This dissertation consists of three separate papers. In the first paper I ask how the influence of global and local forces on Mexican maize prices changed during a period when markets became more open to inter and intra-national trade. I find that the influence of global forces do vary over the study period, and counter to expectation, are highest at the beginning and middle of the period rather than the end. In contrast, the influence of local forces follows expectation and decreases over time. However the effect sizes diminish in a model that accounts for unobserved state-year variation, which could include policy interventions designed specifically to cushion against local influences. Overall, the results suggest that opening agricultural markets can result in regionally distinct outcomes and that government price supports have a stronger impact on the influence of local forces than they do on global forces.

In the second paper I evaluate characteristic based clustering (CBC) as a tool to both facilitate forecasts of cross-market grain price data and conduct exploratory analysis of price behavior dissimilarities across multiple markets. Characteristic based clustering refers to the practice of identifying similarities in multiple time series where the distances among the time series are a function of the structural characteristics in the data. I conduct a simulation experiment to determine if the characteristic clustering approach can be used to improve the accuracy or computational efficiency of grain price forecasts. I find that forecasts from characteristic based clusters of time series are sometimes as accurate as forecasts from individual time series and can be completed in 60-80% of the computational time. In the second half of the paper I use characteristic based clustering to explore the similarity of price behavior among Kenyan maize markets. I find that price behavior in remote markets has become increasingly dissimilar from markets in other Kenyan cities, and that these dissimilarities cannot be explained solely by geographic distance.

The final paper focuses on the appropriate method of inference when modeling balanced spatial panel data with unobserved time varying spatial correlation. I use simulation experiments to compare the performance of estimation techniques that use prespecified spatial weights matrices and those that do not. The results suggest that the pattern of time varying correlation does have some impact on the choice of method, but not as much as the spatial weights misspecification literature would suggest. I also find that choosing the appropriate inferential method is less of a concern if the data generating process follows a hub and spoke correlation structure. Finally, I confirm earlier results that the cluster robust modifications proposed by Bester, Conley, and Hansen (2011) perform well as long as group sizes are chosen appropriately.