Nowcast — methodology

How the composite works

Single indicators contradict each other and lag. A transparent composite turns the noise into one forward read you can defend, component by component. Black-box indices ask for trust — this shows exactly how each input moves the needle.

Normalize. Each indicator maps to a comparable signal in [-1, +1] around a neutral reference:
```
signal = clamp( (value − neutral) / scale × direction, −1, +1 )
```
direction is +1 when higher-is-good (e.g. PMI) and −1 when higher-is-bad (e.g. jobless claims). Extreme values clamp at ±1.
Weight. The composite is the weighted average of the normalized signals, so each component's contribution is signal × weight ÷ Σweights. The weight sum is always shown.
Read. The composite score maps to a regime:
- score > 0.2 → EXPANSION
- -0.2 ≤ score ≤ 0.2 → CAUTION
- score < -0.2 → CONTRACTION

Worked example (seed mix)

method.example

$ nowcast compute --explain
Yield curve (10y–3m)        signal=-0.27  contrib=-0.07
Initial jobless claims      signal=+0.08  contrib=+0.02
Manufacturing PMI           signal=-0.08  contrib=-0.02
Consumer sentiment          signal=-0.12  contrib=-0.02
Housing starts              signal=-0.13  contrib=-0.02
score=-0.11  →  CAUTION

Transparency & limits

No hidden weighting — every weight and input is editable.
Non-partisan framing; this is a method, not a forecast.
Illustrative only. Out of scope for v1: live data feeds and official forecasting.
Payment-dark — nothing here is gated or monetized.