Modelling technology in agriculture and manufacturing using cross-country panel data

Why do we observe such dramatic differences in labour productivity across countries in the macro data? This thesis argues that the growth empirics literature oversimplifies the complexity of the production process across countries and neglects data cross-section and time-series properties, leading to bias in the empirical estimates. Chapter 1 presents two general empirical frameworks for cross-country productivity analysis and demonstrates that they encompass the growth empirics literature of the past decades. We introduce our central argument of cross-country heterogeneity in the impact of observables and unobservables on output and develop this against the background of the pertinent time-series and cross-section properties of macro panel data. Chapter 2 uses data from 48 countries to estimate manufacturing production functions. We discuss standard and novel estimators, focusing on their treatment of parameter heterogeneity and data time-series and cross-section properties. We develop the Augmented Mean Group (AMG) estimator and show its similarity to the Pesaran (2006) Common Correlated Effects (CCE) approach. Our results confirm parameter heterogeneity across countries in the impact of observable inputs on output. We check the robustness of this finding and highlight its implications for empirical measures of TFP. Chapter 3 investigates the heterogeneity of agricultural production technology using data for 128 countries. We develop an extension to the CCE estimators which allows us to suggest that TFP is structured such that countries with similar agro-climatic environment are influenced by the same unobserved factors. This finding offers a possible explanation for the failure of technology-transfer from advanced countries of the temperate 'North' to developing countries of the arid/equatorial 'South'. Our Monte Carlo simulations in Chapter 4 investigate the performance of the AMG, CCE and standard (micro-)panel estimators. Failure to account for cross-section dependence is shown to result in serious distortion of the empirical estimates. We highlight scenarios in which the AMG is biased and offer simple remedies