EDDIE for Investment Opportunities Forecasting Michael Kampouridis http

EDDIE for Investment Opportunities Forecasting Michael Kampouridis http: //kampouridis. net/ Email: mkampo [at] essex [dot] ac [dot] uk

Outline Presentation of EDDIE 8 -TEACH demonstration Comprehensive exercises

EDDIE’s goal EDDIE is a GP tool that attempts to answer the following question: “Will the price of the X stock go up by r% within the next n days”? Users specify X, r, and n

How EDDIE works Training Data Financial Expert 1. Suggestion of indicators 5. Approval / rejection EDDIE Testing Data 4. Apply Training Data 2. Output 3. Evaluate Genetic Decision Tree (GDT)

How the training data is created Expert More Given adds: input: Daily 50 days 12 days Vol closing M. A. 80 50 90 82 52 99 83 53 87 82 51 82 …. . Define target: 4% in 20 days? 1 0 1 1 …. .

A typical GDT: EDDIE 8 Terminal s Function s If-then-else < Buy (1) Var. Constructor MA 12 6. 4 > If-then-else Not Buy (0) Buy (1) Var. Constructor 5. 57 Momentum 50

EDDIE 8: Technical Indicators Technical Indicator (Abbreviation) Moving Average (MA) Trade Break Out (TBR) Filter (FLR) Volatility (Vol) Momentum (Mom) Momentum Moving Average (Mom. MA)

GP Process Initialise population Calculate fitness of each tree in the population Selection of individuals for producing new offspring by the means of different genetic operators (e. g. crossover, mutation). These offspring form the new population Repeat the previous two steps for a number of generations N

Performance Measures Predictions Negativ e True Negativ e n n Reality Positive False Positive True Positive Negativ e Positive False Negativ Rate of Correctness (RC) = (TN + TP) Total e Rate of Failure (RF) = FP (FP + TP) Rate of Missing Chances (RMC) = FN (FN+TP) Fitness Function (ff) = w 1*RC-w 2*RMC-w 3*RF

Thanks • You can find these slides on my website, under the teaching tab: – http: //kampouridis. net/teaching/cf 963 • Any other material that we use today (EDDIE 8 Teaching, Lab sheet) can also be found there • If you have any questions, feel free to email me. I’m happy to arrange a meeting • EDDIE 8 -Teaching Demo + Comprehensive exercises

MSc dissertation topic • There a couple of extensions to EDDIE 8, which would fit very well as an MSc dissertation topic • You would be given the source code of EDDIE and be asked to add some new java code, which would be related to heuristic search methods – Java knowledge is required – No need to have implemented heuristics algorithms before. • You would then apply EDDIE 8 to a different stocks and investigate on the advantages of the introduction of heuristics to the search process of EDDIE 8 • Opportunity for those who are interested in a project that has real-life/industry application – Attract industry’s interest – Do actual research – Possibility of publishing the results in a paper

Supplementary Material

Constraints in the Fitness Function • ff = w 1’*RC-w 2*RMC-w 3*RF • Constraint R = [Cmin, Cmax] where Cmin = (Pmin/Ntr) x 100%, Cmax = (Pmax/Ntr) x 100%, 0<= Cmin <= Cmax <= 100% Ntr is the total number of training data cases Pmin is the minimum number of positive predictions required Pmax is the maximum number of positive predictions required If the percentage of positive signals predicted falls in the range of constraint R, then w 1’ = w 1. If not, then w 1’ = 0. In the latter case, the GDT is heavily penalized and ends up with a negative fitness function