What data is needed to assess False Breakout Filtering?

False breakout filtering: what it is

False breakout filtering is a way to reduce or exclude breakout attempts that quickly fail. A “breakout” must be defined mechanically (for example, a price moving beyond a level for a specified lookback window). A “false” breakout typically means the move does not follow through and instead returns back into the prior range within a time or distance condition.

To assess whether a filtering method is coherent and measurable, you need data that lets you (1) identify breakouts, (2) classify them as false or not, and (3) reproduce the same results under clearly stated assumptions.

Data inputs you need

Start with the minimum dataset that supports the definitions.

Instrument and market identifiers

The exact trading instrument name and trading venue or pricing source.
The quote convention (for example, whether you observe bid/ask or mid prices).

Price series with timestamps

Open, high, low, close (OHLC) and/or tick or bar data.
Full timestamp coverage with a clear timezone and bar boundaries.
Consistent sampling: the same timeframe definition for breakout detection and for the “false” window.

Breakout rule parameters

The lookback window used to define the prior range or level.
The threshold for “breakout” (e.g., close beyond level versus any touch).
Whether you require multiple confirmations (such as more than one bar) and how long they last.

False-failure rule parameters

The condition for “false”: return distance (how far back) and return timing (how many bars or how much time).
Whether the filter uses close prices, intrabar highs/lows, or bid/ask.

Assumptions needed for calculations

How you handle spreads, slippage, and execution timing. If you do not have trade/quote data, you must still state what substitute you used (e.g., mid price) and that this changes results.
How you treat missing bars, outliers, and corporate-like changes (rare but possible in some datasets).

Provenance, timeliness, and quality checks

Two datasets can look identical while hiding incompatible details. For assessment, you need provenance (where the data came from), timeliness (when it reflects reality), and quality checks.

Provenance you can verify

Data source identity: which provider or platform produced it.
Whether the series is reconstructed, curated, or directly captured.
Documentation for how the series was aggregated into bars (especially for OHLC).

Timeliness and alignment

The period covered by the dataset.
Any known preprocessing such as timezone conversion or daylight saving handling.
Evidence that the breakout detection and false classification use the same clock and bar definitions.

Data-quality checks (what to look for)

Missing data rate and how gaps are filled or excluded.
Consistency between OHLC and intrabar/tick data (if both exist).
Outlier handling rules: for example, whether spikes are removed or kept.
Replayability: you should be able to rerun the same computation on the same input and get the same classification.

Evidence or example: turning rules into measurable classifications

An assessor needs to connect the data to outcomes in a reproducible way.

Example framework (assumptions must be stated):

Define the breakout level from the prior N bars’ range.
Mark a breakout when the close exceeds that level.
Mark it as “false” if the price closes back inside the prior range within M bars.

With the required data, you can compute counts such as:

Total breakouts detected under the rule.
Proportion classified as false under the timing/distance condition.
Sensitivity to parameters (e.g., changing N or M).

Important: this shows measurability of the filter definition, not guaranteed improvement. Historical classification quality may change with volatility regimes and market microstructure.

Limitations and risks (material failure modes)

Even with correct data, false breakout filtering can fail in identifiable ways.

Definition mismatch If the “breakout” definition uses close prices but “false” uses intrabar extremes (or vice versa), classifications become inconsistent and difficult to compare.
Microstructure effects If you ignore bid/ask or execution timing, the classification may not reflect what a trader could realistically achieve. This limitation is data-dependent.
Parameter overfitting Trying many N and M values on the same period can create an appearance of effectiveness that does not generalize. Use separate time periods for evaluation.