StoQuant — Quantitative Stock Research

Top feature contributions behind a ticker's ML outperformance prediction — which signals pushed the probability up or down, with signed magnitudes. Use after stoquant_get_ml_prediction when the user asks WHY the model is bullish/bearish. horizon must match the prediction you are explaining.

Public, unauthenticated platform scorecard from /public/accuracy/summary (no Pro gate). Key fields: letter `grade`, `hitRate` (do NOT treat hitRate as skill — a high hit-rate on a negatively-skewed book still loses money), `excessSpreadPp` (top-minus-bottom quintile 90-day *alpha*; this is the real edge), `spreadPp` (raw return spread, not alpha), `dataSufficient` (false = sample too small, ignore the grade), `independentWindows` (independent 90-day windows behind the Wilson CI), `sampleSize`, Wilson CI. When `dataSufficient` is true AND `excessSpreadPp` < 0 the live book has an inverted tail — auto-portfolio BUY recommendations are suppressed with gateReason `no_edge`. Call this before any buy-oriented auto-portfolio or optimizer read.

Generic cached alternative-data for a ticker under `data`, keyed by source: google_trends (search interest), wikipedia_pageviews (attention), options_iv (implied-vol metrics), openinsider (insider feed), edgar_8k. Each source carries its latest values and fetchedAt. Complements stoquant_get_alt_signals with retail-attention and options-positioning reads.

The full alternative-data dossier for a ticker in one call, under `data`. Aggregates 15 sources: sec8k (8-K event items + counts), secLate (NT-10K/Q late filings), sec13dg (13D/13G activist/passive ownership), secDilution (S-1/S-3/424B shelf + active-dilution flag), secGovernance (DEF 14A, Form 25 delisting, Form 15 deregistration flags), secInsiderIntent (Form 144/Form 3 counts), secS8 (employee stock plans), fdaEnforcement (Class I/II/III actions), cpscRecalls (consumer recalls), clinicalTrials (phase/recruiting — biotech), github (stars/forks/issues/push activity), appStore (iOS rank deltas), hackerNews (story velocity), gdelt (news tone). Sources with no recent activity for the ticker are null (e.g. clinicalTrials for a non-biotech). This is the platform's differentiated edge — use it to surface catalysts and red flags an ordinary data feed misses.

Recent analyst rating changes for a ticker — upgrades, downgrades, reiterations, and initiations — with a 90-day action breakdown. Use to see the direction of sell-side revisions (momentum in sentiment), not just the static consensus.

Sell-side analyst forward estimates for a ticker under `data` (per fiscal period: estimated EPS avg/high/low, estimated revenue avg/high/low, number of analysts) with a `count`. Use for consensus expectations and dispersion. Some future periods may have null estimates until coverage fills in.

Analyst price targets for a ticker with a `consensus` block (avgTarget, highTarget, lowTarget, numAnalysts) and the underlying `data`. Use to gauge implied upside/downside vs the current price and the dispersion of targets.

Read-only recommendation history for an auto-portfolio (ticker, action, status, timestamps). Use to inspect what the engine last proposed — including rows that stayed pending or were gated. Does not execute or cancel anything. Power-tier; requires portfolio ownership.

Read-only snapshot of an auto-portfolio: last evaluation (`evalStatus.lastEvalResult`, `lastEvalAt`, `recsGenerated`) plus diagnostics (cash, holdings tickers, cascade/drawdown breakers, risk params, recent recommendations). `lastEvalResult` is `success` or `gated:<reason>`. Important gate: `gated:no_edge` means the public accuracy scorecard has dataSufficient && excessSpreadPp < 0, so BUY recommendations were suppressed — do not treat that as a buy signal. Other gates include `static_snapshot`, `health_critical`, `cascade_breaker_paused`, `no_cash_buys_suppressed`, `no_changes`. Requires Power-tier API key and ownership of portfolioId. This does NOT run a new evaluation and does NOT place orders.

Reload a previously stored backtest run by numeric id, including metrics and bias disclosure. Use after stoquant_run_backtest or to inspect history. Power-tier. Read-only.

Metadata for past backtest runs (ids, strategy, dates) without the full equity curve. Filter by strategyId. Power-tier. Read-only.

Current performance of major index ETFs (SPY, QQQ, IWM, and peers): 1-day and year-to-date return. Use to frame an individual stock against the market. Early in the trading day the response may be `{error:'Benchmarks not yet computed'}` until the daily compute runs — treat that as 'not ready', not a failure.

Market-wide list of tickers with large recent insider purchases (default >= $1M) under `items`, each with the buy detail plus a current quote and Q-Score. Use for 'where are insiders putting real money right now' discovery across the whole market.

Cross-ticker hot list of the highest-impact near-term catalysts the platform is tracking (FDA decisions, activist stakes, dilution events, clinical readouts, etc.). Use for market-wide 'what's about to move' discovery, distinct from per-ticker stoquant_get_catalysts.

Upcoming and recent catalysts under `data`, each with ticker, type (e.g. earnings_soon, activist_stake, dilution, fda), date, label, direction (bullish/bearish/neutral), and source. Omit ticker for the market-wide feed; pass a ticker to filter to one name (may be empty if that stock has no tracked catalysts right now). Use to find date-driven, event-based opportunities.

The named, two-sided evidence behind a ticker's Q-Score — the 'bull case vs bear case' as concrete facts rather than scores. Returns five grids (`momentum`, `sentiment`, `earnings`, `valuation`, `risk`), each with `cells`: every cell has a human-readable `label` (e.g. 'Price above 50-day average', 'Short interest'), a formatted `value`, and a `tone` ('bullish'|'bearish'|'neutral'|'unknown'). To assemble a balanced case, collect cells with tone 'bullish' (the bull case) and 'bearish' (the bear case) across all five grids. Use this AFTER stoquant_get_qscore when the user asks WHY the score is what it is, or wants the specific positives and negatives to weigh — it is the evidence layer the research page shows above the dimension breakdown.

Quote plus computed technical indicators for one ticker: RSI(14), MACD, Bollinger Bands, and moving-average context. Use when you need momentum/overbought-oversold context, not just price. Heavier than stoquant_get_quote; can occasionally time out on cold cache — retry once if so.

Undervalued small-caps with confirming signals (the 'hidden gem' screen: low valuation, healthy balance sheet, low analyst coverage, plus momentum/ML/insider confirmation). Returns `{gems:[...], totalCount, lastScanAt}`. freshness filters by how recently a gem's signals fired. Use for value-oriented idea generation distinct from momentum.

Insider-buying cluster metrics for a ticker under `data`: distinctFilers14d, officerCount14d, purchaseCount90d, totalValue90d, clusterScore, and a clusterMomentum flag. Cluster buying by multiple insiders (especially officers, large dollar value) is a strong bullish signal — this scores it. Pairs with stoquant_get_insider_trades (the raw transactions).

Recent insider (Form 4) transactions for a ticker under `data`, plus a `summary` (net buy/sell counts and dollar value). Cluster buying by officers/directors — especially large dollar buys — is a notable bullish signal. Use to check whether insiders are accumulating or distributing.

Top institutional holders for a ticker: organization name, shares held, and percent ownership. Use to gauge institutional concentration and who the big holders are.

Aggregated institutional sentiment for a ticker: holder additions vs reductions and a net `signal` direction. Use as a 'smart money' flow read distinct from analyst ratings.

Forward (~14 day) economic calendar: scheduled releases with forecast, prior value, and expected market impact. Use to flag upcoming macro events (CPI, FOMC, jobs) that could move positions.

Full top-down macro picture in one call: `regime` classification, key `indicators`, `macroStress` gauge, `sectorRotation` (which sectors are favored, with valid GICS sector names), forward `calendar`, and current `themes`. Best first call for 'what's the macro backdrop' before drilling into stocks.

Latest values for one FRED macro series under `data` (recent observations) with a `count`. Common ids: DGS10 (10y Treasury yield), DGS2 (2y), T10Y2Y (10y-2y spread), UNRATE (unemployment), CPIAUCSL (CPI), VIXCLS (VIX), FEDFUNDS (fed funds). Use for a single specific indicator; for the full picture use stoquant_get_macro_dashboard.

Current macro investment themes with their related sectors — the narrative layer (e.g. rate-cut beneficiaries, AI capex). Use for thematic idea generation that ties stocks to a macro thesis.

Current market-regime classification from the Hidden Markov Model over S&P 500 returns: regime label (e.g. Bull/Range/Bear) and the probability vector across states. Use to set top-down context before stock-level calls. If the model reports it could not fit, probabilities fall back to roughly uniform — read that as 'regime undetermined', not a confident Range call.

ML model's probability that the ticker will OUTPERFORM the S&P 500 over the horizon (NOT a buy/hold/sell call). Returns `data.prediction` with: `probability` (0-1, chance of outperformance), `classification` ('bull' = predicted to outperform, 'bear' = underperform), `modelVersion`, `modelAuc`, `plainLanguage` (headline/summary/topDrivers), and `isHeuristic` (true when the trained model is gated for distribution drift and a heuristic fallback is used — weight it less). Use stoquant_explain_ml_prediction for the feature-level 'why'. horizon is in trading days. Note: early in the US trading day (before ~10am ET), the prediction may not yet be refreshed with intraday data — a `null` or empty prediction should be treated as 'not yet computed' rather than an error.

Small/mid-caps flagged with high (>5x) upside potential: revenue acceleration, improving fundamentals, and pre-discovery setups. Higher risk/variance than hidden gems. Use for aggressive growth idea generation.

Recent news headlines for a ticker under `items`: each with publishedAt, headline, source/provider, category, and a sentiment label. Use for the latest narrative around a stock. For news plus filings plus earnings in one combined view, use stoquant_get_news_events.

Combined event view for a ticker under `data`: `news` (headlines), `filings` (SEC), and `earnings` (calendar). The single best call for 'what is happening with this stock' — recent coverage, regulatory activity, and upcoming/just-reported earnings together.

System-wide auto-portfolio health report (subsystem statuses, degraded mode, last check time). Use to see whether the engine is healthy before trusting a last-eval snapshot. Read-only — does not trigger healing or reset the drawdown breaker.

The Q-Score is StoQuant's headline 0-100 composite for a ticker, blending fundamentals, momentum/technicals, ML, analyst, and 'hidden gem' value dimensions on SECTOR-RESIDUALIZED forward returns (sector beta removed). Returns: score (0-100), signal ('strong_buy'|'buy'|'hold'|'sell'|'strong_sell'|'unrated'), per-dimension `components` (each with rawValue, normalizedScore, weight, and an `available` flag — unavailable dims contribute 0), `confidence`/`confidenceInterval`, `dimensionContributions`, a `verdict` and `plainLanguage` summary, and `riskAlerts`. This is the best single tool for 'is this a good stock right now'. period selects the scoring horizon. Note: early in the US trading day (before ~10am ET), the score may not yet be refreshed with intraday data — a `null` or empty score should be treated as 'not yet computed' rather than an error.

Quantitative analytics for a ticker under `data`: factor exposures, correlation/quality metrics, and risk decomposition. Use for portfolio-construction and factor-tilt questions. Complements stoquant_get_qscore (verdict) with the underlying quant detail.

Real-time quote for ONE ticker: last price, day change (absolute and percent), bid/ask, day high/low, and volume. Use for a single symbol; for 2+ symbols use stoquant_get_quotes_batch (one call, far fewer requests). Prices are delayed per the data vendor, not tick-by-tick.

Current quotes for up to 50 tickers in a single call. Preferred over calling stoquant_get_quote in a loop — same data, one request, stays well under rate limits. Returns one quote object per ticker (price, change, volume).

Find peers of a ticker under `peers`, each with qScore, quote, and market cap. `by` selects the relation: 'sector' (same sector), 'tag' (shared thematic tags), 'setup' (same technical setup firing), or 'correlation' (highest 60-day return correlation). Use to build a comparison set or find alternatives to a name.

Recent SEC filings for a ticker under `items` (with `count`/`total`), unioning the whole suite: 8-K events (with item labels like 'Item 2.02 Results of Operations'), NT-10K/Q late filings, 13D/13G activist/passive ownership, S-1/S-3/424B dilution, DEF 14A / Form 25 / Form 15 governance, and Form 144/3 insider intent. Each item has accession, formType, filedDate, category, and a human-readable label. Use to read the actual regulatory record behind a thesis.

Top and bottom performing stocks within a GICS sector, by recent return. Use to find what is leading or lagging inside a sector. sector must be a full GICS name (see stoquant_get_macro_dashboard sectorRotation for valid names), e.g. 'Technology', 'Health Care', 'Financials'.

Short-interest history for a ticker: append-only `snapshots` plus a `summary` (latestFloat = % of float short, latestRatio = days-to-cover, and window deltas). Rising short interest with a price squeeze can signal a short squeeze; falling can signal covering. Use for crowded-short reads.

Historical track record per signal type: an array of records with `signalType`, `hitRate` (0-1), `totalObservations`, `avgReturnPercent`, `precision`, and `windowDays`. Use to weight how much to trust a given signal before acting on it — prefer signals with high hitRate AND large totalObservations.

Hit-rate heatmap of signals over time/ticker for a date range — which signal types worked when. Use for backtest-style 'has this signal been reliable lately' questions. Dates are YYYY-MM-DD; omit for the default recent window.

Returns both the raw Q-Score (`originalQScore`) and a `stage`/regime-adjusted variant (`adjusted`) for a ticker. Use this when you want the score adjusted for the stock's lifecycle stage and current market regime rather than the raw composite. For the full breakdown use stoquant_get_qscore.

Tickers with a sharp (2x+) surge in social mentions in the last hour vs the prior hour — the earliest social-momentum signal, with sentiment and a recent post snippet per name. Use to catch breakouts in attention before they are widely noticed.

Current social-media buzz for a ticker: `totalMentions`, `recentPostCount24h`, `sentimentBreakdown` (bullish/bearish/neutral), `topPlatforms`, and a sample of `recentPosts`. Use to read retail attention and crowd sentiment. Sources are credibility-weighted platform aggregates, not raw noise.

Daily social mention-volume and sentiment trend for a ticker over the last ~30 days. Use to see whether attention and bullishness are building or fading, not just the current snapshot.

Leaderboard of social sources ranked by true directional accuracy (Wilson lower-bound on their calls). Use to judge how much weight to give a platform or author's sentiment — high-credibility sources earn more trust than raw mention volume.

Historical price series (~90 days) for up to 50 tickers, for charting or trend/return computation. interval controls granularity (1d daily, 1wk weekly, 1mo monthly). Returns an array of close prices per ticker, not full OHLCV.

Qualitative competitive intelligence for a ticker: economic moats, structural weaknesses, upcoming catalysts, and peer comparison. Best for mid/small-caps; mega-caps and thinly-covered names may return `{data:null}` (no edges computed) — treat that as 'no data', not an error.

Read-only strategy performance for a portfolio. `view` selects the slice: `kpis` (headline returns), `equity-curve` (vs SPY, normalized to 100), `monthly-returns`, `risk-metrics` (Sharpe/Sortino/max drawdown/capture), `attribution` (P/L by signal type), `comparison` (all active portfolios). Pro-tier. Does not trade.