Nowflation Methodology: 1,684 Series, 26 Sources, Base January 2018 = 100

689,676 observations. Shelter: Nowflation Realtime prices it at today's new-lease rents; the Nowflation CPI Gauge rolls those rents through the CPI's slow lease turnover (the all-tenant rent index), as the CPI itself does. Historical values are vintage-true and never silently revised; every methodology change ships in a dated public changelog.

What Nowflation Realtime measures

Nowflation Realtime measures the prices being set in the US economy today. Where a live market price exists we use it. That means pump prices, vehicle auction and transaction prices, and rents on leases signed this month. Where no credible live source exists we hold the latest official reading and label it as held. We don't tune this number to match CPI or any other index. CPI runs on inputs that are weeks to years old, so the gap between Realtime and CPI is the point of the series. We judge Realtime on how much of the basket is live and how old its inputs are. We also check whether official CPI moves toward it over the following 5 to 12 months.

What the Nowflation CPI Gauge measures

The CPI Gauge estimates what official CPI would read if its inputs were refreshed today. It keeps CPI's construction, including the slow way shelter moves as old leases roll over. It's the bridge between Realtime and the print, and it feeds our CPI forecast.

Nowflation Realtime: method v2.2

Each of the 21 CPI components uses a live market price where a credible source exists and holds the latest official BLS subindex otherwise, labeled as held. A live source is stale after twice its publishing cadence; the component then holds the official reading. Monthly sources enter as their own year-over-year change for their latest reference month; pump prices compare the same calendar day a year earlier. Weights, base period and chain-linking are the CPI Gauge's. Shelter in the published number: rent and owners' equivalent rent both from Zillow ZORI (all homes, SA) and Apartment List national rent (new leases). No public renewal-rent series with monthly cadence and usable terms exists, so Realtime shelter is new-lease only. Why Realtime shelter uses the Zillow and Apartment List blend. Both shelter lines, Shelter: Rent and Shelter: Owned, use an equal blend of Zillow ZORI (all homes) and Apartment List. Apartment List measures transacted new-lease rents, which net out concessions that Zillow's asking rents do not (39.2% of Zillow rental listings offered a concession in August 2026, per Zillow), and pricing all of shelter on one index matches how the CPI Gauge prices shelter. Apartment List publishes no single-family series, so the owners' weight uses a blend that leans toward apartment rents. Zillow single-family is tracked in shadow for comparison. Apartment List publishes its Rent Estimates for public use. If that changes, Realtime moves to Zillow all-homes on both shelter lines as a logged method change. Source: Apartment List National Rent Report. Schedule: Nowflation Realtime is computed at 7:55 AM ET and 6 PM ET. The 7:55 AM run writes the day's single live reading once its own checks pass, and the 8:05 AM publish shows it; the 6 PM run is recorded with its checks but is not published and never replaces the live reading. The Nowflation CPI Gauge is recomputed at 7:55 AM and 6 PM ET and published at 8:05 AM ET. Live readings from 2026-09-15 are stored as published and never rewritten. Method v2.2 since 2026-09-28 (v2.0 before; readings published under v2.0 are unchanged). Values before 2026-09-15 are a reconstruction under method v2.0. Rents: Data Provided by Zillow Group; Apartment List. The Nowflation Realtime page lists every component's source and input age.

Nowflation CPI Gauge: method v1.5 (since September 28, 2026; v1.4 before)

What this page documents

This page is the engine's own account of how the Nowflation CPI Gauge is computed, written out of the same run that produced today's number (method v1.5, as of October 18, 2026) so the description cannot drift from the calculation. The Gauge is a daily price index on the CPI basket: 21 components carrying the December 2025 BLS relative-importance weights, indexed to January 2018 = 100, with live market prices spliced onto the official subindex wherever a source has a better record than the survey, and the official subindex held between prints everywhere else. Today 4 components (39.7% of the basket) carry a live source and 17 (60.3%) hold their last BLS print. The inventory behind it is 1,684 series, 689,676 observations, 26 sources.

The five steps, as the engine runs them

Method v1.5 since September 28, 2026: BLS relative-importance aggregation (price relatives since December), monthly live sources compare the same month a year apart and step aside once the BLS has published a newer month; the index is linked to the v1.4 level on the day before the switch, and readings published before it are unchanged.

1. Collect. 1,684 public series from 26 sources, plus 4 licensed series that are not redistributed, all pulled on a fixed ET schedule (FRED, FMP, Zillow, AAA, Mortgage News Daily, Manheim, Cox, USDA and others); the Cleveland Fed nowcast and Kalshi prices are also pulled, for the scoreboard, not the index. Every observation is stored with a vintage date — we archive what was known when, so history can't be silently rewritten. The basket is 21 components: 20 official BLS subindexes (seasonally adjusted, with their NSA twins ingested for validation) plus a 0.94% residual that rides core CPI as a proxy.
2. Rebase. Every input series is indexed to January 2018 = 100. Series that start later are spliced: scaled to match the component's index level on their first overlap date.
3. Blend & splice. Volatile components ride live high-frequency data spliced onto official BLS history: shelter = an all-tenant rent index built from new-lease rents (80% Zillow ZORI + 20% Apartment List) — a tenure-weighted average of the last 48 months of new leases (monthly turnover 0.045, mean tenure ~22 months) — a measurement of what the CPI's slowly rotating rent panel pays this month, not a forecast (vs the BLS all-tenant research index: r 0.987, ours minus BLS +0.69pp over its last three published quarters); fuel = AAA daily pump price (validated weekly vs EIA); used vehicles = the Manheim wholesale index, entering 7 days after month-end when Cox publishes it; new vehicles = the Kelley Blue Book / Cox average transaction price, monthly (what buyers paid — mix-affected, so it runs hotter than the quality-adjusted CPI subindex when the sales mix shifts up-market). Food at home stays official: the USDA advertised-price composite (beef, pork, chicken, eggs, dairy, fresh produce, weekly) is ingested and graded as a diagnostic but not spliced — it correlates 0.71 with CPI food at home in the same month but reads ~1.7pp below the shelf on average and worsened the print-eve grade at both horizons. A live source that goes stale (no new data for its staleness window — 14 days for daily feeds, 60 for monthly) stops counting as live: the component holds its last live level and moves with the official BLS subindex until the feed returns. From method v1.5 (2026-09-28) a monthly live source also steps aside while the BLS has published a newer month than its latest: the component takes that newer official print and counts as held, not live, until the source publishes the same month or later. Every other component (medical, recreation, education & communication, transportation services, household furnishings, personal care, electricity, utility gas…) carries its official BLS subindex at its last measured print between releases — no momentum, no extrapolation.
4. Quality control. Any component moving more than 5% in one day from a live source is held one day at its prior value and flagged — one bad reading from a source can't print. Official BLS steps are exempt (already vetted). Publication never blocks: missing inputs carry forward and reduce the coverage score instead. The October-2025 CPI hole (BLS collected no data during the shutdown) is closed for year-over-year bases only: each carried subindex's October level is imputed as the geometric mean of its September and November prints, dated available on the November release, so the vintage-true daily history is unchanged (imputed months in use: 2025-10; components: 18). A QA check fails the run if any carried component's base month has no observation.
5. Aggregate. Headline = the BLS December chain (method v1.5 since 2026-09-28): the CPI-U relative-importance weights — the December 2025 table governs 2026 (the December table of the prior year governs each year; every row is the published table, none is hand-seeded) — multiply every component's price change since the previous December, as the BLS aggregates its own index, so no component is re-weighted by its cumulative price change since 2018 and the January weight switch never moves the level. YoY joins the year-ago-to-December segment, on the base year's weights, to the December-to-today segment, on this year's; each component enters on its own base: for a carried subindex the level 12 months before the print it holds, for a monthly live source its own month a year earlier, for a daily one the same calendar day a year earlier. Each component's contribution is its exact share of the change, so the contributions add up. Readings before 2026-09-28 stay as published under v1.4 (Σ weightᵢ × component indexᵢ, Laspeyres-style, chain-linked at December 31, with calendar-day bases for live overlays); the level is linked on the last day before the switch. MoM composes the same way. Every monthly input enters the daily history on its publication date — the BLS release calendar for CPI subindexes, each source's observed publication lag otherwise (Apartment List 2 days after month-end, Manheim 7 days after month-end, EIA weekly gasoline 2 days after its reference date) — never on its reference date. Four variants are published: CPI-comparable (the all-tenant rent index carrying both the rent and the owners'-equivalent weight, 34% of the basket), its seasonally adjusted-fuel twin for the pace, supercore (services ex-shelter, now including transportation services), and PCE-weighted (same components under BEA PCE relative-importance shares — out-of-pocket scope, graded against official PCE).

The basket on October 18, 2026: weight, source, lag and reading

Weights are the BLS CPI-U relative importance table, December 2025 (bls.gov/cpi/tables/relative-importance), in force for 2026; refreshed each January; 0.94% of the basket has no ingested subindex and rides core CPI as a proxy. Year-over-year for each component is measured over exactly twelve months on its own base (headline YoY = the BLS December chain: the year-ago-to-December change on the base year’s weights times the December-to-today change on this year’s, each component on its own base (the held print’s month−12 for carried subindexes, the source’s own month a year earlier for a monthly live source, the calendar day for daily ones)). Shelter: Owned is priced in the Gauge on the all-tenant rent index, so its row shows that reading; its own official subindex is named in the source column.

How shelter is measured

Shelter is the largest and slowest part of the CPI, so it gets its own model. what the CPI's slowly rotating rent panel pays this month (tenure-weighted new-lease rents), measured, not forecast. The parameters are a monthly turnover of 0.045 over 48 months of leases (mean tenure about 22 months), chosen by mean of MAE at k=0 and k=1 of the measured all-tenant YoY vs CPI rent+OER YoY (weights from the BLS table), common sample (score 0.857). Against CPI rent plus owners' equivalent rent it reads with a correlation of 0.846 in the same month (mean absolute gap +0.92pp, bias +0.49pp, 79 months since January 2020) and 0.889 one month ahead (gap +0.79pp, bias +0.51pp). Against the BLS's own all-tenant regressed rent index (atrr_yoy_us, 23 quarters since January 2018) the correlation is 0.987 with a mean gap of +1.29pp; ours minus BLS, last three published quarters: 0.7pp (positive = we read above the BLS all-tenant index) (2025-07: BLS 2.9%, Nowflation 3.58%). Last calibrated September 20, 2026.

Shelter: Owned (home-buying affordability measure, shown on /housing). P(t) = L·r(1+r)^360 / ((1+r)^360 − 1), L = 0.80 × ZHVI, r = 30yr rate / 12. Inputs: Zillow Home Value Index (mid-tier SA) — the home price in the payment calc (zhvi_us, through 2026-09-30); Freddie Mac PMMS weekly 30yr (official fallback) (MORTGAGE30US, through 2026-10-15). Marginal-buyer construction: current price × current mortgage rate, the rate from the Freddie Mac weekly survey for every day of the history; the year-ago payment is the Zillow month held today minus 12 at the rate a year earlier. Runs hotter than CPI when rates rise, colder when they fall. The CPI-comparable variant instead applies the market-rent blend to the OER weight.

Validation against the official print

Each published variant is scored monthly against the official series it is meant to track over the full history since 2018 (Correlations computed monthly against official CPI YoY (NSA, latest vintage) over the full 2018-to-now history. Beware single-number correlation claims: Pearson ignores constant level offsets, so a series can advertise 0.99 correlation while sitting 2pp away from the print.).

Lead curve — the CPI Gauge's monthly YoY against official CPI YoY k months later: k = 0: 0.9915 (92 months); k = 1: 0.9768 (91 months); k = 2: 0.9427 (90 months); k = 3: 0.9035 (89 months); k = 4: 0.8592 (88 months). Peak correlation is at k = 0. Whether that beats CPI's own persistence is the question the lead-test page answers; a high correlation alone does not settle it.

Live coverage today: 39.7% of basket weight is priced from a market source ahead of the print; the rest holds the last official reading.

Known limitations

• Roughly two-thirds of basket weight rides official BLS subindexes between monthly prints (sticky categories move slowly; this costs little accuracy but means those tiles only update on CPI day).
• Zillow/Apartment List publish revised histories; our own vintage archive is clean from launch (2026-07) forward, earlier history uses their latest published revisions.
• Weights are the published BLS December relative-importance values (December 2025 for 2026), renormalized to sum to 1; a new year uses the newest table until the January refresh. About 0.94% of the basket has no ingested subindex and rides core CPI as a proxy.
• Shelter is the all-tenant rent index measured from new-lease rents, which has read above the BLS all-tenant research index in recent quarters (ours minus BLS +0.69pp over its last three published quarters, r 0.987); it measures the CPI's rotating rent panel, not the marginal tenant.
• Construction changes are graded in-sample on the print-eve record since 2019 before adoption; misses are autocorrelated, so the record holds far fewer independent observations than prints (the effective sample is on the lead-test page) and gains under ~0.1pp are not provable within three years of live prints. The live scoreboard accumulates from the adoption date.
• Component-level readings on this dashboard are Nowflation-computed (SA-based) series; official figures are quoted separately on the top cards, NSA at 1 decimal, exactly as BLS publishes them.
• The CPI Gauge's print-eve record since January 11, 2019, as graded on the lead-test page: 93 prints ≈ 20 independent observations (gap autocorrelation 0.65); the MAE is 0.3 ± 0.1pp.

Reading the Gauge next to the CPI

Today the Nowflation CPI Gauge reads 3.45% year over year against the official 3.7% for September 2026 — a gap of -0.25pp. Today that gap is mostly shelter (−0.41pp), with motor fuel pulling the other way (+0.17pp); the gap page breaks it down component by component. A positive gap means the Gauge's live prices are running ahead of the print, a negative one that they are running behind it; over time shelter, a third of the basket, is the largest structural source of the difference, because the survey carries leases signed up to a year earlier. The October 2025 CPI gap (no data collected during the federal shutdown) is closed for year-over-year bases only, by imputing 18 carried components as the geometric mean of their September and November prints. Nowflation CPI Gauge: method v1.5 since September 28, 2026, v1.4 before that (the archive was recomputed once under v1.4, on September 8). Readings are kept as published, as logged on the corrections page. They have not changed since.

Freshness formulas

• live share: sum of basket weights of components whose value today comes from a non-BLS market source that is not stale (older than 2× its expected cadence) and not older than the month its official subindex already holds
• stale: a live component whose every source is older than 2× its expected cadence; it holds its official BLS subindex until a fresh observation arrives
• stepped aside: a monthly source inside its freshness window steps aside once the BLS has published a newer month; the component holds that newer official print and counts as held, not stale
• avg input age days: basket-weighted average of as-of date − observation date: daily inputs use the price date; monthly inputs and held BLS subindexes use the last day of the reference month; the CPI Gauge shelter adds the mean lease age implied by its tenure memory

Terms on this page

All-tenant rent index

The Nowflation CPI Gauge's shelter measure: new-lease rents from Zillow and Apartment List averaged over the trailing 48 months of leases with weights that decay at the lease-turnover rate, so it reflects what the whole renting population pays this month rather than only the marginal tenant signing a lease. The Gauge uses it as measured on both the rent and the owners'-equivalent weight; its lead over the official rent index is tested separately.

Owners' equivalent rent (OER)

OER is the CPI's measure of housing costs for homeowners: what an owner would pay to rent their own home, imputed from the rents of comparable rental units in the BLS rent panel rather than from mortgage payments or house prices. It is the single largest CPI component, which is why the rent panel's six-month repricing cycle matters so much for headline inflation.

Vintage (vintage-true data)

A vintage is a statistic exactly as it stood on a given date, before later revisions; a vintage-true backtest uses only the data that existed the day before each release, so it cannot borrow information from the future. Nowflation stores its own history vintage-true and never silently revises a published value.

Seasonal adjustment (SA vs NSA)

Seasonal adjustment removes the regular calendar pattern (holiday sales, summer gasoline, January price resets) so month-to-month changes can be compared; BLS quotes the monthly change on the SA index and the year-over-year rate on the NSA index, where seasonality cancels. Nowflation follows the same convention, and its pace figures use the BLS gasoline factors.

The gap

The difference between the Nowflation CPI Gauge and the latest official CPI print, in percentage points. Over time its largest structural source is shelter timing — rents measured this month versus a survey that reprices each unit every six months — while on a given day live pump and vehicle prices can move it more; it changes sign, and the gap page breaks today's figure down component by component. Nowflation Realtime has its own gap to CPI, shown on its page.

Relative-importance weight

The share of consumer spending the BLS assigns each CPI item, published each December and used for the following year; the Gauge uses the same table so its basket is the CPI basket.

Carry-forward

The rule for a component with no live source: it holds the level of its last official subindex print until the next CPI release, with no extrapolation.

Chain-linking

Joining index segments computed under different weights at a common date (here December 31) so a weight change moves neither the level nor the year-over-year rate.

Related: How the Gauge is built · Guide · Corrections log · Open data · System status