A strategy without a falsifiable hypothesis is a story. Before you curve-fit history, write what you believe, why it might persist, and what would prove you wrong.
Mechanism first. Are you claiming a behavioural bias, a microstructure quirk, a macro regime effect, or a seasonal pattern? Each has different data needs and decay paths. “It worked on the chart” is not a mechanism.
Scope next. Instruments, session, horizon, and whether the edge is directional or relative-value. A rule that works on EUR/USD at London open may fail on the same symbol at Asia open.
Features must be knowable at decision time. No peeking at closes you could not have traded. Align bar timestamps, corporate actions (where relevant), and realistic execution delays.
Costs are part of the hypothesis. Model spread, slippage, financing, and turnover. If edge dies after costs, it was never deployable—only photogenic.
Regime dependence is a feature, not an embarrassment. State when the idea should be off: crisis volatility, policy pivots, illiquid holidays. If you cannot name off-regimes, you cannot size risk honestly.
Only then run backtests—and publish failure modes alongside any summary. Walk-forward and holdout samples reduce overfitting; they do not eliminate it.
VegaDeck does not rank, sell, or copy-trade strategies in v1.x. Use this for learning and internal research.
Educational only—not investment advice.
中文版：/academy/strategy-hypothesis-checklist-zh
Educational only · not investment advice · Risk disclosures