Modeling Near-Term Recession Risk in the United States: Evidence from a Regularized Logit Approach

The aim of this paper is to develop a statistical framework with the purpose to model a
near-term recession risk in the United States, using a regularised logistic regression.
Based on monthly data from 1990 to 2024, the analysis consisting of a set of financial
macroeconomic indicators, like the yield curve spread, labor market conditions, real
economic activity and consumer expectations, aims to estimate the probability of a
recession within a twelve-month time frame. The dependent variable in the model is a
forward-looking binary recession indicator based on the National Bureau of Economic
Research (NBER) classification.
To improve the model reliability and mitigate overfitting, L2 regularization is used into the
logistic specification. The performance of the model is evaluated using classification
metrics like precision, recall, F1 score and the area under the ROC curve (AUC). Instead of
using a fixed threshold, the classification boundary is determined by optimizing for the F1
score, resulting with a balanced treatment of false positives and false negatives in a policyrelevant setting.
The results point to a framework with strong predictive performance, with an AUC of
around 0.78 and recall above 60% for recession periods. The findings point out the value
of combining financial market signals with real-sector and sentiment-based indicators
when it comes to evaluating short-term recession risk. In general, this study offers an
interpretable and practically applicable approach to recession forecasting, with strong
potential to be used in macroeconomic monitoring and risk assessment.
Keywords: Recession Probability Modeling, Term Spread, Business Cycle Analysis,
Macroeconomic Forecasting, Regularized Logit Model