| --- |
| pretty_name: Options Yearly Transaction Cost Params |
| language: |
| - en |
| license: other |
| task_categories: |
| - tabular-regression |
| tags: |
| - finance |
| - quantitative-trading |
| - backtesting |
| - options |
| - derivatives |
| - transaction-costs |
| - bid-ask-spread |
| - market-microstructure |
| size_categories: |
| - n<1K |
| extra_gated_prompt: "This dataset is free to browse and gated for download. Approval is tied to a Papers With Backtest subscription, which also covers the other datasets of the organisation: https://paperswithbacktest.com/pricing" |
| dataset_info: |
| features: |
| - name: symbol |
| dtype: string |
| - name: tier |
| dtype: int64 |
| - name: year |
| dtype: int64 |
| - name: moneyness |
| dtype: string |
| - name: price_min |
| dtype: float64 |
| - name: price_max |
| dtype: float64 |
| - name: half_spread |
| dtype: float64 |
| - name: n_quotes |
| dtype: int64 |
| - name: n_underlyings |
| dtype: int64 |
| - name: method |
| dtype: string |
| - name: volume_min |
| dtype: float64 |
| - name: volume_max |
| dtype: float64 |
| - name: underlying_volume |
| dtype: float64 |
| - name: period_start |
| dtype: string |
| - name: period_end |
| dtype: string |
| - name: n_sessions |
| dtype: int64 |
| splits: |
| - name: train |
| num_examples: 880 |
| --- |
| |
| # Options Yearly Transaction Cost Params |
|
|
| The parameters of the cost model a backtest charges on a listed US option: **half the quoted bid-ask spread, in dollars**, by option price, distance to the money and liquidity of the underlying, with a row of its own for each of the 50 underlyings with the largest option volume. The options counterpart of [Stocks-Yearly-TransactionCostParams](https://huggingface.co/datasets/paperswithbacktest/Stocks-Yearly-TransactionCostParams) and [Forex-Yearly-TransactionCostParams](https://huggingface.co/datasets/paperswithbacktest/Forex-Yearly-TransactionCostParams). |
|
|
| <!-- MEASURED:START - rewritten by dataset_cards/refresh.py --> |
| |
| **880 rows**, 16 columns: 5 liquidity tiers and 50 underlyings, each a grid of 2 moneyness bands by 8 price bands, measured in **2026** on 4.4 million quotes of 5,416 underlyings. |
| |
| <!-- MEASURED:END --> |
| |
| ## Why It Matters |
| |
| [Options-Daily-Price](https://huggingface.co/datasets/paperswithbacktest/Options-Daily-Price) holds trades, not quotes, so a backtest fills at the last trade and must add the spread itself. On options that spread is not a rounding error: the median half spread is **2.6 % of the option's price** for the ten most traded underlyings, 5.3 % for the next 80, 9.3 % for the next 470, 13 % below them and 24 % for underlyings trading under a thousand contracts a day. A strategy that pays nothing for it, or a flat fee, is measuring something else. |
| |
| ## Load It |
| |
| ```bash |
| pip install --upgrade pwb-toolbox |
| ``` |
| |
| ```python |
| from pwb_toolbox import datasets as pwb_ds |
| |
| params = pwb_ds.load_dataset("Options-Yearly-TransactionCostParams") |
| spy = params[params["symbol"] == "SPY"] |
| ``` |
| |
| From `pwb-toolbox` 0.1.47, `pwb_toolbox.backtesting.options_half_spread(price, otm, tier=..., underlying=..., year=...)` reads this table and returns the dollar half spread of one contract, with the fallbacks below. |
|
|
| ## Columns |
|
|
| | Column Name | Description | |
| | :--- | :--- | |
| | **symbol** | Underlying ticker for an underlying's own row (`SPY`, `AAPL`); `tier_0` to `tier_4` for a liquidity-tier row. | |
| | **tier** | Liquidity tier, 0 to 4, by mean daily option volume of the underlying: `<1k`, `1k-10k`, `10k-100k`, `100k-1M`, `>1M` contracts. On an underlying's row, the tier it sat in over the measurement window. | |
| | **year** | Calendar year of the measurement. | |
| | **moneyness** | `near`: strike within 10 % of the spot (`K/S` for a call, `S/K` for a put, between 0.9 and 1.1). `far`: beyond. | |
| | **price_min**, **price_max** | Option price band, in dollars per share: 0, 0.10, 0.50, 1, 2, 5, 10, 25, and no upper bound (`price_max` null) for the last. | |
| | **half_spread** | Median of `(ask - bid) / 2` over the cell, in dollars per share, to $0.001. Multiply by the contract multiplier (100) for the cost of one contract, on each side. | |
| | **n_quotes** | Quotes in the cell. On a `tier_fallback` row, the underlying's own count, under 20. | |
| | **n_underlyings** | Underlyings contributing to the cell (1 on an underlying's row). | |
| | **method** | `measured`, or `tier_fallback` when the underlying had fewer than 20 quotes in that cell: the row then carries its tier's value. | |
| | **volume_min**, **volume_max** | Bounds of the tier, in contracts a day (`volume_max` null for the top tier). | |
| | **underlying_volume** | On an underlying's row, its mean daily option volume over the window; null on a tier row. | |
| | **period_start**, **period_end** | First and last snapshot of the measurement window. | |
| | **n_sessions** | Snapshots in the window. | |
|
|
| ## How It Was Measured |
|
|
| End-of-day snapshots of the whole US listed chain: the last quote of each contract before the close (15:59 ET for equity options, 16:14 for the ETFs that close late). Two-sided quotes only, and only contracts that **traded** that session, the ones a backtest on daily bars can actually fill. Each quote is placed in a price band by its mid, in a moneyness band by its strike against the underlying's price, and in a tier by its underlying's mean daily option volume over the window; the half spread of a cell is the median over its quotes. |
|
|
| The 50 underlyings with the largest option volume get a grid of their own, measured on their quotes alone. Their rows are not always tighter than their tier (Apple near the money at $1-2: 0.105 against 0.03 for the top tier); they are more accurate, not more lenient. |
|
|
| ## How to Read a Contract's Cost |
|
|
| 1. Use the underlying's own row if it has one, else its tier from its trailing option volume (21 sessions is what the published strategies use). |
| 2. Pick `near` or `far`, then the price band holding the option's price. |
| 3. Never charge more than the price itself: a $0.05 option cannot cost $0.05 on top of its price to sell. |
| 4. **A year without a measurement takes the nearest measured year**, and the cost should be labelled as extrapolated. Today only 2026 is measured, so every earlier year falls back to it. |
|
|
| `pwb-toolbox` applies these four rules, and maps `FB` to `META`'s row (the options table names an underlying as it was named at the time). |
|
|
| ## Caveats |
|
|
| - **One window, applied to every year.** The 2026 level is the only measurement so far. Option spreads were wider in 2014-2016, so the cost of the early years is understated; the tier, which follows the underlying's volume at the time, absorbs part of it. Years measured later are added as rows of their own and take precedence over the fallback. |
| - **Quoted, not effective.** A patient order inside the spread pays less than half the quoted spread; this is the cost of crossing it, the conservative side. |
| - **Five snapshots.** It measures the level of a cell, not its day-to-day variation, and the close of a quiet week rather than a stressed market. |
| - **No SPX, RUT, XSP or VIX quotes** in the snapshots: those options are priced by their tier. NDX is measured, and wide at 16:15. |
| - **Contracts that did not trade are left out**, so a far, deep or long-dated contract is measured only on the days it traded. |
|
|
| ## Access |
|
|
| Browsing the card and the schema is open to anyone. Downloading the files needs an approved request, tied to a subscription: [what each plan includes](https://paperswithbacktest.com/pricing). The same approval covers every dataset of the organisation. |
|
|
| ## Elsewhere |
|
|
| - [Options-Daily-Price](https://huggingface.co/datasets/paperswithbacktest/Options-Daily-Price), the contracts these costs apply to |
| - [The strategy catalogue](https://paperswithbacktest.com/strategies) |
| - [`pwb-toolbox`](https://github.com/paperswithbacktest/pwb-toolbox), the loader and the backtest harness |
| - [Every dataset in this organisation](https://huggingface.co/datasets/paperswithbacktest) |
|
|
