Nonlinear generalized method of moments (GMM) is a standard econometric tool, but it becomes computationally prohibitive when the model involves many nonseparable moments and parameters. The new SLIM algorithm recasts estimation as a stochastic approximation problem, enabling it to handle settings that are out of reach for conventional batch estimators.
The authors prove that SLIM produces consistent and asymptotically normal estimates under weak conditions: it does not require a consistent starting value or a convex GMM criterion. They also show that random-scaling and plug-in confidence sets are valid when computational noise is non-negligible relative to sampling uncertainty. An optional second-order refinement can achieve full-sample GMM efficiency, and the paper develops J-tests for overidentifying restrictions.
SLIM is extended to clustered data with unequal cluster sizes. Monte Carlo experiments on a nonlinear demand system with 576 moments and 380 parameters demonstrate computational gains and scalability to one million observations. A real-data application uses LinkedIn records on 4.8 million college graduates to perform clustered estimation and inference for first employment destinations.