Recession Indicators: Yield Curve, Sahm Rule, Claims and the LEI

The business cycle dating committee of the National Bureau of Economic Research announced that a recession had begun in December 2007 on a day in December 2008, twelve months later, by which point the financial system had already come apart in public. It moved faster in 2020, naming the February peak within a few months, because the collapse was too abrupt to argue about. Neither announcement was useful to anybody managing money. Neither was meant to be. The committee writes history. It issues no warnings.

That delay is the reason a handful of forward-looking measures attract so much attention, and each has a record, each has failed at least once, none produces a date, and the overlap between them is larger than it appears. What follows is how each is built and where it has misled.

What is actually being predicted

A recession is a broad decline in activity spread across the economy and lasting more than a few months, which is close to the committee’s own wording and is deliberately vague about thresholds. None of that is observable in real time. Employment, income, spending and production are all published with a lag and revised afterwards, so even the question of whether a recession has already started cannot be settled on the day.

The indicators below measure one of two things: what the bond market expects, or how fast something has already deteriorated. Confusing the two starts most of the trouble.

Indicator	What it measures	Typical lead	Main weakness
Yield curve inversion	Bond market expectations of future short rates	Long and highly variable	Says nothing about timing, and has been early by up to two years
Sahm rule	Speed of deterioration in the unemployment rate	Short, and it confirms a turn already under way	Triggered by rising labor supply as well as by job losses
Initial jobless claims	Weekly filings for unemployment benefits	Short and very timely	Noisy, and needs a four-week average
Conference Board LEI	Ten leading components combined into one index	Medium	Weighted toward manufacturing and goods

The yield curve

The curve inverts when a shorter maturity yields more than a longer one, and two spreads get quoted: the 10-year minus the 2-year, which captures the market’s medium-term view, and the 10-year minus the 3-month bill, which research published by the Federal Reserve Bank of New York in the 1990s found the more reliable predictor of the two.

The logic is that a long yield is roughly the average of expected short rates plus a term premium, so when investors expect the central bank to be cutting a year or two out, they are expecting an economy that needs help, and long yields fall below the policy rate to say so. The shapes and what each implies are in the yield curve explained, and the current shape is on the yield curve tool.

The record is the strongest of any single indicator, with an inversion preceding every American recession since the 1960s. It is also the most abused. Lead times have run from several months to around two years, equity markets have delivered large gains while the curve was inverted, and the inversion that began in 2022 ran unusually long without the downturn arriving on the historical schedule, an episode I wrote about in the inversion that took two years to mean anything. An indicator with a two-year error bar is a reason to check an allocation. It is not a reason to trade.

The Sahm rule

Claudia Sahm designed this as a trigger for automatic stimulus payments, which is why it is built for real time. It rests on a regularity in the data: unemployment drifts down slowly through an expansion and rises quickly once a downturn has begun, so the speed of the rise carries information that the level does not.

The rule triggers when the three-month moving average of the unemployment rate rises 0.5 percentage points or more above its lowest three-month average of the preceding twelve months.

Work through an illustration. Suppose the unemployment rate over the last three months printed 4.1 percent, 4.2 percent and 4.3 percent. The current three-month average is:

(4.1 + 4.2 + 4.3) / 3 = 4.2%

Suppose the lowest three-month average over the previous twelve months was 3.6 percent. The gap is:

4.2 - 3.6 = 0.6 percentage points

That clears the half-point threshold, so the rule has triggered. The inputs come straight from the household survey described in the jobs report, which is also where the weakness lives. The unemployment rate can rise because people are losing jobs or because more people have entered the labor force to look for work, and those are very different economic situations that produce an identical trigger. Sahm has said publicly that unusual labor supply conditions can distort her own rule, which is a more honest position than most indicator designers take about their work.

Weekly claims

Initial claims for unemployment insurance are published every Thursday for the week ending the previous Saturday, and they come from state administrative records, which makes them the most current labor market data available anywhere.

The weekly figure is noisy. Holidays, auto plant retooling, hurricanes and state processing backlogs all move it, sometimes substantially. The four-week moving average is what analysts actually watch, and a sustained rise well above the range of the preceding year has historically been among the earliest signs of a labor market turning, while continuing claims, which count people still receiving benefits, answer a different and equally useful question: whether the people who lost jobs are finding new ones.

The Leading Economic Index

The Conference Board combines ten components into one index, among them average weekly manufacturing hours, initial claims, new orders, building permits, the interest rate spread and stock prices, and the conventional reading is the three Ds, depth, diffusion and duration: declines of reasonable size, spread across a majority of components, sustained for several months.

Its composition is its limitation. The index leans toward manufacturing and construction, which have shrunk steadily as a share of a service economy, and it has produced long stretches of decline with no recession following. An index designed around the cyclical sectors of the 1960s will keep describing the cyclical sectors of the 1960s.

Credit spreads and the survey measures

Credit spreads are the bond market’s fear gauge. When the extra yield demanded to hold high-yield debt over Treasuries widens sharply, lenders are pricing higher default risk, and credit markets have often moved before equities did, in part because the people trading them are paid to worry about the downside, with the mechanics in corporate bonds and credit spreads.

Purchasing manager indexes are diffusion measures where 50 separates expansion from contraction, and they arrive on the first business days of the month and give a quick read on new orders and hiring intentions, with the same manufacturing bias as the LEI. Housing responds to interest rates before anything else does, so building permits and new home sales turn early in both directions, which makes them useful and also makes them noisy.

Equity market internals belong here, with a caveat. Paul Samuelson’s remark in 1966 that the stock market had predicted nine of the last five recessions remains the correct level of scepticism. A narrowing advance, where the index holds up while participation falls away, is worth noticing, and the market breadth tool is where to see whether it is happening.

Why counting signals overstates the evidence

This is the part that gets skipped. None of them is independent. The LEI contains initial claims, the interest rate spread and stock prices, so an LEI decline driven by those three components is not a second opinion about anything, and treating it as confirmation means counting the same information twice.

The same problem runs through the whole list. Credit spreads widen partly because equities fell. Purchasing managers read the same newspapers. When somebody tells you that four separate indicators are flashing, the useful question is how many distinct pieces of underlying data sit behind those four, and the answer is frequently two.

What a recession call would actually mean

Markets do not wait for the dating committee. Equities have generally fallen before a recession is declared and begun recovering before it ends, which is why the announcement itself has almost never been a useful trading event, and government bonds have historically done their job in these episodes, as the central bank cuts and yields fall, though 2022 demonstrated that stocks and bonds can fall together when inflation is what has gone wrong.

Earnings estimates decline in a downturn, which pressures valuations through the numerator while lower discount rates support them through the denominator, exactly as set out in how interest rates affect stocks, and which force wins depends on the depth of the downturn and on where valuations started. Nobody knows either in advance.

The practical use of all this is preparation. If a set of signals turning would change what you hold, the time to think about the mix is while the signals are quiet, and the framework for that decision is in the asset allocation guide.

See how growth is actually measured in GDP explained, and test the Sahm arithmetic in the Fed and the economy quiz.