`docs/rent-multiplier-1978-1984.md`

63 lines
# Rent multiplier — LA 1978–1984 anchor

Do NOT use a canned inflation multiplier (CPI, 3%/yr, etc.). The rental
projection model must be **derived from real LA County data** over the
1978–1984 window — the inflection that turned a stable housing market
into the modern LA price regime.

## Why this window specifically

- **1978** — pre-Proposition 13 passage (June 1978) + stable single-family
  median prices through most of 1977–78
- **1984** — LA Olympics year; median LA-MSA home prices had roughly
  doubled in nominal terms, rents moved with them
- The pair frames the fastest rate-of-change event in post-war LA housing
  and gives us an empirical **time-to-double** for both sale and rent

## What we actually need from the data

1. **LA-MSA median home price, monthly, 1978-01 → 1984-12**
   Sources to ingest:
   - CAR (California Association of Realtors) median price series
   - Federal Reserve FRED series for LA MSA
   - LA County Assessor aggregated sale data (our own after ingest)
2. **LA-MSA median contract rent, same window**
   Sources:
   - BLS CPI-U rent-of-primary-residence, LA-Long Beach-Anaheim
   - US Census decennial + American Housing Survey
3. **Derived metrics** (what the app actually uses):
   - `rate_of_increase(year, year+1)` — observed yoY %, not assumed
   - `months_to_double(start_year)` — empirical
   - `rent_to_price_ratio(year)` — how rents tracked sales
   - A piecewise function: stable → acceleration → super-heat phases

## How the multiplier plugs into the app

Per-address rental projection never uses a flat annual %. Instead:
```
projected_rent(year) = base_rent * multiplier_curve(year_relative_to_1978)
```
where `multiplier_curve` is the empirical LA 1978–1984 trajectory fit
to the current local market's position on the equivalent cycle. This is
what lets the site say "at this address, based on LA's documented
1978–1984 acceleration pattern, the projected rent trajectory is X"
instead of a generic "assumes 3%".

## Data ingestion TODO

- [ ] Pull FRED series: `CASTHPI`, `MSPUS`, `LXXRSA`
- [ ] Pull BLS CPI-U rent series for LA MSA
- [ ] Pull CAR historical median series (requires login)
- [ ] Store in `price_series` table: `(series_id, period_start, value, source)`
- [ ] Compute `multiplier_1978_1984.json` — a single-file snapshot the app reads
- [ ] Flag all numbers in UI with year + source, never display raw forecast

## Editorial rules (from the research report)

- Every projected number is tagged with its source tier (A govt / B archive /
  C verified secondary / D social/lead)
- Never display a projection as "truth" — always show the anchor window
  ("projected via LA 1978–1984 trajectory, CAR series")
- No forward projections past the current year without flagging
  `confidence_score < 0.8`