Yujun Zhou is a Ph.D. student in Agricultural & Consumer Economics at the University of Illinois at Urbana-Champaign, under the supervision of Professor Kathy Baylis. His research focuses on agricultural policy and food security in Sub-Saharan African countries. He is proficient in causal inference and machine learning and makes use of the best tools in both worlds.
Ph.D. in Agricultural and Consumer Economics, 2020
University of Illinois, Urbana-Champaign
M.S. in Applied Economics, 2014
University of California, Davis
B.A. in Economics, 2013
Hunger is on the rise throughout Africa, with famine threatening millions across several countries. Rapid and accurate identification of food insecurity crises can enable humanitarian responses to mitigate casualties from hunger and save lives. We develop a predictive model of food security based on readily available, spatially granular data on prices, geography, and demographics. Using machine learning techniques, we are able to improve the accuracy of predicting those villages that face a potential threat of hunger. As with any rare event, one challenge with predicting food insecurity is the low rate of severe food insecurity in the baseline data. We use several different approaches to address this imbalance to allow us to capture a higher fraction of these rare events. We apply our procedure to three sub-Saharan African
The growth in available geospatial data, along with the rise of machine learning methods, have let themselves to numerous spatial-temporal forecasting applications to solve real-world problems such as deforestation, pollution, and food security. Choosing the right performance evaluation matters for generating accurate and trustworthy out-of-sample predictions. However, with spatial-temporal dependencies between observations in both the training and testing data, the independence assumption of the testing set is violated. As a result, model performance evaluated using cross-validation (CV), and out-of-sample (OOS) can be over-optimistic. In this study, we show the changes in CV and OOS performance when we adjust for different types of spatiotemporal correlations in both simulated data and real-world panel data. We also show how the model selection is affected by the performance evaluation process to prefer overfitting models. Lastly, we propose and compare solutions such as blocking and clustering to improve performance evaluation procedures in both simulated and real-world data with spatiotemporal structures.
Globally, over 800 million people are food insecure. Current methods for identifying food insecurity crises are not based on statistical models and fail to systematically incorporate readily available data on prices, weather, and demographics. As a result, policymakers cannot rapidly identify food insecure populations, hampering responses to mitigate hunger. We develop a replicable, near real-time model incorporating spatially and temporally granular market data, remotely-sensed rainfall and geographic data, and demographic characteristics. We train the model on 2010-2011 data from Malawi and forecast 2013 food security. Our model correctly identifies the food security status of 77% of the most food insecure village clusters in 2013 while the prevailing approach fails to correctly classify any of these village clusters. Our results show the power of modeling food insecurity to provide early warning and suggest model-driven approaches could dramatically improve food insecurity responses.
Many countries, particularly in the developing world, use public stockholding programs to stabilize price for both farmers and consumers. Governments directly purchase and store staple grains, and then sell them to processors or consumers often at a substantial subsidy. Despite the substantial costs of these stockholding programs, little is known about their effectiveness in mitigating the retail price swings. This paper estimates the effects of the purchase and sales activities of the Zambian Food Reserve Agency (FRA) on maize market prices across more than thirty markets in Zambia using monthly price data from 2003 to 2008. To deal with the endogeneity in purchases and sales, we use predicted FRA purchase and sales targets as instrumental variables. Controlling for other policies in place, we find evidence that FRA activities stabilize retail prices in the major district markets within the cropping year. Results show that FRA purchases raise local prices for surplus maize producers for about 5% on average during the time of harvest and FRA sales help to lower the price during the lean season up to 7%. On the other hand, we are only able to find evidence of the FRA reducing price volatility between years in a few district markets.