Methodology¶

Version 1.0.0 · Release date 2026-05-24

1. Purpose¶

A reference industry classification for digital assets, usable for:

1. Cross-sectional sector neutralization of alpha signals (group_neut).
2. Risk attribution — decomposing returns into sector, chain, and idiosyncratic components.
3. Peer-group comparison — fundamental and technical metrics are only meaningful relative to peers performing a similar economic function.
4. Comparability with established institutional research — the 6-digit CCSSXX positional code format was chosen to make potential future crosswalks to institutional classification systems tractable. See §3.2 for the affiliation disclaimer.

2. Design principles¶

3. Hierarchy¶

Three levels:

Class (2-digit, 4 entries)            ── reporting & risk-regime boundary
  └── Sector (4-digit, 14 entries)    ── primary group-neutralization axis
        └── Sub-sector (6-digit)      ── attribution & fine-grained neutralization

3.1 Why three levels (and not two, or four)¶

The choice tracks the precedent set by established equity-market classification systems (cited in §7) — both are four-level systems with 5,000+ constituents — adjusted downward by one level because the digital-asset universe is roughly an order of magnitude smaller.

Each level serves a distinct downstream purpose: Class for stablecoin-vs-rest risk boundaries and UI navigation; Sector as the primary group-neutralization axis (empirically the strongest cohesion signal); Sub-sector for attribution and the largest-bucket niche neutralization (DEX, Lending, L2, LST).

3.2 Code structure¶

Six-digit codes follow CCSSXX:

• CC = class (2-digit)
• SS = sector-within-class (4-digit when concatenated as CCSS)
• XX = sub-sector-within-sector (6-digit when concatenated as CCSSXX)

Sub-sector codes ending 90–99 are community extensions. They are visually distinguishable (e.g., 301090 for Liquid Staking) and reserve the 00–89 range for potential future alignment with institutional classification baselines.

Uses a 6-digit positional code format (CCSSXX) as a structural convention. This project has no affiliation with, and makes no claim of mapping compatibility to, Datonomy, GICS, or ICB. The format choice exists to make potential future crosswalks tractable, not to guarantee one today.

3.3 Orthogonal tags¶

Tags can be added in future versions for further orthogonal axes (token economic model, settlement-layer dependency, regulatory regime) but each addition requires evidence that the new axis captures meaningful variance not absorbed by hierarchy or existing tags.

4. Classification process¶

4.1 New asset¶

1. Read whitepaper, protocol documentation, and a recent Token Terminal or DefiLlama snapshot of fee flow.
2. Classify on primary economic function, breaking ties in this order:
3. Fee or revenue source (where does the value flow that accrues to the token?)
4. Largest TVL or transaction-volume component
5. Marketing positioning (lowest weight — tie-breaker only)
6. Assign chain_ecosystem by largest-TVL deployment chain at snapshot date.
7. Submit a PR adding one row to classification/snapshot.csv and one file decisions/<symbol>.md with the rationale (template in CONTRIBUTING.md).

4.2 Reclassification¶

Point-in-time (PIT) immutability contract. The effective_from field in classification/snapshot.csv records the date from which each classification row applies. Historical published panels are immutable. Reclassifications create new rows in snapshot.csv with a later effective_from; they never modify rows for past dates. A backtest pinned to a release tag (e.g. git checkout v1.0.0) will reproduce exactly across future taxonomy bumps.

Warning: running a historical backtest against main after a reclassification PR is merged WILL show different results for pre-reclass dates. For reproducible historical backtests, always check out a release tag.

effective_from semantics: cells in the wide matrix dated before an asset's effective_from are NaN, meaning "asset not yet classified under this schema version" — not "asset does not exist". See SCHEMA.md for the full null policy.

4.3 Cadence¶

Continuous PR review against main. The repository is tagged once per quarter (v2026.Q2, v2026.Q3, …) for consumers who require a stable reference. Between tags, main is always usable; tagged versions are immutable.

5. Out of scope (intentionally)¶

The following are deliberately excluded from v1.0 to keep the scope tight. Each is a potential future extension only if community demand emerges:

• Investable filters (mcap tier, ADV, listing venue) — these are properties of an asset's tradability, not its classification. This repository defines a coverage universe (which assets are classified); the investable universe (which assets a specific strategy trades) is determined downstream by each consumer applying their own filters. See UNIVERSE.md for coverage criteria.
• Regulatory tags — jurisdiction-specific, fast-changing, and outside the scope of an industry classification.
• Stablecoin internals — peg mechanism details, reserve composition. We classify the type of stablecoin; deeper attribution is a separate dataset.
• NFT collection-level classification — collections are not fungible assets.
• Per-asset Sharpe / IR / drawdown statistics — these are derived analytics, not classification.

6. Versioning¶

A consumer should pin to a specific release tag for reproducible backtests.

7. References¶

Academic and industry references that informed the design. Inclusion here is academic citation; no endorsement or affiliation is implied or claimed.

• MSCI, Coin Metrics, and Goldman Sachs (2022). Datonomy Methodology. (Digital-asset classification reference. Uses a 6-digit positional code format (CCSSXX) as a structural convention. This project has no affiliation with, and makes no claim of mapping compatibility to, Datonomy, GICS, or ICB. The format choice exists to make potential future crosswalks tractable, not to guarantee one today.)
• Bhojraj, S., Lee, C., and Oler, D. (2003). "What's My Line? A Comparison of Industry Classification Schemes for Capital Market Research." Journal of Accounting Research, 41(5).
• S&P Global and MSCI. Global Industry Classification Standard (GICS) Methodology. (Cross-asset precedent for multi-level hierarchical industry classification.)
• FTSE Russell. Industry Classification Benchmark (ICB) Ground Rules.
• Hoberg, G., and Phillips, G. (2016). "Text-Based Network Industries and Endogenous Product Differentiation." Journal of Political Economy.

Trademark notice¶

GICS is a registered trademark of S&P Global and MSCI Inc. Datonomy is a trademark of MSCI Inc. ICB is a trademark of FTSE International Limited. WorldQuant and BRAIN are trademarks of WorldQuant LLC. All other trademarks are the property of their respective owners. This project is independent of, and has no business relationship with, any of these organizations.