EstateFi Price Index (EFPI)

Methodology

Verbatim, as recorded in the daily build:

Repeat-sales price index (Bailey-Muth-Nourse regression, the method behind S&P/Case-Shiller). Built only from real on-chain YAM fills. Each observation is the log price change between two CONSECUTIVE sales of the SAME token, so property-specific price levels cancel and only genuine price change remains. Monthly log-index levels are estimated by OLS with the base month pinned to 100. Weighting: EQUAL-WEIGHT (each repeat-sale pair counts once) — 'notional' here is trade size, not property value, so value-weighting would let whales steer the index; a value-weighted variant is computed as a cross-check only. Dust rule: fills below $1 notional, non-positive price/amount, or price below 10% of the token's trailing median are dropped (a lone fat-finger cheap print), while genuine declines survive because the trailing median adapts. Pairs beyond 8x / (1/8) are dropped as residual data error; same-month pairs carry no monthly signal and are dropped. Sample discipline: a month is published only with >= 10 repeat-sale pairs touching it, else it is SUPPRESSED (not faked). Per-issuer sub-indices are each self-based at 100 in their own first solid month and emitted only where data allows. Limitations: RealT dominates the tape so the all-market series is effectively RealT-led (see coverage); thin issuers (Lofty/Blocksquare on YAM) get no index; only tokens that trade twice are visible; monthly granularity ignores within-month timing.

Why repeat-sales? A naïve average price would mix apples and oranges: a month with more trades of expensive properties would look like a price rise even if nothing actually got dearer. The repeat-sales method sidesteps this by measuring only the price change between two consecutive sales of the very same token. Because it is the same property both times, its idiosyncratic level — location, size, condition — cancels out, leaving only genuine price movement. This is the same design that underpins the S&P/Case-Shiller Home Price Indices (the Bailey-Muth-Nourse regression, 1963), the accepted benchmark for US housing.

Sample discipline. A month is published only when at least 10 repeat-sale pairs touch it; thinner months are suppressed, not fabricated. Fat-finger prints and dust are dropped by the rules above (sub-$1 notional, prices below 10% of a token's trailing median, or pairs beyond 8× / 1/8), while genuine declines survive because the trailing median adapts.

Equal-weight, by design. equal-weight per repeat-sale pair (value-weighted variant computed as cross-check). Here "notional" is trade size, not property value, so value-weighting would let a few large trades (whales) steer the whole index. A value-weighted variant is computed only as a cross-check.

Limitations, stated plainly. RealT dominates the on-chain tape, so the all-market series is effectively RealT-led (see coverage). Thin issuers get no index at all. Only tokens that trade at least twice are visible, and monthly granularity ignores within-month timing. We would rather publish a narrow, auditable benchmark than a broad one taken on trust.

By issuer

Each issuer sub-index is self-based to 100 in its own first solid month and emitted only where the data supports it. Issuers below the sample threshold are marked too thin — they are tracked, but no index is published.

"Solid mo." = months meeting the ≥10-pair threshold. "Pairs" = all retained repeat-sale pairs for that issuer. Latest and Δ are shown only for published sub-indices.

Citation & sources

The EFPI is free to use with attribution. Exact formulas and contract addresses are on the methodology page. Please cite:

EstateFi Price Index (EFPI) — EstateFi, estatefi.io — retrieved 2026-07-24

Source: EstateFi — estatefi.io. Index recomputed daily from public-blockchain data. Last generated 2026-07-24.