Core Concepts
The ideas behind the backtesting engine and how to think about results.
Event-driven backtesting
Quantinger uses an event-driven backtesting model. Signals fire on bar close and trades execute at the next bar's open. This is the only correct behaviour — using the signal bar's close price for entry is look-ahead bias, which inflates backtest results and leads to real losses when the strategy goes live.
Every trade in our engine is filled at next-bar open with slippage applied on top. Your backtest performance is therefore always slightly worse than the raw signal would suggest, and that is exactly right.
Slippage and fees
Default slippage is 0.05% on entry and 0.10% on exit. Default fees are 0.075% taker and 0.020% maker, matching standard Binance and Bybit rates. These defaults mirror real-world conditions for retail traders.
You can adjust both in strategy settings. If you are testing a strategy that trades during high-volatility periods, increase slippage. If you have a maker-only approach, use the maker fee.
Fees and slippage are always on. There is no switch to turn them off, because a backtest without costs is not a backtest.
Walk-forward efficiency
A single in-sample backtest tells you almost nothing. Walk-forward analysis splits your history into rolling train and test windows, optimises on each train window, then measures performance only on the untouched test window that follows.
Walk-forward efficiency is out-of-sample return divided by in-sample return. Below 0.50 means the strategy is largely fitted to the past. Above 0.70 suggests the edge generalises.
Monte Carlo and the bad path
One equity curve shows what happened once. Monte Carlo resamples your trade sequence a thousand times to show what could reasonably have happened instead, producing a distribution rather than a headline.
The number worth sizing against is the P95 drawdown — the loss you would suffer in the worst five percent of paths. If you cannot survive that number emotionally and financially, the strategy is too large for your account regardless of its average return.
Position sizing should be derived from P95 drawdown, not from the backtest's observed maximum drawdown.
Overfitting probability
The Probability of Backtest Overfitting (PBO) estimates how likely it is that your strategy's apparent edge is an artefact of selection across the parameter space you searched.
We gate at PBO below 0.15. A strategy with a beautiful curve and PBO of 0.71 is not a strategy — it is a memory of the data you tested on.