Methodology
How The Signal Index works
Fourteen leading indicators, weighted by historical predictive power, normalized across cities of vastly different sizes.
01 The 14 indicators
| Category | Indicator | Source | Weight |
|---|---|---|---|
| Capital | VC inflow (24-mo trailing) | PitchBook + Crunchbase | 14% |
| Capital | Round size distribution | PitchBook | 6% |
| Capital | Follow-on rate | PitchBook | 5% |
| Talent | BLS tech occupation growth | BLS QCEW | 9% |
| Talent | LinkedIn skill density | LinkedIn Economic Graph | 8% |
| Talent | Senior eng migration | 7% | |
| Innovation | USPTO filings (24-mo) | USPTO bulk data | 10% |
| Innovation | arXiv publications | arXiv | 5% |
| Innovation | GitHub repo creation | GitHub Archive | 6% |
| Demand | Job posting growth | BLS + scraped boards | 8% |
| Demand | Salary inflation | levels.fyi + BLS | 5% |
| Network | Meetup density | Meetup.com + Eventbrite | 4% |
| Network | Accelerator throughput | YC, Techstars, others | 6% |
| Network | University output | NSF + Open Syllabus | 7% |
02 Normalization
Raw values are normalized by metro population (or for very small metros, by labor force) to make a 200k-person metro comparable to a 4M-person metro. We then z-score against a 36-month historical baseline so a "score" reflects current momentum, not cumulative size.
03 Validation
We backtest by hiding 6 months of data and asking the model to predict outcomes that subsequently happened. Current model has 72% directional accuracy at the 12-month horizon and 64% at 24 months — better than chance (50%) and better than the consensus analyst (61% / 57%).
04 What we don't claim
- We don't claim point estimates (e.g. "Austin will raise exactly $X in 2027")
- We don't claim category certainty (e.g. "Quantum will / won't be big")
- We don't claim founder-level predictions
- We claim relative momentum and category-direction, with calibrated confidence