Research
Saugat Khanal is a Ph.D. student in Agricultural Economics at Oklahoma State University. His research spans spatial econometrics, policy analysis, time-series econometrics, and risk, uncertainty, and insurance programs. His current work addresses interpolation-induced bias in the climate variables used to estimate crop yield response, developing spatial errors-in-variables models that correct for measurement error introduced when weather station data is interpolated onto county-level administrative units. He holds an M.S. in Agricultural Economics from Oklahoma State University and a B.Sc. in Agriculture from the Agriculture and Forestry University, Rampur, Nepal.
Hierarchical Spatial Errors-in-Variables Model
Khanal works under Dr. Dayton Lambert, Sparks Chair in Agricultural Sciences at Oklahoma State University, on a Hierarchical Spatial Errors-in-Variables Model (HSEVM) for crop yield estimation. The project jointly estimates kriged interpolation surfaces for climate variables alongside county-level yield regressions, using Bayesian inference through the Integrated Nested Laplace Approximation (INLA). Lambert has directed key modeling decisions on the project, including the use of county centroid polynomial coordinates as instruments to address the Modifiable Areal Unit Problem and spatial support misalignment between weather stations and county boundaries.
Wine grape economics in the Southern Great Plains
During his Master’s degree, Khanal worked under Dr. Courtney Bir, Associate Professor of Agricultural Economics at Oklahoma State University. His thesis examined the economic feasibility of wine grape and vine production in the Southern Great Plains region of Oklahoma, evaluating establishment costs, yield expectations, and breakeven economics for prospective growers. That work developed into the manuscript “Economic Feasibility of Wine Grape Cultivation in the Southern Great Plains,” coauthored with Bir and Aaron Essary, currently under review at the Journal of Agricultural and Applied Economics.
Modeling expertise
Spatial errors-in-variables
Climate variables used in crop yield models are rarely observed at the same spatial resolution as the outcome itself. Weather stations report point measurements, while yield data is recorded at the county level, creating a mismatch that ordinary regression ignores. Spatial errors-in-variables models treat the interpolated climate surface as a noisy proxy for the true underlying variable, explicitly modeling the measurement error introduced by interpolation rather than assuming it away. In the Hierarchical Spatial Errors-in-Variables Model (HSEVM), this approach corrects substantial attenuation bias in temperature-related yield damage estimates, strengthening the estimated effect of extreme heat by more than three times relative to conventional regression.
Kriging & interpolation
Weather stations are unevenly distributed across the landscape, and most econometric applications need climate values at locations where no station exists, such as a county centroid. Kriging provides a statistically principled way to interpolate these values, using the spatial correlation structure of the observed data to predict values at unobserved locations along with an estimate of prediction uncertainty. This research uses kriging to construct interpolated surfaces for growing degree days, killing degree days, and precipitation-based drought indices across Iowa counties, then examines how the choice of interpolation method and station density affects the accuracy of downstream yield regressions.
Simulation & bias diagnostics
Because the true value of an interpolated climate variable is never directly observed, it is difficult to know from real data alone how much bias a given modeling choice introduces. This work addresses that using Monte Carlo simulation, comparing model estimates against a known "oracle" data-generating process across varying station densities and replications. This makes it possible to quantify how measurement error propagates through a regression model under controlled conditions before applying the same methods to real yield data. In the HSEVM work, simulations spanning station densities of 20, 60, and 120 per region and 1,000 replications, alongside SIMEX-based bias correction, characterize how much conventional approaches understate climate effects on yield.
Bayesian inference (INLA)
Fully Bayesian estimation of spatial models is often computationally expensive using standard Markov Chain Monte Carlo methods, particularly once a model includes both a spatial interpolation component and a separate outcome regression. Integrated Nested Laplace Approximation (INLA) provides a faster, deterministic approximation to the posterior distribution for a broad class of latent Gaussian models, making it practical to jointly estimate interpolation surfaces and yield regressions within a single hierarchical framework. The HSEVM work relies on INLA with the Stochastic Partial Differential Equation (SPDE) approach to represent spatial random fields, allowing joint estimation to run in minutes rather than hours.
Climate–yield modeling
Climate-yield models link weather exposure, such as growing season temperature and precipitation, to agricultural output, typically at the county or field level. These models inform crop insurance pricing, climate adaptation policy, and long-run agricultural risk assessment. This work focuses specifically on how the climate data entering these models is constructed, since most applications rely on interpolated weather-station data without accounting for the resulting measurement error. Using county-level corn yield data for Iowa from 2016 to 2024, joint estimation of climate and yield is compared against the conventional two-step approach of interpolating first and regressing second, finding meaningfully different estimates of heat sensitivity between the two.
Time-series econometrics
Beyond spatial applications, this work uses time-series methods to study macroeconomic relationships that unfold over years or decades. This includes cointegration analysis and error-correction modeling, which distinguish between long-run equilibrium relationships among variables and short-run deviations from that equilibrium. In a manuscript on U.S. inward foreign direct investment, an Autoregressive Distributed Lag (ARDL) error-correction framework separates the long-run relationship between macroeconomic conditions and FDI inflows from short-run adjustment dynamics, using a parsimonious dynamic specification to keep the model interpretable while still capturing meaningful adjustment behavior.
Discrete choice & limited dependent variable modeling
Many questions in agricultural economics involve outcomes that are not continuous: whether a farmer purchases insurance, and if so, how much coverage to buy; whether a household adopts a new technology; whether a producer participates in a given market at all. These questions call for discrete choice and limited dependent variable models rather than standard linear regression. In work on willingness to pay for cattle insurance in Nepal, a double-hurdle model separately estimates the decision to participate in the insurance market and, conditional on participating, the amount a farmer is willing to pay, since these two decisions can be driven by different factors.
Risk, uncertainty, and insurance program design
Agricultural producers face weather and price risk that public and private insurance programs are designed to manage, but the effectiveness of these programs depends on how well their triggers and payouts match the risk farmers actually face. This work examines that question from two directions: farmer-level willingness to pay for insurance products in the case of Nepali cattle insurance, and the design of rainfall index insurance programs, where a mismatch between an index and a farmer's actual losses, known as basis risk, can undermine the value of coverage even when the underlying index is well constructed.
Research interests
- Policy analysis
- Climate risk and agricultural productivity
- Spatial econometric modeling
- Measurement error and errors-in-variables models
- Climate–yield modeling
- Agricultural decision-making under climate uncertainty