%0 Conference Paper %1 JinLi:2000:CEF %A Li, Jin %A Tsang, Edward P. K. %B Computing in Economics and Finance %C Universitat Pompeu Fabra, Barcelona, Spain %D 2000 %K algorithms, genetic programming %T Reducing Failures in Investment Recommendations using Genetic Programming %U http://ideas.repec.org/p/sce/scecf0/332.html %X FGP (Financial Genetic Programming) is a genetic programming based system that specialises in financial forecasting. In the past, we have reported that FGP-1 (the first version of FGP) is capable of producing accurate predictions in a variety of data sets. It can accurately predict whether a required rate of return can be achieved within a user-specified period. This paper reports further development of FGP, which is motivated by realistic needs as described below: a recommendation "not to invest" is often less interesting than a recommendation "to invest". The former leads to no action. If it is wrong, the user loses an investment opportunity, which may not be serious if other investment opportunities are available. On the other hand, a recommendation to invest leads to commitment of funds. If it is wrong, the user fails to achieve the target rate of return. Our objective is to reduce the rate of failure when FGP recommends to invest. In this paper, we present a method of tuning the rate of failure by FGP to reflect the user's preference. This is achieved by introducing a novel constraint-directed fitness function to FGP. The new system, FGP-2, was extensively tested on historical Dow Jones Industrial Average (DJIA) Index. Trained with data from a seven-and-a-half-years period, decision trees generated by FGP-2 were tested on data from a three-and-a-half-years out-of-sample period. Results confirmed that one can tune the rate of failure by adjusting a constraint parameter in FGP-2. Lower failure rate can be achieved at the cost of missing opportunities, but without affecting the overall accuracy of the system. The decision trees generated were further analysed over three sub-periods with down trend, side-way trend and up trend, respectively. Consistent results were achieved. This shows the robustness of FGP-2. We believe there is scope to generalise the constrained fitness function method to other applications.

@inproceedings{JinLi:2000:CEF, abstract = {FGP (Financial Genetic Programming) is a genetic programming based system that specialises in financial forecasting. In the past, we have reported that FGP-1 (the first version of FGP) is capable of producing accurate predictions in a variety of data sets. It can accurately predict whether a required rate of return can be achieved within a user-specified period. This paper reports further development of FGP, which is motivated by realistic needs as described below: a recommendation {"}not to invest{"} is often less interesting than a recommendation {"}to invest{"}. The former leads to no action. If it is wrong, the user loses an investment opportunity, which may not be serious if other investment opportunities are available. On the other hand, a recommendation to invest leads to commitment of funds. If it is wrong, the user fails to achieve the target rate of return. Our objective is to reduce the rate of failure when FGP recommends to invest. In this paper, we present a method of tuning the rate of failure by FGP to reflect the user's preference. This is achieved by introducing a novel constraint-directed fitness function to FGP. The new system, FGP-2, was extensively tested on historical Dow Jones Industrial Average (DJIA) Index. Trained with data from a seven-and-a-half-years period, decision trees generated by FGP-2 were tested on data from a three-and-a-half-years out-of-sample period. Results confirmed that one can tune the rate of failure by adjusting a constraint parameter in FGP-2. Lower failure rate can be achieved at the cost of missing opportunities, but without affecting the overall accuracy of the system. The decision trees generated were further analysed over three sub-periods with down trend, side-way trend and up trend, respectively. Consistent results were achieved. This shows the robustness of FGP-2. We believe there is scope to generalise the constrained fitness function method to other applications.}, added-at = {2008-06-19T17:35:00.000+0200}, address = {Universitat Pompeu Fabra, Barcelona, Spain}, author = {Li, Jin and Tsang, Edward P. K.}, biburl = {https://www.bibsonomy.org/bibtex/276590580778a4b4db26d7fbc921f5835/brazovayeye}, booktitle = {Computing in Economics and Finance}, interhash = {c7734dd833ce35dc24f1a3e9701b0600}, intrahash = {76590580778a4b4db26d7fbc921f5835}, keywords = {algorithms, genetic programming}, month = {6-8 July}, notes = {http://enginy.upf.es/SCE/index2.html}, timestamp = {2008-06-19T17:45:31.000+0200}, title = {Reducing Failures in Investment Recommendations using Genetic Programming}, url = {http://ideas.repec.org/p/sce/scecf0/332.html}, year = 2000 }