Print

Events

Past events:
Unknown
Workshop on Agent-based Modelling of Algorithmic Finance, 24-25 May 2018 Program 9:30-10:00 Christian Borch: Welcome 10:00-11:00 Brian G. Peterson: "Machine Learning and Simulation in Training" 11:00-11:15 Coffee 11:15-12:15 Blake LeBaron: "Self-reinforcing Market Heterogeneity" 12:15-13:45 Lunch 13:45-14:45 Björn Hertzberg: "An Introduction to the Ecosystem of a Modern Exchange" 14:45-15:00 Coffee 15:00-16:00 Vijay Prabhakar: "Market Making and Index Arbitrage Strategies without Customer Flow" 16:00-16:15 Coffee 16:15-17:15 Nicholas Skar-Gislinge, Pankaj Kumar, and Zach David: "The AlgoFinance ABM Framework: Background and Outlook" Friday, 25 May 9:30-10:30 Tony Guida: "Training ML Equity Models" 10:30-10:45 Coffee 10:45-11:45 David Fellah: "Practical Concerns in Algorithmic Trading" 11:45-14:00 Lunch and Closing Discussion Abstracts and Bios                       Brian G. Peterson: "Machine Learning and Simulation in Trading" Abstract: Machine Learning (ML) and Artificial Intelligence (AI) are impacting every aspect of our everyday lives from the phones in our pockets to the news we read and the cars we drive. We will briefly cover proprietary trading, and how prop trading is similar and different from traditional asset management, as well as the implications for the strategies deployed. This talk will then predominantly look at how ML and AI methodologies are being employed in trading strategies. We will examine the most common use cases, including highlighting cases where simpler models would likely return similar results with lower model risk. We will also examine the challenges of using backtests and more advanced market simulations for evaluating prospective trading strategies. We will conclude with case studies utilizing ML and agent models for trading strategy development, including measurement, feature design, and simulation. Throughout, we will highlight pitfalls and misconceptions between the public literature and the proprietary implementations in trading firms. We hope to foster a robust discussion of how academic models may be used to explain and inform production trading systems to improve the quality, interpretability, and safety of deployed strategies. Bio: Brian Peterson has more than a decade of experience researching, designing, developing, and deploying production quantitative trading systems. He has been the lead executive for quantitative trading in multiple Chicago proprietary trading firms where his personal assets have been at risk every day. Brian is co-author or maintainer of over 10 packages for using the R statistical language in finance, and acts as the organization administrator for R’s participation in the prestigious annual Google Summer of Code program, and is on the organizing committee of the annual R/Finance conference. In addition, Brian has continued to research, publish, and teach, and holds an appointment as a Lecturer in the University of Washington Applied Mathematics, Computational Finance and Risk Management graduate department. At UW, Brian developed the trading sequence of courses, and teaches quantitative trading systems design. Brian has deep experience delivering large, technically complex production systems utilizing the latest technologies and techniques, including advanced optimization, machine learning and artificial intelligence, low latency execution, and algorithm design, judged directly by the performance of these systems in live markets.    Blake LeBaron: "Self-reinforcing Market Heterogeneity" Abstract: Modeling heterogeneity in both financial markets, and other economic settings is still a challenge that is not well understood. This talk will present a few empirical examples both with market data, and the much more challenging information from individual forecast surveys. All suggest a world in which heterogeneity across market participants changes over time. There is no indication of any convergence over either short or long time scales. A simplified agent-based financial model will be will be used to demonstrate that these properties may be somewhat fundamental in all multi-agent learning environments. Though the model is relatively simple, it is still uses a rich computational framework that is necessary to capture the complete dynamics of heterogeneous beliefs.  Bio: Blake LeBaron has a Ph.D. in Economics from the University of Chicago. He is the Abram L. and Thelma Sachar Chair of International Economics at the International Business School, Brandeis University. LeBaron was at the University of Wisconsin from 1988-1998, and also served as director of the Economics Program at The Santa Fe Institute in 1993. He was a Sloan Fellow, and is a recent recipient of the Market Technician’s Association Mike Epstein award. He recently spent two years as a visiting researcher with the Office of Financial Research in the U.S. Treasury Department. He currently directs the Masters of Science in Business Analytics program at Brandeis, and is part of a Brandeis interdisciplinary research and teaching group interested in modeling dynamics in a wide range of fields. LeBaron’s research has concentrated on the issue of nonlinear behavior of financial and macroeconomic time series. He has been influential both in the statistical detection of nonlinearities and in describing their qualitative behavior in many series. LeBaron’s current interests are in understanding the quantitative dynamics of interacting systems of adaptive agents and how these systems replicate observed real-world phenomenon. Also, LeBaron is interested in understanding some of the observed behavioral characteristics of traders in financial markets. This behavior includes strategies such as technical analysis and portfolio optimization, along with policy questions such as foreign exchange intervention. In general, he seeks to find out the empirical implications of learning and adaptation as applied to finance and macroeconomics.    Björn Hertzberg: "An Introduction to the Ecosystem of a Modern Exchange" Abstract: An exchange's trading rules is the formalization of the permissible interactions of agents in the exchange of assets. Traditionally these agents have been human but increasingly these humans are augmented or replaced by software and the traditional characterization of agents is blurring. We take a look at some dimensions on how to characterize members of a modern stock exchange and give some examples from Nasdaq Stockholm Stock Exchange.  Bio: Björn Hertzberg has a passion for taking data and distilling into business value. He has an extensive background in applied mathematical market microstructure and an adamant believer successfully hybrid of traditional statistical modeling and modern machine learning and AI. Björn has worked both as an ETF fund manager in Quantitative Risk Management, software consultancy and high frequency trading (HFT) and has for the past 7 years been the Nordic head of Economic & Statistical Research at Nasdaq.    Vijay Prabhakar: "Market Making and Index Arbitrage Strategies without Customer Flow" Abstract: Market making and index arbitrage strategies make up a large portion of the market volume in automated markets. Many firms act as broker/dealers and receive institutional and retail customer orders which influence their activities in this space. However, for many market participants, trading is done a proprietary basis without the benefits or obligations of customer order flow. For these participants, their activities need to be profitable on their own with no expectation of a secondary source of revenue. We will discuss some of the issues concerning algorithms, signals, and risk for these participants as they design strategies in this space.  Bio: Vijay Prabhakar is the Director of Trading and Technology at A.R.T. Advisors, LLC, a quantitative hedge fund based in New York. Vijay studied chemistry and machine learning at Columbia University and joined the finance industry instead. Prior to joining A.R.T. Advisors, Vijay was a partner at a few trading firms, including Tradeworx/Thesys Technologies LLC, a NJ-based hedge fund, Headlands Technologies LLC, a Chicago-based proprietary trading firm and Indy Research Labs, LLC, a NY-based proprietary trading firm. Vijay’s main area of focus is in the short-term, quantitative strategies in equities and futures markets around the world.   Nicholas Skar-Gislinge, Pankaj Kumar, and Zach David: "The AlgoFinance ABM framework: background and outlook"  Abstract: Within the algo-finance project at CBS, we aim to study the interactions between automated trading algorithms. We have built a simulation framework in order to simulate these interactions, using an ecology of representative agents trading in a simulated exchange. While this is not a 1:1 replication of a financial market, it allows for studying “what if” scenarios and for cutting through the clutter of trades not relevant for our main research questions. In this talk, we will present the simulation framework and the current state of the modelling work. We will then discuss our coarse graining approach, where we hypothesize that it is possible to describe common trading strategies as “families” and to model market actors using similar trading strategies through representative agents. Bio Nicholas Skar-Gislinge:  Nicholas is a physicist with a background in Nanoscience and a PhD in Structural Biophysics from the Niels Bohr Institute, University of Copenhagen, where he has also previously worked as a post-doc. He has vast experience with mathematical modelling and is currently working on an agent-based model of algorithmic finance. Bio Pankaj Kumar: Pankajhas a background in electrical and electronics engineering (Birla Institute of Technology, Mesra, India) and in data analytics/computer science (Shiv Nadar University, India). He has experience with execution in algorithmic trading and has previously worked as a data scientist in Prognoz, Moscow. His research interests include agent-based models, high-frequency trading, large-scale data mining, and deep learning. He is currently working on an agent-based model of algorithmic finance. Bio Zachary David: Zach iscurrently creating trade execution analytics software for institutional investors. Previously, he developed multi-strategy simulation environments and platforms for high-frequency execution in the Chicago trading industry. He provides the technical and theoretical foundation for the agent-based system underlying project AlgoFinance.   Tony Guida: "Training ML Equity Models"  Abstract: In this presentation, we apply a popular Machine Learning approach (extreme gradient boosted trees) to train different ML equity agents. Gradient boosting has proved to be very effective to enhance signals for diversified equity portfolios. Our ambition is to present the methodology and the steps necessary from a practitioner point of view to train ML equity model according to different tenor of prediction. We created short term and long term ML agents and analyzed what the main most important features used are and contrast our results with the asset pricing theory. We then discuss the main key elements that should be applied to use those ML agents in an ABM environment and provide some elements regarding how they can interact together. Bio: Tony Guida is Senior Investment Manager – Quantitative PM, managing multi-factor equity portfolios for the asset manager of a UK pension fund in London. Prior to that Tony was Senior Research Consultant for Smart Beta and Risk allocation at EDHEC RISK Scientific Beta, advising asset owners how to construct and allocate to risk premia. Before joining EDHEC Tony worked eight years at UNIGESTION as a Senior Research Analyst. Tony was a member of the research and Investment Committee for Minimum Variance Strategies and he was leading Factor Investing research group for institutional clients. Tony is the editor and co-author of the forthcoming book: Practical Applications of Machine Learning and Big Data for Quantitative Investment (summer 2018) Tony holds Bachelors and Master degrees in Econometry and Finance from the University of Savoy in France. David Fellah: "Practical Concerns in Algorithmic Trading" Abstract: I will be interested in discussing how ABM’s can be of practical application as surrogate, generative models for training reinforcement learning algorithms, in particular for on-line learning cases. Notably, how can we ensure interaction dynamics are consistent with empirical market data; can ABM’s be used in a hybrid mode where market data provides background ‘noise,’ for example? I would like to offer a point of view from the field and hopefully provide some practical anecdotes in trading so that research can be directed both towards understanding of market structure and improved algorithms for investors. It would also be interesting to discuss the problem of liquidity fragmentation, price discovery, and peer to peer networks that are commonplace in FX and post MiFID2. Finally, can ABM’s be used to augment our understanding of how normal Auctions, and volatility auctions could be improved?  Bio: David’s background was originally in Chemistry, working in the Morphogenesis Laboratory at Manhattan College, New York. He joined ITG to be part of the Algorithmic Strategy Group in 1998, and later ran the algorithmic quant group at Miletus Trading, a start-up that was purchased by Liquidnet in 2007. In 2010 he joined J.P. Morgan as co-head of (Linear, non-derivatives) Quantitative Research Group, based in London. At this time the focus was on developing the Central Risk Book at J.P. Morgan. Finally, he returned to ITG in 2018 and is currently heading up the global Trading Analytics Group. His main areas of focus are on high-frequency market impact modelling, order routing, optimal trade scheduling for portfolios, and reinforcement learning algorithms for single securities mainly in cash products.

Last updated by: Christian Borch 26/05/2018