Markov Chain-Based Financial Prediction System

Github Link: github.com/MetinUnlu/markovchain_finance

This project implements a comprehensive financial prediction system using Markov chains to model stock market behavior and make trading decisions. The system analyzes historical stock data, creates state-based models, and generates predictions for future price movements with threshold-based decision making.

Project Overview

The system utilizes Markov chains to model transition probabilities between different market states, calculate the expected value of each state, and predict outcomes. If the predicted value exceeds a predefined threshold, trading orders are executed. The project demonstrates both the potential and limitations of Markov chain applications in financial markets.

Mathematical Foundation

The core of this system is based on mathematical modeling of state transitions:

State Transition Probability: P_ij = (Number of transitions from i to j) / (Total transitions from i)
Expected Value Calculation: E(s_i) = Σ P_ij × V_j
Threshold-Based Decision Making: Orders are placed when O(s_i) > θ

Technical Implementation

The system processes Google stock data spanning from 2004 to 2024, implementing several key components:

Technical Indicators

Short-term EMA (14-day): Tracks recent price movements
Long-term EMA (50-day): Reflects established price trends
RSI (14-day): Identifies overbought/oversold conditions
Close Gap: Percentage change in closing prices

State Creation and Pattern Recognition

The system creates complex state patterns by combining:

Short and Long EMA gap classifications (Low/Medium/High)
RSI signals (Overbought/Oversold/Normal)
Binary state determination based on price movement direction

Dataset and Preprocessing

The project uses comprehensive Google stock data with the following features:

Date, Open, High, Low, Close, Volume
Calculated technical indicators (EMA, RSI)
State patterns generated from indicator combinations
Transition matrices for state-to-state probability calculations

Results and Performance Analysis

Key Findings

The system was tested on the last 10% of the dataset (validation period), revealing important insights:

Strategy Performance: Achieved 16% profit using threshold-based decisions
Comparison with Buy-and-Hold: Simple buy-and-hold strategy yielded 58% profit over the same period
Accuracy: The prediction model achieved 53% accuracy
Threshold Impact: Without threshold (threshold=0), the system resulted in 92% loss

Honest Assessment

While the Markov chain approach shows promise, this implementation demonstrates several limitations:

The strategy underperformed compared to simple buy-and-hold
The model favored 'buy' predictions due to Google's general upward trend
False positives in trading decisions significantly impacted returns
The threshold mechanism improved performance but couldn't overcome fundamental limitations

Technologies and Tools

Python: Primary programming language
Pandas & NumPy: Data manipulation and numerical computations
Matplotlib: Data visualization and result plotting
Jupyter Notebook: Interactive development and analysis
Technical Analysis Libraries: For calculating financial indicators

Key Insights and Lessons Learned

This project provided valuable insights into both the potential and limitations of Markov chains in financial prediction:

State Design is Critical: The success of Markov chain methods heavily depends on creating meaningful states that capture market behavior
Threshold Implementation: Risk management through thresholds is essential for practical trading applications
Market Complexity: Simple technical indicators may not capture the full complexity of market dynamics
Benchmark Comparison: Always compare against simple baseline strategies

Future Improvements

The project identifies several areas for enhancement:

Reinforcement Learning Integration: Implement RL to create adaptive expected states
Enhanced Indicators: Incorporate more sophisticated technical analysis tools
Multi-Asset Analysis: Extend the approach to multiple securities
Risk Management: Implement proper position sizing and stop-loss mechanisms

Conclusion

This Markov chain-based financial prediction system serves as an educational exploration of quantitative finance methods. While the current implementation shows limitations compared to simple strategies, it demonstrates the importance of rigorous testing and honest evaluation in financial modeling. The project highlights that successful algorithmic trading requires sophisticated state modeling, robust risk management, and realistic performance expectations.

The methodology is repeatable with any stock and time range, making it a valuable learning tool for understanding both the potential and pitfalls of mathematical approaches to financial prediction.

Markov Chains
Financial Modeling
Technical Analysis
Quantitative Finance

State Modeling
Risk Management
Trading Systems
Time Series Analysis