Project Overview
Objective: Analyze historical stock prices and optimize trading strategies using time-series techniques and Evolutionary Algorithms. Focus: Feature engineering, data transformation, discretization, and optimization workflow for stock price prediction.
Data Sources and Data characteristics
Simulated or publicly available financial datasets. Stocks and assets include: Gold, USD/EUR, Crude Oil, S&P 500. Irregular intervals (15m, 30m, 1h, 2h) and missing values handled.
Project Workflow
Data Ingestion & Preprocessing
Cleaned raw time-series data and aligned irregular intervals. Prepared datasets in time-series format for analysis.
Feature Engineering & Transformation
Calculated absolute and percentage price changes across time windows. Created lagged features for predictive modeling. Prepared labeled datasets for supervised learning using prediction horizons.
Discretization & Labeling
Converted percentage changes into categories (‘Rise’, ‘Fall’, ‘Steady’) using TAU matrices. TAU matrices generated via multiple modes: constant, quantile, random. Final labels standardized for modeling purposes.
Optimization & Evolutionary Algorithms
Integrated with Evolutionary Algorithms to optimize trading strategies. Developed fitness functions to evaluate TAU matrices and track strategy performance (budget_left). Managed EA population initialization and optimization loop.
Key Technical Highlights
Time-series manipulation and step-based historical calculations. Pseudo-code implementation for ABSOLUTE_CHANGE, PERCENTAGE_CHANGE, SHIFT_DATA, and ADD_LABELS functions. Workflow supports reproducibility for strategy optimization and data preparation.
Results / Insights
Labeled datasets ready for modeling. Framework demonstrates integration of time-series features with Evolutionary Algorithms. Provides a reproducible methodology for trading strategy analysis.
Explore alternative optimization algorithms (e.g., genetic programming, particle swarm).
Add visualizations and dashboards for strategy analysis.