Rankings screen (`POST /api/rankings/screen`)

POST /api/rankings/screen runs server-side filters over the full ranking cross-section at one trading day (latest by default) and returns up to 500 rows sorted by rank_ordinal (1 = best, rank_percentile 100 = best). This is the stat-arb cross-section in one call — replaces N per-ticker /rankings round-trips.

What it gives you

For one (metric, cohort, window) triple at one as_of date, the endpoint returns the ranked subset of the universe that passes your filter — each row carrying the ticker, the rank position, the percentile, the cohort size at that teo, and the underlying metric value.

This is how you ask questions like:

"Show me the top 5% of names by 21d subsector-residual return today."
"Which 50 stocks have the highest 252d-window market-cap rank inside the XLK cohort?"
"Give me decile 1 (best) by er_l3 cross-sectionally, across the full universe."

Request

POST /api/rankings/screen
{
  "metric":  "subsector_residual",   // mkt_cap | gross_return | sector_residual |
                                     // subsector_residual | er_l1 | er_l2 | er_l3
  "cohort":  "subsector",            // universe | sector | subsector
  "window":  "21d",                  // 1d | 21d | 63d | 252d
  "as_of":   "2026-05-26",           // optional (default: latest teo)
  "min_percentile": 95,              // optional (0–100, 100 = best)
  "decile":  1,                      // optional (1–10, 1 = best)
  "sector_filter": "XLK",            // optional (sector ETF ticker)
  "limit":   100                     // optional (1–500, default 100)
}

Filters combine — min_percentile AND decile AND sector_filter all apply. limit caps the post-filter result set; rows are returned in rank_ordinal order.

Response shape

{
  "as_of": "2026-05-26",
  "metric": "subsector_residual",
  "cohort": "subsector",
  "window": "21d",
  "rows": [
    {
      "ticker": "NVDA",
      "rank_ordinal": 1,
      "rank_percentile": 100.0,
      "cohort_size": 32,
      "metric_value": 0.084
    },
    ...
  ],
  "count": 47,
  "_metadata": { "data_source": "zarr", ... }
}

Use-case patterns

Top-decile residual screen (stat-arb)

curl -sS -X POST "https://riskmodels.app/api/rankings/screen" \
  -H "Authorization: Bearer $RISKMODELS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "metric":"subsector_residual",
    "cohort":"subsector",
    "window":"21d",
    "decile":1,
    "limit":50
  }' \
  | jq '{count, top: .rows[:5]}'

Best-of-sector cap-rank screen

curl -sS -X POST "https://riskmodels.app/api/rankings/screen" \
  -H "Authorization: Bearer $RISKMODELS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "metric":"mkt_cap",
    "cohort":"sector",
    "window":"1d",
    "sector_filter":"XLF",
    "limit":20
  }'

Python

from riskmodels import RiskModelsClient
client = RiskModelsClient.from_env()

df = client.screen_rankings(
    metric="subsector_residual",
    cohort="subsector",
    window="21d",
    min_percentile=95,
    limit=100,
    as_dataframe=True,
)
df[["ticker", "rank_percentile", "metric_value"]].head()

Filter rules

metric must be one of: mkt_cap, gross_return, sector_residual, subsector_residual, er_l1, er_l2, er_l3. Other ranking metrics aren't queryable through this endpoint.
window must be 1d / 21d / 63d / 252d. For point-in-time (PIT) metrics like mkt_cap / er_l1 / er_l2 / er_l3, only 1d is meaningful — the others return the same value.
cohort picks the peer set for the percentile: universe (cross-sectional across all in-mask names), sector (within bw_sector cohort), subsector (within subsector ETF cohort).
rank_percentile is the upper-tail percentile: 100 = best, 0 = worst. min_percentile: 95 → top 5%.
decile is 1 = best, 10 = worst. Mutually consistent with min_percentile (decile 1 ≡ min_percentile >= 90).

Pricing & limits

0.02/request (billing_code: rankings_screen_v1). Max 500 rows per call. For larger pulls or historical traversals, use the per-ticker /api/rankings endpoint or the daily zarr export.

Where it sits

The rankings screen reads ds_rankings_{etf}_{universe}.zarr directly. That zarr is rebuilt daily by the ERM3 pipeline (1d step's _compute_and_save_rankings). 96 ranking variables exist; this endpoint serves the 7 most-requested metrics × 3 cohorts × 4 windows = 84 cells of the full grid.

GET /api/rankings — per-ticker rank lookup (single ticker, all cohorts/windows).
GET /api/industry-panel — industry-level peer-β cross-section (complementary to stock-level rankings).
POST /api/batch/lstar — per-ticker Lstar history for a panel of tickers you've already selected via this screen.
OpenAPI RankingsScreenResponse — full schema.

Rankings screen (POST /api/rankings/screen)