README.md

---
license: odc-by
pretty_name: "Implied volatility surface for on-chain crypto options"
tags:
  - options
  - implied-volatility
  - derivatives
  - greeks
  - order-book
  - bitcoin
  - ethereum
  - solana
  - cryptocurrency
  - time-series
task_categories:
  - time-series-forecasting
size_categories:
  - 1M<n<10M
---

# Crypto options implied volatility surface

Every listed option on one venue, priced, with its greeks, sampled through the day.

An option quote is computed on demand and kept by nobody. The chain that existed at any past
moment is not recoverable from the venue or from the chain it settles on: only trades leave a
record, and most of these strikes never trade.

## Contents

| name | one row is |
|---|---|
| `e22_options_surface` | one instrument at one moment: strike, expiry, mark, forward, and the full greek set |
| `e22_options_book` | top of book for a near-the-money ladder: bid, ask, their sizes, and both quoted in vol terms |

## Reading it

`iv`, `delta`, `gamma`, `vega`, `theta` and `rho` are the venue's own values, published
alongside its mark price rather than recomputed here. That is deliberate: re-deriving implied
volatility needs a rate and dividend assumption, and a number the venue itself margins against
is the more useful one.

Each asset's chain is fetched in one request and every row from it shares a `sampled_ts`,
so grouping on `(asset, sampled_ts)` recovers one surface and adding `expiry` recovers one
smile. `round_ts` is shared by the assets collected in the same pass, which is what to
group on when comparing across assets at a moment.

`strike` and `expiry` are parsed from the instrument name, and both are null where the name
does not match the expected shape rather than being guessed. `expiry_ts` comes from the venue.

## What the mark costs you

The surface carries the venue's mark. `e22_options_book` carries what was actually quoted:
`best_bid_iv` and `best_ask_iv` give the two sides in vol terms and `iv_spread` the distance
between them. Across one ladder that distance ran from 4.2 to 18.6 vol points, and the mark sat
inside the band every time, near the middle rather than at it.

The ladder is deliberate rather than exhaustive, because a book costs one request per
instrument. It takes the strikes nearest the forward across several expiries and skips the
expiring contract, which was measured to have no book at all. So `e22_options_book` covers a
slice of `e22_options_surface` and never all of it; the two join on `instrument_name` within a
round.

A one-sided book is a real state and is recorded as one, with the missing side null while
`error` stays empty. `error` is set only where the request itself failed.

## Before you build on this

- One venue, and not the largest one. This is what a single order book quoted, not a
  market-wide consensus. Anything inferred about crypto volatility generally needs a second
  source.
- Every listed instrument appears, including deep out-of-the-money strikes that never trade.
  A mark price is published for those too, so filter before treating the surface as tradeable.
- Put and call at the same strike and expiry carry the same implied volatility by construction.
  Two rows agreeing is parity, not confirmation.
- Sampled at an interval, so a move that reverses between samples is invisible. `sampled_ts`
  is ours, taken at observation, and carries our network distance to the venue.
- A fetch that fails writes one explicit error row for that asset with the measurements null.
  Check `error` before reading an absent chain as a delisting.
- Coverage per asset differs and changes as the venue lists and expires instruments; count
  distinct `expiry` per day rather than assuming a fixed ladder.

Partitions are parquet, one file per collection window, under `dataset/YYYY/MM/`. Every dataset here carries a FIXED 7-day sample WINDOW starting at its own first day of collection, together spanning 2026-08-29 to 2026-09-04, so you can check schema, coverage and quality before asking for more. It does not advance, so there is nothing to gain by re-downloading it. The full history is held privately, available on request.

```python
from huggingface_hub import snapshot_download
import pandas as pd, glob

path = snapshot_download("dataforge-labs/crypto-options-surface", repo_type="dataset",
                         allow_patterns="e22_options_surface/**")
df = pd.concat(map(pd.read_parquet,
                   glob.glob(f"{path}/e22_options_surface/**/*.parquet", recursive=True)))
```

## Coverage

`e0_run_manifest` lists every collection window with its poll counts and failure counts, and is
published in full rather than windowed. Gaps between windows are real, cannot be filled in
afterwards, and nothing here is interpolated.

## License and contact

ODC-BY: use it freely, credit "DataForge (dataforge-labs)". Questions and requests for the full
history via the discussions tab.