This paper evaluates the impact of major natural resource discoveries since 1950 on GDP per capita. Using panel fixed-effects estimation and resource discoveries in countries that were not previously resource-rich as a plausibly exogenous source of variation, I find a positive effect on GDP per capita levels following resource exploitation that persists in the long term. Results vary significantly between OECD and non-OECD treatment countries, with effects concentrated within the non-OECD group. I further test GDP effects with synthetic control analysis on each individual treated country, yielding results consistent with the average effects found with the fixed-effects model.
Gregory Clark argued in A Farewell to Alms that preindustrial societies, including England, were Malthusian. Day wages show incomes were trendless: as high in Europe in the medieval era as in 1800, even in England. The opposed view is that England and the Netherlands grew substantially from 1200 to 1800. Early day wages overestimate living standards. Here we show that preindustrial farm employment shares can be estimated from probate occupation reports. These imply only 60 percent employed in farming in England in 1560–1579 and 1653–1660, consistent with the high incomes indicated by wages. Day wages do measure preindustrial living standards.
Over the past decade, the production of tight oil and shale gas significantly increased in the United States. This paper uniquely examines how this energy boom has affected regional crime rates throughout the United States. There is evidence that, as a result of the ongoing shale-energy boom, shale-rich counties experienced faster growth in rates of both property and violent crimes including rape, assault, murder, robbery, burglary, larceny and grand-theft auto. These results are particularly robust for rates of assault, and less so for other types of crimes. Examining the migratory behavior of convicted sex offenders indicates that boomtowns disproportionately attract convicted felons. Policy makers should anticipate these effects and invest in public infrastructure accordingly.
How are oil revenues shared throughout society? We combine high-resolution geo-coded data on night-time lights and population to construct global measures of rural poverty from 2000-2013. We find that oil booms, due either to high prices or new discoveries, increase light intensity and GDP. However, the increase in output is limited to cities and towns, with no evidence that it benefits the rural poor. We also find that while urbanization is occurring throughout the developing world, there is no evidence that it is hastened by oil wealth.
This paper examines the impact of the oil price boom in the 1970s and the subsequent bust on non-oil economic activity in oil-dependent countries. During the boom, manufacturing exports and value added increased significantly relative to non-oil dependent countries, along with wages, employment, and capital formation. These measures decreased, though to a lesser and more gradual extent, during the bust and subsequent period of low prices, displaying a positive relationship with oil prices. However, exports of agricultural products sharply decreased during the boom. Imports of all types of goods displayed strong pro-cyclicality with respect to oil prices. The results suggest that increased local demand and investment spillovers induced by the oil revenue windfall resulted in increased manufacturing activity.
"Diagnostics and Corrections for Synthetic Controls When Near the Edge of the Convex Hull", with Joseph Cummins, Douglas Miller and David Simon
Synthetic control (SC) methods are becoming increasingly popular among applied researchers across a number of fields. However, researchers implementing SC currently lack tools for assessing the suitability of their data to the use of SC. We focus on a fundamental assumption in SC models – that treated unit covariates and trends are within the “convex hull” of the control unit values. We show that large time trends or covariate values in the treatment group (relative to control group) can generate a substantial bias in SC treatment effect estimates even when pre-period treatment group outcomes fall within the convex hull of control group outcomes. We describe the problem and provide practitioners with empirically implementable diagnostics for detecting when trends and covariates are likely to cause bias. We then describe two bias correction procedures that can reduce the bias when treatment group predictor values or trends are large relative to those in the control group.