Research

Working Papers

1- Public Insurance, Incentives, and Climate Risk: Evidence from U.S. Agriculture  (job market paper)

This paper examines how public subsidies for crop insurance affect the sensitivity of agricultural yields to extreme weather, highlighting the trade-offs inherent in social insurance design. Using a novel instrumental variable based on the internal rating parameters of the U.S. Federal Crop Insurance Program, we provide causal evidence that a one-percentage-point increase in insurance enrollment intensity reduces average yields by 6.2 percent, with significantly larger effects under extreme heat. At the 95th percentile of heat exposure, yield losses rise to nearly 8 percent. These f indings suggest that while premium subsidies stabilize farmer income in the short run, they gener ate moral hazard by discouraging costly adaptive investments, thereby increasing vulnerability to climate extremes. By situating crop insurance within the broader framework of social insurance, our results underscore a central Pigouvian concern: subsidies designed to provide risk protection may simultaneously distort production behavior and amplify exposure to aggregate climate risk. 

Papers

1- Inclusive Growth Diagnosis of Iran, 2018 (in Farsi).

2- An Analysis of Access of Iranian Households to Micro Finance Using Data Mining, 2018 (in Farsi)

3- Evaluation of Universal Health Insurance in Iran, 2018 (in Farsi)

4- An Analysis of Unequal Access of Social Groups to the Benefits of Economic Growth in Iran, 2018 (in Farsi)

Instrumental Variables

Literature Review of Federal Crop Insurance

1- Understanding Crop Insurance

Section 2 of Chapter 3 in "Rate Methodology Handbook Actual Production History (APH)" (2009) is a great resource for understanding crop insurance and theories behind its pricing. Pages 205- 210 in chapter 12 of  Gardner and Kramer (1986) covers all alternatives proposed to improve the Federal Crop Insurance program up to the point of publication.

2- Policy Shocks

Table A-1 of Appendix to the Rosch (2021) paper, "Federal Crop Insurance: A Premier" provides an overview for the selected legislation that have affected the Federal Crop Insurance program. Other policy shocks are as following:

• 1947 Amendments

The FCIC reduced its operations from 2500 to 375 counties in 1947 (Gardner and Kramer, 1986, page 199). 

• Experimental program in 1948 for dry edible beans in four widely separated counties with different types of farming.

 Gardner and Kramer, 1986, page 199

• Experimental program in 1948  of multiple crop contract, indemnities based on their combined coverage

 Gardner and Kramer, 1986, page 199

• Acreage Reduction Program (1950 - 1996)

1- Yu, Smith and Sumner (2017) used this policy for the sensitivity analysis. 

• Removal of fourteen counties from FCIP in 1956

Beginning in 1956, FCIP announced insurance would no longer be sold in fourteen counties in Colorado, New Mexico, and Texas. These were considered high risk farming areas not suitable for insurance, because total indemnities had substantially exceed total premiums. If insurance had not been sold in these counties since 1948, the national program would have experiences a surplus of premiums over indemnities, rather than a deficit (Gardner and Kramer, 1986, page 200).

• Soil Bank Program (1955-1973)

• Establishment of Disaster Payments Program through The Agriculture and Consumer Protection Act of 1973, and The Rice Production Act of 1975.

Government started payments for prevented planting and payments for abnormally-low yields for producers of selected crops (Gardner and Kramer, 1986, page 201).

• Federal Crop Insurance Act of 1980 

It expanded the crop insurance program to become the major form of disaster protection in the United States, authorized expansion of the program to all counties with significant agriculture, with a priority for those counties with substantial acreages of crops earlier covered by the disaster payments program  (Gardner and Kramer, 1986, page 202).

• Catastrophic Risk Protection Program (1994 - now)

1- Yu, Smith and Sumner (2017) used this policy for the sensitivity analysis. 

• Farm Bills (1938 - now)

1- Yu, Smith and Sumner (2017) used Farm Bills of 1994, 2000 and 2008 for the sensitivity analysis. Their attention to choosing the pre and post years is instructive.

2- O'Donoghue, Roberts and Key (2009) used farm Bill 1994 to study the impact of increase in crop insurance coverage on farmers' diversification.

• Mandatory Premium-Free Catastrophic Crop Yield Insurance Policy (enacted in 1994 and rescinded in 1996)

For one year, only farmers who enrolled in the catastrophic insurance became eligible for other farm program benefits. 

3- Instrumental Variables

The Tsiboe and Turner (2023) paper (replication files) provides a great overview of endogeneity problem in modeling of crop insurance and reports on four general approaches of using Instrumental Variables for solving the identification problem in this context. Other resources are as following:

• Log of the initial premium divided by the maximum premium as an instrument for the log difference in coverage as measured by premiums per acre

     Weber, Key and O’Donoghue (2016) uses this IV. 

• Number of degree days greater than 30 degrees Centigrade as an IV for insurance indemnity.

Schoengold, Karina, Ya Ding, and Russell Headlee (2014) uses this IV.

• Membership on the House Appropriations Committee as an IV for disaster payments.

Schoengold, Karina, Ya Ding, and Russell Headlee (2014) uses this IV.

Data Cleaning Steps for Research Assistants

The following list is the data cleaning steps using the Stata software that I found to be standard across my research projects: 

1- Use relative folder addresses in projects shared among team members using the "C:\Users\\`c(username)\rest_of_address'" format.

2- Run the "codebook" command as the first step to learn about the 1) variable labels, 2) value range of variables, 3) irregularities in their values and 4) degree of variation in values (the higher the variation, the better are variables for regression modelling). Open the log file (explained in step 3) and look for these information in the log file.

3- Use log files for monitoring steps of processing codes. Use the following lines when running the initial "codebook" command, as well as other bodies of code, when necessary.

cap log close

log using "filename.log", replace

ds

codebook `r(varlist)'

log close

4- In case of creating new variables, use one part of the original variables in the naming of the new variables. It helps with better understanding where the variables came from, with less need to go and read the code. For example, if you create a variable "gender_respondent" based on the original variable "a1", it is better to name the new variable as "a1_gender_respondent".

5- In case of creating new variables, keep the original variables in the dataset, unless you have space issues on your computer system. It helps with cross-checking the variables later, and making sure there are no mistakes in coding. In the previous example, keep both "a1" and "a1_gender_respondent" in the data set.

6- In case of creating new variables, add labels to them immediately. Labels become an asset when sharing codes with others, as well as for understanding the data set by others when running the "codebook" command.

7- Keep all the documentations in the .do files. No side files for documenting the process, unless asked to do so.

8- Keep number of .do files at minimum.

9- Do your best to merge different data sets using the 1:1 command, and do not opt into using either of the 1:m or m:1 or m:m merging easily.

10- Use proper names for all files, and use the date of last update and your initials in the files names.