A protocol for evaluating quantitative strategies

This is the evaluation protocol used to assess whether a result is ready for publication in this library. It is a disclosure standard, not a statement that every Wealthoak model has completed each test.

1. Establish the data and capital basis

Record source, symbol, venue, timestamp convention, observation frequency and exact available dates. Identify missing observations and how they are handled. Futures require an explicit contract-roll rule; equities may require corporate-action and delisting treatment. Test history varies by instrument and dataset.

Record initial capital, position sizing, leverage, deposits and withdrawals. Returns from different capital bases cannot be compared simply by comparing currency profit.

2. Separate selection from evaluation

Record all parameter candidates considered and the rule used to select a candidate. Reserve an evaluation period that did not inform selection. If a result influences a design change, it becomes part of the development process and cannot continue to serve as untouched evidence.

For walk-forward evaluation, define each training window, evaluation window and update frequency before examining the results. Report the sequence of held-out periods rather than selecting the most attractive one.

3. Model the mechanics of execution

Disclose fees, spread, slippage, order type, fill timing, financing and liquidity assumptions. Historical bar data does not uniquely identify every possible intrabar price sequence. An emulator’s fill rules must be described, including any lower-timeframe data used.

TradingView documents its broker emulator, commissions, slippage and historical-testing limitations in its strategy documentation. Report settings and data exports with a result so those assumptions can be inspected.

4. Look for instability

Examine nearby parameter values, different periods, changing cost assumptions and adverse market regimes. A result that disappears after a small parameter or cost change deserves a narrower interpretation than a headline return suggests.

5. Publish a reproducible package

The package needs a versioned model description, source series or an explanation of access restrictions, calculation code, test settings, benchmark, limitations and report date. A checksum identifies the exact artifact; it does not certify the artifact’s accuracy.

Use the report template. Review performance disclosures and the current publication status.