Asset Allocation: A Statistical Learning Approach

点击翻译按钮获取中文摘要

This award will contribute to the advancement of national prosperity and economic welfare by developing new and improved methods to solve optimal asset allocation problems. Portfolio optimization problems involving allocation of resources across multiple assets are of fundamental importance in financial management and are faced daily by financial institutions, asset managers, pension plans, university endowments, insurance companies, and individual investors. Beyond finance, portfolio optimization problems arise in other industries, such as portfolios of drug development projects in the pharmaceutical industry, portfolios of research and development projects in the technology industry, portfolios of energy generating assets in the energy industry. This award aims to improve out-of-sample performance of asset allocation methods by building on recent advances in machine learning. The award contributes to the education and development of human resources by providing research opportunities to students and enriching student experience at undergraduate and graduate levels by bringing challenging applied problems in asset allocation to the classroom.The pioneering work of Markowitz on mean-variance portfolio optimization laid the theoretical foundations of applying optimization to the asset allocation problem. Notwithstanding the enormous importance and influence of these classical contributions, the fact remains that, from an empirical point of view, Markowitz-style portfolio policies are often outperformed by the naive 1/N equal weights portfolio policy out of sample due because of limitations associated with estimating statistical parameters from historical data. This project modifies the classical Markowitz portfolio optimization by introducing a new portfolio optimization framework inspired by recent advances in machine learning and econometrics. A partially egalitarian least absolute shrinkage and selection operator (PELASSO) building on the celebrated LASSO regression was recently introduced in the econometrics literature in the context of solving the problem of optimally combining different forecasts. Inspired by these recent developments, this project employs the PELASSO portfolio optimization approach by regularizing the portfolio optimization problem to assign portfolio weights of some of the assets to zero and to select and shrink weights of the surviving assets in the portfolio towards equal weights to hedge against parameter estimation risk. The PELASSO portfolio optimization aims to achieve improved out-of-sample performance and is evaluated empirically using rich data from high frequency econometric sources. The project utilizes methods at the intersection of optimization, statistics, machine learning and economics.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.