Topic: Self-Selected topic in evolutionary computation for finance.
The purpose of this final topic is to let you choose and explore some area of interest, carry out some practical investigations into the area, and produce a report, which discusses your findings.
Possible options for this include:
? Applying some of the techniques we have looked at in the course so far to a different financial problem (Predictions, Portfolio Optimisation, Algorithmic Trading). Included at the end of this topic is a copy of the paper, "Natural Computing in Finance: A Review" by Brabazon et al.
This is a longer version of the paper distributed at the start of the course and identifies a number of possible application areas.
? Trying a new technique on an existing problem. Possible techniques to explore are Differential Evolution
https://en.wikipedia.org/wiki/Differential_evolution
a GA type approach which is well supported in R (take a look at the DEoptim package and the paper, "Differential Evolution with DEoptim" by Ardia et al. included at the end of this topic), or Particle Swarm Optimisation (PSO) which is implemented by the PSO package. Again the Brabazon et al. identifies some techniques but you could also take a look at section 2 the attached paper by Katherine N. Mullen entitled "Continuous Global Optimization in R" which overviews a number of R implementations of algorithms you might wish to consider.
? Extend one of the techniques we have looked at so far. For example, we have only really touched the surface of algorithmic trading. As mentioned in the lectures there are various possible extensions here from using GAs to optimise the GP parameters through to evolving the actual complete trading rules.
? Use a different package for an existing technique. There are other implementations of the algorithms we have been using available - e.g.
https://cran.r-project.org/web/packages/GA/index.html
(and also the paper, "GA: A Package for Genetic Algorithms in R" by Luca Scrucca), but many others are available (again, take a look at the Mullen paper for some examples). This is a simpler option and probably only worth exploring if the packages offer significantly different functionality to the ones used already.
To ensure that your choice is a reasonable one, please send a short single slide or one-page document outlining what you plan to do first. This should be just a short overview of what your plans are: a brief summary of the problem you intend to tackle and the solution approach or technique you plan to investigate, and any other important information such as the package you intend to use, the availability of data etc.
The submission for this assignment will take the form of a 10-page (approximately) report. The structure of your report will vary according to your topic, but will typically contain the following sections:
? Introduction and overview of the study
? Background to the problem (e.g. technique, package, problem, dataset etc.)
? Experimental design (representation, fitness function, key parameters etc.)
? Results (and maybe comparison with other strategies)
? Analysis (comments on and insights into the results)
Conclusions