crypto-sectors

A single flat sector field for the 232-coin OKX perpetual universe, empirically validated for cross-sectional group-neutralization (group_neut).

---

What this is

crypto-sectors publishes one thing: a single flat sector label for every coin in the 232-coin OKX USDT-margined perpetual universe, validated for use as a group-neutralization (group_neut) input in cross-sectional crypto factor research.

Universe: 232 crypto-native coins. Identity anchored by cg_id (CoinGecko slug) + okx_instid to prevent ticker-collision mis-resolution. Daily log-returns span 2019-11-27 to 2026-06-13 (2391 rows).

This is not a covariance model or a risk model. It is a validated classification — an input to an exposure matrix. Do not use sector labels as regression variables or factor returns directly.

---

The 15-label sector field

One layer, no hierarchy, no numeric codes. Each label is defined economics-first in taxonomy.yaml with ordered precedence rules.

FACTOR sectors — within-group demean removes a real shared return factor

STRONG (mean intra-group market-residual pairwise correlation, 10k-perm):

Sector	Mean residual corr
privacy	+0.55
captive_franchise	+0.30
value_transfer	+0.23
compute_storage	+0.18
meme	+0.10

WEAK but significant:

Sector	Mean residual corr
eth_scaling	+0.04
metaverse_gaming	+0.04
interoperability	+0.03

MARGINAL — significant only because large n tightens the permutation null; cohesion is mostly market beta. Treat with caution:

Sector	Mean residual corr
smart_contract_platform	+0.013
information_technology	+0.011

RESIDUAL sectors — intra-group correlation indistinguishable from null

Demean within these groups is approximately a no-op. They are pooled into OTHER in the production group_neut kernel. Kept as organizational labels; candidates for sub-division once each sub-leaf has >= 3 live members.

Sector	Mean residual corr
defi	+0.003
ai_infrastructure	-0.002
oracles_data	+0.008
media_content	+0.005
OTHER	(catch-all)

Full per-label statistics: validation/sector_okx/cohesion_sector.csv.

---

Validation — 10,000-perm, production OTHER-collapse kernel

Method: per-day cross-sectional variance reduction (VR) vs a size-preserving label-shuffle null, averaged over days with >= min_assets valid coins.

Slice	n coins	min_assets	VR	Null mean	Excess	p
ALL	232	30	12.91%	8.22%	+4.69 pp	0.0001
TOP30	top 30	20	22.78%	16.03%	+6.75 pp	0.0008
TOP50	top 50	30	17.97%	11.50%	+6.48 pp	0.0001
TOP100	top 100	40	17.46%	11.75%	+5.71 pp	0.0001

Full run log: validation/sector_okx/run_10k.log. Summary: validation/sector_okx/REPORT.md.

---

Production OTHER-collapse kernel

This is the conservative/honest kernel — what group_neut does and what the VR numbers above measure. A naive "demean within every label" overstates VR.

Per day:

1. Identify FACTOR sectors with >= 3 live members on that date.
2. Demean returns within each such group.
3. Pool RESIDUAL-labelled coins and any FACTOR group with < 3 live members on that date into a single OTHER bucket.
4. Demean the OTHER pool (market-neutralize the remainder).

---

Honesty notes

• ai_infrastructure and oracles_data are new carve-outs that are RESIDUAL — they do not pass cohesion alone. Kept as organizational labels, pooled into OTHER in group_neut. Candidates for sub-division (e.g., defi -> dex/lending/lsd/rwa/perps; ai_infrastructure -> gpu-compute/agents/data) once each sub-leaf has >= 3 live members.

• Big buckets (smart_contract_platform, information_technology, defi) are mostly market beta. Their statistical significance comes from large n tightening the null, not from strong cohesion.

• NOT A RISK MODEL. This is a validated classification — an input to a risk model's exposure matrix. It is not a covariance model or factor risk model.

• Deployment gate. Validated for research and group-neutralization. Alpha-capital deployment is paper-trading-gated.

---

Repository contents

Path	Description
classification/sector.csv	Committed source of truth. Columns: symbol, sector, role, cg_id, okx_instid, audit_from, prev_label, note. 232 rows.
classification/sector.parquet	Parquet mirror (derived by scripts/build_sector_field.py).
classification/sector_panel.parquet	PIT (date × symbol) label matrix, pandas StringDtype, <NA> before listing. The direct group_neut input.
classification/sector_roles.json	label -> {role, n, mean_resid_corr, cohesion_p}
classification/universe_tiers.json	TOP10/20/30/50/100 membership lists.
data/returns.parquet	232-symbol daily log-returns.
taxonomy.yaml	Flat sector label definitions (economics-first), precedence/boundary rules, FACTOR/RESIDUAL governance note.
decisions/	Per-coin economic rationale. decisions/sector-field.md is the master memo; decisions/sector-field-okx-2026-06.md is the audit record.
research/	128 per-coin economic-rationale notes.
validation/sector_okx/	varred_sector.json, cohesion_sector.csv, run_10k.log, REPORT.md.
scripts/	build_sector_field.py, validate_sector_okx.py, validate_schema.py, README.md.
.github/workflows/ci.yml	CI: build + schema-validate + smoke-validate on published files.

---

Quick start

import pandas as pd

# Load the PIT label panel and returns
panel   = pd.read_parquet("classification/sector_panel.parquet")   # date x symbol, StringDtype
returns = pd.read_parquet("data/returns.parquet")                   # date x symbol, float64

# Align
common_dates  = panel.index.intersection(returns.index)
common_assets = panel.columns.intersection(returns.columns)
panel   = panel.loc[common_dates, common_assets]
returns = returns.loc[common_dates, common_assets]


def group_neut_day(ret: pd.Series, lbl: pd.Series) -> pd.Series:
    """
    Production OTHER-collapse group_neut for one cross-section.

    - FACTOR groups with >= 3 live members: demean within group.
    - RESIDUAL labels + any FACTOR group with < 3 live members:
      pooled into OTHER, then market-neutralized.
    """
    # Drop assets missing either return or label
    valid = ret.dropna().index.intersection(lbl.dropna().index)
    r = ret[valid].copy()
    g = lbl[valid].copy()

    # Load FACTOR labels (from classification/sector_roles.json in practice)
    FACTOR_LABELS = {
        "privacy", "captive_franchise", "value_transfer", "compute_storage", "meme",
        "eth_scaling", "metaverse_gaming", "interoperability",
        "smart_contract_platform", "information_technology",
    }

    # Collapse RESIDUAL labels and under-populated FACTOR groups into OTHER
    counts = g.value_counts()
    def collapse(label):
        if label not in FACTOR_LABELS:
            return "OTHER"
        if counts.get(label, 0) < 3:
            return "OTHER"
        return label

    g = g.map(collapse)

    # Within-group demean
    group_means = r.groupby(g).mean()
    return r - g.map(group_means)


# Apply to a single date
date = "2025-01-15"
neutralized = group_neut_day(returns.loc[date], panel.loc[date])

---

Reproduce

# Derive parquet + panel + roles from sector.csv (committed source of truth)
python scripts/build_sector_field.py

# Full 10,000-perm validation (~1.5 h)
python scripts/validate_sector_okx.py --n-perm 10000

# Smoke test (fast)
python scripts/validate_sector_okx.py --n-perm 200

# Referential integrity check
python scripts/validate_schema.py

---

Status

Single flat sector field on the 232-coin OKX perpetual universe. Pre-release.

Validated for research and group-neutralization. Alpha-capital deployment is paper-trading-gated.

---

License

Code (scripts/) — MIT. Data (taxonomy.yaml, classification/, decisions/) — CC BY 4.0.

Attribution:

crypto-sectors contributors (2026). crypto-sectors: a validated sector classification for the OKX crypto perpetual universe. https://github.com/quantbai/crypto-sectors