고려대학교 경제학과

다가오는 경제학 세미나 관련하여 안내드립니다:

Title: Scalable BLP: Stochastic Nested Fixed Point Algorithm

Abstract: In this paper, we propose a scalable estimation method for the random coefficients logit model, commonly known as the BLP model (Berry, Levinsohn, and Pakes, 1995). The standard nested fixed point (NFP) algorithm for estimating the model is computationally intensive and often infeasible for large datasets or in online data stream environments. We propose a stochastic nested fixed point (SNFP) algorithm that updates parameters sequentially using stochastic gradient methods. By avoiding avoiding full-sample inversion and optimization, SNFP substantially reduces memory usage and computational burden. We establish the asymptotic properties of the estimator under standard regularity conditions and implement an online inference procedure based on the random scaling method of Lee et al.\ (2022). Monte Carlo simulations demonstrate that SNFP achieves statistical accuracy comparable to the standard NFP estimator while remaining stable and scalable to over one million markets. An empirical application to scanner data further highlights the practical benefits of SNFP for large-scale structural demand estimation.