REPLYOPSAI RESEARCH NOTE
RESEARCH NOTE

Backtest vs PAPER Trading

A backtest asks how rules would have behaved on historical data. A forward PAPER test asks what the frozen system does as new data arrives. They answer different questions.

Why a backtest can look better

Repeated optimization can fit noise in historical data. Data leakage, unrealistic fills, omitted fees and selecting only successful variants can also inflate apparent performance.

What forward PAPER adds

The future sequence is unknown when the version starts. That makes forward observation useful for detecting regime sensitivity, operational failures and behavior that was hidden by historical optimization.

What each evidence stage can answer

STAGEUSE IT FOR
BacktestRejecting weak ideas, estimating trade frequency and finding obvious historical failure.
Forward PAPERTesting a frozen version on unseen sequence, operational behavior and recorded net outcomes.
LiveMeasuring real fills, latency, liquidity constraints and capital-risk behavior.

Passing one stage is permission to investigate the next stage—not proof that the next stage will work.

PAPER still has limitations

PAPER execution is simulated. Real slippage, liquidity, latency and exchange constraints can differ, so PAPER profitability should not be presented as live-money profitability.

Our evidence ladder

BACKTEST → PAPER → LIVE

Each stage can answer additional questions, but success at one stage does not guarantee success at the next.

See current PAPER experiments →

Continue with forward evidence