tradevodata

Point-in-time US equity fundamentals from SEC EDGAR. Every value carries the date it actually became public, so a backtest can only see what was knowable at the time.

pip install tradevodata

The free sample needs no key

import tradevodata as tv

rows = tv.sample()        # 40 large caps, ~3,280 rows, CC0, no signup

Returns a pandas DataFrame if pandas is installed, otherwise a plain list of dicts — so the line above works on a bare install. Force either shape with to_pandas=True/False.

Same columns and the same semantics as the paid dataset, so you can write and verify your join logic before paying for anything.

as_of is required, on purpose

import tradevodata as tv

client = tv.Client(api_key="tvd_...")     # or set TRADEVODATA_API_KEY

client.fundamentals("AAPL", as_of="2024-06-30")

There is no way to ask this library "what is Apple's revenue?" — only "what was Apple's revenue knowable on this date?". On 2024-06-30 the answer is FY2023, because the FY2024 10-K had not been filed yet. That distinction is the entire product, and a default value for as_of is how lookahead bias gets written by accident.

Whole-universe cross-section

One call per rebalance date, instead of looping tickers:

snap = client.snapshot(as_of="2024-06-30", concept="Revenue", to_pandas=True)

Bulk

client.download("tradevodata.csv.gz")     # entire dataset, one file

Doing the point-in-time join yourself

If you're working from sample() or a bulk download, the as-of join is yours to do — as_of_filter does it correctly:

rows = tv.sample()
knowable = tv.as_of_filter(rows, as_of="2020-03-31")

Keeps only rows whose first_filed <= as_of, then the newest fiscal period per ticker/concept. That is the same logic the API applies server-side.

What each row tells you

The client raises a UserWarning when rows come back flagged or when your as_of runs past our data cutoff. Silence it with Client(warn_on_flags=False) if you're handling qa_status yourself.

Honest limits

• Annual only (10-K and 10-K/A). Quarterly is on the roadmap, not shipped.
• US only, and no delisted companies — so mind survivorship bias if you build universes from this alone. We fix lookahead bias; that is a different problem.
• 7 concepts, up to 12 fiscal years.
• Filing lag averages 66 days across the universe (median 60, max 120). The 40-company sample averages 43 — large caps file fastest, so the sample is better than the whole.

If you need quarterly, delisted coverage, or breadth today, Sharadar is genuinely good and you should buy that instead.

Zero dependencies

Standard library only. pandas is opt-in:

pip install "tradevodata[pandas]"

See the bug for yourself

The Colab notebook runs the experiment on the free sample — no key, no signup, no install. It joins fundamentals both ways at every month-end and counts the disagreements. On the 40-company sample: 47 of 413 ticker-months (11%) use a revenue number that was not yet public.

Links

• Docs — https://tradevodata.com/docs
• Free CC0 sample — https://github.com/christianpichichero-max/pit-fundamentals
• Methodology — how each number is derived from raw filings, so you can check any row

MIT licensed. Data sourced from SEC EDGAR (public domain). This is a dataset, not investment advice.