A system for Monte Carlo experimentation

A new computer system for Monte Carlo Experimentation is presented in this thesis. The new system speeds and simplifies the process of coding and preparing a Monte Carlo Experiment; it also encourages the proper design of Monte Carlo Experiments, and the careful analysis of the experimental results. A new functional language is the core of this system. Monte Carlo Experiments, and their experimental designs, are programmed in this new language; those programs are compiled into Fortran output. The Fortran output is then compiled and executed. The experimental results are analyzed with a standard statistics package such as S, Isp, or Minitab or with a user supplied program. Both the experimental results and the experimental design may be directly loaded into the workspace of those packages. The new functional language frees programmers from many of the details of programming an experiment. Experimental designs such as factorial, fractional factorial or latin square are easily described by the control structures and expressions of the language. Specific mathematical models, such as arima(p,n,q) models, regression models with specific collinearity properties, tabular data generated by logit or log-linear models are generated by the routines of the language. Numerous random number generators and many standard statistic routines are included. It is easy to use standard variance reduction techniques, such as common or antithetic variables, conditional Monte Carlo, weighted samples, importance sampling or control variates.