The leading indicator league table, and why it reshuffles every cycle

Textbooks give a fixed order: building permits and factory hours lead, industrial production and employment coincide, unemployment lags. The claim is checkable, because the series are free and the turning points are dated by one body. Measured against the December 2007 and February 2020 peaks, the order comes out completely differently in the two episodes — one supposed leader led by 25 months in one and 3 months in the other. This page shows both, names the six indicators the dating committee actually uses, and reports two leading indices that simply stopped being published.

The six the dating committee actually uses

Before ranking indicators it is worth knowing which ones decide the answer. The National Bureau of Economic Research names them: real personal income less transfers, nonfarm payroll employment, employment as measured by the household survey, real personal consumption expenditures, manufacturing and trade sales adjusted for price changes, and industrial production. It adds that there is no fixed rule about what measures contribute information to the process or how they are weighted, and that in recent decades, the two measures we have put the most weight on are real personal income less transfers and nonfarm payroll employment.

Two things follow immediately. Not one of the six is a “leading” indicator — they are all coincident by construction, because the committee is dating the turn rather than anticipating it. And gross domestic product is not on the list at all; it enters only in the separate determination of the peak or trough quarter.

When each series actually turned, in two episodes

For each indicator, the method is: take the highest monthly value in the two years up to and just past the NBER peak, and report where it falls relative to the peak month. A series that peaks well before the business cycle peak led; one that peaks in the same month coincided; one that peaks after lagged.

Building permits, the classic leading indicator, led by 25 months in 2007 and by 3in 2020 — a difference of nearly two years in how much warning the same series gave. New orders for capital goods peaked after the December 2007 peak and 19 months before the February 2020 one. Industrial production, a coincident indicator by the committee’s own list, peaked in the December 2007 peak month exactly and 17 months beforeFebruary 2020 — so in the second episode a coincident indicator “led” by far more than a leading one.

Only two rows behave consistently. Nonfarm payroll employment peaked in the peak month in both episodes, which is exactly what the committee’s stated weighting would predict. And real personal consumption expenditures peaked one month before, in both. Everything described as leading moved by 22 months between the two episodes; the two coincident measures moved by nothing.

The 2020 column carries a caution that has to be stated. Four of its rows — manufacturing and trade sales, industrial production, capital goods orders and factory hours — put their local maximum in 2018, seventeen to twenty-two months before the peak. A slowdown in manufacturing through 2018 and 2019 is a fact about those series; calling it a signal of what happened in February 2020 is a claim this page does not make, and the reason to print the column rather than a verdict is that the reader can see the ambiguity for themselves.

Two leading indices that stopped, without saying so

A failure mode no back-test can capture is the indicator that ceases to exist. Two turned up in the files read for this page.

The first ends in February 2020— the exact month of the last NBER peak, which is a coincidence with an unpleasant quality to it. The second ends in January 2024. Neither file announces that it has finished; each simply runs out. A chart of either looks entirely normal until the last date is checked, and a page updated automatically from one of them would keep drawing the same flat line for years.

The most-cited leading index of all, published by a private conference board, was not read for this page at all and no value from it appears. Where a series could not be opened at its publisher, this site leaves it out rather than taking it from somewhere else.

The current picture is assembled from four different months

A second problem with indicator dashboards is invisible on the dashboard. On 13 August 2026 the nine series read for this page had latest observations spread across four different reference periods.

The newest number describes the week ending 1 August 2026; the oldest describes May 2026. Any composite that averages these is averaging observations up to three months apart, and the fastest series in the set is the one the NBER says it normally place[s] little weight on. Speed and reliability run in opposite directions here, and that is not a defect to be engineered away: it is what the data are.

Two composites are worth naming for what they are. The Chicago Fed National Activity Index read −0.02 for June 2026, against an observed range from −18.32 in April 2020 to +6.32in June 2020 — two months apart, and the two most extreme values in a series of 712 running back to March 1967. Any indicator whose whole historical range was traversed twice inside one quarter carries a warning about what “normal” means in it.

What the record supports, and what it does not

Supported: the two indicators the NBER weights most heavily behaved identically in both episodes, peaking in the peak month. If you want to know whether the turn has happened, those are the series that answer, and they answer late by construction.

Not supported: a stable ordering of leading indicators. Six of the eight series moved their position by three months or more between the two episodes, and three by seventeen months or more. A table that assigns each indicator a fixed lead in months describes an average of episodes that did not resemble each other.

Untestable here: whether any of this would have been usable at the time. Every ranking above was computed with the current vintage of each series and against peak dates the NBER announced 11 and 4 months after the fact. A real-time version of this exercise would need the vintages available on each date, which is a different and much larger piece of work, and this page does not pretend to have done it.

What these sources do not say

• Two episodes are not a general result. The finding here falsifies a fixed ordering; it does not establish a new one, and this page claims only the first.
• Nothing was measured in real time. Current vintages were used against peak dates announced 11 and 4 months after the fact.
• The best-known private leading index was not read. No value from it appears here, because its publisher was not opened.
• Two of the series read have stopped being published, in February 2020 and January 2024, and neither file says so.
• No causal claim. A series peaking before another is a fact about two files, not a mechanism, and none of the publishers asserts one.
• No market series was read. Nothing here says what any indicator does to any price.

How to verify this yourself

1. Download fredgraph.csv?id=PERMIT and find the highest value between December 2005 and March 2008. It should fall in November 2005.
2. Do the same for the two years to February 2020. The maximum should fall in November 2019.
3. Repeat for PAYEMS around both peaks. Both maxima should fall in the peak month itself.
4. Download USSLIND and look at the last row. It should be dated February 2020.
5. Download USALOLITONOSTSAM and look at the last row. It should be dated January 2024.
6. Request all nine series in one session and list the last date of each. They span four different reference periods.

Where to go next on this site

• Recession indicators — four signals tested against every recession, with the announcement lag measured.
• PMI and survey indicators — what a diffusion index is, and how often below-zero months were not recessions.
• Economic cycles — the chronology every timing measurement on this page is scored against.
• All macroeconomics guides — the area index, one click away.

Sources opened for this page

Each source below was opened and read on 13 August 2026. Verified on 13 August 2026.

• National Bureau of Economic Research — Business Cycle Dating and the dating procedure FAQUsed for: the six monthly indicators the committee names, the statement that there is no fixed rule about weights, the sentence naming real personal income less transfers and payroll employment as most heavily weighted, the assessment of unemployment insurance claims, and the peak months of December 2007 and February 2020. Read 13 August 2026.
• Federal Reserve Bank of St. Louis — FRED series PERMIT, AWHMAN, NEWORDER, CMRMTSPL, INDPRO, PCEC96, PAYEMS and W875RX1Used for: the full monthly history of each series, the local maximum of each in the two years around the December 2007 and February 2020 peaks from which every entry in the timing table was computed, and the latest reference period and value of each on 13 August 2026. Read 13 August 2026.
• Federal Reserve Bank of St. Louis — FRED series CFNAI and ICSAUsed for: the 712 monthly National Activity Index values from March 1967 with the extremes of minus 18.32 in April 2020 and plus 6.32 in June 2020 and the reading of minus 0.02 for June 2026, and the weekly claims value of 199,000 for the week ending 1 August 2026. Read 13 August 2026.
• Federal Reserve Bank of St. Louis — FRED series USSLIND and USALOLITONOSTSAMUsed for: the 458 monthly values from January 1982 ending in February 2020, and the 829 monthly values from January 1955 ending in January 2024, establishing that both series have stopped without any notice in the file. Read 13 August 2026.
• National Bureau of Economic Research — US Business Cycle Expansions and ContractionsUsed for: the peak and trough months used to place every turning point above, and the note that the business cycle data were last updated on 14 March 2023. Read 13 August 2026.

Change log

• 13 August 2026— page rewritten from scratch against the NBER dating pages and twelve FRED series, all opened that day. Removed from the previous version: a fixed table of leading, coincident and lagging indicators with lead times in months, which the two measurable episodes contradict; a private leading index quoted without being read at its publisher; two discontinued series presented as current, one of which stopped in February 2020; a composite 'current reading' assembled from series whose latest months differ by up to three; and the omission of the six indicators the dating committee actually names.

Vextor Capital is not authorised under MiFID II as an investment firm, is not registered with the SEC, FINRA, the CFTC or the NFA, is not an investment adviser, holds no position in any instrument named on this page, and receives no payment from any bank, broker, exchange, index provider or data vendor. Every outbound link on this page points to a statistical agency, a central bank or a public data series.