| Abstract [eng] |
This research is dedicated to comparing two trading strategies in the context of multi-criteria optimization. Input data – stock closing prices taken from Nasdaq Baltic over the span of three years. Optimization criteria: transaction count and portfolio value at the last time period. One trading strategy simply trains from the training set and tests on the test data. While second – executes training and testing following the rolling walk-forward method. Indicators – cross-overs of closing price and parametrized moving averages. Optimization is achieved using NSGA-II algorithm with different input parameters: mutation, cross-over, number of generations and population size. Within the research it is proved that financial data is stochastic and that re-learning (walk-forward) strategy is on average superior based on portfolio value criteria and superiority grows with the number of generations of genetic algorithm because of the overfitting observed by the non-relearning strategy. In regards to the number of transactions criteria, the non-relearning strategy was on average superior. |
