Recession Signal Index

Which signals actually led

This is the part most recession dashboards leave out. Below is how many months before each recession began that each indicator first crossed its threshold, looking back up to 36 months. "Never" means it did not cross at all before that recession started.

The yield curve went 3 for 3, inverting 9, 23, 10 months before the three recessions in the sample. Nothing else comes close. The Sahm Rule and the Chicago Fed activity index never crossed their thresholds before a recession began, in any of the three.

That is not a flaw in those indicators. It is what they are built to do. The Sahm Rule exists to identify a recession that has already started, quickly and from real-time data. Judging it as a predictor is judging a smoke alarm for failing to prevent fires.

What the coincident gauge is good at

Once you stop asking it to predict, the four-signal gauge is strikingly clean. Across 305 months, it averaged 3.19 of 4 during NBER recessions and 0.23 of 4 outside them. That is roughly a 14-fold separation between "a downturn is running" and "it is not".

So the honest way to use this page is as two different questions. Is a recession plausibly coming? Watch the curve, and expect a lead measured in quarters, not weeks. Is one already here? Watch the four coincident signals, which is the question they can actually answer.

The thresholds

• Yield curve (10y minus 3m): active when the spread is below zero.
• Sahm Rule: active at 0.50 percentage points or more.
• Smoothed recession probability: active at 20% or more. The 20% line is ours.
• Chicago Fed activity index: active when the three-month average is at or below -0.70.
• Initial jobless claims: active when more than 10% above the same month a year earlier. The 10% line is ours.

Two of those five lines are editorial and are labelled above. The rest are the conventional thresholds each indicator is normally read against. Recession dates are the official NBER dates. A month only enters the index when all five indicators can be evaluated, so the count is never depressed just because one series has not published yet.

The honest limits

• Three recessions is a tiny sample. The yield curve going three for three is encouraging and nowhere near statistical proof. It has also inverted without a recession following, which is why this page shows every inversion on the chart rather than only the ones that worked out.
• Lead time is wildly variable. Nine months before 2001, twenty-three before 2008. An indicator that is right but two years early is very hard to trade.
• This is not a forecast. No probability is being assigned to a future recession here, and none should be inferred. It is a tally of what published indicators are doing, with their record attached.
• Revisions move the past. Several of these series get revised, so a backtested lead time can shift after the fact.

For the current reading on all five indicators individually, with sparklines and thresholds, see the recession indicators dashboard. Full sourcing for every figure is on the methodology page.