What's the actual probability this stock rises 5% this week?

Trade on probabilities, not predictions.

Most tools give you a story. Data Investors is an end-to-end analytics platform that turns SEC filings, market prices, and macro data into calibrated probability estimates for short-horizon moves — rebuilt nightly, tested end to end, and graded against what actually happened.

See it in action How it works
14,164
daily price bars, split-adjusted
223
SEC quarterly filings parsed
17,941
news articles scored for sentiment
209+
tests behind the pipeline
What it does

One nightly loop, end to end

Every market evening the platform runs the whole journey — raw scrape to graded prediction — with no hands on the wheel.

1

Scrape & warehouse

Prices, SEC fundamentals, macro series, news sentiment, insider and institutional flow are collected after every close, then transformed through tested dbt layers — source to staging to facts to model-ready datasets.

2

Model & backtest

Gradient-boosted models retrain on a walk-forward schedule and answer one auditable question per stock: how likely is a +5% move within five sessions? Every retrain is scored out-of-time before its numbers count.

3

Publish & verify

Morning predictions are out by 9:15 ET. Every evening, past predictions are graded against realized prices and the grades are published — including the misses. Then the loop re-renders this site.

The data

Five sources, joined honestly

Daily market prices

~26 years of split-adjusted OHLCV per ticker, cross-checked between two independent vendors before anything downstream sees it.

SEC XBRL fundamentals

Structured filings straight from EDGAR with exact acceptance timestamps — a filing only touches a trading day it was truly public for.

FRED macro series

Every vintage of every series is stored — the model only ever sees the value that was knowable that day, not today's revision.

News & insider flow

17,941 sentiment-scored articles and 17,457 insider transactions, joined after their public timestamps (Form-4 deadline included).

Integrity rules

The three commitments

No lookahead, ever

Every input joins the dataset only after the moment it became public knowledge. Warehouse tests enforce it on every build, so backtests can't cheat and live results match tested ones.

Every claim is backtested

Models are graded out-of-time on dates they never saw, and calibration is checked by decile — stocks given a 20% probability should hit about 20% of the time.

Probabilities meet their base rates

A probability is never shown without the base rate it has to beat. The gap between them — not the number itself — is the signal.

Why it exists

This platform started as a personal question: could one person build a market-data system rigorous enough to trust with real decisions? The answer had to be verifiable — so every number on this site can be traced from a raw filing or price bar, through tested transformations, to a graded prediction. Meet the person behind it →

Explore the preview Read the engineering story