While there has been longstanding interest in quantitative finance in the use of models from machine learning and related fields, they are often applied towards the
attempt to predict directional price movements, or in the
parlance of the field, to “generate alpha” (outperform the
market). Here we have instead focused on a problem in
what is often called algorithmic trading—where one seeks
to optimize properties of a specified trade, rather than
decide what to trade in the first place—in the recently
introduced dark pool mechanism. In part because of the
constraints imposed by the mechanism and the structure
of the problem, we have been able to adapt and blend
methods from statistics and reinforcement learning
in the development of a simple, efficient, and provably
effective algorithm. We expect there will be many more
applications of machine learning methods in algorithmic
trading in the future.
We are grateful to Curtis Pfeiffer and Andrew Westhead
for valuable conversations and to Bobby Kleinberg
for introducing us to the literature on the newsvendor
1. akritas, m.g. nonparametric survival
analysis. Stat. Sci. 19, 4 (2004), 615–623.
2. alon, n., spencer, J. The Probabilistic
Method, 2nd Edition. Wiley, new york,
3. bogoslaw, d. big traders dive into dark
pools. business Week article, available
4. brafman, r., tennenholtz, m.
r-maX—a general polynomial
time algorithm for near-optimal
reinforcement learning. J. Mach.
Learn. Res. 3 (2003), 213–231.
5. carrie, c. illuminating the new dark
influence on trading and u.s. market
structure. J. Trading 3, 1 (2008), 40–55.
6. cesa-bianchi, n., lugosi, g. Prediction,
Learning, and Games. cambridge
university Press, 2006.
7. domowitz, i., finkelshteyn, i.,
yegerman, h. cul de sacs and
highways: an optical tour of dark pool
trading performance. J. Trading 4, 1
Kuzman Ganchev (email@example.com.
edu), university of Pennsylvania.
Yuriy nevmyvaka ( firstname.lastname@example.org),
university of Pennsylvania.
8. foldes, a., rejto, l. strong uniform
consistency for nonparametric
survival curve estimators from
randomly censored data. Ann. Stat. 9,
1 (1981), 122–129.
9. ganchev, k., kearns, m. nevmyvaka, y.,
vaughan, J. W. censored exploration and
the dark pool problem. in Proceedings
of the 25th Conference on Uncertainty
in Artificial Intelligence, 2009.
10. huh, W.t., levi, r., rusmevichientong, P.,
orlin, J. adaptive data-driven inventory
control policies based on kaplan–meier
estimator. Preprint available at http://
11. kaplan, e.l., meier, P. nonparametric
estimation from incomplete
observations. J. Am. Stat. Assoc. 53
12. kearns, m., singh, s. near-optimal
reinforcement learning in polynomial
time. Mach. Learn. 49 (2002), 209–232.
13. Peterson, a.v. kaplan-meier estimator.
in Encyclopedia of Statistical
Sciences. Wiley, 1983.
Michael Kearns (email@example.com.
edu), university of Pennsylvania.
Jennifer Wortman Vaughan (jenn@seas.
harvard.edu), harvard university.
© 2010 acm 0001-0782/10/0500 $10.00
Magazine Seeks Its Next
For more information, see
As ACM’s premier magazine on applied computer-human interaction
(CHI), interactions is designed to keep developers, designers, managers,
researchers, and users abreast of the latest tools and ideas emerging from the CHI community—and beyond. This colorful, bi-monthly
magazine shares timely articles, stories, and practical content related
to the interactions between experiences, people, and technology.
Its primary objective is to trace innovative technologies from their
R&D beginnings to real-world applications and future potential.
To apply, please send
a résumé, letter of motivation,
Search Committee Chair
interactions is also the membership magazine for ACM’s SIGCHI and
as such is distributed to more than 7,000 members worldwide.
The role of co-editor-in-chief is a three-year volunteer position and is
assisted by ACM professional staff for the production of the magazine.
The position starts in the third quarter of 2010 in preparation for the
January-February 2011 issue.