Making crop insurance rates more accurate

A new study by agricultural economics experts from Texas A&M AgriLife and Virginia Tech makes a strong case for using historical weather information in crop insurance programs to price policies even more accurately.

The study, “Incorporating Historical Weather Information into Crop Insurance Rating,” by Yong Liu, an agricultural economist at Texas A&M AgriLife Research and assistant professor in the Texas A&M Department of Agricultural Economics, Bryan-College Station, was recently published online at American Journal of Agricultural Economics. He is co-author with Ford Ramsey, Department of Agricultural and Applied Economics, Virginia Tech.

About crop insurance rates

Crop insurance is the most expensive agricultural policy in the U.S. with over $110 billion in liabilities in 2020. Farmers and others buy crop insurance to protect themselves against either crop loss due to natural disasters or loss of income due to a downturn in agricultural commodity prices.

In the US federal crop insurance program, a key principle in the design of crop insurance policies is that they must be actuarially fair, meaning that the expected benefit under the policy must equal the premium.

“Achieving this goal requires accurate pricing of policies, and accurate pricing depends on accurate modeling of all loss-causing variables,” Liu said.

Traditionally, he said, known or fixed historical yield data or historical loss cost data were used to estimate loss income or cost.

“For example, soil information is fixed or known at the time the policy is sold,” he said. “Incorporating this type of known information is conceptually similar to dealing with time trends and other fixed determinants of yields or loss costs.”

Liu said loss probabilities and expected losses are then used to calculate premiums. Many rating procedures use exclusively fixed or deterministic variables in determining expected losses.

“But it is widely recognized that much of the observed variation in yields and loss costs is due to changes in weather and other variables,” Liu said. “Current loss variables used in setting crop insurance rates may be amended to include other applicable variables such as weather.”

Stochastic variables, like time, have a random probability distribution or pattern that can be analyzed statistically but not accurately predicted. Unlike fixed variables, stochastic variables are unknown when the policy is sold.

“The inclusion of these variables, especially the inclusion of long-term weather data, will allow a more thorough and accurate assessment of the distribution over time,” Liu said.

The case for using historical weather information

Liu noted that in the federal crop insurance program, historical weather information is already incorporated to some extent through post-event rate adjustments. He also noted that reinsurers often use weather information when evaluating crop and risk insurance portfolios.

“The distribution of yields that is correlated with time has been shown to roughly approximate the distribution of yields based on observed yields,” he said. “And several previous studies have discussed the potential benefits of using weather or climate information in crop insurance rating.”

He also noted that weather data is often available over a longer period of time than yield data or loss and cost data.

“This is particularly true at the farm level where yield records are notoriously short-lived, in areas where production is sporadic or for crops with limited historical production,” he said.

Liu said that if weather data is useful for making predictive assumptions about yields and loss costs, then incorporating historical weather information when setting crop insurance rates should provide additional accuracy.

“Our approach uses observations where the loss variables are missing,” he said. “The inclusion of historical weather data necessarily includes observations with missing dependent variables.”

About the research

In this study, Liu and Ramsey apply a Bayesian approach to incorporate historical weather information into crop insurance ratings. The Bayesian paradigm has the advantage of reflecting uncertainty from all unknowns instead of only known information.

“We treated the cases of weather information as a stochastic predictor of both yields and loss cost ratios,” Liu said. “In the case of yields, we used county-level corn yields from seven states in the Midwest. For loss cost ratios, we used Illinois and Iowa county-level corn and soybean cost ratios for the federal crop insurance program.

The models are embedded in a Bayesian algorithm that uses historical weather information to estimate the necessary actuarial factors to determine crop insurance premiums, he said.

Liu said that in the case of mining, the study was able to demonstrate the following:

— A private insurer incorporating weather information can develop rates that give them a competitive advantage over government-set crop insurance rates.

— This advantage is enhanced when additional historical weather information is available. Using more informative data that spans a longer period will improve overall accuracy.

— This advantage is slightly stronger at lower coverage levels.

He said that in the case of loss costs, the study was able to demonstrate:

—Historical distributions of weather-related loss costs differ modestly from those without historical information.

— Time weighting can be incorporated through a streamlined one-step process.

Liu said the study makes two main contributions to the crop insurance discussion. The first was the application of a theoretically consistent Bayesian approach to incorporating historical meteorological data into the estimation of conditional forecast distributions of yield.

“Herein, we show that incorporating historical weather information leads to economic benefits for private insurers, demonstrating the efficacy of the proposed approach,” Liu said.

He said the second contribution applies the same approach to allocating loss costs.

“This includes a single algorithm for bounded loss costs, and we find that historical time-dependent distributions differ modestly from empirical distributions based on observed loss costs,” he said.

Liu said the study’s findings have implications for the design of crop insurance programs both in the U.S. and around the world.

“This study suggests that increasingly large and often disparate data sets can be combined and used to improve agricultural policy,” he said. “As measurement and modeling of weather and crop production continue to evolve, so will crop insurance products and actuarial methodologies.”

He said that by developing rates that reflect heterogeneous risk exposure across locations, the methods developed in the study could encourage increased program participation and minimize adverse selection.

Source: is AgriLife TODAY, which bears full responsibility for the information provided and is wholly owned by the source. Informa Business Media and all its subsidiaries are not responsible for any content contained in this information asset.

Leave a Comment

Your email address will not be published.