Backtest (high-level API)

Tutorial for NautilusTrader a high-performance algorithmic trading platform and event driven backtester.

View source on GitHub.

Overview

This tutorial walks through how to use a BacktestNode to backtest a simple EMA cross strategy on a simulated FX ECN venue using historical quote tick data.

The following points will be covered:

• How to load raw data (external to Nautilus) into the data catalog
• How to set up configuration objects for a BacktestNode
• How to run backtests with a BacktestNode

Prerequisites

• Python 3.11+ installed
• JupyterLab or similar installed (pip install -U jupyterlab)
• NautilusTrader latest release installed (pip install -U nautilus_trader)

Imports

We'll start with all of our imports for the remainder of this tutorial.

import shutil
from decimal import Decimal
from pathlib import Path

import pandas as pd

from nautilus_trader.backtest.node import BacktestDataConfig
from nautilus_trader.backtest.node import BacktestEngineConfig
from nautilus_trader.backtest.node import BacktestNode
from nautilus_trader.backtest.node import BacktestRunConfig
from nautilus_trader.backtest.node import BacktestVenueConfig
from nautilus_trader.config import ImportableStrategyConfig
from nautilus_trader.core.datetime import dt_to_unix_nanos
from nautilus_trader.model.data import QuoteTick
from nautilus_trader.persistence.catalog import ParquetDataCatalog
from nautilus_trader.persistence.wranglers import QuoteTickDataWrangler
from nautilus_trader.test_kit.providers import CSVTickDataLoader
from nautilus_trader.test_kit.providers import TestInstrumentProvider

As a once off before we start the notebook - we need to download some sample data for backtesting.

For this example we will use FX data from histdata.com. Simply go to https://www.histdata.com/download-free-forex-historical-data/?/ascii/tick-data-quotes/ and select an FX pair, then select one or more months of data to download.

Once you have downloaded the data, set the variable DATA_DIR below to the directory containing the data. By default, it will use the users Downloads/Data/ directory.

DATA_DIR = "~/Downloads/Data/"

path = Path(DATA_DIR).expanduser() / "HISTDATA"
raw_files = list(path.iterdir())
assert raw_files, f"Unable to find any histdata files in directory {path}"
raw_files

Loading data into the Parquet data catalog

The FX data from histdata is stored in CSV/text format, with fields timestamp, bid_price, ask_price. Firstly, we need to load this raw data into a pandas.DataFrame which has a compatible schema for Nautilus quotes.

Then we can create Nautilus QuoteTick objects by processing the DataFrame with a QuoteTickDataWrangler.

# Here we just take the first data file found and load into a pandas DataFrame
df = CSVTickDataLoader.load(raw_files[0], index_col=0, datetime_format="%Y%m%d %H%M%S%f")
df.columns = ["timestamp", "bid_price", "ask_price"]

# Process quotes using a wrangler
EURUSD = TestInstrumentProvider.default_fx_ccy("EUR/USD")
wrangler = QuoteTickDataWrangler(EURUSD)

ticks = wrangler.process(df)

See the Loading data guide for further details.

Next, we simply instantiate a ParquetDataCatalog (passing in a directory where to store the data, by default we will just use the current directory). We can then write the instrument and tick data to the catalog, it should only take a couple of minutes to load the data (depending on how many months).

CATALOG_PATH = Path.cwd() / "catalog"

# Clear if it already exists, then create fresh
if CATALOG_PATH.exists():
    shutil.rmtree(CATALOG_PATH)
CATALOG_PATH.mkdir(parents=True)

# Create a catalog instance
catalog = ParquetDataCatalog(CATALOG_PATH)

# Write instrument to the catalog
catalog.write_data([EURUSD])

# Write ticks to catalog
catalog.write_data(ticks)

Using the Data Catalog

Once data has been loaded into the catalog, the catalog instance can be used for loading data for backtests, or simply for research purposes. It contains various methods to pull data from the catalog, such as .instruments(...) and quote_ticks(...) (shown below).

catalog.instruments()

start = dt_to_unix_nanos(pd.Timestamp("2020-01-03", tz="UTC"))
end =  dt_to_unix_nanos(pd.Timestamp("2020-01-04", tz="UTC"))

catalog.quote_ticks(instrument_ids=[EURUSD.id.value], start=start, end=end)[:10]

instrument = catalog.instruments()[0]

Add venues

venue_configs = [
    BacktestVenueConfig(
        name="SIM",
        oms_type="HEDGING",
        account_type="MARGIN",
        base_currency="USD",
        starting_balances=["1_000_000 USD"],
    ),
]

Add data

data_configs = [
    BacktestDataConfig(
        catalog_path=str(ParquetDataCatalog.from_env().path),
        data_cls=QuoteTick,
        instrument_id=instrument.id,
        start_time=start,
        end_time=end,
    ),
]

Add strategies

strategies = [
    ImportableStrategyConfig(
        strategy_path="nautilus_trader.examples.strategies.ema_cross:EMACross",
        config_path="nautilus_trader.examples.strategies.ema_cross:EMACrossConfig",
        config={
            "instrument_id": instrument.id,
            "bar_type": "EUR/USD.SIM-15-MINUTE-BID-INTERNAL",
            "fast_ema_period": 10,
            "slow_ema_period": 20,
            "trade_size": Decimal(1_000_000),
        },
    ),
]

Configure backtest

Nautilus uses a BacktestRunConfig object, which enables backtest configuration in one place. It is a Partialable object (which means it can be configured in stages); the benefits of which are reduced boilerplate code when creating multiple backtest runs (for example when doing some sort of grid search over parameters).

config = BacktestRunConfig(
    engine=BacktestEngineConfig(strategies=strategies),
    data=data_configs,
    venues=venue_configs,
)

Run backtest

Now we can run the backtest node, which will simulate trading across the entire data stream.

node = BacktestNode(configs=[config])

results = node.run()
results