What we're about
Upcoming events (4)
FULL TITLE Learning Interest Rate Interpolation ABSTRACT The usual methods for interest rate interpolation consider only the values and time to maturity of spot rates as the inputs, and differ mainly on the continuity of the implied forward rates. We treat the interpolation problem as a replication problem, where a bond (or interest rate future/swap) is priced as a function of the minimum variance replicating portfolio of the traded bonds (or derivatives). In this view, the hedging ratios determined by the interpolation are as important (if not more) than getting the “right” interpolated rate; this is similar to the adjustments to the Black and Scholes delta as a consequence of the joint dynamics of the asset price and volatility in the different volatility models. We show how to learn the parameters of the weight functions and apply this method to the overnight rate indexed interest rates derivatives in Brazil. We then extend the concept from interpolating broken dates to the market references, in order to determine which points are key to the shape and dynamics of the curve and which points can be replicated by these real anchors. BIO Marcos C. S. Carreira, a PhD candidate at École Polytechnique, is the co-author of the book "Brazilian Derivatives and Securities: Pricing and Risk Management of FX and Interest-Rate Portfolios for Local and Global Markets". He was Derivative Products Officer and later Technical Modeling Officer at BM&FBovespa, where he contributed to risk management, derivatives pricing, exchange fees, microstructure and HFT functions. At Credit Suisse Brazil, he was a Managing Director in charge of the FX and IR Options desk, after being the Risk Manager responsible for Market, Counterparty and Liquidity Risks. Marcos holds an engineering degree from Instituto Tecnológico de Aeronáutica (ITA) and a Masters in Economics at Insper. Marcos also lectured for the MECAI Professional Masters course in Mathematical Finance at ICMC-USP and is a regular speaker at quantitative finance conferences.
Financial Applications of Machine Learning Terry Benzschawel Seminar Program 5:45pm Registration 6:00pm Seminar 7:30pm Wine and cheese reception Abstract In this talk, I describe a variety of machine learning models that I have built and applied to problems in business and finance. I begin with an historical introduction to neural networks, including brief descriptions of the perceptron, and methods of gradient descent, backpropagation and regularization. I then describe single hidden-layer perceptrons built in the early 1990s to detect fraud on credit card portfolios, identify customers who will give up their credit cards, and later, for trading US Treasury bonds. I then describe recent work with deep learning networks that predict spread changes for corporate bonds, price moves from trade flows, and a natural language processing model that predicts market moves from sentiment data. Finally, I provide some thoughts on how artificial intelligence/machine learning is changing the fixed income trading business. Biography Terry Benzschawel has recently left a thirty-year career on Wall Street to start his own firm. Prior to that, Terry was a Managing Director in Citigroup's Institutional Clients Business. Terry headed the Quantitative Credit Trading group which developed quantitative tools and strategies for credit market trading and risk management, both for Citi's clients and for in-house applications. Terry received a Ph.D. in Experimental Psychology from Indiana University (1980) and his B.A. (with Distinction) from the University of Wisconsin (1975). His Ph.D. thesis concerned development of a neural network model of the human visual system. Terry has done post-doctoral fellowships in Optometry at the University of California at Berkeley and in Ophthalmology at the Johns Hopkins University School of Medicine. He also was a visiting scientist at the IBM Thomas J. Watson Research Center prior to embarking on a career in finance. He currently serves on the steering committees of the Masters of Financial Engineering (MFE) Programs at the University of California at Berkeley and serves there as an Executive in Residence. In 1988, Terry began his financial career at Chase Manhattan Bank, training genetic algorithms to predict corporate bankruptcy. In 1990, he was hired by Citibank to build neural network models to detect fraudulent card transactions and to predict credit card attrition. In 1992 he moved to investment banking at Salomon Brothers where he built models for proprietary trading for Salomon's Fixed Income Arbitrage Group. In 1998, he moved to the fixed income strategy as a credit strategist with a focus on client-oriented solutions across all credit markets and has worked in related roles since then. Terry was promoted to Managing Director at Citi in 2008. Terry is a frequent speaker at industry conferences and events and has lectured on credit modelling at major universities. In addition, he has published over a dozen articles in refereed journals and has authored two books: CREDIT MODELING: FACTS, THEORIES AND APPLICATIONS and CREDIT MODELING: ADVANCED TOPICS. In addition, Terry has been the instructor for courses in credit modelling for Incisive Media, the Centre for Finance Professionals, the Machine Learning Institute and has taught in UCLA’s MFE program last Fall. Finally, Terry has taught a course on credit modelling at Russia's Sberbank in Moscow. Acknowledgements Special thanks to the Fordham University Gabelli School of Business for hosting and sponsoring the seminar.
FULL TITLE Deep Learning Volatility: On pricing and calibration of (rough) stochastic volatility models with deep neural networks ABSTRACT We present a powerful neural network based calibration method for a number of volatility models including the rough volatility family. The aim of neural networks in this work is an off-line approximation of complex pricing functions, which are difficult to represent or time-consuming to evaluate by other means. We highlight how this perspective opens new horizons for quantitative modelling: The calibration bottleneck posed by a slow pricing of derivative contracts is lifted. This brings several model families (such as rough volatility models) within the scope of applicability in industry practice. As customary for machine learning, the form in which information from available data is extracted and stored is crucial for network performance. With this in mind we discuss how our approach addresses the usual challenges of machine learning solutions in a financial context (availability of training data, interpretability of results for regulators, control over generalisation errors). We present specific architectures for price approximation and calibration and optimize these with respect different objectives regarding accuracy, speed and robustness. We also find that including the intermediate step of learning pricing functions of (classical or rough) volatility models before calibration significantly improves the generalisation performance compared to the performance of deep calibration networks that are trained directly on data. BIO Blanka Horvath is a Lecturer at King's College London in the Financial Mathematics group, and an Honorary Lecturer in the Department of Mathematics at Imperial College London. Blanka holds a PhD in Financial Mathematics from ETH Zurich, a postgraduate degree (Diplom) in Mathematics from the University of Bonn, and an MSc in Economics from The University of Hong Kong. In her research she lays a particular emphasis on the applicability of her research and maintains close collaborations with the industry, including: JP Morgan, Deutsche Bank, Zeliade Systems and AXA. Her research interests are in the area of Stochastic Analysis and Mathematical Finance. They include (but not limited to): * Numerical methods as well as machine learning techniques for option pricing, forecasting and simulation. * Laplace methods on Wiener space and heat kernel expansions. * Smile asymptotics for local- and stochastic volatility models with a particular emphasis on rough volatility models and also SABR-type models.
Options Portfolio Selection Paolo Guasoni Seminar Program 5:45pm Registration 6:00pm Seminar 7:30pm Wine and cheese reception Abstract We develop a new method to optimize portfolios of options in a market where European calls and puts are available with many exercise prices for each of several potentially correlated underlying assets. We identify the combination of asset-specific option payoffs that maximizes the Sharpe ratio of the overall portfolio: such payoffs are the unique solution to a system of integral equations, which reduce to a linear matrix equation under suitable representations of the underlying probabilities. Even when implied volatilities are all higher than historical volatilities, it can be optimal to sell options on some assets while buying options on others, as hedging demand outweighs demand for asset-specific returns. Biography Paolo Guasoni holds the Stokes Chair in Financial Mathematics at Dublin City University since 2009 and specializes in Mathematical Finance. His research investigates the effects of market frictions, incentives, and preferences, in portfolio choice and asset pricing, and has appeared in the Journal of Financial Economics, Finance and Stochastics, Mathematical Finance, and Annals of Applied Probability. He has attracted funding by the European Research Council, the National Science Foundation, Science Foundation Ireland, and the European Commission. He serves as Associate Editor for Finance and Stochastics, Mathematical Finance, SIAM Journal in Financial Mathematics, Applied Mathematical Finance, and the European Journal of Finance. Acknowledgments Special thanks to the Fordham University Gabelli School of Business for hosting and sponsoring the seminar.