Stepan Gordeev

We study the costs of misallocation of inputs between multi-product firms that endogenously choose among heterogeneous products. Misallocation of inputs between firms has been shown to be a significant drag on aggregate productivity: it is especially severe between farms in the agricultural sectors of low-income economies. Existing estimates of its costs have relied on models of single-product firms using a single aggregate production function. Using rich farm-crop-level data from India, we estimate product-level production functions and find that they are meaningfully different from one another and from the aggregate one. We build a general equilibrium model of firm-level misallocation in which multi-product firms (or farms) are able to choose the set and mix of heterogeneous products. Misinterpreting product heterogeneity as evidence of distortions and missing the endogenous product choice response to real distortions biases single-product models to overstate misallocation, while ignoring returns-to-scale heterogeneity and within-firm productivity dispersion biases them to understate it. On net, the single-product model understates the aggregate productivity cost of misallocation between Indian farms by 28%.

In many of the poorest countries, agriculture is unproductive and subsistence farming is widespread. I propose nutrition demand as a mechanism that drives the production decisions of subsistence farmers and ultimately contributes to low aggregate agricultural productivity. I explore this mechanism in a model of farm-operating households facing explicit caloric needs and costly domestic trade, and test the model’s predictions on Malawian household-level data. In the model and in the data, the smallest farmers focus their consumption on obtaining calories and specialize their production in unsold staple crops; medium farmers diversify both their diet and their subsistence production; the largest farmers shift consumption to purchased goods by producing and selling marketable farm products. I quantify the aggregate implications of this farm-level product choice using the model. It suggests that lowering trade frictions enough for the average share of output sold by farmers to reach even 50% would make the country’s agricultural sector 42% more productive. Half of this increase is caused by the mechanically reduced erosion of output, and the other half by a better alignment of individual farmers’ product choice with their comparative advantage rather than their family’s nutritional needs or food preferences.

The role of intangible capital in production is growing relative to conventional capital. This paper considers the implications of this shift on the misallocation of inputs across public US firms. I show that ignoring intangibles leads to an overestimation of misallocation costs by 54%. The degree of this overestimation gets worse over time, which explains most of the measured deterioration in allocative efficiency in the US in recent years. I find that misallocation is almost twice as severe in sectors that use comparatively more intangibles as in sectors relying more on tangible capital. I calibrate a variable markup model in which the outcome of intangible investments is uncertain and markups increase with firm productivity. I find that it can generate a significant portion of the measured misallocation.

We study whether heterogeneity in the intrinsic features of production inputs may generate heterogeneity in production technologies optimally chosen by firms, leading existing estimates of misallocation to overstate its costs. Existing estimates of the severity of the misallocation of inputs across firms rely on assuming homogeneous production technology and interpreting deviations from that technology as evidence of misallocative distortions. We use a state-of-the-art clustering algorithm for ordinal data to group Indian agricultural plots into land types by the intrinsic physical features of each plot. We find that production functions are significantly heterogeneous across land types, which we confirm with placebo-like randomization inference. Some types of land are better suited to land-intensive technology, others to labor-intensive technology, etc. We build a model in which heterogeneous farmers face distortions and choose the type of land to rent. We use the model to quantify the cost of misallocation for India’s aggregate agricultural productivity and compare it to conventional models that assume homogeneous production technology.

Preferential Trade Agreements (PTAs) have emerged as the dominant form of international trade governance. Provisions included in PTAs are increasingly numerous, broad in their purview, deep in their scope, and varied between agreements. We study the economic, political, and geographic determinants of PTA design differences. For each of the hundreds of classified PTA provisions, we consider hundreds of country-pair characteristics as potential determinants, covering many individual mechanisms the literature has studied. To deal with this high-dimensionality and complexity, we employ Random Forests, a supervised machine learning technique. We use a robust variable importance measure to identify the most important determinants of the inclusion of each PTA provision. Contagion due to competition for export markets, geographic proximity, and governance quality emerge as the most important determinants of PTA design.

This paper applies machine learning methods to identify provisions in preferential trade agreements (PTAs) that enhance agricultural trade and to determine the socio-economic and political factors influencing their inclusion. Using a theory-consistent three-way gravity model, we employ Lasso regularized regression to identify PTA provisions impacting agricultural trade. The Lasso estimates highlight the relevance of competition policies, export taxes, intellectual property rights, capital movement, state enterprises, and technical trade barriers. Next, we employ Random Forests to determine the economic, political, social, and geographic factors driving the inclusion of these PTA provisions. The results emphasize the importance of contagion, governance quality, energy use, and geographic proximity.

Economic and non-economic transactions in Indian society are often influenced by caste. The trade of goods and services is more likely to occur within the same caste than across different castes. Such restrictive transactions generate inefficiency in the local market and affect the economic outcomes of households and the overall welfare of society. This study explores the prevalence and consequences of these identity-based transactional patterns, particularly within the agricultural sector. Leveraging a rich dataset comprising detailed transaction-level information from India, we rigorously quantify the misallocative effects of caste-related friction. First, we document widespread caste-based transactions in the local market. Then, to assess the malleability of these caste-related frictions, we utilize exogenous economic shocks as natural experiments within local markets. This allows us to investigate the potential for reducing or permeating caste-based barriers to trade, shedding light on the adaptability of identity to external economic shocks.