Skip to content

Martingale42/nautilus-dev / skills/nt-backtest

nt-backtest

Verified

Use when working with backtesting engine, fill models, matching engine, simulated exchange, or backtest configuration in NautilusTrader.

nt-backtest

What This Skill Covers

NautilusTrader backtesting domain — backtest engine, simulated exchange, fill models, and matching logic.

Python modules: backtest/, backtest/models/, execution/matching_core (simulated exchange context) Rust crates: nautilus_backtest, nautilus_execution (matching subset)

When To Use

  • Setting up and running backtests (BacktestNode, BacktestEngine)
  • Configuring simulated exchanges and venues
  • Customizing fill models, latency models, fee models
  • Backtest configuration (BacktestRunConfig, BacktestVenueConfig)
  • Understanding matching engine behavior
  • Benchmarking backtest performance

When NOT To Use

  • Strategy/order logic → use nt-trading
  • Data loading/catalog → use nt-data
  • Live deployment → use nt-live
  • Indicator logic → use nt-signals

Python Usage

BacktestNode (Recommended)

BacktestNode is the high-level API for running backtests with configuration:

from nautilus_trader.backtest.node import BacktestNode
from nautilus_trader.config import (
    BacktestRunConfig,
    BacktestDataConfig,
    BacktestVenueConfig,
    BacktestEngineConfig,
)

config = BacktestRunConfig(
    engine=BacktestEngineConfig(
        strategies=[
            ImportableStrategyConfig(
                strategy_path="my_module:MyStrategy",
                config_path="my_module:MyStrategyConfig",
                config={"instrument_id": "ETHUSDT-PERP.BINANCE", ...},
            ),
        ],
    ),
    data=[
        BacktestDataConfig(
            catalog_path="/path/to/data",
            data_cls="nautilus_trader.model.data:Bar",
            instrument_id="ETHUSDT-PERP.BINANCE",
            bar_type="ETHUSDT-PERP.BINANCE-1-MINUTE-LAST-EXTERNAL",
        ),
    ],
    venues=[
        BacktestVenueConfig(
            name="BINANCE",
            oms_type="NETTING",
            account_type="MARGIN",
            base_currency=None,
            starting_balances=["10_000 USDT"],
        ),
    ],
)

node = BacktestNode(configs=[config])
results = node.run()

BacktestEngine (Direct API)

BacktestEngine provides lower-level control, useful for strategy testing:

from nautilus_trader.backtest.engine import BacktestEngine
from nautilus_trader.config import BacktestEngineConfig

engine = BacktestEngine(config=BacktestEngineConfig())

# Add venue
engine.add_venue(
    venue=Venue("SIM"),
    oms_type=OmsType.HEDGING,
    account_type=AccountType.MARGIN,
    base_currency=USD,
    starting_balances=[Money(1_000_000, USD)],
)

# Add data
engine.add_data(bars)
engine.add_instrument(instrument)

# Add strategy
engine.add_strategy(strategy)

# Run
engine.run()

# Get results
engine.trader.generate_order_fills_report()
engine.trader.generate_positions_report()

Venue Configuration

BacktestVenueConfig(
    name="SIM",
    oms_type="HEDGING",       # HEDGING or NETTING
    account_type="MARGIN",     # CASH or MARGIN
    base_currency="USD",
    starting_balances=["1_000_000 USD"],
    fill_model=FillModel(),    # Optional custom fill model
    # latency_model=LatencyModel(), # Optional latency simulation
)

Python Extension

Custom FillModel

from nautilus_trader.backtest.models import FillModel

class MyFillModel(FillModel):
    def __init__(self, ...):
        super().__init__()
        # TODO: Initialize model parameters

    # Override fill probability/slippage methods as needed

See templates/fill_model.py for full template.

Custom Fee Models

Configure fee structures per venue:

from nautilus_trader.model.objects import Money

# Via BacktestVenueConfig
BacktestVenueConfig(
    ...,
    fee_model=MakerTakerFeeModel(
        maker_fee=Decimal("0.0002"),
        taker_fee=Decimal("0.0004"),
    ),
)

Rust Usage

NautilusTrader provides two Rust APIs for backtesting: BacktestEngine (low-level) and BacktestNode (high-level with catalog streaming). Both run without Python.

Dependencies

Add to your Cargo.toml:

[dependencies]
nautilus-backtest = { version = "0.55", features = ["streaming"] }
nautilus-execution = "0.55"
nautilus-model = { version = "0.55", features = ["stubs"] }
nautilus-persistence = "0.55"
nautilus-trading = { version = "0.55", features = ["examples"] }

ahash = "0.8"
anyhow = "1"
tempfile = "3"
ustr = "1"

Drop streaming, nautilus-persistence, tempfile, ustr if only using the low-level BacktestEngine.

Feature flags:

FlagCrateEffect
high-precisionnautilus-model16-digit fixed precision (default 9). Required for crypto.
stubsnautilus-modelTest instrument stubs (audusd_sim, etc.)
examplesnautilus-tradingExample strategies (EmaCross, GridMarketMaker)
streamingnautilus-backtestCatalog-based data streaming via BacktestNode

BacktestEngine (Low-Level API)

Direct control: build engine, add venues/instruments, load data in memory, register strategies, run.

use ahash::AHashMap;
use nautilus_backtest::{config::BacktestEngineConfig, engine::BacktestEngine};
use nautilus_execution::models::{fee::FeeModelAny, fill::FillModelAny};
use nautilus_model::{
    enums::{AccountType, BookType, OmsType},
    identifiers::Venue,
    instruments::{Instrument, InstrumentAny, stubs::audusd_sim},
    types::{Money, Quantity},
};
use nautilus_trading::examples::strategies::EmaCross;

// 1. Create engine
let mut engine = BacktestEngine::new(BacktestEngineConfig::default())?;

// 2. Add venue (31 args — most are Option with None defaults)
engine.add_venue(
    Venue::from("SIM"),
    OmsType::Hedging,
    AccountType::Margin,
    BookType::L1_MBP,
    vec![Money::from("1_000_000 USD")],
    None,            // base_currency
    None,            // default_leverage
    AHashMap::new(), // per-instrument leverages
    None,            // margin_model
    vec![],          // simulation modules
    FillModelAny::default(),
    FeeModelAny::default(),
    None, // latency_model
    None, // routing
    None, // reject_stop_orders
    None, // support_gtd_orders
    None, // support_contingent_orders
    None, // use_position_ids
    None, // use_random_ids
    None, // use_reduce_only
    None, // use_message_queue
    None, // use_market_order_acks
    None, // bar_execution
    None, // bar_adaptive_high_low_ordering
    None, // trade_execution
    None, // liquidity_consumption
    None, // allow_cash_borrowing
    None, // frozen_account
    None, // queue_position
    None, // oto_full_trigger
    None, // price_protection_points
)?;

// 3. Add instrument and data
let instrument = InstrumentAny::CurrencyPair(audusd_sim());
let instrument_id = instrument.id();
engine.add_instrument(&instrument)?;
let quotes = generate_quotes(instrument_id);
engine.add_data(quotes, None, true, true);

// 4. Register strategy and run
let strategy = EmaCross::new(instrument_id, Quantity::from("100000"), 10, 20);
engine.add_strategy(strategy)?;
engine.run(None, None, None, false)?;

Run the example: cargo run -p nautilus-backtest --features examples --example engine-ema-cross

BacktestNode (High-Level API)

Loads data from ParquetDataCatalog and streams in configurable chunks. Requires streaming feature.

use nautilus_backtest::{
    config::{BacktestDataConfig, BacktestEngineConfig, BacktestRunConfig, BacktestVenueConfig, NautilusDataType},
    node::BacktestNode,
};
use nautilus_model::enums::{AccountType, BookType, OmsType};
use nautilus_persistence::backend::catalog::ParquetDataCatalog;
use ustr::Ustr;

// 1. Write data to catalog
let catalog = ParquetDataCatalog::new(catalog_path, None, None, None, None);
catalog.write_instruments(vec![instrument])?;
catalog.write_to_parquet(quotes, None, None, None)?;

// 2. Configure the run (configs use builder pattern)
let venue_config = BacktestVenueConfig::builder()
    .name(Ustr::from("SIM"))
    .oms_type(OmsType::Hedging)
    .account_type(AccountType::Margin)
    .book_type(BookType::L1_MBP)
    .starting_balances(vec!["1_000_000 USD".to_string()])
    .build();

let data_config = BacktestDataConfig::builder()
    .data_type(NautilusDataType::QuoteTick)
    .catalog_path(catalog_path.to_string())
    .instrument_id(instrument_id)
    .build();

let run_config = BacktestRunConfig::builder()
    .venues(vec![venue_config])
    .data(vec![data_config])
    .maybe_chunk_size(Some(100))
    .build();

// 3. Build, add strategies, run
let mut node = BacktestNode::new(vec![run_config])?;
node.build()?;

let engine = node.get_engine_mut("ema-cross-run").context("engine not found")?;
engine.add_strategy(EmaCross::new(instrument_id, Quantity::from("100000"), 10, 20))?;

node.run()?;

Run the example: cargo run -p nautilus-backtest --features examples,streaming --example node-ema-cross

Registering Actors

Actors register with add_actor (same pattern as strategies):

let actor = SpreadMonitor::new(instrument_id);
engine.add_actor(actor)?;

Warmup window: reserve the first day for calibration

Strategies that need historical data on on_start (indicator seeding, parameter calibration) request a window that ends at the run start. If your run starts at the catalog's earliest data, the request returns nothing.

Convention: ingest N+1 days, but set --date-from to the second day. The first ingested day becomes the 24h warmup window:

# Ingest 7 days
as-mm-ingest --date-from 2026-04-06 --date-to 2026-04-12 ...

# Backtest only 6 days — first day is warmup buffer
as-mm-backtest --date-from 2026-04-07 --date-to 2026-04-12 ...

request_quotes / request_trades issued at run start now have data to deliver. See nt-trading/references/guides/rust_patterns.md §2 for the warmup pattern itself.

Pre-load instruments before data

Backtest data cannot be loaded for an instrument the engine doesn't know about. Always:

// 1. Read instruments from catalog
let instruments = catalog.query_instruments(Some(vec![instrument_id]))?;
for inst in &instruments { engine.add_instrument(inst)?; }

// 2. Then load data
let quotes = catalog.query_quote_ticks(Some(vec![instrument_id]), None, None)?;
engine.add_data(quotes, None, true, true);

Same order in BacktestNode: instruments are loaded from catalog at build time; if your catalog lacks them, the run aborts before the strategy starts.

Fill model honesty

The default FillModelAny::default() is aggressor-style: every limit order is filled at the touch as soon as the market reaches it, ignoring queue priority. For market-making strategies this overstates fill rates by 2–10× vs production.

Implication: backtest PnL is not predictive of live PnL until you either (a) implement a queue-aware fill model, or (b) externally validate fill rates with shadow-mode live runs.

Document this loudly in your README — anyone reading backtest reports should know they're optimistic by construction.

Three-binary pattern

For real strategies, separate concerns into three binaries sharing one strategy:

strategies/my-strat/
├── src/strategy.rs   ← used by both backtest and live
├── bin/ingest.rs     ← venue archive → ParquetDataCatalog
├── bin/backtest.rs   ← BacktestEngine + strategy
└── bin/live.rs       ← LiveNode + strategy

Strategy code is identical between backtest and live — no #[cfg] branches. The only difference is which engine hosts it. See nt-trading/references/guides/rust_patterns.md §6 for the full pattern.

Rust Extension (PyO3 Path)

Performance-Optimized Fill Models

Rust fill models for complex matching logic (order book simulation, market impact):

use nautilus_model::types::Price;

pub struct MyFillModel {
    prob_fill_on_limit: f64,
    prob_slippage: f64,
}

impl MyFillModel {
    pub fn is_limit_filled(&self) -> bool { /* ... */ }
    pub fn is_slipped(&self) -> bool { /* ... */ }
}

Custom Matching Logic

The matching engine core lives in crates/execution/src/matching_core/. Extend it for custom matching behavior.

PyO3 Binding Conventions

  • Use #[pyclass] and #[pymethods] for Python-visible types
  • Register in crates/pyo3/src/lib.rs
  • Wrap FFI functions in abort_on_panic(|| { ... })
  • Use workspace dependency inheritance (serde = { workspace = true })

Key Conventions

BacktestNode vs BacktestEngine

  • BacktestNode: Configuration-driven, supports multiple runs, recommended for most use cases
  • BacktestEngine: Direct API, better for strategy unit tests and programmatic control

Benchmarking Practices

  • Use BacktestEngine timing for consistent benchmarks
  • Profile with py-spy or cProfile for hotspot identification
  • Compare against baseline runs with same data
  • See references/guides/benchmarking.md for detailed practices

Backtest Reproducibility

  • Same data + same config = same results (deterministic)
  • Set random_seed in fill model for stochastic fills
  • Log all configuration for reproducibility

Time Advancement

  • Backtest engine advances time event-by-event
  • Clock timers fire at their scheduled times during replay
  • ts_event on data determines processing order

References

  • references/concepts/ — backtesting, order book
  • references/api/ — backtest API
  • references/guides/ — benchmarking practices, benchmarking review checklist, run_rust_backtest
  • references/examples/ — clock timer, portfolio, cache usage, Rust backtests (engine_ema_cross, node_ema_cross), model configs
  • templates/ — fill_model.py
  • Cross-skill: nt-trading/references/guides/rust_patterns.md for strategy-side patterns (warmup, atomic state, three-binary), nt-trading/references/guides/troubleshooting_rust.md for debug workflow.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 197kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall