The Marta Walsh Index

Every Friday we count two things across Arizona’s $2 million-plus market: how many homes are exposed for sale, and how quickly buyers are putting them under contract. The ratio, corrected for the season and compared to the tier’s own recent history, is the Marta Walsh Index. At 100, the luxury market is behaving like its usual self. Above 100, demand is outrunning supply; below, supply is outrunning demand.

City-level market gauges exist and are useful — but no published index isolates the luxury tier, and the tier does not follow the city. In Scottsdale, we measured the two directly: the mass market ($1–2M) and the luxury tier ($2M+) agree about eras (level correlation 0.96) and disagree about moments — their three-month trends point in opposite directions in 31% of months, including May 2020 (mass at twice its normal pace, luxury exactly normal) and late 2022 through 2023 (mass cooling, luxury running 90+ points hotter than the city read implied). A luxury seller navigating by a city index is steering by an instrument that cannot see their market one month in three. This page documents the instrument that can.

The backtest

Construction

Universe. Every ARMLS residential listing with an asking price of $2,000,000 or more, 1998 to present — the same universe as One Shot — refreshed weekly by full-record reconciliation, so late entries and corrections are absorbed rather than frozen. A price-sanity guard excludes data-entry errors.

Demand is the count of listings going under contract in the trailing four weeks — dated to the contract event itself, not to when the status was entered, which keeps the index in event time rather than data-entry time. Supply is the count of listings actively exposed on the measurement Friday. A listing whose sale falls out of escrow is returned to the exposed pool for the interval it was, in fact, back on the market. ARMLS represents a home under contract and accepting backups not as its own status but as status “Active” plus a flag field, with the contract date entered only at conversion; the pipeline reads the flag, ends a flagged home’s exposure at its status-change date, and uses that date as the contract event until the true date supersedes it. One structural honesty note: a ratio cannot distinguish demand arriving from supply leaving — the index rises either way. The components below (raw contract counts, active counts, debut vs. returning listings) are published precisely so a reader can make that decomposition for any given week.

Seasonal correction. Weekly absorption is divided by a week-of-year factor estimated from the trailing ten years and refit annually — because the seasonal cycle itself has changed: the spring surge factor grew from roughly 1.0× in the mid-2000s to 1.5× in the 2020s, and a fixed correction would mis-read every recent February as hot.

The scale. Each week’s seasonally-adjusted absorption is divided by the median of its own preceding five years and multiplied by 100. So 100 means “normal for the market’s recent past” — 2018 is judged against 2013–2018, today against its own prior five years, and the norm rolls forward automatically so the index never anchors to an era that ages out. This is deliberate: a fixed $2 million line marked an estate tier in 2010 and marks ordinary luxury in 2026 (a repeat-sales look at same-home pairs shows the upper tier appreciating roughly 1.8× over the last six or seven years), so comparing today’s raw absorption to 2010’s against a single fixed baseline would mislead. The cost of the rolling norm, stated plainly: 100 means “normal versus the recent market,” not “balanced in some absolute sense.” Once a week is published it is frozen and never revised, even as the norm moves on.

Discipline rules. The published series is smoothed over four weeks. A weekly move within ±6% is reported as unchanged — the index does not manufacture narrative from noise. The headline is the week just measured, with data capped at the day before the Friday pull, so the reading is identical whatever hour it is run. Once a week publishes, its value is frozen and never revised — a number quoted from this index stays that number, permanently.

Components

Anticipated inventory. Alongside visible supply, we track every parcel that left the market after a failed campaign within the trailing year and has not returned or sold — currently 1,148 dark parcels against 1,140 active listings, about 1.0 per active. Historical return curves (Kaplan–Meier, with censoring, on 18,704 failed campaigns) say 29% of such homes relist within 13 weeks, 44% within six months, 53% within a year, and about 60% ever. The component is published as a measurement with a historical expectation attached — deliberately not as a forecast, for the reason given under What failed.

Debut vs. returning listings. New listings are split by whether the same parcel had a listing end within the prior two years. A rising share of returns means failed campaigns recycling through the market rather than fresh supply arriving.

Fast-lane share. The percentage of new contracts signed within 14 days of listing — the tier’s most sensitive temperature gauge (21% in 2019, 35% at the 2021–22 peak).

Price changes. The MLS export does not record when a price changed, so this component cannot be backtested; it is measured live, by comparing every active listing’s asking price between consecutive weekly issues — count of cuts, count of increases, and median cut size.

Shelf composition (added Sept 4, 2026; first published with the Sept 11 issue). What a buyer actually meets on the active shelf: the share of active listings that are either past 90 days on market or a return of a parcel whose prior campaign ended within two years, plus the count of fresh listings that are neither. Homes flagged UCB/CCBS are excluded as not shoppable. Because shelf age is strongly seasonal — early September is the annual stale peak — each issue carries the same-calendar-week value for 2019 and for the 2021–22 frenzy as reference. The finding that motivated the component: the shelf is always mostly rejected asks (87% stale-or-returning in the first week of September 2019, 60% at the frenzy’s equivalent week), so a rising active count usually measures the sitting market, not the clearing one.

What the index means for a single listing

We linked roughly 22,000 launch campaigns (2008–2024) to the index reading in their launch week and followed them to resolution. The relationship is real — monthly launch-cohort sell rates correlate with absorption at launch — but its shape is the finding, and the finding cuts against intuition: the market never guarantees a sale. Even in the frenzy of 2021, with the index above 400, roughly one listing in three still failed to sell. In the 2009 crash, four in five failed. Between those extremes, across the entire wide normal band, the sell rate moves far less than the market’s temperature would suggest. The market sets the odds; it does not set them to certainty at any point in the record.

The interpretation we place on this is deliberately conservative: sellers collectively adapt their asking prices to conditions, spending the market’s extra demand on ambition rather than banking it as probability — which is why conditions pay out in achieved price and speed rather than in the odds of selling. Stated as a rule: normal market fluctuation rarely changes a launch’s odds; regime changes change them enormously but arrive perhaps twice a decade; the asking price matters every day. This is why fifty Fridays a year the index’s honest reading is “the outcome is in the seller’s hands,” and why the rare weeks it exits the band carry real information. The evidence linking asks to outcomes is associational — documented in One Shot — and this section inherits that caveat.

What failed, and stays in

Two negative results are part of this methodology and we consider them load-bearing. First, we tested whether the off-market pool predicts future listing volume: a hazard-model forecast built from the pool lost decisively to the naive benchmark of “same quarter last year” (correlation 0.67 vs 0.91) and added no incremental accuracy when combined with it. The pool is therefore published as a nowcast, never a forecast. Second, listing-date seasonality moves speed, not odds: in the current era the best calendar week improves time-to-contract by roughly two weeks while changing the probability of selling by approximately nothing. Claims this index will not make are as much a part of its definition as the ones it does.

Flow leads price — and what this index is not

The MWI measures how fast the market is clearing, not where prices are going. The distinction matters most at turning points. In our own record, tier-wide absorption bottomed in the first quarter of 2009; the tier’s price level — measured by our repeat-sales index — did not bottom until the fourth quarter of 2011, thirty-three months later. Through 2009–2011, absorption ran strong while prices kept falling, because returning buyers were clearing a distressed pipeline that kept refilling; prices turned only when the pipeline exhausted. Housing clears on volume before it clears on price, and any supply-demand gauge is structurally early on price for that reason.

There is also an asymmetry worth stating plainly. Demand cannot hide: a buyer who does not bid does not exist, so when demand withdraws, a ratio index sees it immediately — flow gauges have historically been sharp at tops. Supply can hide: foreclosure pipelines, discouraged owners, and failed campaigns waiting off-market can feed a recovering market for years, which is why flow gauges run early at bottoms. The Anticipated Inventory component exists to shrink exactly that blind spot — it is our estimate of the hidden reservoir — while making no claim to see all of it. Accordingly: this index is not a price forecast, and sustained readings should be interpreted through the components, not the headline alone.

Soft-launch revision, Aug 29, 2026: readings computed before this date predate the UCB/CCBS flag handling described above and read low; the series was recomputed under the final method. Formal amendment logging begins at public release.

The full record, scored month by month →  ·  Published issues →  ·  Weekly series, CSV → (CC BY 4.0 — free to reuse with attribution)

Data and methods notes

The MWI publishes here every Friday. Questions about the construction are welcome — email Marta or call 480-274-5710.

Technical questions about construction, data, or replication: Matt Walsh · matt.walsh@russlyon.com · 480-277-1117.