measured productivity and other key indicators of development and welfare. We use the framework to derive an econometric specification and a summary measure of distortions based on input-output tables, as well as pinpoint identification challenges and potential solutions to these challenges.
A central ingredient of the framework is the input-output table. In a massive data effort, we have constructed a novel database of input-output tables for 106 countries at different levels of development (from Uganda to US) and in different time periods (from 1950s to present). For example, our database includes such rare gems as input-output tables for Peru in 1955, Bangladesh in 1960, and Ghana in 1965. The broad time-series and cross-sectional coverage is essential for identifying the systematic relationship between linkages and development. These input-output tables come from national statistical offices and central banks, various international statistical agencies (e.g., OECD, Eurostat, United Nations), and academic/commercial initiatives (e.g., Global Trade Analysis Project, GTAP).
- We show that the strength of linkages – measured as the average output multiplier (AOM) from an input-output table – is strongly and positive related to measured output per worker and total factor productivity (see Figure 1).
Linkages are quantitatively important, a one standard deviation increase in the average output multiplier is associated with a 15-35% increase in output per worker depending on the specification, most of which stems from gains in productivity rather than accumulated factors of production. We subject this result to a battery of robustness checks. We consider additional controls and subsamples, use methods robust to outliers, exploit the panel dimension of the data, allow for nonlinear effects, and utilise alternative measures of linkages. Although there is some variation in the estimated strength of the relationship, the qualitative and quantitative results largely survive these checks. As a part of this robustness analysis, we also shed new light on why previous attempts to empirically relate linkages and productivity have been unsuccessful.
Figure 1. Average output multiplier (AOM) and output per worker
Notes: AOM is calculated as where is the Leontief inverse of the input-output (IO) matrix , is the summer vector and is the number of industries. IO data are from GTAP7.
While cross-country regressions are subject to doubts about omitted variables and measurement issues, we can evaluate our findings using a calibrated version of our model and a more structural approach to identifying distortions. We use the input-output data and two different identifying assumptions to extract industry-level distortions for each country, then compute the productivity gains associated with eliminating these distortions.
- We find that eliminating distortions would result in gains of roughly 6-10% for the median country in the sample, rising to 13-20% for countries at the 75th percentile, and higher for a number of poor countries (Figure 2).
Most of these gains come from eliminating intermediate input distortions in agriculture, with a smaller fraction coming from services and the smallest from manufacturing. These results are consistent with the fact that cross-country variation in total factor productivity (TFP) is highest in agriculture (Restuccia and Rogerson 2008, Gollin et al. 2012).
Figure 2. Gains from eliminating distortions
Notes: The gains from eliminating distortions are computed using the model and the input use patterns in the 20 richest economies to predict technology and identify distortions for middle income and poor countries. The regression line is drawn using least median squares to ensure robustness against outliers. Primary sector includes agriculture, mining, utilities and construction.
These gains are broadly in line with the quantitative relationships we found in the country-level regressions. When we regress the model-implied TFP on average output multiplier, we estimate slope coefficients similar to those we found in the data. The results indicate that the data is both qualitatively and quantitatively consistent with the hypothesis that distortions in intermediate goods account for a modest but tangible fraction of cross-country variation in productivity. This finding challenges the view that intermediate goods linkages can be neglected when studying the process of economic development. At the same time, our results do not support the view that distortions in intermediate goods markets are the primary cause of low productivity in poor countries.
In a famous example, Adam Smith illustrated gains from specialisation in a pin factory. While his focus was on gains from specialisation within a factory, further economic development proved that the scope of gains extends far beyond the boundaries of a plant or a firm. For example, just in North America, Honda has a network of more than 600 direct suppliers. The history of auto industry with the rise and fall of the Ford Rouge factory demonstrates that successful firms increasingly rely on networks of suppliers scattered all over the world. The omnipresent specialisation in modern economies is a genuine marvel which surely has gone beyond the wildest dreams of the famous Scottish economist.
While a great deal of empirical research has been done on specialisation at the very micro level (e.g., division of labour in a pin factory) and the very macro level (e.g., international trade between countries), linkages across firms and industries within a country – i.e., the middle level – have been much less studied. Most of the analysis at this middle level is theoretical and qualitative but the predictions are clear, these linkages should play an important role in economic development and are likely to be an important source of productivity gains. Having built a database of input-output tables for a broad spectrum of countries and times, we provide evidence consistent with these predictions – countries with stronger linkages have higher productivity. This relationship is quantitatively strong and robust. We also show that the empirically observed sensitivity of productivity to the strength of linkages is in line with the results from a calibrated multisector neoclassical model.
Admittedly, we cannot completely rule out potentially confounding factors in cross-country regressions or model misspecification, and hence our findings call for more research on the workings of linkages between firms and industries. Various works in economics grapple with the importance of these linkages and specialisation. However, these efforts lack a unifying framework with a macroeconomic perspective. We hope that future research will take up these challenges.
Acemoglu, D, P Antras, and E Helpman (2007), “Contracts and technology adoption,” The American Economic Review, 97 (3), 916–943.
Bartelme, D, and Y Gorodnichenko (2015), “Linkages and Economic Development”, NBER Working Paper No. 21251.
Ciccone, A (2002), “Input chains and industrialization,” Review of Economic Studies, 2002, 69 (3), 565-587.
Gollin, D, D Lagakos, and M E Waugh (2013), “The Agricultural Productivity Gap in Developing Countries”, The Quarterly Journal of Economics, December.
Hirschman, A O (1958), The strategy of economic development, Yale University Press, 1958.
Jones, C I (2011), “Intermediate Goods and Weak Links in the Theory of Economic Development,” American Economic Journal: Macroeconomics, 3 (2), 1–28.
Restuccia, D and R Rogerson (2008), “Policy distortions and aggregate productivity with heterogeneous establishments,” Review of Economic Dynamics, 11 (4), 707–720.