B1247
Title: Multivariate distributional stochastic frontier models with missing values
Authors: Rouven Schmidt - Clausthal University of Technology (Germany) [presenting]
Alexander Ritz - Clausthal University of Technology (Germany)
Benjamin Saefken - Clausthal University of Technology (Germany)
Abstract: The main objective of the stochastic frontier analysis is to separate the composed error term into noise and inefficiency components, making the estimation of this term crucial. However, in real-world scenarios, missing values of covariates are common and significantly affect the estimated distribution of the composed error term. To address this issue, the Distributional Stochastic Frontier Model is extended to handle missing data. The idea is that missing values contribute zero to the additive predictor from their corresponding smooth. Instead, each missing value has its own Gaussian random effect, with specific random effect variances per input. Separate Gaussian random effects are generated for each missing value, replacing the original smooth effect in the model. In the next step, the model is extended to accommodate multiple outcomes, using a copula approach to represent the joint distribution of the composed error terms, thus capturing the intricated interdependencies among the multiple outcomes. A comprehensive Monte Carlo simulation is performed to demonstrate the effectiveness of the proposed method. In addition, the approach is used to estimate sorghum and millet production in Burkina Faso, demonstrating its practical applicability. The implementation of our methodology is readily available in the dsfa package.
