Comments and Discussion

Pinelopi Koujianou Goldberg:

Erdem and Tybout's paper accomplishes two goals. First, it provides a critical and thought-provoking summary of empirical studies of the effects of trade liberalization on the industrial sector in developing countries. The fact that Jim Tybout has been one of the most significant contributors in this area makes the views expressed in the paper particularly valuable. Second, the paper proposes an ambitious new approach toward evaluating trade reforms in these countries based on calibration of a dynamic industry evolution model. I comment on these two aspects of their analysis in sequence.

Empirical Findings on Industrial Sector Responses to Trade Reform

The paper's table 1 summarizes the existing findings of empirical work regarding the effects of trade reform in developing countries. Overall, these findings seem to support the import-discipline hypothesis that price-cost margins fall following trade reform while firm productivity increases. The efficiency improvements are attributed to a large extent to market share reallocations from inefficient to more efficient firms, but intrafirm productivity gains also play a significant role. Erdem and Tybout point out three main deficiencies of these studies:

• mdash The data used in empirical studies of market power and productivity suffer from serious shortcomings. The evidence based on such studies is thus potentially flawed.

• mdash Even if one took the evidence at face value, existing studies still do not tell anything about the mechanisms through which efficiency improvements are achieved.

• mdash Existing studies do not estimate the welfare effects of trade reforms. Short-run improvements in productivity and reductions of domestic market power do not necessarily translate to welfare improvements in the long run.

Let me start by unequivocally expressing agreement with the third point mentioned above. The authors' paper and subsequent simulation of the dynamic model make clear that the enthusiasm with which results of productivity studies are often received by trade reform proponents is a bit premature. Quality downgrading, high firm and job turnover, and loss in producer surplus can potentially undo the welfare benefits of increased productivity in the short run. And this is still abstracting from general equilibrium effects of trade liberalization on factor prices and potential increases in inequality. The absence of a comprehensive welfare evaluation of trade policies is not so much a problem for the existing empirical studies themselves, which, in an effort to provide a rigorous and convincing empirical analysis, have focused on particular welfare components rather than general welfare analysis as it is for the way the results of these studies are (mis)used. The authors suggest caution in the general conclusions drawn from available empirical results.

In contrast, the first and second points of the critique listed above deal specifically with the data and econometric methodology employed by existing empirical studies. Inference on productivity is usually based on estimating versions of the following production function:

where q_it is output, x_it is inputs, and θ_it is productivity (unobserved by the econometrician, but observed by the firm).

All variables denoted with small letters are in logs. Using various, often very sophisticated econometric methods, researchers estimate θ_it. In a second step, they then relate θ_it to measures of trade liberalization. As the authors point out, the usual finding is that trade liberalization increases productivity.

The main problem with the above studies is that neither output q nor physical inputs x are actually observed in the data. Instead, we observe sales (p * q), and input expenditures (w * x), where p denotes the product price and w stands for factor prices. The traditional approach employed in the empirical literature is to deflate sales by industry-specific price deflators and pretend that output is observed. Similarly, on the input side expenditures are deflated by input deflators, and the resulting variables are treated as physical inputs. This approach works well when inputs and outputs are homogenous goods, an assumption that is rarely valid in practice. In the presence of product differentiation and market power, however, the resulting productivity estimates are problematic.

To understand the nature of the problem, let us abstract from potential measurement problems in input markets for the moment...