Help centre

The recession diffusion index

Methodology

The recession overlay is a diffusion index: eleven monthly US series, each reduced to a yes or no against a fixed threshold, then counted.

IndicatorFRED seriesTransformFlashes when
Nonfarm payrollsPAYEMSSix-month changebelow 0
Real income less transfersW875RX1Six-month changebelow 0
Industrial productionINDPROYear on yearbelow 0
Real manufacturing and trade salesCMRMTSPLSix-month changebelow 0
Sahm ruleSAHMREALTIMELevel0.50 or above
Chicago Fed activity index, 3-month averageCFNAIMA3Levelbelow -0.70
Yield curveGS10 minus TB3MSLevelbelow 0
Building permitsPERMITYear on yearbelow 0
Credit spreadBAA minus GS10Levelabove the 80th percentile of the training period
Financial conditionsNFCILevelabove 0
Initial claimsICSA, 4-week averageYear on yearabove 10%

States

Five or more flashing reads Recession, three or four Warning, otherwise Expansion.

Live rule and history

Since 26 August 2026 each indicator trips on its own latest monthly print, and the page shows the month beside each. The historical series, used for the chart, applies the older rule of the last month in which all eleven had printed. So a live reading can differ from what the history later records for the same month; the live reading is the more current, the history the more complete.

Validation

Backtest against NBER recession dates from 1971, with everything from 2000 onward held out. The coincident model reached an out-of-sample AUC of 0.976. At the five-indicator cut it recalled 93% of recession months with one false alarm since 2000. Per indicator, out of sample since 2000:

IndicatorAUC
Initial claims0.919
Nonfarm payrolls0.863
Real manufacturing and trade sales0.859
Financial conditions (NFCI)0.842
Chicago Fed activity index, 3-month average0.836
Sahm rule0.831
Industrial production0.804
Real income less transfers0.786
Credit spread, Baa minus 10-year0.715
Building permits0.698
Yield curve, 10-year minus 3-month0.424

AUC is the chance the indicator ranks a random recession month above a random healthy one; 0.5 is a coin toss. The yield curve's low score on a month-by-month test is consistent with an indicator that leads by a year or more, which is why it sits among the leading five rather than the coincident six. Past backtest results do not guarantee future results.

Past performance does not guarantee future results. Trading and investing involve significant risk of loss. Nothing on this page is financial advice or a recommendation to buy or sell any financial product. Everything on the platform is identical for every subscriber on a plan and is never tailored to an individual.