Recession Risk Through a Real-Economy Lens

Forecasting economic recessions remains a fundamental challenge in macroeconomic research and investment decision-making. Financial markets often signal recessions before economic data visibly deteriorate, making indicators such as yield spreads and credit spreads valuable early-warning tools. However, market-based indicators can also generate costly false alarms when financial conditions reflect temporary shocks rather than sustained economic weakness.

To capture both market expectations and underlying economic conditions, we develop a framework that integrates financial indicators with a broad set of macroeconomic variables. By integrating financial indicators with measures of consumption, housing, labor markets, production, and financial health, our framework improves explanatory power from 0.38 to 0.54 and increases classification accuracy from 84% to 89%, while reducing false recession signals. Our analysis suggests that recession forecasts become substantially more reliable when financial market signals are combined with measures of real economic activity.

In the United States, recession dates are determined by the National Bureau of Economic Research (NBER) Business Cycle Dating Committee, which evaluates a broad range of economic indicators to assess the depth, duration, and diffusion of economic downturns.

While widely regarded as the definitive record of business cycles, the NBER process is inherently backward-looking. Historically, official recession announcements have been delayed by four- to twenty-one months, with an average lag of approximately eleven months (see Exhibit 1).

By the time a recession is officially identified, markets and economic conditions have often already adjusted, highlighting the need for forward-looking models that can assess recession risk over investor-relevant horizons.

Financial indicators, including the yield-term spread (Term), near-term forward spread (NTFS), and excess bond premium (EBP), provide timely information about market expectations and financing conditions. These measures are complemented by macroeconomic indicators that reflect the evolving state of the real economy.

To align with the NBER's emphasis on the depth, duration, and diffusion of economic downturns, we focus on medium-term (six- to twelve-month) changes in economic conditions rather than short-term fluctuations.

The macroeconomic dataset is organized into five categories: financial health, consumption, housing, labor markets, and production, each representing a distinct dimension of economic activity.

Principal Component Analysis (PCA) is applied within each category to extract latent factors that summarize the common variation across the underlying variables. This approach reduces dimensionality while preserving the key information contained in the broader dataset.

The modelling framework is based on the NBER recession indicator, which is represented as a binary time series, as illustrated below, taking the value of one during recession periods and zero otherwise.

Statistically, this process can be approximated as a Bernoulli-type outcome, where each time period is associated with a probability of recession,$P_r$ .

The dependent variable is constructed as a forward-looking indicator. Specifically, the recession variable,$〖NBER{〗}_{(}t,t+12)$, equals one if at least one recession month occurs within the subsequent twelve months, and zero otherwise.

Recession probabilities are estimated using a generalized linear model (GLM) with a probit specification (Equation [1]), in which the probability of entering a recession is modeled as a nonlinear function of a set of explanatory variables.

As shown in Exhibit 2,the empirical results demonstrate that integrating financial indicators with macro factors substantially improves the performance of recession prediction models. The model with integrated variables achieves higher explanatory power relative to traditional financial based models, with an R² of approximately 0.54.

In addition, the model delivers the highest accuracy score of approximately 89%, defined as the proportion of correctly classified recession and non-recession periods.

Exhibit 3 shows historical recession probability estimates in the last 53 years. A comparison with benchmark model based exclusively on financial indicators highlights several advantages of the integrated approach.

The model generates fewer false positives, indicating improved robustness to transient shocks in financial markets. This is particularly important during periods of elevated market volatility, when purely financial indicators may produce misleading signals.

The inclusion of macroeconomic variables provides greater stability by anchoring model outputs in observed economic conditions, rather than relying exclusively on market-implied expectations.

Empirical findings indicate that incorporating real-economy information significantly enhances predictive accuracy and reduces false signals. The resulting framework provides a more reliable and interpretable assessment of recession risk by combining market expectations with evidence from underlying economic conditions. For institutional investors and portfolio managers, such insights can help identify cyclical turning points earlier and support more informed portfolio allocation decisions.

Harvey, Campbell R. “Forecasts of Economic Growth from the Bond and Stock Markets.” Financial Analysts Journal 45, no. 5 (September 1989): 38–45. https://doi.org/10.2469/faj.v45.n5.38.

Sharpe, Steve, and Eric Engstrom. “The Near-Term Forward Yield Spread as a Leading Indicator: A Less Distorted Mirror.” Finance and Economics Discussion Series 2018.0, no. 55 (August 2018). https://doi.org/10.17016/feds.2018.055.

Favara, Giovanni, Simon Gilchrist, Kurt F. Lewis, and Egon Zakrajšek. “Recession Risk and the Excess Bond Premium.” FEDS Notes 2016.0, no. 1739 (April 2016). https://doi.org/10.17016/2380-7172.1739.

Kiley, Michael T. “Financial and Macroeconomic Indicators of Recession Risk.” FEDS Notes, no. 6/21/2022 (June 2022). https://doi.org/10.17016/2380-7172.3126.