Managed runtime environments have become popular targets for compilers of high-level programming languages. They provide a high-level type system with enforced runtime safety, as well as facilities such as garbage collection, possibility sandboxed access to services of the underlying platform, multithreadng, and a rich library of data structures and algorithms. But managed runtime environments lack a clear performance model, which hinders attempts ar optimizing the compilation of any language that does nor have a direct mapping to the runtime environments semantics. This is aggravated if the language is dynamically typed. We assert that it is possible to build a compiler for a dynamic language that targets a managed runtime environment so that it rivals a compiler that targets machine code directly in efficiency of the code it generates. This dissertation presents such a compiler, describing the optimizations that were needed to build it, and benchmarks that validate these optimizations. Our optimizations do not depend on runtime code generation, only on information that is statically available from the source program. We use a novel type inference analysis to increase the amount of information available.