Monte Carlo Simulation for Econometricians

Many studies in econometric theory are supplemented by Monte Carlo simulation investigations. These illustrate the properties of alternative inference techniques when applied to samples drawn from mostly entirely synthetic data generating processes. They should provide information on how techniques, which may be sound asymptotically, perform in finite samples and then unveil the effects of model characteristics too complex to analyze analytically. Also the interpretation of applied studies should often benefit when supplemented by a dedicated simulation study, based on a design inspired by the postulated actual empirical data generating process, which would come close to bootstrapping. This review presents and illustrates the fundamentals of conceiving and executing such simulation studies, especially synthetic but also more dedicated, focussing on controlling their accuracy, increasing their efficiency, recognizing their limitations, presenting their results in a coherent and palatable way, and on the appropriate interpretation of their actual findings, especially when the simulation study is used to rank the qualities of alternative inference techniques. Preface and Overview Since many decades much of the research in econometric theory is supported or illustrated by Monte Carlo simulation studies. Often the design of such studies follows particular patterns that have become traditional. Performing Monte Carlo studies is usually not taught as such in graduate schools. As a novice one is simply expected to imitate and extend relevant earlier studies published in the recent literature. Many scholars seem to think that setting up a Monte Carlo study is basically too self-evident to bother much about; apparently, it can be done without requiring a manual, because that does not seem available. Therefore, we try to present and illustrate the fundamentals of executing such studies here, pointing to opportunities not often utilized in current practice, especially regarding designing their general setup, controlling their accuracy, recognizing their shortcomings, presenting their results in a coherent and palatable way, and with respect to an appropriate and unprejudiced interpretation of their actual findings. Monte Carlo simulation (abbreviated as MCS from now on) produces from random experiments rather straight-forward statistical inference on the properties of often very complex statistical inference techniques. So, it has an intrinsic recursive nature, because it employs