Momentum Breakouts by Market Capitalization

Methodology

Registered before analysis. A replay answers the question under default settings; records from Momentum Lab, the Elite plan's momentum engine for US stocks, then test that answer against real quotes and fills.

1. Market cap as of the trigger. The in-app Momentum Lab backtest applies today's market cap to past days, leaking later information into the band. Here each cap is rebuilt as of the trigger date with the rules of production's SEC-derived source: shares outstanding from the latest 10-K or 10-Q filed on SEC EDGAR before that date (at most 200 days old, no reverse-split 8-K since), times the lower of the last price and the previous close. Stocks without a qualifying figure are excluded, just as an unknown cap never passes in production.
2. Triggers without lookahead. Default PREMARKET_BREAKOUT rules run on completed 1-minute bars only: reference = previous close; armed from +3%; a trigger when a bar's close reaches +5% after an earlier bar finished below +5%; skipped above +8%. The live engine checks quotes about once a second, and 1-minute closing prices approximate that. Only the first trigger per stock and day enters the primary analysis.
3. Filters. Price ≥ $1, premarket volume ≥ 50,000 shares and premarket dollar volume ≥ $250,000, all measured up to the trigger minute. The 1.0% spread filter and halt detection need live quote and halt data, which bars lack: spread enters through costs, and halted days are excluded.
4. Outcomes and costs. Entry at the next bar's open. Costs follow Momentum Lab's backtest assumptions: a 0.2% spread and 0.05% slippage, no commission. Spreads are generally wider in smaller companies, so a stress case doubles both in the $100M–$500M band.
5. Records. Aggregated, anonymised trigger, order and fill records report skip reasons, fill rates and slippage against the trigger price by band and parameter version, with paper and live kept apart. No figure drawing on fewer than five accounts is published.
6. Statistics. A calendar split holds back the last 30% of trading days for one confirmatory run: a two-sided test, at the 5% level, for an ordered trend across the three bands on the primary outcome. Pairwise contrasts are Holm-adjusted; a below-$100M band, possible only in replay, is descriptive. Success rates carry 95% Wilson intervals; net R is shown as a full distribution with quartiles and tails; intervals come from a bootstrap over trading days, since triggers cluster on busy days. Results are also shown per calendar half-year and per listing exchange.
7. Reproducibility. Published with the results: the strategy parameter version, a hash of all thresholds and costs, the filing snapshot date and bar-data checksums.

How success is defined

Primary outcome. After entry at the open of the 1-minute bar that follows the trigger, a trigger succeeds when price trades 3% above the fill before it trades 3% below it, within 60 minutes. The window is clock time: a premarket minute without trades has no bar, and it still counts. The 3% distance equals the default hard stop, so the bracket is 1R in each direction; for a driftless price the no-edge reference is about one half, and lower once costs and the tie rule below are applied. A bar spanning both levels counts as a loss. If neither is reached, the trade closes at the closing price of the last bar inside the 60 minutes and is not a success.

Net R. R is 3% of the fill price. Net R is the trade's return after the assumed spread and slippage, divided by 3%. Gaps at the 09:30 open or after a halt can produce losses beyond −1R; these are kept and reported, never truncated.

Band assignment. Bands follow the preset floors, and a cap exactly at a floor belongs to the band above it: exactly $500M falls in the $500M–$2B band, as it would pass the BALANCED preset.

Strategy outcome (secondary). The net return of the default exit model: a 0.5% trail on the last price, a 3% hard stop from the fill, no take-profit, positions carried into the regular session and closed 2 minutes before the close. Within each 1-minute bar the adverse extreme is assumed to come first. A trail is not a maximum loss, so this distribution is shown in full.

Execution outcomes (records only). Slippage from the trigger price to the average fill; the share of entry orders filled within the 10-second timeout; the mix of skip reasons per band, including market cap unavailable and moved too far.

Context. Maximum favourable and adverse excursion within the 60 minutes, and the return from the fill to the 09:30 open for triggers before 09:00.

Pattern and strategy criteria

Momentum Lab's Premarket Momentum Breakout (PREMARKET_BREAKOUT) with its default settings, as documented in the Momentum Lab guide:

• Reference: the previous close. The alternative, the first fresh premarket price PatternX observed, is not used.
• Direction: LONG, the default. SHORT and BOTH are out of scope.
• Arming and trigger: armed from +3% (a 2% band below the trigger); triggers at +5% after 1 confirmation tick, and only if the price was first seen below the trigger.
• Chase limit: 3 points, so a stock first seen above +8% is skipped ("Skipped — Moved Too Far"). The strategy re-arms after the move falls below +4%.
• Session: PREMARKET, 04:00–09:30 New York time.
• Default filters: price ≥ $1; premarket volume ≥ 50,000 shares; premarket dollar volume ≥ $250,000; spread ≤ 1.0% of the mid, with no two-sided quote counting as a fail; halt detection on; RVOL filter off. An unknown input never passes an enabled minimum.
• Mandatory market-cap filter: a minimum of $100M (BROAD), $500M (BALANCED, the default) or $2B (LARGE_CAP). A cap exactly at the minimum passes; an unknown cap never passes ("Skipped — Market cap unavailable"). The study's bands are the ranges between these floors, so each band is exactly what one preset adds to the next stricter one.
• Production market-cap sources: IBKR's market-cap field; otherwise SEC EDGAR shares outstanding × the lower of the last price and the previous close, labelled derived; otherwise Polygon. The replay uses only the SEC method, point-in-time.
• Default exits: 0.5% trail on the last price; 3% hard stop from the average fill; no take-profit; session policy CONTINUE_REGULAR; flat 2 minutes before the close; no overnight positions.

Why the floor exists, and how the presets differ, is explained in the market-cap filter guide.

Exclusions

• No point-in-time market cap: funds and ETFs; foreign issuers filing 20-F or 40-F; secondary listings; stocks with no 10-K or 10-Q within 200 days of the trigger, or with a reverse-split 8-K since the filing used.
• Filters not met at the trigger minute: price below $1, premarket volume below 50,000 shares, or premarket dollar volume below $250,000.
• Halts and pauses: stock-days with a trading halt or volatility pause in the outcome window. Bar data has no halt flag, so a regular-session gap of 5 minutes or more without bars serves as the proxy; live records use the engine's own halt detection where the data source supports it.
• Corporate actions: stock-days with a split, reverse split or symbol change.
• Repeat triggers: re-armed triggers after the first one per stock and day, reported separately.
• Records left out of outcome metrics (kept for execution metrics): positions closed manually or through the kill switch's exit option, since their outcome reflects a person's decision rather than the strategy.
• Small cells: any band-by-subgroup figure drawing on fewer than five accounts is withheld.

Limitations

• Market cap travels with other traits. Smaller companies tend to have lower share prices, wider spreads, thinner premarket trading and more frequent share issuance. This is an observational comparison, so a difference between bands cannot be attributed to market cap alone.
• Shares outstanding lag. Filings can be up to 200 days old, and companies that issue shares between filings have understated caps, which blurs band edges. A sensitivity run drops stocks within 10% of an edge.
• Coverage gaps. Foreign issuers and stocks without recent filings are excluded here, although live trading may cover some of them through IBKR or Polygon figures.
• Bars are not quotes. One-minute bars carry no spread or quote size, the trigger timing only approximates an engine that checks about once a second, and a 0.5% trail is coarse on 1-minute data.
• Scope of the live engine. Live Momentum Lab only considers stocks on its movers lists (50 gainers and 50 losers per data source, refreshed every 15 seconds). The replay considers all qualifying stocks, so it can include triggers the live engine would never have seen.
• Records reflect choices. Accounts choose presets and parameters, so live records are not a random sample of triggers; an account on LARGE_CAP never produces a small-cap trigger. Paper fills are kinder than real fills, and a small number of accounts means wide intervals.
• Survivorship. Smaller companies delist more often; if the bar source omits delisted stocks, the smallest band may look better than it was.
• Period dependence. Premarket activity in small and mid caps comes in waves with market conditions, so band differences in one period may not persist. Read this design alongside the relative volume study and the extended-hours trading risks guide.