Backtesting
Backtesting replays a strategy’s rules over historical data to see how it would have performed. Mandatory before going live — but an overfit backtest is more dangerous than none at all.
A backtest answers one question: had you traded these exact rules all along, what would have happened? Encode entries, exits and sizing into executable logic, simulate them bar by bar over history, and read off the equity curve, max drawdown, win rate and payoff ratio — the mandatory passage from “idea” to “system.”
The biggest trap is overfitting: parameters tuned until history looks perfect have simply memorized noise, and live trading exposes them instantly. Tell-tale signs: results collapsing under small parameter changes, too few trades, samples covering only one market regime. Antidotes: out-of-sample testing, parameter sensitivity analysis, and only using rules with a logical rationale.
The second trap family is execution bias: look-ahead (using data unknowable at the time), survivorship bias (testing only symbols that survived), and ignoring slippage and fees. Look-ahead is the sneakiest — signals revised after the fact score perfectly in backtests and simply do not exist in live trading.
CycleMaster’s cycle classification is structurally immune to look-ahead: every state label freezes the moment it prints and never repaints, so historical replay shows exactly the signals that were genuinely available at the time. Strategies backtested on those labels test the real decision flow — not hindsight.
FAQ
Coverage matters more than length: the sample should span at least one bull, one bear and one ranging market. A “holy grail” tested in a single regime is most likely overfit to that regime.
The usual four killers: overfitting, look-ahead bias, underestimated slippage and fees, and regime change. Live performance at 70% of backtest is normal; at 30%, audit those four — especially whether your signals repaint.
Related terms
CycleMaster labels every symbol with its cycle state and macro resonance verdict in real time — and signals never repaint.