QuarterOscillate

Abstract

This project develops a comprehensive backtesting and optimization method for mean reversion trading strategies applied to the VN30 Index Futures market. The system combines technical indicators (Bollinger Bands with 1.8 standard deviations, 13-period RSI, and 14-period ATR) to identify overbought and oversold conditions that signal potential price reversals. Through modular architecture, the framework enables systematic testing of the hypothesis that prices revert to their short-term moving average after reaching extremes. Validation on historical market data demonstrates the system's effectiveness in capturing mean-reverting price movements while maintaining controlled risk parameters with 4×ATR take-profit and 1×ATR stop-loss levels.

Introduction

This project explores the application of mean reversion trading strategies to the VN30 Index Futures market (VN30F1M) as part of the Computer Finance course study. Mean reversion strategies operate on the principle that asset prices tend to revert to their historical mean or average over time. When prices deviate significantly from this mean—reaching overbought or oversold conditions—they create potential trading opportunities as the market corrects itself.

The system is built with a modular architecture that separates concerns into distinct components:

• Data loading and processing for VN30F1M futures
• Technical indicator calculation (SMA, Bollinger Bands, RSI, ATR)
• Signal generation based on band crossovers and momentum conditions
• Risk management with volatility-adjusted position sizing
• Backtesting engine with precise entry and exit execution
• Performance evaluation using standard financial metrics
• Parameter optimization to enhance strategy robustness

This separation allows for methodical strategy development and rigorous performance testing across different market conditions, enabling both academic analysis and practical trading applications.

Trading Hypotheses

The strategy aims to capture short-term price reversals using indicators like Bollinger Bands and RSI, with ATR-based dynamic risk management. It is built on the observation that VN30F1M prices tend to revert to the mean after reaching extreme levels.

Target Market

• Ticker: VN30F1M
• Timeframe: 15-minute
• Strategy Type: Mean Reversion

Entry Conditions

Buy Signal (Bullish Hypothesis)

• Price was below the lower Bollinger Band (BB20, 1.8) and has crossed back above it
• RSI(13) is below 30 (oversold)

Sell Signal (Bearish Hypothesis)

• Price was above the upper Bollinger Band and has crossed back below it
• RSI(13) is above 70 (overbought)

Indicators Used

• SMA(20): 20-period Simple Moving Average
• BB20 (1.8 STD): Bollinger Bands with 1.8 standard deviations
• RSI(13): 13-period Relative Strength Index
• ATR(14): 14-period Average True Range for volatility-based exits

Order Execution

• Entry: Market order upon signal confirmation
• Position Size: 1 contract per trade

Exit Conditions

Take-Profit

• Exit when unrealized profit reaches +4 × ATR(14)

Stop-Loss

• Exit when unrealized loss reaches -1 × ATR(14)

---

This systematic approach aims for consistent trade execution with clear entry/exit logic, leveraging volatility-based rules to adapt to changing market conditions.

Data

Data Collection

The system retrieves tick-by-tick data for VN30 Index Futures from a PostgreSQL database using the DataLoader class. The core functionality is implemented through the get_active_contract_data method, which:

1. Queries active front-month contract data by joining:

   • quote.matched table: price and timestamp data (columns: datetime, tickersymbol, price)
   • quote.futurecontractcode table: contract mapping information (futurecode='VN30F1M')
   • quote.total table: volume data (columns: datetime, tickersymbol, quantity)

2. Executes the following SQL query:

   SELECT m.datetime, m.tickersymbol, m.price, v.quantity
   FROM quote.matched m
   JOIN quote.futurecontractcode f ON m.tickersymbol = f.tickersymbol 
        AND DATE(m.datetime) = f.datetime
   LEFT JOIN quote.total v ON m.tickersymbol = v.tickersymbol 
        AND m.datetime = v.datetime
   WHERE m.datetime BETWEEN %s AND %s
   AND f.futurecode = 'VN30F1M'
   ORDER BY m.datetime

3. Implements automatic data caching using the CacheManager class to:

   • Store retrieved data in pickle format for improved performance
   • Generate unique cache keys based on date range and contract identifier
   • Automatically load cached data when available, bypassing database queries

Database configuration setup:

{
  "host": "<host or ip>",
  "port": <port>,
  "database": "<database name>",
  "user": "<username>",
  "password": "<password>"
}

Save as config/database.json

Data Processing

The DataProcessor class handles the transformation of raw tick data into analytical-ready format:

1. Contract Rollover Management: Automatically handles VN30 futures contract rollovers at month boundaries
2. OHLCV Generation: Converts tick data to configurable timeframe candles (default: 15-minute)
3. Trading Hours Filtering: Restricts data to regular trading hours (09:00-15:00)
4. Technical Indicator Calculation: Adds Bollinger Bands, RSI, and ATR indicators
5. Caching Implementation: Implements two-level caching:
   • Tick data caching in data_cache/contract_data/
   • OHLCV data caching in data_cache/ohlcv/

Implementation

Environment Setup

1. Clone the repository:

git clone https://github.com/algotrade-course/Group2.git
cd group2

2. Create and activate virtual environment:

python -m venv venv
source venv/bin/activate  # Unix/macOS
venv\Scripts\activate     # Windows

3. Install dependencies:

pip install -r requirements.txt

4. Configure database connection: Create config/database.json with your Algotrade PostgreSQL credentials. You can find this information in ps2-financial-data of the course in the database.json file. The format should be something like this:

{
  "host": "example.vn",
  "port": 8080,
  "database": "database",
  "user": "username",
  "password": "password"
}

Running the Strategy

Running In-Sample Backtest

python src/run_backtest.py

Parameter Optimization

python src/run_optimization.py

Running Out-of-Sample Backtests

1. Using Reference Parameters (Default)

python src/run_backtest_outsample.py

Note: This will produce identical results to those shown in the README. For reproducible purpose, running this script is enough but you are open to try different scripts and parameters

2. Using Custom Optimized Parameters

python src/run_backtest_outsample.py --custom

3. With Custom Date Range

python src/run_backtest_outsample.py --start 2024-06-01 --end 2025-01-01

The reference optimized parameters are:

{
    "bb_window": 23,
    "bb_std": 1.1,
    "rsi_period": 11,
    "rsi_lower": 37,
    "rsi_upper": 64,
    "atr_period": 19,
    "take_profit_mult": 5.6,
    "stop_loss_mult": 1.2,
    "default_timeframe": "5min"
}

Notebooks

For interactive analysis:

• notebooks/updated_backtest.ipynb - Step-by-step backtest walkthrough
• notebooks/updated_optimization.ipynb - Optimization process visualization
• notebooks/run_backtest_outsample.ipynb - Outsample backtesting and the notebook includes an option to toggle between reference and custom parameters:

# Choose whether to use reference parameters (True) or custom optimized parameters (False)
use_reference_parameters = True

Configuration

Strategy parameters are configured in config/strategy_config.json:

{
    "parameters": {
        "bb_window": 20,
        "bb_std": 1.8,
        "rsi_period": 13,
        "rsi_lower": 30,
        "rsi_upper": 70,
        "atr_period": 14,
        "take_profit_mult": 4.0,
        "stop_loss_mult": 1.0,
        "commission":0.001,
        "risk_free_rate": 0.03
    }
}

Reproducibility Note:

All the illustration and metrics shown here should be found in the results folder along with additional illustration. Please note that the result from the optimization will vary. To produce the same results as this README, please run the run_backtest.py / updated_backtest.ipynb as it is and run the run_backtest_outsample.py / run_backtest_outsample.ipynb with the reference parameters. or in short: For python:

python src/run_backtest.py

and

python src/run_backtest_outsample.py

For notebook:

• notebooks/updated_backtest.ipynb
• notebooks/run_backtest_outsample.ipynb

In-sample Backtesting

Parameters

Using default configuration:

• Bollinger Bands: 20-period, 1.8 std dev
• RSI: 13-period, 30/70 thresholds
• ATR: 14-period
• Take-profit: 4×ATR, Stop-loss: 1×ATR

Results

Period: 2024-01-01 to 2024-06-01

Metric	Value
Win Rate	27.27%
Profit Factor	0.68
Sharpe Ratio	-2.23
Maximum Drawdown	-8.35%
Average Trade Return	-108.17
Total Return	-5.95%
Total Trades	55
Average Win	854.07
Average Loss	-469.02
Expectancy	-108.17

Default parameters equity curve showing portfolio value over time

Detailed price chart with entry/exit signals and equity curve

Trade Analysis

Profit/loss distribution of individual trades

Breakdown of trade exits by reason (take-profit, stop-loss, time-exit, Time-exit)

Optimization

Optimization Objective Function

The optimization process uses Optuna to maximize a custom scoring function implemented in the StrategyOptimizer.objective method. The scoring algorithm employs a multi-factor approach:

1. Primary Score Calculation:

   score = sharpe_ratio - |0.1 * max_drawdown| + (total_trades / 1000)

   This combines:

   • Sharpe ratio as the primary performance metric
   • Penalty for maximum drawdown (coefficient: 0.1)
   • Bonus for trade count to avoid overly passive strategies

2. Conditional Scoring Logic:

   • If total trades < 10: Score = -1.0 (invalid strategy)
   • If Sharpe ratio ≤ 0: Score = total_return × 10 (fallback to return-based scoring)
   • Otherwise: Uses the multi-factor formula above

3. Error Handling: Returns -1.0 for any failed trials to ensure robust optimization

This scoring function balances risk-adjusted returns (Sharpe ratio), risk control (drawdown penalty), and strategy activity level, ensuring the optimization process identifies robust trading parameters.

Process

Using Optuna for optimization

• Number of trials: 100
• Objective: Maximize Predefined score
• Timeframes tested: 5min, 15min, 30min, 1h
• Parameter ranges:
  • BB window: 10-50
  • BB std: 1.0-3.0
  • RSI period: 5-30
  • ATR period: 5-30
  • Take-profit multiplier: 2.0-6.0
  • Stop-loss multiplier: 0.5-2.0

Results

Best parameters found:

{
    "timeframe": "5min",
    "bb_window": 23,
    "bb_std": 1.1,
    "rsi_period": 11,
    "rsi_lower": 37,
    "rsi_upper": 64,
    "atr_period": 19,
    "take_profit_mult": 5.6,
    "stop_loss_mult": 1.2
}

Optimization Analysis

Optimization score progression over 100 trials (Best score: ~1.5)

Parameter impact on optimization score (correlation coefficients)

Parameter importance interpretation:

• take_profit_mult (+0.44): Strong positive correlation
• timeframe_num (+0.42): Strong positive correlation
• bb_std (-0.50): Strong negative correlation
• bb_window (-0.37): Moderate negative correlation
• rsi_upper (-0.33): Moderate negative correlation
• stop_loss_mult (-0.17): Slight negative correlation
• rsi_lower (+0.26): Mild positive correlation
• atr_period (+0.10): Minimal positive impact
• rsi_period (-0.03): Negligible negative impact

Stronger coefficients (closer to ±1) indicate more influential parameters.

Timeframe Performance Summary

Timeframe	Total Trades	Win Rate	Sharpe Ratio	Max Drawdown	Score (Mean)	Score (Max)
15min	63.90	0.434	-1.264	0.0398	0.5272	1.3842
1h	13.13	0.331	-8.255	0.0200	-0.5675	-0.0077
30min	24.29	0.493	-4.274	0.0300	-0.1174	0.3505
5min	97.92	0.413	-2.854	0.0642	0.0521	1.5008

Out-of-sample Backtesting

Parameters

Using optimized parameters from previous step

Results

Period: 2024-06-01 to 2025-01-01

Metric	Value
Total Return	15.26%
Sharpe Ratio	-1.32
Maximum Drawdown	-6.90%
Win Rate	39.21%
Profit Factor	1.20
Total Trades	227
Average Trade Return	67.24
Average Win	1029.24
Average Loss	-553.19
Expectancy	67.24

Optimized parameters performance on validation data

Comparison with Default Parameters

Metric	Default	Optimized	Improvement	Improvement %
Win Rate	36.84%	39.21%	+2.37%	+6.42%
Profit Factor	1.03	1.20	+0.17	+16.40%
Sharpe Ratio	-1.74	-1.32	+0.41	+23.84%
Max Drawdown	-5.33%	-6.90%	-1.57%	-29.37%
Total Return	0.68%	15.26%	+14.59%	+2159.68%
Total Trades	76	227	+151	+198.68%

Performance metrics: Default vs. Optimized parameters

Portfolio value comparison between default and optimized parameters

Conclusion

Since the initial hypothesis, the team has successfully developed and implemented a comprehensive trading system that incorporates backtesting, optimization, and visualization components. The strategy, based on mean reversion principles using Bollinger Bands, RSI, and ATR, has been effectively applied to the VN30 Index Futures market, yielding results that align with the original hypothesis of price reverting to the mean after significant deviations.

The optimization process has proven valuable, enhancing performance by fine-tuning parameters. However, it is clear that optimization doesn't always lead to the best set of parameters. At times, it can generate false signals, particularly when market conditions change or when applied to different timeframes. The results show that timeframes play a crucial role in the strategy's performance, with shorter timeframes providing more frequent signals but also more noise, while longer timeframes offer fewer trades but higher win rates.

While the system has been implemented successfully, it can still be considered somewhat naive in its approach. There is room for improvement, particularly in refining the optimization process to avoid overfitting and reducing the risk of false signals. Additionally, further work is needed to make the strategy more adaptive to varying market conditions.

In conclusion, while the strategy is viable and demonstrates potential, future improvements could focus on refining the optimization process, incorporating market condition filters, and exploring different timeframes for better performance consistency.

---