Quantitative Trading with Bayesian Methods

Add to your list(s) Download to your calendar using vCal

• Dr Bluford Putnam, Chicago Mercantile Exchange
• Wednesday 28 March 2012, 11:00-12:00
• Lecture Room 3B, Inglis Building, Department of Engineering, Trumpington Street.

If you have a question about this talk, please contact Hugh Christensen.

Chief Economist of the CME Group (www.cmegroup.com).

Bayesian inference methods are especially suited for analyzing financial markets for several reasons. Financial markets are all about forming expectations, determining one’s confidence in those expectations, then absorbing new information and revising one’s expectations and confidence. Bayesian inference goes through the same process of forming an hypothesis and confidence assessment, receiving new information, and revising that hypothesis and related confidence. Moreover, Bayesian methods are designed to accept expert information and to update its usefulness in a step by step manner through time. And finally, Bayesian methods are well-suited for forecasting problems with time-varying parameters which are common in the analysis of financial markets. For all of these reasons and more, Bayesian methods are extremely appealing for use in forecasting financial market expected returns, as well as estimating volatility and correlations between pairs of exposure returns.

The practical application of Bayesian methods to trading and investing in financial markets, however, has to integrate a Bayesian forecasting and confidence assessment process for market exposures with a portfolio construction and risk control process. One approach is to use mean-variance optimization systems of the kind pioneered by Professor Harry Markowitz for use in developing portfolios for investment in financial markets. The Bayesian forecasting focus on revising both projections of returns and estimating the confidence (volatility) in those projections fits neatly into the mean-variance framework which requires as inputs expectations for returns, volatility, and correlations, as part of the process of designing a portfolio to maximize returns relative to the volatility of the expected return stream.

This presentation, “Quantitative Trading with Bayesian Methods”, explores the lessons learned from over 20 years of practical experience in using Bayesian methods in asset management. The presentation will examine the investment challenges associated with using next-step ahead Bayesian time series analysis with daily, weekly, and monthly data for currencies, fixed income securities, and equity indices, with special attention to the problems of investing during and after the financial crisis of 2008. Careful attention is paid to the underlying assumptions and whether they are sufficiently robust for use in ever-changing financial markets.

Much of the presenter’s research has been of a proprietary nature. The types of Bayesian models being utilized in the presenter’s investment experience, however, were described in a sequence of four articles in published in the Proceedings of the American Statistical Association in 1994, 1996, 1997, and 1998. Also, the presenter has written a number of business press articles related to the practical application of quantitative techniques to investing. While the presenter is no longer directly involved with asset management, two companies using Bayesian-based quantitative systems with several billions of dollars of assets use systems which descended intellectually (in part) from the work of the presenter and his colleagues, and they have established continuous investment track records since 1998.

This talk is part of the Division F Financial Modelling Group series.

This talk is included in these lists:

• Information Engineering Division seminar list
• Lecture Room 3B, Inglis Building, Department of Engineering, Trumpington Street

Note that ex-directory lists are not shown.