You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
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 in this organisation and the strategy catalogue at https://paperswithbacktest.com. Plans and what each one includes: https://paperswithbacktest.com/pricing
Log in or Sign Up to review the conditions and access this dataset content.
Stocks Daily Liquid Universe
Which large, liquid US equities were actually trading on a given day, and how they ranked. One row per session per name, ranks 1 to 1000.
5,789,995 rows over 3,954 symbols, 4 columns, covering 2003-10-07 to 2026-10-12. Refreshed monthly.
Why It Matters
A cross-sectional equity study has to answer one question before it computes anything: which names could have been held on that date. Answer it with today's index membership and the result is not a backtest. The winners of 2026 get replayed back to 2003, the names that mattered then and have since been bought out or gone bankrupt never appear, and the measured premium is the premium of having known the ending.
This table is the answer, measured at each session:
- 2,013 distinct names pass through the top 500 over the 23 years, against 500 on any one day. A study that loads only today's 500 is missing three quarters of the names it should have been able to hold.
- 705 of those 2,013 (35%) are delisted today. For comparison, only 3.5% of the symbols in
Stocks-Daily-Priceare companies that have since been delisted. The dead are in here, which is the whole point. - The frozen approach is wrong by a measurable amount, and it gets worse the further back you go. Of the 500 names ranked here at the end of 2005, 220 are in the current S&P 500 list — so a study using that list at that date is wrong about 56% of its universe. At the end of 2015 it is 39% wrong, at the end of 2020 34%.
How It Is Built
At each session, the names whose median dollar volume over the preceding 60 sessions is the highest, restricted to ordinary equity.
Liquidity, not index membership. No historical index-constituents feed is available to this pipeline: "the S&P 500 as of 2010" cannot be reconstructed from the data at hand, and inventing it is worse than not having it. A point-in-time liquidity screen can be built, is reproducible from the daily aggregates, and is what the literature means by "large, liquid US equities". It is not the S&P 500 and does not claim to be: even on the most recent session, 326 of these 500 names are current index members. The index excludes foreign-domiciled names that trade in enormous size here (BABA), and includes members that do not rank in the top 500 by dollar volume.
The median over 60 sessions, not the day's volume, so that one exceptional session — an earnings release, an index add, a short squeeze — does not pull a name in for a single day and out again the next. Measured turnover is 74 entries and exits a year on 500 names, 15%, the order of magnitude a real index turns over at.
Ordinary equity only. Common stock, ADRs, ordinary shares, GDRs and NY registry shares; not funds, ETFs, preferred shares, warrants, rights, units or index tickers. Exchange test symbols are excluded by name — ZVZZT was among the 500 largest dollar volumes of 2010-06-30, listed and heavily traded and entirely uninvestable.
Load It
Installation/Upgrade:
pip install --upgrade pwb-toolbox
Most users never load this table directly. pwb-toolbox resolves the "sp500" pseudo-symbol to it and enforces membership per date, so a strategy gets the point-in-time universe without asking:
from pwb_toolbox import datasets as pwb_ds
# every name that was in the top 500 at any point, each one priced only over
# the sessions where it was a member
df = pwb_ds.load_dataset("Stocks-Daily-Price", symbols=["sp500"])
To read the membership table itself:
from pwb_toolbox import datasets as pwb_ds
univ = pwb_ds.load_dataset("Stocks-Daily-LiquidUniverse")
print(univ[univ["date"] == "2010-06-30"].nsmallest(10, "rank"))
Example Output:
date symbol rank dollar_volume_median
2010-06-30 AAPL 1 6.904693e+09
2010-06-30 C 2 3.140464e+09
2010-06-30 BAC 3 2.811561e+09
2010-06-30 GS 4 2.159579e+09
2010-06-30 XOM 5 1.976669e+09
Columns
| Column Name | Description |
|---|---|
| date | Trading date (YYYY-MM-DD). US equity sessions only. |
| symbol | Stock ticker, as the consolidated tape carried it on that date. |
| rank | 1 is the most liquid name of that session. Ranks run to 1000 so a consumer can take the top 300, 500 or 1000 from the same file; pwb-toolbox takes 500. |
| dollar_volume_median | The median of volume * close over the 60 sessions up to and including this one, in dollars. The quantity the rank is on, published so a study can apply its own threshold instead of a rank. |
What This Data Does Not Cover
It is not index membership. See How It Is Built. A paper that specifies "S&P 500 constituents" is being approximated here, not reproduced, and the approximation is wrong about roughly a third of the names on any given day. For a paper whose result depends on the index rather than on size and liquidity — index-addition effects, index-rebalancing flow — this table is the wrong instrument and no screen built from prices would be the right one.
It begins in 2003-10-07, not 1990. The daily aggregates it is built from start there, and the 60-session median needs a quarter of history before the first ranking. A strategy specified over 1990-2026 runs over 2003-2026 here. The frozen list reaches 1990 only because it is not measuring anything.
A name needs 20 sessions of volume before it can be ranked, so a large IPO is absent from the universe for its first month even while it trades in size.
Ticker reuse is not resolved. A symbol is the string the tape carried, and a handful have been reassigned to a different company after a delisting. Joining this table to prices by (date, symbol) is correct; treating a symbol as a company identifier across decades is not.
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. The same subscription covers the other datasets in this organisation.
Elsewhere
Stocks-Daily-Price, the prices this universe selects from- The strategy catalogue
pwb-toolbox, the loader that resolves"sp500"to this table- Every dataset in this organisation
- Downloads last month
- 12