How Hedge Funds Use Job Posting Data (and the Traps in It)

Why funds buy job postings

A company has to post a job before it can hire for it. That makes job postings one of the few public signals that moves before the headcount, revenue and guidance numbers a company eventually reports. Funds use them for a handful of recurring questions:

• Hiring velocity: is a company posting more or fewer jobs than it did a year ago?
• Mix shift: is the hiring moving from engineering to sales, or from stores to warehouses?
• Geographic expansion: which metros and states are gaining openings?
• Wage pressure: are posted salaries for the same roles rising faster than peers?
• Technology adoption: which skills are showing up in a company's requirements?

Each question only works if a posting count means the same thing every month. Most of this post is about why that is harder than it sounds. Figures below come from Canaria's production archive (build of August 2026): unique US postings after semantic deduplication, mapped to listed employers.

The raw material: 7,889 tickers

Across our archive since 2022, 89,070,364 unique postings map to 7,889 stock tickers. In the US, the ticker-mapped share has been steady at about a fifth of all unique postings:

Most postings come from private employers, which is itself useful: a listed company's hiring can be read against private competitors in the same occupation and metro. Postings carry SOC occupation codes (99.3% of unique US postings since 2022) and NAICS industry codes (86.0%), so that comparison needs no manual mapping. The schema lists every field.

A worked example: the largest listed US posters

Here are the listed employers with the most unique US postings in the first seven months of 2026, against the same months of 2025:

We left Amazon out on purpose. It is the second-largest poster in 2026, but it publishes one role once per city, and those copies need collapsing before a year-over-year number means anything. That is the first of five traps.

Five traps in posting counts

1. One job, many copies

The same opening appears on the employer's site, several job boards and aggregators, and some employers post one role separately in dozens of cities. Count raw rows and you measure distribution, not demand. Canaria collapses cross-source copies with two-stage semantic deduplication: 1B+ collected postings reduce to 400M+ unique jobs. Multi-location postings of one role are the case to check by hand for your largest names.

2. Staffing firms post for their clients

Robert Half and AMN Healthcare both rank in the table above, but most of their postings are for client assignments. AMN's 39.8% decline is more likely a read on its clients' demand for travel staff than on AMN's own hiring. Treat staffing tickers as a read on their clients' industries, which the occupation and industry codes on the postings let you do.

3. Some postings are not jobs in the usual sense

Nerdy (NRDY) posted 134,068 unique US postings in the first seven months of 2026, more than Lowe's, largely tutoring roles. High-volume gig and marketplace posters swamp any ranking by count. Normalize each company against its own history before comparing it to peers.

4. History has to be point-in-time

A backtest breaks if the dataset was reconstructed after the fact: postings that were removed, companies that were renamed and tickers that changed all leak the future into the past. Our postings carry firstSeenDate and lastSeenDate from collection, a jobID that persists across deliveries, and a separate ticker history module, so a signal can be rebuilt as it would have looked on each date. Our archive starts in 2022, which gives roughly four years of history.

5. Company-to-ticker mapping is its own model

Postings name employers, not securities. A subsidiary has to be linked to its listed parent, and a foreign company to the line that trades in the US. Each posting's ticker comes with tickerRelation and tickerParentName fields showing how it was matched, plus a tickerMatchConfidence score, so you can drop weak matches rather than trust every link equally.

What a quant-grade job dataset needs

We built Canaria's delivery around that list. More detail is on our page for investment teams, and the methodology documents how each field is produced and measured.

How to start

1. Pick a universe (a sector, an index, a watchlist) and a window starting in 2022.
2. Pull unique postings for those tickers and inspect the largest names for the five traps above.
3. Normalize each company against its own trailing average before ranking.
4. Test the signal against the fundamentals you care about before trusting it in a model.

A free 5,000-record sample built on your own universe is the quickest way to see whether the data answers your question. If you are still comparing vendors, our guide to the best job posting data providers covers archive depth, deduplication and price.

This post describes data and methods. It is not investment advice.