Skip to content

What Predicts US Recessions?: summary and key takeaways

The paper tests many leading indicators against the Treasury term spread for predicting NBER recessions, using probit models scored by area under the ROC curve. At short horizons the annual S&P 500 return adds the most. At horizons beyond one year, capital goods orders and broker-dealer margin debit add the most.

Liu, W.; Moench, E. · 2016 · International Journal of ForecastingSummary updated 20 November 2026
Published2016Journal article
Asset focusMacroRecession prediction
TypeResearchInflation & Macro
Reading time4 minthis summary

What Predicts US Recessions? at a glance

The question
Do other leading indicators predict US recessions better than the Treasury term spread, and at which horizons?
Dataset & period
Monthly US data, January 1959 to December 2011, covering seven recessions as dated by the NBER. Out-of-sample test: estimation on January 1959 to August 1985, forecasting on September 1985 to December 2011. Horizons of 3, 6, 12, 18 and 24 months.
Method
Each model is a probit that maps predictors onto the NBER recession indicator. The benchmarks are the 10-year minus 3-month term spread, alone and with its six-month lag. Each other indicator is added to the spread models in turn. Models are ranked by area under the ROC curve (AUROC), with a block bootstrap that allows for serial correlation, and a missing-data probit handles series that start late.

Key findings of What Predicts US Recessions?

  • At the 3-month horizon, adding the six-month lagged spread lifts the mean AUROC in-sample from 0.67 to 0.86.
  • Also at 3 months, adding the annual S&P 500 return raises the in-sample mean AUROC to 0.96. Broker-dealer debit margins and total payroll employment come next.
  • At 6 months, the lagged spread lifts the in-sample mean AUROC from 0.77 to 0.88, and the S&P 500 return, building permits and the 5-year Treasury minus fed funds spread all improve on that significantly.
  • The term spread alone is strongest at 12 months, with a mean AUROC of 0.87 in-sample. The best added variable, the 5-year Treasury minus fed funds spread, reaches 0.91, but the gain is not clearly significant against the two spread models.
  • At 18 and 24 months, broker-dealer margin debit is the main variable that adds significantly. At 18 months it takes the in-sample mean AUROC to 0.88. At 24 months the spread alone gives 0.70, while margin debit, non-defense orders and an ISM or commodity price index give 0.79 to 0.81.
  • Out-of-sample at 3 months, the spread alone scores 0.73 and adding the lagged spread gives 0.87. With the S&P 500 return the mean AUROC is 0.95. At 24 months the spread alone falls to 0.57.

Robustness, as reported in the paper

  • The out-of-sample test estimates parameters on 1959 to August 1985 and forecasts the later period. It confirms the short-horizon gains from the S&P 500 return and the long-horizon gains from non-defense orders and margin debit.
  • The best added variables at 6 and 12 months differ between the in-sample and out-of-sample runs. The authors read this as a sign that the drivers of the cycle changed within the sample.
  • Block-bootstrap results use an average block of eight years. Blocks of five and eleven years give very similar results.
  • Real-time or first-release data are used where available, and series without real-time data before 1985 are left out of the out-of-sample test.

Caveats and limits

  • Only seven recessions are in the full sample, and three in the out-of-sample period, so the results rest on few events.
  • The paper restricts itself to probit models and does not report trading rules, transaction costs or any portfolio application (inferred).
  • At the 12- and 18-month horizons the incremental gains over the term spread are small and often not significant at conventional levels.
  • Recession dates come from the NBER, which announces them with a delay, and the sample ends in 2011 (inferred).

Items marked (inferred) are our reading of the design, not statements by the authors.

Why it matters

The term spread is a hard benchmark for recession risk at one year, and the paper shows where extra inputs help. Equity returns matter near term, while funding and leverage measures such as broker-dealer margin debit help at longer horizons. Macro and risk teams can use this to choose inputs by forecast horizon.

Questions about this paper

Which horizon is the Treasury term spread most useful for?

The paper finds the spread predicts recessions best about one year ahead, with a mean AUROC of 0.87 in-sample. Lagged values of the spread help mainly at shorter horizons.

What improves on the term spread for short-term recession forecasts?

The annual return on the S&P 500 stands out at 3 and 6 months. At 3 months it raises the in-sample mean AUROC to 0.96.

What helps at horizons beyond one year?

New orders of non-defense capital goods and broker-dealer margin debit improve forecasts significantly beyond one year. At 24 months the models using them score between 0.79 and 0.81 in-sample.

Where can I read the original paper?

The original is at its publisher: https://doi.org/10.1016/j.ijforecast.2016.02.007. We link to the paper and do not host the PDF.

Follow new quant research

Download MacroFinance market data as CSV or Excel, or pull it by API, and test the ideas in these papers yourself. New summaries are added as we review them.

Create a free account
Summaries are written by MacroFinance from the published papers, in our own words, for information only. They are not investment advice and may omit detail. Papers are not hosted here: each page links to the original at its publisher.