OptionChainAnalytics

OptionChainAnalytics provides point-in-time option-chain containers, feed normalisation, chain reconstruction, queries, and visualisation in Python for quantitative research.

It is the data-container layer: provider credentials and data rights, pricing models, portfolio backtests, and proprietary datasets remain separate. Pricing and implied-volatility inversion are delegated to vanilla-option-pricers; generic time-series and plotting utilities come from qis.

Install

OptionChainAnalytics requires Python 3.10 or newer. CI covers Python 3.10 through 3.14.

Install the published package:

pip install option-chain-analytics

For development from a clone:

pip install -e .

Provider-specific integrations are optional:

For example, pip install "option-chain-analytics[cboe]" installs the CBOE file dependency without installing unrelated network providers.

First success: no data or credentials

The authoritative offline example constructs a deterministic Black-Scholes-Merton option panel, reconstructs a historical chain, queries its front-expiry ATM strike and volatility, and selects a weekly roll maturity:

python examples/first_success.py

Expected evidence:

ticker=SYNTH
observation_times=2
contracts_at_first_time=30
expiries=['12Jan2024', '19Jan2024', '16Feb2024']
first_expiry_atm=100.00, vol=0.2057
weekly_roll_expiries=['12Jan2024']

See examples/first_success.py for the executable source. The documentation includes that file directly, so the tutorial cannot drift into a second implementation.

Supported examples

OCA keeps six repository examples with explicit data boundaries:

• deterministic offline construction and one-expiry ThetaData-shaped analytics;
• resumable ThetaData EOD cache construction, cache-first ATM/skew plots, and PDF chain reports;
• standardized local SPX, VIX, BTC, and ETH cache construction.

The examples guide lists every script, its required data, whether it makes a network request, and its output. Removed research prototypes are not retained as runnable examples.

Data model

• OptionsDataDFs holds an option-observation panel (chain_ts) plus an aligned spot-price frame.
• SlicesChain reconstructs all available expiries at one exact observation time.
• ExpirySlice provides call/put, ATM, delta-strike, volatility, open-interest, and execution-price queries for one expiry.
• SliceColumn defines the common option-feed schema, including source time, contract, forward, discount factor, strike, expiry, quote, implied volatility, Greeks, volume, and open interest.

Observation and expiry timestamps are timezone-aware. Exact lookup is the reconstruction default; scheduled studies can explicitly select the latest previous observation, but never a later one. Volatilities are decimals (0.20 means 20%), time to maturity is in years, and each adapter must preserve and document its price/multiplier convention.

Provider adapters own feed-specific choices such as report alignment, rate symbols, settlement timestamps, product scope, and admissible bounds. Reusable numerical kernels live separately under option_chain_analytics.utils; the call-put parity utility has no provider or optional-solver dependency. Black-Scholes analytics remain delegated to vanilla-option-pricers.

Empirical feeds

Local adapters cover Deribit/Tardis crypto histories and SPX/VIX CBOE fitted-chain files. These datasets are not distributed. OCA_DATA_PATH holds raw provider archives, OCA_CACHE_PATH holds normalized reusable chains, and OCA_OUTPUT_PATH holds generated reports. Their source-checkout defaults are the ignored data/, resources/, and outputs/ directories. The data catalogue defines every supported provider directory, cache filename, and loader. CBOE files can be mapped with:

from option_chain_analytics import OptionsDataDFs
from option_chain_analytics.data.cboe import load_local_cboe_options_data

options_data = OptionsDataDFs(
    **load_local_cboe_options_data(
        ticker='SPX',
        start='2023-01-03',
        end='2023-01-03',
    )
)

The CBOE mapper always infers bid/ask implied volatilities from the source bid/ask prices using the contemporaneous forward, discount factor, and time to maturity. This keeps every CBOE-backed OptionsDataDFs instance on the same complete schema.

For repeated empirical studies, build one normalized Parquet cache per underlying after installing the cboe extra:

from option_chain_analytics.data.cboe import build_local_cboe_options_cache

build_local_cboe_options_cache(ticker='SPX')
build_local_cboe_options_cache(ticker='VIX')

This creates ignored resources/cboe_options/spx_options_oca.parquet and resources/cboe_options/vix_options_oca.parquet files by default. The normal loader uses a valid cache automatically and still accepts start/end filters. OCA embeds its cache schema and source-file fingerprint in each Parquet file and rejects stale caches. Use overwrite=True to rebuild deliberately.

CBOE data supplies implied forwards but no independent spot series. Pass spot_data, or use is_use_front_forward_as_spot=True only for visualisation; a forward proxy is not a valid spot return series for backtesting.

ThetaData EOD workflow

ThetaData national EOD reports provide historical US equity and ETF option quotes. OCA converts them to the same OptionsDataDFs schema used by the local CBOE and Tardis adapters. Index options with product-specific or AM settlement conventions are deliberately outside this adapter.

1. Install and authenticate

Install the optional official client integration:

pip install "option-chain-analytics[thetadata]"

The official client reads THETADATA_API_KEY or its supported credentials file. For example, a PowerShell session can keep the key in process memory without writing it into the repository:

$thetaKey = Read-Host "ThetaData API key" -AsSecureString
$env:THETADATA_API_KEY = [System.Net.NetworkCredential]::new('', $thetaKey).Password

Credentials and raw provider responses are never written to OCA's normalized cache.

2. Build or resume a local history

The supported builder requests one month at a time and checkpoints each completed partition. An interrupted request can therefore be rerun safely; compatible existing months are skipped. The default end date is yesterday, which is suitable for delayed-data accounts:

python examples/build_thetadata_eod_cache.py \
    --ticker NVDA \
    --start-date 2023-06-01

By default this stores normalized files under resources/thetadata_options/nvda/{options,spot}/YYYY-MM.parquet. Set OCA_CACHE_PATH or pass --output-dir to choose another private cache root. The default request keeps expiries from 0 to 60 calendar DTE and 20 strikes around spot; pass --all-strikes only when the larger download is actually required.

The same workflow is available through the package API:

from option_chain_analytics import build_thetadata_eod_cache, load_thetadata_eod_cache

cache_root = build_thetadata_eod_cache(
    ticker='NVDA',
    start_date='2023-06-01',
    min_dte=0,
    max_dte=60,
    strike_range=20,
)
options_data = load_thetadata_eod_cache(cache_root)

print(options_data.ticker)
print(len(options_data.chain_ts), 'option observations')
print(len(options_data.get_timeindex()), 'EOD chains')

Monthly Parquet files are only the resumable physical layout: the loader returns one continuous OptionsDataDFs. Date bounds avoid scanning the full cache when an analysis needs a shorter window:

nvda_july = load_thetadata_eod_cache(
    cache_root,
    start_date='2026-07-01',
    end_date='2026-07-31',
)

3. Reconstruct a chain and extract volatility

Every provider report retains its actual timestamp. Select one of those timestamps and reconstruct the chain exactly—there is no implicit borrowing from a later observation:

import pandas as pd

from option_chain_analytics import create_chain_at_time

value_time = nvda_july.get_timeindex()[-1]
chain = create_chain_at_time(nvda_july, value_time)

# Select the first listed expiry at least seven calendar days away.
slice_id = chain.get_next_slice_after_date(value_time + pd.Timedelta(days=7))
atm_vol = chain.get_atm_vol(slice_id=slice_id)
skew_25d = chain.get_skew(slice_id=slice_id, delta=0.25)

print({'value_time': value_time, 'expiry': slice_id, 'atm_vol': atm_vol, 'skew_25d': skew_25d})

Volatility values are decimals: 0.25 means 25%. ATM volatility averages the available call and put mark IVs at the nearest forward strike. OCA's delta skew is (call IV - put IV) / log(call strike / put strike) at the requested absolute delta.

4. Extract and plot rolling ATM volatility and skew

The following installed-package example reconstructs every exact EOD chain and rolls to the first expiry at least seven calendar days away. Because it uses only the public OCA API, it works after a normal pip install; the repository examples/ directory is not required:

import matplotlib.pyplot as plt
import pandas as pd

from option_chain_analytics import create_chain_timeseries

# Reconstruct exactly the reports present in the bounded cache. Using the
# observed timestamps avoids introducing a synthetic schedule or look-ahead.
observation_times = nvda_july.get_timeindex()
chains = create_chain_timeseries(
    options_data=nvda_july,
    dates_schedule=observation_times,
    time_selection='exact',
)

records = []
for value_time, chain in chains.items():
    roll_boundary = value_time + pd.Timedelta(days=7)
    eligible = [
        (expiry_slice.expiry_time, slice_id)
        for slice_id, expiry_slice in chain.expiry_slices.items()
        if expiry_slice.expiry_time >= roll_boundary
    ]
    if not eligible:
        continue

    expiry_time, slice_id = min(eligible)
    atm_vol = chain.get_atm_vol(slice_id=slice_id)
    skew_25d = chain.get_skew(slice_id=slice_id, delta=0.25)
    if atm_vol is None or not pd.notna(atm_vol):
        continue
    records.append(
        {
            'value_time': value_time,
            'expiration': expiry_time,
            'dte': (expiry_time - value_time).total_seconds() / 86_400.0,
            'atm_vol': float(atm_vol),
            'skew_25d': None if skew_25d is None or not pd.notna(skew_25d) else float(skew_25d),
        }
    )

vol_data = pd.DataFrame(records).set_index('value_time').sort_index()
if vol_data.empty:
    raise RuntimeError('the selected cache window has no eligible rolling expiries')

# Volatilities are stored as decimals, so multiply by 100 for percentage axes.
atm_figure, atm_axis = plt.subplots(figsize=(11, 5), tight_layout=True)
atm_axis.plot(vol_data.index, 100.0 * vol_data['atm_vol'], marker='o')
atm_axis.set_ylabel('ATM implied volatility (%)')
atm_axis.set_xlabel('ThetaData EOD observation time')
atm_axis.set_title('NVDA rolling ATM implied volatility')
atm_axis.grid(alpha=0.3)
atm_figure.autofmt_xdate()
atm_figure.savefig('nvda_atm_vol.png', dpi=160)

skew_figure, skew_axis = plt.subplots(figsize=(11, 5), tight_layout=True)
skew_axis.plot(vol_data.index, 100.0 * vol_data['skew_25d'], marker='o', color='tab:orange')
skew_axis.axhline(0.0, color='black', linewidth=0.8)
skew_axis.set_ylabel('25-delta skew (vol points / log-strike)')
skew_axis.set_xlabel('ThetaData EOD observation time')
skew_axis.set_title('NVDA rolling 25-delta implied-volatility skew')
skew_axis.grid(alpha=0.3)
skew_figure.autofmt_xdate()
skew_figure.savefig('nvda_25d_skew.png', dpi=160)

vol_data is the empirical table used by the plot. It retains the selected expiration and actual calendar DTE beside each ATM-volatility and skew observation, so maturity rolling is inspectable.

The figures below are generated internally from OCA's deterministic simulated data, using the same schema, chain reconstruction, and seven-calendar-day roll as the ThetaData example. They illustrate the expected outputs without redistributing licensed market observations:

Regenerate both documentation assets offline with python tools/generate_readme_figures.py.

The repository also provides cache-first command-line wrappers for separate ATM and skew figures:

python examples/fetch_thetadata_atm_timeseries.py \
    --ticker NVDA --start-date 2026-07-01 --end-date 2026-07-31 \
    --metric atm --output nvda_atm_vol.png
python examples/fetch_thetadata_atm_timeseries.py \
    --ticker NVDA --start-date 2026-07-01 --end-date 2026-07-31 \
    --metric skew --delta 0.25 --output nvda_25d_skew.png

5. Plot the complete option chain at one exact date

Load only the required report date, reconstruct its actual provider timestamp, and create the strike-space and delta-space figures for every live expiry. The figures can be inspected in memory or persisted as one multi-page PDF:

from matplotlib.backends.backend_pdf import PdfPages

from option_chain_analytics import (
    create_chain_at_time,
    load_thetadata_eod_cache,
    run_chain_report,
)

report_date = '2026-07-17'
one_day = load_thetadata_eod_cache(
    cache_root,
    start_date=report_date,
    end_date=report_date,
)
if len(one_day.get_timeindex()) == 0:
    raise RuntimeError(f'no ThetaData report is cached for {report_date}')

# The date identifies a cache partition; value_time is the actual provider
# report timestamp retained by OCA.
value_time = one_day.get_timeindex()[0]
chain = create_chain_at_time(
    options_data=one_day,
    value_time=value_time,
    time_selection='exact',
)
if chain is None:
    raise RuntimeError(f'no exact option chain is available at {value_time}')

print('value_time:', chain.value_time)
print('expiries:', list(chain.expiry_slices))
print('contracts:', len(chain.options_df))

figures = run_chain_report(chain)
with PdfPages('nvda_chain_20260717.pdf') as pdf:
    for figure in figures.values():
        pdf.savefig(figure)

The equivalent repository command is:

python examples/run_chain_report.py \
    --ticker NVDA --date 2026-07-17 --output nvda_chain_report.pdf

Snapshot and direct-live alternatives

fetch_thetadata_eod.py is a small single-expiry example. Without --live it is synthetic, deterministic, credential-free, and does not contact ThetaData:

python examples/fetch_thetadata_eod.py
python examples/fetch_thetadata_eod.py --live --ticker AAPL \
    --value-date 2026-07-24 --expiration 2026-08-21 --metric both

For a short range that should not be cached, pass --live to fetch_thetadata_atm_timeseries.py, or call fetch_and_plot_thetadata_atm_vols and fetch_and_plot_thetadata_skew from that module. Cache-first access is recommended for repeated research because it avoids downloading the same reports again.

The adapter records the provider's actual EOD report time, joins only an underlying report available at or before that time, and preserves both calls and puts. Direct snapshot/time-series loads retrieve ThetaData SOFR EOD by default to anchor the discount factor and then robustly infer forwards from parity; pass rate_symbol=None for a parity-only fit. The partitioned cache builder uses the parity-only policy. Every bid, mark, and ask IV and every mark Greek is computed by vanilla-option-pricers, rather than mixing provider and OCA analytics.

Documentation and development

Start with the documentation site, then read the schema contract, point-in-time reconstruction, and data-source guide.

pytest -q
ruff check src tests examples tools docs/conf.py
sphinx-build -W -b html docs docs/_build/html
python -m build

The installable package lives under src/option_chain_analytics/; repository-only scripts live in examples/. Raw datasets, normalized caches, agent reports, and generated outputs live in ignored data/, resources/, agents/, and outputs/ directories.

Research and licensing boundary

OCA can provide a public, auditable input layer for empirical studies and replication. Strategy logic and the QF-paper backtests remain in SigmaStrats, and a public example is not expected to reproduce results computed from a private production dataset exactly.

The software is released under the MIT License. Dataset licences and access terms are separate from the software licence. Citation metadata is provided in CITATION.cff, and contribution guidance is provided in CONTRIBUTING.md.