Introduction

This chapter is about development accounting. It is widely known, and will be found again to be true here, that cross-country differences in income per worker are enormous. Development accounting uses cross-country data on output and inputs, at one point in time, to assess the relative contribution of differences in factor quantities, and differ­ences in the efficiency with which those factors are used, to these vast differences in per-worker incomes.

Conceptually, development accounting can be thought of as quantifying the relation­ship

Like growth accounting, this is a potentially useful tool. If one found that Factors are able to account for most of the differences, then development economics could focus on explaining low rates of factor accumulation. There would of course be ample scope for controversy over the policies better suited to engineering higher investment rates in various types of capital, but there would be consensus over the fact that the intermediate goal of development policy is to engineer those higher rates. Instead, should one find that Efficiency differences play a large role, then one would have to confront the addi­tional task of explaining why some countries extract more output than others from their factors of production. Experience suggests that this additional question is the hardest to crack.

The consensus view in development accounting is that Efficiency plays a very large role. A sentence commonly used to summarize the existing literature sounds something like “differences in efficiency account for at least 50% of differences in per capita in­come”.

Operationally, the key steps in development accounting are: (1) choosing a functional form for F, and (2) accurately measuring Income and Factors. Efficiency is backed out as a residual. As for the Solow residual, this residual is a “measure of our ignorance” on the causes of poverty and under-development. And, as in growth accounting, one poten­tially promising research strategy is to try to “chip away” at this residual by improving on steps (1) and (2), i.e. by looking at alternative functional forms, and by attempting a more sophisticated measurement of Income and Factors. For example, one could try to include information on quality differences in the capital stock - instead of relying exclusively on quantity.

The bulk of this chapter aims at outlining strategies for such a chipping away.^[375] It investigates the potential for different functional forms, and different ways of estimating inputs and outputs, to reduce the measure of our ignorance. Rather than reaching firm conclusions, it tries to classify ideas into more or less promising. Its contribution is to formulate sentences such as “improvements in the measurement of x are unlikely to significantly reduce the unexplained component of per-capita income differences”, or “the unexplained component is somewhat sensitive to the measurement of z, so this is a potentially fruitful area for further research”.

The experiments I perform fall in five broad categories. The first is a fairly mechanical set of robustness checks with respect to the choice of parameters in the basic model used in the literature, as well as with respect to possible measurement errors in output, labor, and years of schooling. I conclude that none of these robustness checks seriously calls into question the conclusions from the current consensus (Section 3).

Second, I consider extensions of the basic development-accounting framework aimed at improving the measurement of human capital. In most development-accounting ex­ercises differences in human capital stem exclusively from differences in the quantity of schooling. One set of extensions I consider exploits cross-country data on school resources and test scores as proxies for the quality of education, and then uses these quality indicators to augment the quantity-based measure of human capital. I find that taking into account schooling quality leads to trivially small reductions in the measure of our ignorance. Another extension replicates existing work that augments human cap­ital by a proxy of the health status of the labor force. There is some indication that this may lead to a significant reduction in the unexplained component of income, but I argue that the bulk of the variance most likely remains unexplained. All the measures of human capital considered are built on the assumption that the private return to hu­man capital accurately describes its social return. I conclude this section with a brief discussion of why and how one may want to try and relax this assumption (Section 4).

Third, I turn to the measurement of physical capital. Here I review contributions that highlight enormous cross-country variation in the composition of the stock of equipment. A simple model shows how to relate variation in capital composition to unobserved quality differences in the capital stock. How much of the responsibility for efficiency differences can be assigned to these differences in the quality of capital de­pends on parameters that are very hard to pin down, but the potential is extremely large. I therefore conclude that the composition of capital should be a key area of future re­search. I also glance at vintage-capital models, but argue that they hold little promise for development accounting, as well as at the distinction between private and public investment, which is instead potentially quite important (Section 5).

The most innovative contributions of the chapter are represented by the fourth and fifth sets of extensions. In the former I explore the role of the sectorial composition of output. The large differences in overall efficiency that are found at the aggregate level could reflect large differences in efficiency within each sector of the economy, but they could also be due to the fact that some countries have more of their inputs in intrinsically less productive sectors than others. I explore this idea by looking at an agriculture/non- agriculture decomposition (poor countries have as much as 90% of their workforce in agriculture, rich countries as little as 3%), but find that only a small fraction of the overall variation in efficiency is due to differences in sectorial composition: Efficiency differences appear to be a within industry phenomenon (Section 6).

The last set of exercises explores a radical departure from the standard framework, and finds radically different answers. In the standard framework, which relies on a Cobb-Douglas specification of the production function, efficiency differences are factor neutral: if a country uses physical capital efficiently, it also necessarily uses human cap­ital efficiently. I argue that this is a pretty restrictive assumption, and propose a simple CES generalization of the basic framework where cross-country efficiency differences are allowed to be non neutral. Stunningly, I find that, when non neutrality is allowed for, the data say that poor countries use physical capital more efficiently than rich coun­tries (while rich countries use human capital more efficiently). Furthermore, when the development-accounting exercise is performed in a context of non-neutral efficiency differences the conclusions on the contribution of these differences to cross-country income inequality become very fragile. In particular, if the elasticity of substitution between physical and human capital is low enough, observed differences in factor endowments become able to explain the bulk of the cross-country income variance.

Before plunging into the data and the calculations, it is worthwhile to stress the lim­its of development accounting. Development accounting does not uncover the ultimate reasons why some countries are much richer than others: only the proximate ones. Like growth accounting, it has nothing to say on the causes of low factor accumulation, or low levels of efficiency. Indeed, the most likely scenario is that the same ultimate causes explain both. Furthermore, it has nothing to say on the way factor accumulation and ef­ficiency influence each other, as they most probably do. Instead, it should be understood as a diagnostic tool, just as medical tests can tell one whether or not he is suffering from a certain ailment, but cannot reveal the causes of it. This does not make the test any the less useful.

2.