nanoback 0.5.1

C++20 event-driven trading backtester with Python bindings, snapshots, audit ledger, and microstructure-aware execution.

BACKTESTER

BACKTESTER is a low-latency, event-driven trading backtesting repo built around a C++20 execution engine and exposed through a Python package named nanoback.

What It Does

• Multi-asset backtests over contiguous matrix inputs
• Market and limit order simulation
• Partial fills, queue blocking, venue caps, and bid/ask-aware execution
• Parent/child order IDs with deterministic audit ledger output
• Snapshot/resume support for long-running simulations
• Risk controls for leverage, drawdown, cash, and per-asset limits
• Financing and borrow-cost accrual
• Compiled research policies and analytics in C++
• Python strategy/plugin fallback
• CSV and Parquet loaders
• Asserted benchmarks and pytest coverage

Repo Layout

• include/nanoback: C++ headers
• cpp: C++ engine, policies, and Python bindings
• python/nanoback: Python API, loaders, ledger utilities, and strategy helpers
• examples: runnable examples
• benchmarks: performance checks
• tests: regression and functional coverage

Quickstart

cd C:\Users\TAPESH\documents\BACKTESTEER
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -e .[dev]
python -m pytest

Core Capabilities

Execution Engine

• Event-driven C++20 core
• Market and limit orders
• Child-order slicing with delay steps
• Cancel/replace semantics
• Session-aware order cancellation
• Snapshot/resume from engine state

Market Realism

• Bid/ask execution path
• Queue-ahead fraction and venue volume share caps
• Slippage and participation-based impact
• Cost calibration helpers from empirical fill data

Risk and Audit

• Gross leverage checks
• Drawdown kill switch
• Cash and borrow accounting
• Per-asset max position and notional limits
• Deterministic ledger export to CSV/JSONL

Research Layer

• Compiled momentum, mean reversion, and moving-average crossover policies
• Compiled rolling volatility, cross-sectional ranking, and minimum-variance weights
• Python strategy hooks when custom event logic is needed

Minimal Example

import numpy as np
import nanoback as nb

result = nb.run_backtest(
    timestamps=np.array([1, 2, 3, 4], dtype=np.int64),
    prices=np.array([100.0, 101.0, 99.0, 102.0], dtype=np.float64),
    signals=np.array([1, 1, 0, -1], dtype=np.int64),
    config=nb.BacktestConfig(
        max_position=2,
        child_order_size=1,
        child_slice_delay_steps=1,
    ),
)

print(result.pnl)
print(len(result.ledger))

Benchmark

.\.venv\Scripts\python.exe benchmarks\benchmark_engine.py --mode latency
.\.venv\Scripts\python.exe benchmarks\benchmark_engine.py --mode stress --baseline benchmarks/benchmark_engine_stress_baseline.json

The benchmark now writes a latency log book to outputs/benchmark_engine_latency.jsonl and checks its results against benchmarks/benchmark_engine_baseline.json.

To track historical runs by release version, enable:

• --record-history --history-file benchmarks/benchmark_results_history.jsonl

Useful flags:

• --log-book to change the JSONL output path
• --baseline to compare against a different baseline file
• --update-baseline to refresh the baseline after an intentional change
• --regression-factor to control how much slower a stage may get before failing

Performance Notes

Measured on this repo with the current 0.2.0 code:

• Standard benchmark, 50_000 x 8
  • elapsed_seconds=0.0609
  • data_generation=0.0324s
  • policy_generation=0.0040s
  • engine_run=0.0246s
  • fills=61033
• Heavier simulation, 200_000 x 16
  • Full pipeline: elapsed_seconds=0.5472
  • data_generation=0.2933s
  • policy_generation=0.0762s
  • engine_run=0.1777s
• Isolated engine path for the same heavy simulation
  • elapsed_seconds=0.2233
  • policy_generation=0.0733s
  • engine_run=0.1499s

The isolated run shows the core backtest engine is fast and scales well, while the full pipeline is more constrained by Python-side data setup.

Institutional Reporting

Use the reporting helpers to generate a performance package for a completed run:

import nanoback as nb

result = nb.run_backtest(...)
report = nb.summarize_backtest(result, symbols=["AAA"])
nb.export_performance_report_json(report, "outputs/performance_report.json")
nb.export_performance_report_markdown(report, "outputs/performance_report.md")

The report includes:

• core PnL, return, drawdown, turnover, and fill-rate metrics
• equity and cash curve summaries
• audit-event counts
• asset-level execution summaries

This is the layer you would use for institutional-style post-trade review and fast root-cause analysis.

Status

The repo is strong as a research and simulation engine. It is not a full OMS/EMS, exchange adapter stack, or compliance platform.