Technical Indicator-Based Portfolio Strategy

This project implements and evaluates simple technical indicator-based trading strategies (SMA, EMA, RSI, MACD, Bollinger Bands) to construct quarterly rebalanced portfolios.

The strategies are benchmarked against the SPY ETF over the period from May 2022 to May 2025.

Objectives

• Build a backtesting framework for quarterly rebalanced portfolios.
• Generate trading signals using different technical indicators.
• Evaluate portfolio performance relative to the SPY benchmark.

Strategy Summary

• Indicators used:

  • Simple Moving Average (SMA)
  • Exponential Moving Average (EMA)
  • Relative Strength Index (RSI)
  • Moving Average Convergence Divergence (MACD)
  • Bollinger Bands (BB)

• Signal Logic:

  • Generate buy/sell/hold signals based on indicator-specific logic.
  • Adjust portfolio weights by ±3% per stock each quarter depending on the signal.
  • Normalise weights to sum to 1 on each rebalance date.

• Rebalance Frequency:

  • Quarterly, on the 3rd Friday of March, June, September, and December to align with S&P's rebalancing dates.

Key Findings (May 2022 – July 2025)

• From May 2022 to May 2024, MACD and Bollinger Bands outperformed the SPY benchmark in cumulative returns.
• After May 2024, the SPY benchmark began to outperform all strategies.
• Among all indicators:
  • MACD showed the most consistent outperformance early on but converged with SPY after mid-2024.
  • Bollinger Bands remained the second-best performing strategy.
  • SMA, EMA, and RSI trailed behind in cumulative return.
• As expected, no single technical indicator consistently outperforms across all market regimes.

Visualisation

Cumulative returns of all indicator strategies compared to SPY:

• Please refer to Visualisations in https://github.com/samsea18/ta-analysis/blob/main/ta-analysis/notebooks/POC%20Script.ipynb

Future Work

• Package POC into kedro application.
• API development on top of kedro application.
• Incorporate more advanced indicators (e.g. ATR, Stochastic Oscillator).
• Combine signals using ensemble or voting logic.
• Explore machine learning classifiers trained on technical indicators.
• Include short positions, transaction costs and slippage in the backtest.