Stock Trading Using Linear Genetic Programming with Multiple Time

Stock Trading using Linear Genetic Programming with Multiple Time Frames Garnett Wilson Afinin Labs Inc. St. John’s, Canada gwilson@afinin.com Derek Leblanc Afinin Labs Inc. St. John’s, Canada dleblanc@afinin.com Wolfgang Banzhaf Afinin Labs Inc. St. John’s, Canada banzhaf@afinin.com ABSTRACT A number of researchers have attempted to take successful GP trading systems and make them even better through the use of filters. We investigate the use of a linear genetic programming (LGP) system that combines GP signals pro- vided over multiple intraday time frames to produce one trading action. Four combinations of time frames stretching further into the past are examined. Two different decision mechanisms for evaluating the overall signal given the GP signals over all time frames are also examined, one

based on majority vote and another based on temporal proxim- ity to the buying decision. Results indicated that majority vote outperformed emphasis on proximity of time frames to the current trading decision. Analyses also indicated that longer time frame combinations were more conservative and outperformed shorter combinations for both overall upward and downward price trends. Categories and Subject Descriptors I.2.8 [Artificial Intelligence]: Problem Solving, Control Methods, and Search—Heuristic methods General Terms Algorithms, Performance, Experimentation Keywords computational finance, linear genetic programming, algo- rithmic trading 1. INTRODUCTION Researchers who apply genetic programming (GP) or evo- lutionarycomputationmethodsforanalysisoffinancialmar- kets have a number of different approaches to discover prof- itable opportunities in the price time series they analyze. Someresearcherstrainasystemonanextendedperiodof time, and then allow the evolved solution of static rules to Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. GECCO’11, July 12–16, 2011, Dublin, Ireland. Copyright 2011 ACM 978-1-4503-0557-0/11/07 ...$10.00. operate on a subsequent extended test period [3], [4]. Other researchers have found it beneficial to continually train on a moving window, and only act in anticipation of the im- mediate future of the price changes based on the immediate past [1], [8]. Others go further than this, combining either of these systems with filters that are used to improve the confidence that a GP signal is actually being evolved on an inherently trending time series [3], [4], [7]. In this work, we combine the notions of filtering GP signals and moving windows of varying length in a linear genetic programming system (LGP). In particular, we use the LGP system to de- termine whether price series in partially overlapping time frames are collectively consistent in producing a particular algorithm signal. The next section describes existing literature on the use of filters with GP and the notion of predictable windows with price series. Section 3 describes the linear GP system applied to stock trading using multiple time frames. Section 4describestradingperformance characteristicsof thesystem and its profitability. Conclusions follow in Section 6. 2. PREVIOUS WORK To the authors’ knowledge there are no GP systems in the...

Website: www.cs.mun.ca | Filesize: 1951kb
No of Page(s): 8
Download Stock trading using linear genetic programming with multiple time ....pdf