Methodology

Positioning index (−100 to +100)

The v2 index is a descriptive, cross-sectional comparison across the saved cohort as of the displayed date. It is not a forecast and it is not a trading signal. Each component uses a robust median and interpolated IQR scale, clipped to ±3σ. Fewer than five valid cohort observations make a component unrankable.

index = Σ(z × weight × 100/3), published only with ≥2 rankable components
weights = reported ownership .30 · foreign flow .25 · broker context .20
          shareholder count .15 · foreign holder-class shift .10

• Reported ownership (30%): signed transaction value in the inclusive 90-calendar-day window.
• Foreign flow (25%): signed net inflow in the inclusive 90-day window, normalized only when an observed positive market cap exists.
• Broker context (20%): eligible institutional or foreign broker net value in the inclusive 14-day window; no rows means missing.
• Shareholder count (15%): latest valid month-over-month change on or before the as-of date.
• Foreign holder-class shift (10%): foreign institutional classes minus foreign individual classes across the same two months.

A null component is missing evidence, not a measured zero. Coverage weight is the sum of rankable weights and is not a probability.

Bounded candidate patterns

• Holder cluster: at least three distinct normalized holders transacting in one direction inside a 30-day event window.
• Buy during decline: reported buys while the pre-anchor close was lower than the prior context close.
• Stealth flow: reported buys with abnormal foreign flow in the bounded pre-anchor window.

Candidate membership and evidence stop at the event anchor. Post-anchor prices can change only the retrospective outcome object.

Retrospective outcomes

Outcomes use the first common issuer and IHSG observed session on or after the anchor (within seven calendar days), then the first common session on or after the 7-, 30-, or 60-day target (also within seven days). A missing full horizon is Pending or Unavailable; it is never formatted as 0%. Excess return is the issuer percentage minus the benchmark percentage over the same sessions.

Exit Watch (v3, 0–100)

Exit Watch is a bounded reading of observed exit-side pressure: which investor cohorts appear to be leaving a name, and which may be absorbing that flow. It combines four robust cross-sectional components, each normalized by the issuer’s observed market cap:

• instExit · 30%: net flow of brokers classified institutional in the broker registry (14-day window). Precision overlays from per-cohort top-broker feeds are preferred when present.
• foreignExit · 25%: net foreign flow over the same window.
• insiderExit · 25%: reported insider sell value over 90 days.
• retailAbsorb · 20%: net flow of brokers classified retail (inverted: retail net buying raises the reading as absorption of exit supply).

Each component is a clipped robust z-score (±3σ) across the scored universe. The score publishes only when at least half of the component weight has usable evidence; otherwise the issuer is listed as suppressed / low coverage, never scored zero. Missing components are shown with their reason.

Broker cohort labels come from the broker registry. Of 88 classified firms, 42 are mixed and 2 unknown: they are not counted as either side, so the cohort view is a labeled subset, not a census. Flags (SUSP ≤14D, CA ±7D, FF<20%, SPARSE) mark context that can distort the reading; they annotate, they do not change the score.

Data lineage and coverage

The public app reads a hash-verified local Sectors snapshot and generated artifacts. The snapshot includes parsed filings, foreign flow, daily prices, broker rows, monthly holders, tickers, a benchmark-observed IHSG series, and a saved subsector-aggregate artifact used by the sector rotation board. v3 Exit Watch additionally consumes the broker registry, per-issuer top-broker feeds, suspensions, corporate actions, and free-float feeds; each hashed into the derived manifest (feedHashes). Retrieval times and historical publication availability are not verified for legacy rows. Ambiguous flat-OHLC or zero-volume observations are marked legacy-unknown.