Basic forms of determining evolutionary fitness for use with evolutionary computing techniques, such as Genetic Algorithms (GA), in the context of financial data mining are often insufficient or contradicted by the financial problem constraints, leading to unwanted evolutionary pressures. A GA is used to construct trading models in a highly leveraged, day-trading environment, utilizing multiple forms of fitness and fitness incentives, dynamically weighted to search more effectively the solution space. The GA utilizes variable period technical indicators and an array of historical data, in addition to current market data, as inputs. The modifications to the fitness function are designed to mitigate or neutralize undesirable training incentives based on problem definition, yielding models that are responsive to problem definition without becoming hampered by the counterproductive evolutionary pressures exerted by that definition. The principles may be extended to GAs in various other domains.10. References 1) Achelis, Steven B., Technical Analysis from A to Z, 2aquot; Edition, McGraw Hill, 2000, Pp. 1-380. 2) Austin M, et al., Adaptive systems for foreign exchange trading. Quantitative Finance Vol 4, August 2004, pp C37-C44 3) Chen , A.
|Title||:||Domain Driven Causal Financial Engineering in the Context of Evolutionary Computing|
|Publisher||:||ProQuest - 2009|