The method is a modified Case-Shiller weighted repeat-sales design, estimated in three stages.
- Assemble repeat sales: collect houses with at least two conforming purchase loans since January 1975; drop condos, multi-unit homes, non-conforming loans, and bad records.
Group sales by month using loan origination date.
- First stage (OLS, simple averaging): fit typical log price change across sale pairs; gains of 0.030, 0.012, and 0.018 average about 0.020.
The example trio last sold in January and resold in April, a gap of 3 months.
- Second stage (variance fitting): estimate how price spread grows with time between sales, using interval length without a fixed baseline.
- Third stage (GLS, variance weighting): re-average giving less weight to long-gap pairs; the example 0.020 becomes about 0.019.
Convert to index with January 1991 as 100: 100 times e (math constant) to the 0.019 gives about 101.92, equal to plus 1.9%.
- Seasonal adjustment: remove normal calendar patterns with Census X-13 software; the MoM headline uses adjusted values, and both adjusted and unadjusted series are published.
- Geographic aggregation: chain state growth rates into divisions, then the nation, weighted by each state's detached housing stock.
Weights update yearly from census and survey data.
- Headline rates: divide the index gain since the prior month (or prior year) by that earlier index.
Index values alone mean little; only ratios between periods matter.
the index tracks price change for the same houses, not average or median (middle) sale prices, which mix price with home types sold.