There are large discrepancies between different panel datasets of cross-country estimates of gross domestic product (GDP). But which set of numbers is most accurate? In the first essay, I develop a general statistical framework for answering this question. I derive two estimators for assessing the relative accuracy of GDP data, and perform Monte Carlo simulations to compare them against existing approaches. Both estimators show that a new panel dataset of GDP from the University of Queensland (UQICD) is generally more accurate than the World Development Indicators (WDI) and the Penn World Tables (PWT), the two most widely-used datasets for growth economics research and policy.
In the second essay, I characterize the differences between several versions of WDI, and PWT, and UQICD. I find that these datasets disagree on the magnitude and direction of growth in a large number of countries. The differences appear to be driven by country geography, economic size and frequency of data collection for purchasing power parity estimates. I quantify the policy impact of GDP differences by simulating counterfactual predictions of World Bank aid allocation and find that development assistance can vary substantially due to changes in underlying income measures. Nevertheless, for the vast majority of the cases, aid allocation is relatively inelastic with respect to changes in GDP data.
The third essay empirically tests Tobin's Q-theory of investment, a foundational model for modern macroeconomics and growth theory: it is shown that extant measures of average Q suffer from measurement error and a new measure of Q is proposed which adds close to 15 percentage points of explanatory power, more than twice as much as existing measures of Q. From theory, a cash flow-based measure of Q is derived which performs as well as extant Q. Lastly, an empirical measure for marginal Q is derived, but its explanatory power is found to be weak. These results provide support for the neoclassical model of firm investment.