广义随机后悔最小化模型的子集抽样估计OACHSSCD

The generalized Random Regret Minimization(RRM)model faces the problem that the computational cost of full-sample maximum likelihood estimation increases exponentially under large-scale choice sets.To reduce computational bur-den and guarantee favorable statistical performance of estimators,this paper proposes a parameter estimation method based on subset sampling of alternatives.The proposed method extracts a subset containing the chosen alternative from the complete choice set,corrects choice probabilities by introducing sampling correction terms,adopts an approximate regret function only relying on subsets,and compensates information loss with scaling factors.The paper further theoretically proves that the estimators obtained by this method satisfy consistency and asymptotic normality and share the same asymptotic efficiency as maximum likelihood esti-mation based on complete alternative data.Therefore,the proposed method can be regarded as an effective extension of maximum likelihood estimation in the generalized RRM model.In addition,the paper adopts Monte Carlo simulation to compare and analyze the performance of full-sample maximum likelihood estimation,truncated model,repeated sampling correction method and popu-lation share correction method in terms of average bias,root mean square error and running time.The results show that the pro-posed method performs well on all three indicators.Empirical analysis based on real data further indicates that parameter esti-mates gradually converge to the results of full-sample maximum likelihood estimation as the size of sampled subsets expands,which verifies the practicability and effectiveness of the proposed method.