LEARN · EN · easyquanttrading.com
How EasyQuant validates strategies — the 8-gate pipeline
A beautiful backtest proves nothing by itself. Any optimizer can find a strategy that looks perfect on history — the question is whether it survives data it never saw. EasyQuant runs every candidate through eight fixed verification gates in a fixed order. Only strategies that pass all of them are listed on the platform. This page explains what each gate checks, and why we refuse to paint untested stages green.
- Eight fixed gates — clients cannot reorder or skip them
- Signal health: empty shells and always-on signals are rejected first
- Look-ahead-free engine: bars computed strictly in order, no future data
- Out-of-sample defense: held-out period must not decay or flip negative
- Walk-forward and anchored walk-forward: rolling time extrapolation
- Monte Carlo: thousands of shuffled replays — profit must not depend on trade order
- DSR / PBO: selection bias priced in — how many trials were tried to get this result
- CPCV: combinatorial purged cross-validation — robust across every train/test split (Lopez de Prado)
Why eight gates, not one exam
A single held-out test is statistically weak: one unlucky or lucky segment can flip the verdict. Institutions therefore use multiple, complementary validations — each one measures a different failure mode. EasyQuant runs all eight on every candidate, in a fixed order, and a strategy is listed only when every hard gate passes.
Gate by gate
- 1. Signal health — rejects empty shells (no signals) and always-on signals (no timing). A strategy that never fires, or fires every bar, is filtered before anything else.
- 2. Discovery recheck — the authoritative re-run must agree with the discovery record. Divergent numbers mean the reported result was luck.
- 3. Defense (OOS + DSR/PBO) — a real backtest on a held-out period: out-of-sample decay above 70%, or a flip from profit to loss, blocks deployment. Deflated Sharpe (DSR) prices in how many trials were tried; PBO estimates overfitting probability.
- 4. Walk-forward — the strategy is trained on one window and tested on the next, repeatedly. A chain of out-of-sample segments that all hold up is far more credible than one full-sample curve.
- 5. Anchored walk-forward — training expands from the start while the test window rolls forward. Only a strategy that keeps earning as time moves on passes.
- 6. Re-optimized walk-forward — parameters re-fitted in each fold. If the edge only exists at the originally chosen parameters, that is an overfitting signal.
- 7. Monte Carlo — the trade sequence is shuffled thousands of times. If profit collapses when order is shuffled, the edge was path luck, not skill.
- 8. CPCV — combinatorial purged cross-validation (Lopez de Prado). The data is split into time-ordered blocks and every train/test combination is run; the strategy must be profitable across the distribution of test segments, not just one. Purge and embargo periods prevent information leakage between folds.
What 'listed' actually means
Passing all eight gates earns a strategy a place on the shelf and a 'deployable for paper trading' label — never a promise of live profit. You still observe it with simulated money first, and export the files to run on your own account. The evidence (passes and failures) travels with the strategy and cannot be rewritten by sellers.
FAQ
- What is CPCV?
- Combinatorial Purged Cross-Validation (from Lopez de Prado's Advances in Financial Machine Learning). Instead of one train/test split, the data is cut into time-ordered blocks and every combination of train/test is evaluated, with purge and embargo periods to stop information leakage. The result is a distribution of test performances, not a single lucky number.
- Why do you also run Monte Carlo if you already test out-of-sample?
- They test different things. Out-of-sample tests whether the strategy generalizes to unseen time. Monte Carlo shuffles the trade order to test whether the profit depends on a lucky sequence. A strategy can pass one and fail the other — both matter before listing.
- Can a strategy be listed without passing all gates?
- No. The eight gates are hard requirements in fixed order; untested stages stay grey and are never painted green. Skipping Monte Carlo is only allowed as a research override and is clearly marked 'not a full pass'.
- Does passing the pipeline guarantee live profits?
- Never. Passing means the strategy survived honest statistical scrutiny on historical data — it does not predict future returns. We promise inspectability, not returns.
More guides
Not investment advice. Historical results do not guarantee future performance. EasyQuant is a research factory — you execute on accounts you control.