Two Time-Scales in Global Optimization and Equilibrium

In recent years, the performance improvement in computer architecture is shifting from making a single core faster to increasing the number of processors. Parallel computing becomes the dominant paradigm in computer architecture. In the global optimization and equilibrium community, parallel optimization algorithms have been developed to solve heavily computational intensive problems. One major associated problem is how to effectively utilize parallel computing power. In this dissertation, we consider two timescales parallelism in which tasks assigned to parallel threads are allowed to operate in two timescales. In chapter 2, we present an algorithmic design with interacting annealing processes in two timescales that guarantee a faster identification of a globally optimal solution. In chapter 3, we consider a parallel computing scheme for global optimization that combines a fast timescale multi-start local search with a slow timescale dynamic reallocation of computational resources. In chapter 4, we modified Kyle's informed trading model to include high frequency traders and show that these traders play a beneficial role in the market in which insider trading activity has also been detected.