Computational Finance and Algorithmic Trading Papers by Michael Kearns and Yuriy Nevmyvaka
Over the past 14 years, we have collaborated on a number of
proprietary and research projects in the areas of computational finance, algorithmic trading and
related topics. In some cases we (along with various colleagues) have published papers on these projects in the open academic literature.
Below we give, in chronological order, brief descriptions of these works and (where applicable) the commercial or trading context in which they
were developed. We also provide links to the papers themselves, and to related talk slides.
The Penn-Lehman Automated Trading Project.
M. Kearns, L. Ortiz.
IEEE Intelligent Systems, Nov/Dec 2003.
The Penn-Lehman Automated Trading Project was run by Kearns for several years at Penn, with support from
Lehman Brothers. At the time Kearns was a consultant to a prop stat-arb group at Lehman, which he eventually led, and the project
was a competition in micostructure-based automated trading. Nevmyvaka participated in this project as a
doctoral student at Carnegie Mellon, which led to his
partnership with Kearns
and later hiring at Lehman.
Competitive Algorithms for VWAP
and Limit Order Trading.
S. Kakade, M. Kearns, Y. Mansour, L. Ortiz.
Proceedings of the ACM Conference on Electronic Commerce, 2004.
A primarily theoretical paper, with some experimental validation, of worst-case models for VWAP and other
execution problems. Grew out of K-N discussions on methodologies for execution evaluation.
Electronic Trading in Order-Driven Markets:
Y. Nevmyvaka, M. Kearns, A. Papandreou and K. Sycara.
IEEE Conference on Electronic Commerce, 2005.
An early conceptual and empirical examination of the trade-off between price and fill rate in limit order markets, and the optimal pricing frontier.
Done in collaboration with Lehman colleagues. Based strongly on ideas that would eventually form the core of Nevmyvaka's CMU doctoral dissertation.
Reinforcement Learning for Optimized Trade Execution.
M. Kearns, Y. Nevmyvaka, Y. Feng. International Conference on Machine Learning, 2006.
Our first of many applications of machine learning methods to trading problems, in this case the use of
reinforcement learning for optimized execution. The same conceptual and code framework was used by Kearns and Nevmyvaka to
develop alpha-seeking machine learning methods at Lehman.
of Limit Order Dynamics.
E. Even-Dar, S. Kakade, M. Kearns, Y. Mansour.
ACM Conference on Electronic Commerce, 2006.
A theoretical and experimental evaluation of dynamic stability of macroscopic properties (e.g.
VWAP, ADV, etc.) of limit order markets. Nevmyvaka, now at Lehman with Kearns, contributed heavily to the empirical investigation.
Censored Exploration and the Dark Pool Problem.
K. Ganchev, M. Kearns, Y. Nevmyvaka, J. Wortman.
Conference on Uncertainty in Artificial Intelligence, 2009.
UAI Best Student Paper Award for K. Ganchev and J. Wortman.
Journal version in Communications of the ACM , May 2010.
Performed while Kearns and Nevmyvaka were now jointly running an equities quant prop trading group at Bank of America,
this project developed and applied a machine learning approach to trading in dark pools, and was implemented in
BofA's Electronic Trading Services platform.
[Peter Bartlett commentary]
[BofA marketing summary]
Empirical Limitations on High Frequency Trading Profitability.
A. Kulesza, M. Kearns, Y. Nevmyvaka. Journal of Trading, Fall 2010. (JoT Best Paper Award for 2010)
Undertaken at SAC Capital, where Kearns and Nevmyvaka were co-PMs in the MultiQuant division, this
empirical estimate of HFT profitabilty highlights the inherent tension between short holding periods
and price volatility. It was conducted using a large-scale microstructure backtesting platform that
includes a machine learning module for learning profitable order book states.
Market Making and Mean Reversion.
M. Kearns, T. Chakraborty.
ACM Conference on Electronic Commerce, 2011.
A primarily theoretical study of market-making algorithms and their profitability under various stochastic timeseries models;
the opening Theorem 2.1 was derived from longstanding K-N discussions on market-making.
Machine Learning for Market Microstructure and High Frequency Trading.
M. Kearns, Y. Nevmyvaka.
In High Frequency Trading, M. O'Hara, M. Lopez de Prado, D. Easley, eds. Risk Books, 2013.
An article surveying three case studies in the practical application of machine learning
to trading problems utilizing microstructure data.
Pursuit-Evasion Without Regret, with an Application to Trading.
L.Dworkin, M. Kearns and Y. Nevmyvaka.
International Conference on Machine Learning, 2014.
An article developing a state-constrained variant of classical no-regret learning algorithms,
movtivated by an application to multi-asset trading under inventory constraints.
Speaking in his role as an academic, Kearns has occasionally been interviewed or quoted in mainstream media articles on algorithmic trading topics,
including some on the research above. Links below are in reverse chronological order.
Bloomberg News article on HFT and hybrid quant funds, March 2014
Australian radio program "Future Tense" on "The Algorithm", March 2012
Fiscal Times article on machine learning and technology in trading, March 2011,
Wired Magazine article on algorithmic trading, January 2011,
more extensive remarks
on the author's blog.
Science News article on light speed propagation delays in trading, October 2010
Economist article on flash crash autopsy, October 2010
WSJ online post on HFT research, September 2010
Atlantic article on HFT "crop circles", August 2010
Wall Street Journal article on machine learning in quant trading, July 2010
The Trade magazine
article natural language processing for algorithmic trading, September 2007
Bloomberg Markets magazine
article on machine learning on Wall Street, June 2007