CppBacktester

Docs

A C++ backtesting framework with Python bindings using pybind11 and ONNX Runtime.

Features

• Load and parse CSV ✅
• Load and save json config ✅
• Bar/data format ✅
• Backtesting engine ✅
• Broker class to manage order ✅
• Strategy class ✅
• Run a back test ✅
• Calculate metrics ✅
• ML Model Loading ✅
• API data fetching (not tested)

Building the Project

Prerequisites

• CMake 3.12 or higher
• Visual Studio 2022 (on Windows)
• Python with development headers (if building Python module)
• pybind11 (included in third_party github submodule)
• ONNX Runtime (automatically downloaded if not found)
• vcpkg (to install curl with openssl) vcpkg install curl[core,ssl] --triplet x64-windows-static

Build Options

The project has two main build targets:

• C++ executable (BUILD_EXECUTABLE=ON)
• Python module (BUILD_PYTHON_MODULE=ON)

Both are enabled by default.

Building with CMake

# Create and enter build directory
mkdir build
cd build

# Configure
cmake .. -G "Visual Studio 17 2022" -A x64

# Build
cmake --build . --config Release

Using the Helper Script

For convenience, you can use the provided batch script to clean and rebuild the project:

# On Windows
build.bat

Directory Structure

After building, the output files will be organized as follows:

build/
  output/
    bin/
      backtest_executable.exe  # C++ executable
      *.dll                    # Shared libraries (ONNX Runtime)
    lib/
      cppbacktester_py.pyd     # Python module

Architecture

This section provides an overview of the main classes in the CppBacktester framework and how they interact in a typical backtest.

Main Classes

• DataSource (abstract): Defines a common interface for data providers. Implemented by:

  • CSVDataSource: Loads market data from CSV files using CSVParserStep.
  • APIDataSource: Fetches data from external APIs.

• ParserStep (abstract): Base parser step. Specialized by:

  • CSVParserStep: Parses CSV input into Bar objects.
  • JSONParserStep: Parses JSON-formatted data or config.

• DataLoader: Orchestrates reading raw data via a DataSource and returns a sequence of Bar objects.

• BacktestEngine: Drives the backtest loop. Innitiate DataLoader, Broker, and Strategy.

• Broker: Manages, process order lifecycle, positions, and TP/SL logic.

• Strategy (abstract): Defines decision logic on incoming Bar data. Current implementations:

  • RandomStrategy: Generates random buy/sell orders with fixed TP/SL for testing.
  • MLStrategy: Loads a model via ModelInterface and issues orders based on model outputs.

• ModelInterface (abstract): Defines model loading and produce model input/output.

  • OnnxModelInterface: Implements ModelInterface using ONNX Runtime to load .onnx files.

• TradingMetrics: Calculates and stores aggregated metrics (sharpe ratio, P&L, drawdown).

Other classes:

• Bar: Represents a single market data bar containing open, high, low, close, volume, and any extra cols.
• Config: Reads and stores configuration settings from config.json.
• ColumnSpec: Defines the specification for CSV columns for parsing.
• Indicator: Base class for technical indicators computed on bar data.
• Order: Represents an order created by the broker during a backtest.
• Position: Tracks the state and lifecycle of an open position.
• Utils: Collection of utility functions used throughout the framework.

Workflow

1. Configuration: Read settings from config.json into Config.
2. Data Loading: DataLoader instantiates a DataSource (e.g., CSVDataSource).
3. Engine Setup: Create BacktestEngine, attach a Broker and chosen Strategy.
4. Backtest Loop: For each Bar:
   • Strategy issues TradingSignal
   • Broker executes orders
   • Metrics updated
5. Results: Export performance via TradingMetrics or Python bindings.