Performing Department
Agricultural & Applied Economics
Non Technical Summary
Summary and Justification:Problem Statement: Agriculture is subject to a number of risks including weather, pests, and other perils. Farmers cope with risky production environments through a variety of mechanisms. In Virginia, and throughout the United States, agricultural risk management is now characterized by high rates of off-farm work, changes in contracting practices, and the prevalence of federally subsidized crop insurance. Because of the role of the government in providing a farm safety net of last resort, accurate analysis of agricultural risks has significant impacts on both government expenditures and the continued economic viability of rural communities. Design of effective risk management programs is further complicated by the diversity of agriculture in Virginia and the Mid-Atlantic. The federal crop insurance program is the most expensive instrument of agricultural policy (aside from nutrition programs) in the United States. Parameters for the insurance policies sold under this program are set by the United States Department of Agriculture, while the policies are administered and marketed by private insurers. The federal government subsidizes insurance premiums, pays a portion of the operating costs for qualified private insurers, and maintains a standardized reinsurance agreement. In 2016, the program had over 100 billion dollars in total liability, of which nearly 500 million dollars was derived from Virginia farms. Farmers in Virginia received over 50 million dollars in indemnities. Roughly 40 million dollars of the total premium of 60 million dollars in Virginia was subsidized. Because of the large amount of funding being directed toward the federal crop insurance program, policymakers, auditors, actuaries, and economists, have become increasingly interested in statistical and econometric methods that quantify agricultural risks. This urgency partly stems from the position of federal crop insurance as a replacement for other commodity support programs. Recent budget proposals have included potential reductions in funding for the program. Government agencies have a legislative responsibility to ensure that the prices for federal crop insurance are as actuarially fair as possible. If policies are actuarially fair, the price of the insurance is equal to the expected loss. When insurance is mispriced, there may be low uptake from farmers and losses may be high; this is commonly referred to as adverse selection. Because crop insurance policies involve losses as measured in terms of yields and prices, accurate pricing of insurance demands proper measurement of the variability of yields and prices. Relevance to Virginia and the region: The prevalence of the crop insurance program places some emphasis on its interaction with other aspects of agricultural production. A large, subsidized program has impacts beyond simply indemnifying a producer when a loss occurs. Production practices may be complementary to crop insurance; when crop insurance is purchased, the adoption of the complementary production practice also tends to occur. Characterizing relationships between different production practices allows policymakers to better assess the impact of policy changes and examine how crop insurance and policy affect farm structure. In Virginia, the majority of liability in the crop insurance program rests with growers of corn and soybeans. Recent expansion of the program, in the sense of providing policies for other crops and smaller, diverse farms, has caused increased uptake of crop insurance in Virginia. Policymakers from the state now have two competing objectives at the federal level: support for constituents in agriculture and stewardship of tax dollars. The Congressional Budget Office has suggested significant budget savings if the crop insurance program is scaled back. Improved actuarial performance of the program would allay fiscal concerns while preserving crop insurance as a risk management option for Virginia producers. Better statistical methods would lead to improved actuarial performance and decreased use of subsidies. Virginia has a diverse agricultural sector; it ranks in the top ten in production of a number of commodities such as leaf tobacco, apples, peanuts, tomatoes, cotton, turkeys, and broilers. Crop insurance is available for almost all of these crops in one form or another. However, the use of insurance has been mostly limited to major row crops. This suggests that growers of other crops are either unwilling to pay for insurance risk protection, unaware of the existence of federal crop insurance, or that transaction costs are too high. It is difficult to price revenue insurance policies for fresh produce, for instance, because these products do not have readily available prices from a futures market. By considering the diversity of agriculture in the Mid-Atlantic, this project will seek to better understand risks and risk management for the full suite of crops grown in Virginia. Approach: The overarching goal of this project is to provide more accurate and effective statistical methods for characterizing the risks facing agricultural producers. The data used in this project is collected by public agencies and will be obtained through cooperation with the USDA Economic Research Service. Different econometric approaches will allow for significant factors affecting agricultural production risk to be identified and then incorporated in the pricing of insurance policies.Anticipated outcomes and impacts: By providing these methods, this project will contribute to more exact pricing in the federal crop insurance program, align policies with the needs of agricultural producers in Virginia, and improve the resilience of farm communities. Project results will contribute to general scientific advances in risk modeling and agricultural economics.
Animal Health Component
30%
Research Effort Categories
Basic
60%
Applied
30%
Developmental
10%
Goals / Objectives
Major Goals and Objectives:This project has three major goals:1. To provide insight into the behavior of yields, prices, and their relationship through the application and development of novel statistical methods.2. To determine the effect of the federal crop insurance program on farm structure and production practices in the United States.3. To use the results of the project to inform policymakers, academics, and stakeholders in identifying actuarially sound risk management approaches.All three goals will be approached with the following project objectives, which are targeted for a five year project duration. The goals will be accomplished holistically through the realization of six objectives:1. Implement a copula model for the relationship between prices and yields that can be applied to data nationwide (Goal 1).2. Apply a Bayesian spatial quantile regression model for the effects of weather and technological change on the distribution of crop yields (Goal 1).3. Develop a formal model of farm production with subsidized insurance and determine the effect of federal crop insurance on farm structure and organization (Goal 2).4. Extend spatial models of crop yields and atmospheric conditions to capture directional correlation and assess the impact of directional effects on the construction of sustainable insurance pools (Goal 1).5. Obtain extensive weather data and use regularization techniques to determine the best variables for inclusion in actuarial models (Goal 1).6. Communicate topic and project-associated information to stakeholders through various approaches including online outlets, in-person Extension programming, and published Extension and scientific publications (Goal 3).
Project Methods
Methods/Procedures:Data Sources: Much of the data used in this project is freely available from government sources. Yield data is available from the National Agriculture Statistics Service, while price data will be purchased from the Commodity Research Bureau. Access to farm level financial data will be obtained from the Economic Research Service of USDA within the first year of the project. In order to access controlled data at USDA, personnel on this project will be embedded at USDA's Economic Research Service under a visiting scholar designation.The variety and scale of weather data will be crucial for this project as agriculture is affected by temperature, rainfall, evapotranspiration, residual soil moisture, humidity, and a number of other factors. In Virginia, the variety of geography causes differences in weather and climate across relatively short distances. Detailed weather data will be obtained from climate centers associated with NOAA. The climate centers relevant for Virginia (Southeast and Northeast) have data from local land-based stations, satellites, and radar. These will be accessed and data will be scaled appropriately for the question at hand. Methods/Procedures: The procedures used in accomplishing project objectives are statistical and actuarial in nature. The development of a model assessing technological change in crop yields, as well as the effects of covariates on yields, will follow developments from Reich, Fuentes, and Dunson 2011, Reich 2012, and Reich and Smith 2013. These studies implement a Bayesian spatial quantile regression that, with modification, is capable of addressing all salient features of crop yields. The conditional distribution of yields is able to vary freely through time and inference is easily conducted in the Bayesian framework. Because the number of weather variables that may affect yields is particularly large - especially if one includes interactions - this project will use regularization techniques to identify the most important covariates for predicting crop yields. Because we are interested in predicting more than just the mean effect, these techniques (LASSO, ALASSO, Early Stopping, etc.) will be applied in quantile regression and density estimation settings (Tibshirani 1996, Zou 2006). Quantile regression would provide a potential approach for weather weighting in crop insurance. This allows for longer time series of weather data to be married to shorter time series of crop yields. Spatial models typically assume that the strength of spatial correlation is the same in all directions. However, models have been developed that allow for directional spatial correlation to be captured (Kyung and Ghosh 2009). Directional differences may matter in crop insurance markets because weather in the United States tends to travel from west to east with the prevailing winds. Directional conditional autoregressive and directional simultaneous autoregressive models will be used to determine if the strength of spatial correlation varies by direction and the implications of this variation for the design of insurance pools. These directional models also allow for more detailed descriptions of dependence that may be more accurate in areas like Virginia where microclimates are often observed. The natural hedge is a relationship between prices and yields. With data on yields in different locations, and a single price, this becomes a spatial relationship as well. Characterizations of the natural hedge will be obtained for all corn producing counties in the United States. Copula models will be used to derive insurance rates for policies in the federal crop insurance program. The effect of relaxing actuarial assumptions with respect to the natural hedge can then be obtained. With the idea that more efficient insurance rates can be obtained by combining information from different locations, Bayesian model averaging will be used to combine copula models from different counties (Hoeting et al. 1999). In all of these cases, the goal is to be as general as possible in specifying the underlying statistical model. No parameters are pre-defined, but are instead estimated directly from the data. Assumptions are minimized. Because the program is federal, this project will target all counties in the U.S. for which there is an adequate history of corn and soybean yields. Different models can be compared through different methods including fit statistics and cross-validation. Cross-validation withholds a portion of the data, estimates the model, and then determines how far model predictions are from the withheld data. By doing this step many times, a best model can be identified. There are many available procedures for the assessment of structural changes in agricultural production as a result of subsidized crop insurance. These include instrumental variables, regression discontinuity designs, and other regression based approaches. A theoretical approach is to view the choice of production practices as one of organizational design (Arora and Gambardella 1996, Athey and Stern 2003, Miravete and Pernias 2006). Statistical inference for organizational design covers many different areas, and no statistical approach can be determined without first determining data availability. In general, there is an expectation that the project will use methods aimed at causal inference: instrumental variables, regression discontinuity designs, and structural approaches. Access to the Agricultural Resource Management Survey (ARMS) and Agricultural Census at ERS will be necessary for completion of this portion of the project. The impact and efficacy of the project, in terms of communicating results, will be assessed through several means. The impact of research papers can be obtained from formal citation indices such as the Social Sciences Citation Index - SSCI and Research Papers in Economics - RePEc author rankings. Citations and references in policy works will also be measured although, this impact has typically been more difficult to quantify. When project results are presented in Extension settings, interest in the results and changes in audience knowledge can be measured using survey methods. This project will collect and record these quantifiable measures of impact. Potential Difficulties and Limitations: The most significant potential difficulties associated with this project involve the acquisition of farm-level data. Although public agencies collect farm-level data it is not always open to outside researchers. County-level data is readily available and is sufficient for meeting many of the project objectives. This provides a fallback in the case that micro data cannot be obtained from NASS and ERS in a timely manner. Another alternative is to conduct a smaller farm survey in Virginia. While it would be difficult to generalize the results of such a survey to the national level, it would still be possible to statistically test the structural and organizational implications of the federal crop insurance program.