Correlworks

What it is

Correlworks answers one question for one stock at a time: across every other US-listed security, how did this one's price move with theirs? A stock page takes its pairwise correlations with up to 6,809 other tickers and prints them as a histogram from −1 to +1, with the mean, the share of positive pairs, the share above 0.7 and below −0.7, its SIC sector and a few sector neighbours. Around the 6,810 stock pages sit 184 sector pages, four ranked lists (strongest positive correlations, most positively and most negatively correlated, most independent) and a blog of data notes that began on 30 October 2026.

The paid layer is an interactive dashboard behind an account. Its public description sorts each stock into one of four shapes — market-aligned, low-correlation diversifier, defensive or inverse, regime-sensitive — with a probability for each and a confidence score that the page says is not a forecast of price.

The site is published by DATAUNION of Seoul, which signs the terms (governed by Korean law, effective 1 January 2026) and the privacy policy, and answers at a Gmail address.

Pricing

The stock, sector and ranking pages are free and need no account. Premium is the dashboard, listed at $10 a month; on 4 November 2026 it cost $0 "for a limited time", and the pricing page said the end date had not been set and would be announced there before any charge. LemonSqueezy takes the payments. No other plan is on the pricing page.

Data & coverage

One coefficient per pair: the Pearson correlation of daily closing prices, adjusted for splits, over a fixed window from 22 June 2020 to 18 June 2025 — about 1,255 trading days. It is not rolling, and it is computed on price levels, not daily returns. Dividend adjustment is not mentioned. Bars, names and SIC sectors come from Polygon.io, now Massive.

The universe is about 6,800 securities from the Nasdaq.com stock list: NASDAQ, NYSE and NYSE American listings, SPACs, units and warrants included, and closed-end funds such as PIMCO Dynamic Income Opportunities sit in the rankings beside operating companies.

The coefficients are counted in 0.2-wide bins and drawn on a 0.1 grid by splitting each bin in half, so neighbouring rows repeat: NVDA's −0.7 to −0.6 and −0.6 to −0.5 rows both read 712 pairs. The mean is estimated from bin midpoints. The vendor says true 0.1 bins arrive with the data update.

The homepage's NVDA chart is a separate dataset — a 30-day rolling correlation against 5,000-odd stocks, one distribution a day, from 11 April to 18 June 2025.

Integrations

None that work today. The API page describes one endpoint, correlations for a ticker over a rolling window of 20 to 500 days, and calls the contract a draft with illustrative examples; keys are requested by email. That endpoint answered 404 on 4 November 2026, and the examples repository on GitHub was last committed to on 25 August 2026. The terms forbid bulk scraping and redistributing the data.

Limitations

• The data is 16 months old. The last data date is 20 June 2025 on every page, with an update "in progress" and no schedule. A reader on 4 November 2026 is looking at a window that closed before the last year and a half of the market happened.
• Price-level correlation measures shared trends, not shared moves. The vendor says so on the methodology page: two stocks that both rose over 2020–2025 score high even if their daily moves are unrelated. A diversification or risk model wants return correlations, which this site does not publish.
• The headline numbers flatten the shape they are there to show. NVDA's page calls a mean of −0.072 "near-neutral" above a histogram with 16% of pairs above 0.7 and 41% between −0.7 and −0.3, and the 0.1 rows are halves of 0.2 bins rather than counts.
• The homepage promises more than the plan sells. It describes a pipeline ending in "weight-adjustment recommendations" and rebalancing triggers, a 0–100 diversification rating, CSV and JSON export and white-label licensing. The pricing page lists only the dashboard; this card was written without an account and cannot say which of those exist.
• Small details disagree. The window ends on 18 June 2025, while every page gives 20 June as the last data date. The privacy policy's effective date read "[DATE OF PUBLICATION]" on 4 November 2026.
• A thinly documented publisher. DATAUNION gives a city and a Gmail address, and no registration number, street address or named person.

Alternatives

Portfolio Visualizer and PortfoliosLab put a correlation matrix beside a backtest of the holdings you choose, on current prices — a handful of assets rather than the whole market, and the shape a portfolio question usually takes. ETF Insider does it for a portfolio of funds, beside holdings overlap. To build the whole-market version yourself, on returns and to date, pull daily bars from yfinance or Massive; Riskfolio-Lib and PyPortfolioOpt take the result into an optimiser.