Methodology

Marsad Markets is a research and education tool for self-directed investors. It screens US stocks with rules-based strategies, analyzes upcoming earnings with a transparent factor model, maps supply-chain relationships, and backtests strategies over history.

It is not investment advice, not a recommendation service, and not personalized to you. We are not registered investment advisers. Outputs are estimates from historical, rules-based computation; they can be wrong, and markets change. You own your decisions.

Each strategy applies fixed filters (quality, momentum, breakout, volatility patterns) to its universe, then ranks what passes by a composite confidence score with explicit weights:

• 45% signal strength (how strongly the stock matches the strategy)
• 25% probability score
• 15% win/loss ratio (volatility-adjusted expected move vs stop)
• 10% liquidity quality
• 5% news sentiment

The probability figure is deliberately clamped to 45–75%: we do not display certainty we cannot have. Entry/stop/target levels on screener rows are strategy parameters derived from each stock's own volatility — educational outputs, not predictions.

Each upcoming reporter gets a composite score from five factors with explicit, fixed weights — shown on every analysis page:

• 35% — past post-earnings moves (how the stock actually behaved after its recent reports)
• 30% — EPS revision momentum (our score: 60% consensus-change direction/magnitude, 40% breadth of analyst revisions)
• 20% — analyst ratings, price targets & recent changes
• 10% — competitor earnings read-through (from our relationship graph)
• 5% — news sentiment (kept low — headlines are noisy)

The weighted score lands in [−1, +1]. Favorable means the composite is at or above +0.25; Unfavorable means at or below −0.10; anything between is Mixed. When a factor has no data, the composite renormalizes over what is available — it never fabricates a value.

Market context (the ✦ mark). A hostile market/sector environment can temper a verdict — dampening the composite by up to 0.45 in the worst case, enough to move a Favorable into Mixed — but by design it softens and never flips a verdict to the opposite sign, and a supportive environment never inflates one. Environment is shown as context; it shifts probabilities, not certainties, and this modifier is not point-in-time validated.

The verdict is only as good as the evidence behind it, so every analysis tracks how much of the factor weighting was actually available:

• Below 35% of the weight available → insufficient coverage: the name still lists on the calendar (it is reporting), but no directional verdict is shown at all.
• 35–60% → the verdict is shown with a visible ltd limited-coverage badge — backed by less evidence, and labeled as such.
• 60% and above → full coverage, no badge.

This gate wraps the output only — it never alters the factor math. We would rather show you “not enough data” than a confident-looking verdict built on one factor.

When (and only when) a verdict is favorable, the analysis shows mechanical zones. The core idea: deriving a target and a stop from the same multiple of the same metric makes risk/reward a mathematical illusion, so the two legs are decoupled by time horizon:

• Stop — the event gap. An earnings report is a discontinuous move, so the stop is placed outside the expected event move priced by the options market (the at-the-money straddle → implied move), at a market-structure level (recent swing low, moving average, 52-week low) or a fallback of entry × (1 − 1.1 × implied move). A stop-out is meant to mean the thesis was wrong, not that normal noise occurred.
• Target — the drift. The final target is the stock's own realized average move over roughly ten trading days after its recent reports (post-earnings drift) — historical capacity, not a forecast. A secondary waypoint marks the top of the expected event move, where momentum often stalls; when the expected event move exceeds the historical drift, that waypoint legitimately sits above the final target and the page says so.
• The R/R gate. Reward/risk to the drift target must clear 1.2. This is drift-to-premium framing: it asks whether the stock's historical drift outpaces the gap risk the options market is pricing. Below roughly 1.0, options are pricing a larger gap than the stock historically drifts — the setup is rejected rather than the target stretched (“target not fabricated”).
• Flags, not gates. Trading below the 20/50-day moving average raises a trend-alignment flag (drift setups favor pre-existing uptrends) but does not suppress levels. A stop closer than 2.5% to entry raises an unusually-tight-stop caution — ordinary noise can hit it and the R/R can overstate the edge.
• Limited-data fallback. With no listed options, levels fall back to an ATR-based stop (1.5 × ATR floor) and a measured-move target, labeled “Limited data” on the page.

All of these are tunable heuristics grounded in market structure — not point-in-time-validated optima, and never advice. Post-earnings drift can underdeliver.

Every backtest ships with the same disclosures it shows in the app:

• Historical and illustrative only — never a prediction or guarantee of future results.
• The universe uses current index members (point-in-time membership isn't available on our data plan), so results carry survivorship bias — typically inflating returns by roughly 1–3% per year.
• Exits are modeled at the signal day's close; intraday stop/target fills are not simulated and gaps fill at the close.
• Default cost model: 0.1% spread + 0.1% slippage per side (≈0.4% round trip), $0 commission — configurable per run; real costs differ.
• Automated sanity checks flag statistically implausible results (excessive Sharpe, win rate, or profit factor; tiny trade samples) instead of letting them stand unqualified.

Screens, earnings analyses, and fundamentals are pre-computed on a daily cycle; pages show honest freshness labels (“Updated today / yesterday / N days ago”) derived from real compute timestamps — never a hardcoded “today”. Quotes are end-of-day or delayed; when a batch snapshot lags, ticker pages overlay a fresh delayed quote and label it as such. Earnings times (“Before open / After close”) refer to the US market session, Eastern Time.

Macro data comes from public-domain US government series (Federal Reserve Board and Bureau of Labor Statistics, retrieved via FRED, Federal Reserve Bank of St. Louis). Market data comes from commercial providers under their terms; headlines link to their original publishers.

• No performance claims. Nothing here promises returns. Probabilities are clamped estimates; verdicts are weighted summaries of historical and current data.
• Not point-in-time validated where stated. Several factors are current snapshots without reconstructable history (analyst data, revisions, sentiment), so the earnings composite cannot be validated out-of-sample. We say so on the pages themselves.
• Free-tier data has gaps. Thin names lack options or analyst coverage; that is exactly what the coverage tiers disclose instead of hiding.
• Markets change. Historical drift, volatility regimes, and factor behavior are not stable properties of the world.
• Trailing P/E uses dividend-adjusted prices. The Revenue + Trailing P/E chart divides split- and dividend-adjusted prices by split-adjusted GAAP EPS. Dividends can't be “un-adjusted” from EPS, so older multiples read somewhat low for dividend payers: typically under ~15% at the 10-year end for average-yield names, and up to ~40–50% for the highest-yield names (6–8% yields). Recent quarters are essentially unaffected. Directionally faithful; treat old absolute levels for high-yield names as understated.
• You own your decisions. This is research and education software. Trading involves substantial risk of loss. See the full Disclaimer.