안녕하세요. 경제연구소에서는 다음과 같은 주제로 특강 및 세미나를 개최합니다.

발표자:  Sokbae Lee (Columbia University)

주제: SA-GMM: Stochastic Approximation for Scalable Estimation and Inference in Nonlinear GMM

장소: 대우관 본관 4층 415호 (곽정환홀)

날짜: 8월 26일 화요일 오후 5시 00분 ~ 오후 6시 30분

자세한 내용은 아래를 확인해주시면 되겠습니다.

교수님들의 많은 참석을 부탁드리며 대학원생들이 참여할 수 있도록 적극 권유해 주시면 감사하겠습니다.

Abstract

We propose SA-GMM, a new scalable framework for estimation and inference in nonlinear generalized method of moments (GMM) based on stochastic approximation. The method iteratively updates unbiased estimators of the moment function and its derivative using U-statistics and multi-pass mini-batch sampling. It does not require a consistent initial estimator or global strict convexity, and accommodates both fixed-data and random-sampling asymptotic regimes. We establish almost sure convergence guarantees and develop simple inference procedures based on random scaling and plug-in methods. In Monte Carlo experiments estimating a system of Exact Affine Stone Index nonlinear demand with 380 unknown parameters and a sample size of n=100,000, our SA-GMM algorithm is solved in under 1.4 hours, whereas standard full sample GMM can manage to converge only after 18 hours.