Improving technical trading systems by using a new MATLAB-based genetic algorithm procedure

Technical analysis (TA) is considered as an “economic” test for the random walk 2 hypothesis and thus for the weak form Efficiency Market Hypothesis (EMH). Advocates of TA assert that it is plausible to forecast future evolutions of financial assets‟ price paths with a bundle of technical tools conditioned on historical prices. Among these tools, we can identify technical patterns, which are specific forms of price paths‟ evolutions which are mainly identified visually. When such pattern is confirmed, a technician expects prices to evolve with a specific way. Although, bibliography on testing the efficacy of TA is massive, only a minor fraction of it deals with technical patterns. Various cognitive biases affecting practitioners‟ trading and investment activities and subjectivity embedded in the pattern‟s recognition process via visual assessment, set significant barriers in any attempt to evaluate the performance of trading strategies including such patterns. In this thesis we propose novel, rule-based, identification mechanisms for a set of well known technical patterns classified in the following three general categories: horizontal, zig-zag and circular patterns. The novelty of the proposed methodologies resides in the manner the identification mechanisms are designed. Core principles of TA, regarding the pattern identification via visual assessment are being quantified and the proposed recognizers outperform already existed ones to the fact that they identify all variations of the examined patterns regardless of their size, in a more objective manner. Thus, we believe that the proposed methodologies can set another basis for the development of more sophisticated automatic trading systems and more comprehensive and robust evaluations of TA in general. Implications for the industry and the finance community are also plausible. Software programs (or packages) of TA can include these recognizers in the bundle of all other technical indicators they provide within their services. Finally, practitioners may include these trading rules within their investment and trading activities, after assessing their performance individually, enhancing them (if necessary), or modifying them according to their idiosyncratic investment profile. We subsequently proceed to the individual and joint evaluation of the examined patterns‟ performance. For this purpose we use a variety of datasets (artificially created, US stocks and worldwide market indices) and assess generated returns with ordinary statistical tests, bootstrapped techniques and artificial neural networks. Our empirical findings are either new or comparable with already existed ones. To our point of view, some of the most significant and interesting are the followings: 1) Technical patterns were successfully identified in stochastically generated price paths. Thus, it is reasonable to expect their appearance in real price series too. 2) For specific patterns, when applied on stochastic price series, frequencies of observations, and returns‟ characteristics were similar with those observed in real price series. 3) Generally, our results are in favour of EMH. 4) Indications of market inefficiencies (if any) were more profound in the earlier sub-periods of examination, but not in recent ones. 5) Indications in favour of TA (if any) were observed when shorter holding periods were used. 6) Technical trading rules may successfully predict trend reversals, trend continuations or the sign of future returns, but they fail to generate systematically, statistically significant excess returns. The latter finding, if combined with a variety of cognitive biases included in investors‟ decision making processes, may reason for the apparent wide-spread implementation of TA within the everyday trading and investment activities of practitioners. This thesis is not the first published attempt to quantify such technical patterns and assess the generalised efficacy of TA. However, to our knowledge, the manner we approached the aforementioned issues is new. We believe that the proposed methodologies outperform already existed ones and implications of this thesis to academia and finance industry are significant.

Download Full-text

Manufacturing cell scheduling using genetic algorithms

Proceedings of the Institution of Mechanical Engineers Part B Journal of Engineering Manufacture ◽

10.1243/0954405001517531 ◽

2000 ◽

Vol 214 (2) ◽

pp. 159-163 ◽

Cited By ~ 6

Author(s):

G C Onwubolu

Keyword(s):

Genetic Algorithm ◽

Flow Shop ◽

Performance Measure ◽

Scheduling Problems ◽

Mean Flow ◽

Manufacturing Cells ◽

Data Set ◽

Manufacturing Cell ◽

Combinatorial Optimization Problems ◽

Genetic Algorithm Procedure

This paper presents a new approach to the scheduling of manufacturing cells which have flow-shop configuration. The approach is based on the genetic algorithm, which is a meta-heuristic for solving combinatorial optimization problems. The performance measure demonstrated in this paper is the optimization of the mean flow time. The procedure developed automatically computes the make-span. A flexible manufacturing cell schedule is used as a case study. The genetic algorithm procedure was used to solve a published data set for simple scheduling problems. The genetic algorithm procedure was further used to solve large flow-shop scheduling problems having machine sizes of up to 30 and job sizes of up to 100 in very reasonable computation time. The results show that the genetic-algorithm-based heuristic is promising for scheduling manufacturing cells.

Download Full-text