Portfolio Management Investment System (NSERC-Engage, Principal investigator)

In partnership with WTZIMI [William T Ziemba Investments Inc.])

One of the most interesting and attractive areas of study that has the potential to provide significant benefit for (company) is in the decision-making process under conditions of uncertainty is financial optimization. Prominent examples include: risk management for large public corporations; hedge fund strategies to capitalize on market conditions; security selection for stock and bond portfolio managers; currency hedging for multi-national coprorations; asset allocation for pension plans and insurance companies. In these situations, time periods and uncertainties play important roles. For example, a pension plan manager must focus on both the long-term and short-term consequences of his/her investment strategy. One must attempt to minimize pension contribution expenses over time, while satisfying the needs of the retirees, and also reducing risks. There are also many uncertainties in financial planning, such as economic factors, prices of the securities considered, amount of cash flow, etc. Multi-stage stochastic programming models provide the best option to address these significant practical issues. Stochastic programming provides a general purpose-modeling framework, which can capture real-world features such as turnover constraints, transaction costs, risk aversion, limits on groups of assets and other considerations. In this project, we studied both static one-period applications and dynamic modeling over time to provide options for the enhancement of (the company's) operations. Previous researchers have shown that parameters are difficult to estimate and are time varying; however, the means has been shown to be the most important to accurately estimate. Hence, probability distributions will be based on models that estimate one-period and multi-period scenarios - one does not have to assume that the parameters in these models are known. We modeled transaction costs, market impact costs, liquidity and other market imperfections as uncertainty and scenario optimization stochastic programming-based, multi-period asset-liability models. Furthermore we studied static single period and dynamic multi-period portfolio selection models and computer implement and test them with both simulated data and live investment data.

A Statistical Model for Social Network Analytics (NSERC-Engage, Principal investigator)

  • Online social networks have attracted the interest of millions of users. Facebook has more than 400 million users while Twitter has more than 40 million users (as of July 2009) that exchange over 50 million tweets per day. Users are able to interact with each other, chat, share thoughts and links, play games and conduct several other activities. The popularity of social networking has also attracted the interest of the research community that tries to understand their structure and user interconnection as well as interactions among users. One of the distinguishing features of online social networks and social media is their potential for information propagation. It has been studied both empirically and theoretically for many years by sociologists, statisticians, and computer scientists. In this project, we used some statistical tools to answer the following questions: When are people most likely to comment or like a post on a wall in Facebook? When are people most likely to retweet content? Is there a language pattern between content receiving more tweets, likes or comments? Which region is more likely to leave comments? At what time is a specific domain, or social network website the most active? At what time in a specific region is it most active? This was a joint project with Prosyna Communication.