Systematic Trading Backtesting Engine

A reusable Python framework for researching rule-based trading strategies with transaction costs, delayed execution, trade-level analytics, parameter selection and out-of-sample testing.

Python · pandas · NumPy · yfinance · matplotlib · pytest · systematic trading research

Project objective

The project is designed to answer a more useful question than “did this strategy make money historically?”:

Does the strategy still look reasonable after realistic execution assumptions, parameter selection and a genuine out-of-sample test?

The engine separates market data, strategy rules, backtesting logic and research/optimisation code into reusable modules. It compares four approaches under the same assumptions:

• Buy and Hold;
• Moving-Average Crossover;
• Momentum;
• Mean Reversion.

Research setup

The saved configuration in main.py uses:

Setting	Value
Instrument	SPY
Full sample	01 Jan 2010 – 01 Jan 2026
Train/test split	01 Jan 2019
Initial capital	$10,000
Transaction cost	0.10% per position change
Execution delay	1 trading period
Trading periods/year	252

Parameters are selected on the training period using Sharpe ratio and are then evaluated on the later testing period. Full-period results are retained only as a descriptive overview.

Selected strategy parameters

The saved optimisation output selected:

Moving Average: 50-day / 150-day
Momentum:       126-day lookback
Mean Reversion: 20-day lookback
                entry z-score = 1.5
                exit z-score  = 0.5

These parameters come from the repository's saved selected_parameters.json, rather than being chosen from the final test results.

Out-of-sample results

The testing output stored in the repository reports:

Strategy	Total return	Annualised return	Sharpe	Max drawdown	Completed trades
Buy and Hold	202.77%	17.16%	0.901	-33.72%	0
Moving Average	89.24%	9.55%	0.661	-33.72%	4
Momentum	67.99%	7.70%	0.647	-23.95%	27
Mean Reversion	11.41%	1.56%	0.177	-30.59%	33

The important portfolio lesson is not that one trading rule “wins”. In this saved SPY test, Buy and Hold produced the strongest return and Sharpe ratio, while the momentum strategy experienced the smallest maximum drawdown of the four. The comparison shows why strategy evaluation should include risk, turnover and trading costs rather than return alone.

Backtesting safeguards

Delayed execution

Signals are shifted before they affect positions. This reduces look-ahead bias by preventing the strategy from trading on information from the same bar that generated the signal.

Transaction costs

The engine charges costs when portfolio exposure changes, so frequent trading is explicitly penalised.

Train/test separation

Strategy parameters are selected using the pre-2019 training period. The post-split section is treated as the main evidence rather than optimising on the full historical sample.

Trade-level analysis

The engine records individual trades and reports metrics beyond portfolio returns, including:

• completed trades;
• trade win rate;
• average trade return;
• profit factor;
• market exposure;
• daily win rate.

Performance metrics

Each strategy is evaluated with a consistent set of portfolio statistics:

• final portfolio value;
• total and annualised return;
• annualised volatility;
• Sharpe ratio;
• Sortino ratio;
• maximum drawdown;
• daily win rate;
• market exposure;
• turnover/orders;
• trade-level performance.

Project structure

Backtesting-Engine/
├── README.md
└── backtesting engine/
    ├── backtester.py
    ├── market_data.py
    ├── strategies.py
    ├── research.py
    ├── main.py
    ├── main.ipynb
    ├── backtesting_engine_complete_theory_guide.ipynb
    ├── requirements.txt
    ├── tests/
    │   ├── conftest.py
    │   └── test_backtester.py
    └── outputs/
        ├── selected_parameters.json
        ├── testing_summary.csv
        ├── full_summary.csv
        ├── parameter-search outputs
        ├── daily strategy results
        ├── trade logs
        └── saved charts

Start here

For a portfolio review, I recommend this order:

1. README.md — project purpose, methodology and results.
2. backtesting engine/main.ipynb — end-to-end research workflow.
3. backtesting engine/backtesting_engine_complete_theory_guide.ipynb — detailed explanation of the backtesting concepts and common errors.
4. backtesting engine/backtester.py — reusable execution and portfolio engine.
5. backtesting engine/tests/ — automated checks for core backtesting behaviour.

How to run

cd "backtesting engine"
pip install -r requirements.txt
python main.py

main.py downloads/loads market data, performs the training-period parameter searches, runs the selected strategies on full/training/testing samples, exports CSV results and saves comparison charts to outputs/.

To experiment with another instrument or date range, edit the clearly marked SETTINGS section at the top of main.py.

Limitations

• This is a research backtester, not a live execution or portfolio-management system.
• Transaction costs are modelled as a fixed proportional rate; market impact, bid/ask spreads, taxes and liquidity constraints are not modelled separately.
• The saved results are for SPY and one train/test split; performance on other instruments or periods can differ substantially.
• Parameter search can still overfit the training period, so the out-of-sample section is more important than the in-sample results.
• Historical performance does not imply future profitability.

Skills demonstrated

Object-oriented/reusable Python design · systematic strategy research · market-data handling · bias-aware backtesting · transaction-cost modelling · signal execution · parameter optimisation · out-of-sample evaluation · risk metrics · trade analytics · automated testing

Educational and research project only — not investment advice.