Skip to content

Backtesting and optimization

Running a backtest

BacktestEngine simulates a strategy over historical data with realistic order handling. Choose the constructor by data type:

  • BacktestEngine.by_klines(strategy, data=(KlineData, ...)) - candle-driven.
  • BacktestEngine.by_ticks(strategy, data=(TickData, ...)) - tick-driven.
from tradetropy import BacktestEngine
from tradetropy.datasets import load_btcusd_1m

bt = BacktestEngine.by_klines(SmaCross(), data=(load_btcusd_1m(),))
bt.run()
print(bt.stats)
bt.plot()

By default the engine builds a simulator session with a starting balance and no commission. To configure commission, balance or spread, pass a pre-built SeshSimulatorBase as sesh=.

Metrics

bt.stats is a Stats object with the standard performance figures: return, annualized return and volatility, Sharpe/Sortino/Calmar, max and average drawdown (depth and duration), profit factor, win rate, SQN and more.

print(bt.stats['Return [%]'])
print(bt.stats['Max. Drawdown [%]'])
print(bt.stats['Sharpe Ratio'])

# Trades vs # Positions

# Trades counts every closed Trade row, including partial closes: a position closed in two partial closes produces two rows there. # Positions (and # Positions Long / # Positions Short) count distinct positions instead - partial closes of the same position are merged into one - so a strategy that scales out of its positions does not look like it trades far more often than it actually opens/closes positions. bt.stats.positions exposes the underlying per-position DataFrame (entry/exit, summed size and PnL, and n_partials), while bt.stats.trades keeps the row-per-close view.

Annualized metrics need enough data

Metrics like annualized return, Sharpe and Sortino are set to NaN when the backtest is shorter than a minimum span, or when there are too few closed trades. The bundled datasets are intentionally small, so you will see this warning in the examples - it is expected.

Warmup

The engine automatically reserves warmup bars so indicators are converged before the first on_data(). Recursive indicators (RSI, MACD) declare a warmup_factor so enough history is reserved. Size your proxy window_size to comfortably exceed the longest lookback you use.

Parallel backtests and optimization

PoolBacktestEngine runs many backtests in parallel across processes - the basis for parameter optimization and sweeps. Strategies and sessions are supplied as factories (picklable, since they are rebuilt in worker processes).

from tradetropy.backtest import PoolBacktestEngine

See the Engines reference for the full API.

Multi-symbol

Pass one data object per symbol. See Working with data for a complete multi-symbol example and how timestamp alignment works.