Stock prices prediction with Genetic Algorithm

Genetic algorithm implementation for investment portfolio optimization. Optimization criterion: sharpe = expected return / risk. The effect of risk free rate has been neglected. Data is collected from http://stooq.pl.

INSTALL

devtools::install_github('rpietrusinski/GA-stock-prices')

EXAMPLE

Clean workspace and load genalg.

rm(list=ls())
library(genalg)
library(corrplot)

Load data for given tickers, starting/ending point and data frequency ('d' -> daily).

tickers <- c("kgh","pko","peo","pkn", "ccc", "jsw", "tpe", "lpp", "eur", "cdr")
start <- 20180626
end <- 20180926
freq <- 'd'
data <- lapply(tickers, loadData, start=start, end=end, freq=freq)

Run prepareData function - it prepares data and global variables setup.

prepareData(data)

Run experiment.

Hyperparameters:

num_iter -> Number of iterations
num_spec -> Number of specimen
num_genes -> Number of bits encoding a single stock
p_cross -> Cross-over probability
p_mutant -> Mutation probability

num_iter = 200
num_spec = 400
num_genes = 10 
p_cross = 0.9
p_mutant = 0.01


result <- genalg(num_iter = num_iter,
                 num_spec = num_spec,
                 num_genes = num_genes,
                 p_cross = p_cross,
                 p_mutant = p_mutant)

ANALYSIS

performancePlot()
violinPlot()
riskVsReturnPlot()
optimumPortfolio()
comparePopulPlot()
corrplot(cor,  order='hclust')

The above functions have been created for algorithm's performance measurement as well as the optimum portfolio quality.

performancePlot - returns best/mean/stdev of results through generations
violinPlot - returns distributions of shares in best portfolios in each generation
riskVsReturnPlot - returns a risk vs return map with analyzed stocks as well as the optimum portfolios in each generation
optimumPortfolio - returns the optimum portfolio structure
comparePopulPlot - returns a risk vs return map for both first and last generation
corrplot - (from corrplot package) - returns correlation between considered stocks