My recent and current research agenda focuses on the modeling and prediction of financial asset returns. In particular, my team and I develop and study multivariate stochastic processes suitable for complex data arising from a variety of financial instruments. By calibrating these constructs to real data with sophisticated statistical methodologies, we can make accurate predictions of risk and return.

This, in turn, leads to algorithms for improved asset allocation. We derive optimal investment strategies and portfolio weights, through time, for a given universe of risky assets, such as stocks, bonds, currencies, options, crypto-currencies, and commodities, accounting also for transaction costs and illiquidity issues.

​

To get an idea of the improvements we obtain, Graphic 1 shows the performance of some of our models when applied to the stocks of the DJIA (top lines; total return over time) versus existing investment strategies (bottom lines) such as equally weighted (1/N), Markowitz, and traditional GARCH.

​

The various methods I work on can be divided into five categories (see Graphic 2). The top four are based on multivariate constructs, and all support large-dimensional asset universes, despite the proliferation of parameters and the "curse of dimensionality". The fifth method is for multivariate portfolio allocation, but makes use of only simple, univariate models that have been augmented with extremely fast statistical and numerical approximation procedures to allow it to work.