Walkthrough
How a signal is found
One real clue, start to finish: how much paperboard U.S. mills make, and why it might say something about Amazon. Every number on this page is read from the site’s own data when the site is built.
- 46monthly checkpoints tested
- 31times it made a call
- 58.1%beat the S&P 500
- 50 dayspublication delay assumed
1. The idea
Amazon ships a staggering number of boxes. Boxes are made from paperboard. So if U.S. mills are making more paperboard than a year ago, more goods are probably moving through warehouses. If they’re making less, fewer packages may be in the system. It’s a hunch, not a fact, and the rest of this page is about testing it.
Why output and not box prices? Prices can rise just because pulp or energy got more expensive. Output measures volume, which is closer to what we care about. Box prices still appear on the Amazon page, but only as context.
2. Where the data comes from
The Federal Reserve publishes an industrial production index for paperboard mills every month. It’s free and public: series IPN32213N on FRED, Federal Reserve Bank of St. Louis. An index just means “output compared with a base year,” so 100 is the base year’s level and the direction matters more than the number itself.
The last 36 months of paperboard output, newest reading 87.0 (June 2026). 3. How it’s pulled on a schedule
A GitHub Actions job runs every 4 hours. It asks FRED for the latest numbers, saves them as a small data file, and commits it. That commit redeploys the site automatically, so the page is never more than a few hours behind the source. Government data comes out slowly, which is why the newest reading is from June 2026.
4. How a number becomes a lean
The site averages the last 3 months (88.1) and compares it with the same 3 months a year earlier (89.7). That’s a change of -1.8%.
The rule is simple: a move bigger than 3% either way counts. Up more than 3% is a tailwind (a clue pointing up), down more than 3% is a headwind, and anything smaller is quiet. Right now this signal is quiet (no call). Amazon’s overall lean blends it with its other signals, like online store sales.
5. How the correlation scout fits in
This signal started as a hand-picked idea, not a scout find. The scout is a separate monthly job that tests lots of offbeat public series against stock prices and ranks the strongest links, checking whether the data moves before the price. It’s a brainstorming tool: with that many comparisons, some links show up by pure luck, so a scout idea still has to pass the same honest test below before anyone should trust it.
6. Testing it honestly
The site rewinds to each month from January 2023 to October 2026, and asks what this signal would have said using only data that was public by then. It assumes a 50-day delay before each reading came out, so it never peeks at the future. Then it checks what Amazon’s stock did over the next month.
Across 46 checkpoints the signal pointed up 8 times, down 23 times, and was quiet 15 times. Quiet months aren’t scored, which leaves 31 calls.
7. The fairer test: beating the market
In a rising market almost everything goes up, so “the stock went up” is too easy a test. The tougher one is whether Amazon did better than the S&P 500 over the same month when the signal pointed up, or worse when it pointed down.
Paperboard called the direction right 58.1% of the time and beat the market 58.1% of the time. That sounds good, but with only 31 calls the honest range is roughly 40.8% to 73.6%, and that range includes 50%. So it might be real, and it might be luck. In the out-of-sample study, which only judges each year using earlier years, the signal was kept.
Zoomed out, the whole site’s leans are much closer to a coin flip: 49.8% right on direction and 47.8% vs the market across 765 calls. I tried 8 ways to improve that and 5 new data sources; 0 made the cut. That’s on the track record page, in full.
8. What it means for Amazon
Amazon’s combined lean made 22 scored calls in the backtest. It was right on direction 68% of the time and beat the S&P 500 59% of the time. That’s encouraging, but the sample is small, and every call so far has been a tailwind. A clue, not a forecast.
Live tracking started in October 2026, so real-time results will build up week by week on the track record.
Education only. This page explains how one public data series is turned into a clue. It is not financial advice and never tells anyone what to trade.
See Amazon’s live signals → · How I built this → · Full method →