Monte Carlo robustness

A single equity curve does not tell you whether a result is robust or just a lucky ordering of trades. The tradetropy.robustness module runs a Monte Carlo test: it generates many randomized variants of the result and reports the distribution of the metrics - confidence intervals, probability of loss, risk of ruin and a composite robustness score.

From a finished backtest

from tradetropy import BacktestEngine
from tradetropy.datasets import load_btcusd_1m

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

mc = bt.montecarlo(
    n_sims     = 1000,
    methods    = ['shuffle_order', 'resample_trades'],
    confidence = (0.95, 0.99),
    seed       = 42,
)

print(mc.summary())              # per-metric table: original, mean, std, p5, p50, p95, CI
print(mc.probability_of_loss)    # P(final equity < initial balance)
print(mc.risk_of_ruin(0.5))      # P(drawdown exceeds 50% of equity)
print(mc.robustness_score)       # composite score in [0, 100]
mc.plot(theme='dark')            # equity cone + metric histograms

Methods and levels

Methods operate at one of three levels and cannot be mixed across levels in a single call:

• Trade level (fast, no re-run; perturb the closed-trade list): shuffle_order, resample_trades, skip_trades, randomize_slippage, random_start_index.
• Data level (re-runs the engine in parallel): randomize_prices, random_start_bar.
• Parameter level (re-runs the engine): randomize_parameters.

The explicit API gives full control and lets you compose methods of the same level:

from tradetropy.robustness import MonteCarlo, MonteCarloConfig
from tradetropy.robustness.methods import ShuffleOrder, ResampleTrades, SkipTrades

result = MonteCarlo(bt, MonteCarloConfig(
    n_sims     = 2000,
    methods    = [ShuffleOrder(), ResampleTrades(replace=True), SkipTrades(prob=0.1)],
    metrics    = ['Return [%]', 'Max. Drawdown [%]', 'Sharpe Ratio', 'Profit Factor'],
    confidence = (0.95, 0.99),
    seed       = 42,
)).run()

result.percentile('Max. Drawdown [%]', 5)
result.confidence_interval('Return [%]', 0.95)
result.to_dataframe()

The composite robustness_score (0-100) is a transparent weighted blend of profitability, drawdown safety and consistency.