Saturday, February 23, 2008

Determinants of subnational growth, what do we know?

Last week Neil McCulloch, Bambang Suharnoko and Sumawah Yuningsih presented their empirical work on determinants of regional growth in Indonesia at the World Bank Office Jakarta (the paper is not yet available).

They did a standard Barro-type growth regression with growth rate on the left hand and a set of explanatory variables on the right hand, using district-level data. They did the regression in three different periods to capture the pre-crisis, crisis and post-crisis era. In addition to the standard variables like initial GDP level, share of population with secondary high school or more, telephone line density and some indicators of infrastructure, they introduced average growth of neighboring districts to capture the 'growth spillover effect.' However, because such data is not present, they were not able to add physical capital stock or investment in the regression.

A few weeks before the presentation I told one of the co-authors that I won't be surprised if they don't find anything significant in the regression. Indeed, except for initial GDP at t-zero, which indicates the convergence across districts, they found virtually nothing else was significant. Even the convergence seems to be driven by the crisis; richer regions got hit more.

This work was what Jeffrey Frankel once considered as 'zen economics' -- where your empirical work just prove... nothing. Or, in Neil's words, "We thought we knew what causes regional growth... In fact we don't."

Well, in empirical works, if nothing is significant then nothing is significant. But how do we explain such 'no finding?' Apart from some measurement error problems, I think of two things:
  1. Before 2000 provinces and districts were not independent units of economic activities. Most of decision-making made by the central government. So what matters for regional growth were a) national growth, and b) the way resources were distributed by the central government. Hence the variation of regional physical or human capital did not explain the variation of regional growth.
  2. Since decentralization has just formally started in 2000, the time frame is too short to see the effect. True, some regions perform better than the others. But perhaps that was due to non-standard variables such as leadership, or even luck. I think regions who do well are those lucky enough to have committed, visionary leaders. An article in Kompas about how Sragen became a pioneer in e-government, mainly because the leader had the vision and commitment to do that, provides an anecdotal evidence (unfortunately now Kompas doesn't provide the link to its previous editions).
I suggested my co-author friend to redo the regression ten years from now, when we have enough post-decentralization data points.


  1. I agree with you.

    What's interesting is they seem to be uninterested in the determinants of subnational growth. I think the logic is growth is increased in earnings this year per earnings last year. Infrastructure indicators may not directly increase earnings. Asking what are the factors that do not cause growth is different from asking what are factors that cause growth. Maybe their approach is asking more to the former than the latter. GE infrastructure usually goes into an environment which already have economic growth.

    I think WB have been neglecting political factors as long as it have been involved in Indonesia and other countries, however, there is an interesting and important study that would relate to your title determinants of subnational growth by Djankov called DBD. Not demam berdarah but Doing Business Index. DBD correctly assume that entrepreneurship is the reason for growth thus it measured comprehensively the institutional conditions in which business takes place both at the de jure and de facto level.
    If autonomy would be make any difference in Indonesia it would be competition between districts in applying a good institutional condition to hatch new entrepreneurs.

    Djankov, historically, try to integrate the political factors into economic analysis (Claessens, Djankov, Lang,1999) where they analyze who owns east asian corporations. They founded, in 1999, almost 60% of the market capitalization in Indonesia is owned by top 10 families. These features may have directly orindirectly impeded legal and regulatory development. I think this fact can occur in subnational level too.I wonder why Djankov is not the head of the WB.

  2. Herbi, I don't know what you meant by "they seem to be uninterested in the determinants of subnational growth." In fact, this is what their paper was about. And one of the suggestions was to look closer on what hinders growth, something like Hausmann-Rodrik growth diagnostic approach.

    You are right about infrastructure doesn't always create growth; maybe the causality goes the other way around. But in terms of econometric testing, we should see a significant correlation. But they in fact found nothing.

    About WB have been neglecting political factors -- well, the other side of the argument is WB and donors have been too political. But this is another discussion.