SBIR Phase I: Next Generation Component Software for Simulation-based Econometric Estimation

This Small Business Innovation Research (SBIR) Phase I project proposes to develop user-friendly component software for classical econometric estimation and inference based on simulation methods. In the last decade, different simulation-based methods have been developed to tackle complex economic/statistical models which cannot be estimated by conventional methods such as maximum likelihood estimation (MLE) and generalized method of moments (GMM). Although these simulation-based estimators have desirable theoretical properties, they have remained as research topics in academia and have not become useful tools for practitioners because of the lack of user-friendly software. This project provides a plan to study three leading applications for simulation-based methods: multinomial probit model for cross-sectional data, multiperiod multinomial probit model for panel data, and stochastic volatility models for time series data. MathSoft will use extensive Monte Carlo experiments to explore finite sample properties of various aspects of estimation and inference, with an aim of improving and stabilizing the current algorithms. The user-friendly component software will be developed using the state-of-art JavaBean technology and provide intuitive graphical user interface. The JavaBeans will also be supplied as S-PLUS functions to gain a broad user base. The software will help worldwide economists and practitioners in other fields such as financial industry, social sciences, and biotechnology to conduct flexible and extensible model estimation and inference.