Essays on Forecasting and International Trade in Grain Markets

Abstract

Essay 1 – Forecasting Long-term Grain Supply of Low- and Middle-Income Countries Forecasting models are powerful tools for predicting food supply trajectories, helping policymakers design strategies to improve food security. This study aims to enhance the accuracy of long-term grain supply forecasts by incorporating country-specific characteristics and evaluating alternative modeling techniques. Using a dataset spanning 1980 to 2021 for 78 countries, we evaluate various models to forecast aggregate grain production in each country. To assess model forecast accuracy, we employed a time-series cross-validation approach. Our analysis reveals that Autoregressive Integrated Moving Average models with exogenous variables, country-specific coefficients, linear trends, and weather variables significantly improve forecast accuracy. Our preferred model achieved a mean absolute error of approximately 10% of average production. While there is room for further improvement, our approach represents a substantial advancement over existing methods used by USDA reports. Essay 2 – In-Season US Corn Acreage Forecasting Using Machine Learning Estimates of US corn acreage planted estimation are released in two reports. First is the “Prospective Plantings Report,” released in March, and the “Acreage Report,” published in June. The acreage values from these two surveys are later incorporated into the WASDE monthly reports. Most of these reports rely on statistical survey methods to gather data directly from the farmers, with information released on established dates throughout the year. Our study aims to develop machine learning models to deliver accurate and timely updates for in-season corn acreage forecasts. Our methodology employs a dataset from 1995 to 2020 with 92 variables on markets, weather, and field conditions to assess if publicly available data up to May can provide additional information to predict acreage allocation. The results reveal that we improve the accuracy level to forecast acreage planted. The RF model yields a Mean Absolute Error (MAE) of 33,440 acres, which is lower than the 88,744-acre MAE generated from USDA’s Acreage Report estimates. Also, our findings demonstrate the significant predictive value added by incorporating the information of USDA’s Prospective Plantings Report estimates into models. This indicates modeling complexity alone cannot compensate for the unique insights embedded in farmer survey data. Our study offers a valuable tool to generate a forecast of acreage planted that complements the information provided by the WASDE reports. Essay 3- Effects of Non-Tariff Trade Barriers in Rice Markets: The Case of Rice Export Bans Imposed by India In 2024, as a consequence of export market restrictions imposed by major rice exporters, the FAO’s price index of rice reached its highest nominal level in 16 years. When international grain market prices surge, national governments frequently intervene to minimize the impact on their domestic food markets. India, the world’s major rice exporter, implemented an export ban on broken rice in 2022 and on non-basmati rice in 2023. This paper investigates the effects of the rice export bans imposed by India on the trade flows in the international market. We exploit the differences in export quantities and values on the two major types of rice . Our results indicate that the broken rice export ban had a larger impact on the international broken rice markets since India had a major market share of exports, and the increased exports from other countries were not statistically significant. Also, the results showed that the milled rice export ban had a smaller negative impact because the increase in exports from other countries was larger and statistically significant.