Assessing Asymmetric Macroeconomic Risk

U.S. Macroeconomic Skewness

In recent years, central banks have increasingly adopted a risk management perspective in the formulation of monetary policy. Rather than considering solely the most likely macroeconomic outcome, policymakers often take into account the full distribution of possible outcomes, placing particular attention on tail risks. Central banks discuss whether they see risks around the outlook as balanced, on the upside or on the downside.

This shift in perspective has spurred the development of empirical tools designed to quantify asymmetric risks to real gross domestic product (GDP) growth. A prominent example is the growth-at-risk framework developed by Adrian, Boyarchenko, and Giannone (2019), which uses quantile regressions to show that risk to real GDP growth becomes more left-skewed when financial conditions tighten. Related work has employed alternative methods, such as stochastic volatility or Markov-switching models.

Although the literature has made important progress in documenting time-varying asymmetric risks to real GDP growth, much less attention has been paid to asymmetric risks for the broader economy. Against this background, I propose and illustrate a framework for the assessment of broad-based macroeconomic risks in a systematic and replicable manner. My framework has three key ingredients. First, I work with a dynamic factor model that treats economic activity as a latent common factor, inferred from a set of macroeconomic indicators. Second, I assume an asymmetric distribution for the factor disturbances. Third, I allow for time variation in the parameters of the factor distribution. Time variation is modeled as a Markov-switching process.

Together, these components allow for capturing business-cycle variation in the conditional distribution and higher order moments of the common macroeconomic factor. In particular, a measure, or index, of macroeconomic skewness can be constructed by deriving analytically the conditional skewness of the macroeconomic factor, which reflects variations in the balance of risks of a set of macroeconomic aggregates. Although the model is neither linear nor Gaussian, it can be estimated by a modified Kalman filter that I develop in the paper. Using synthetic data, I demonstrate the effectiveness of the filtering and estimation procedure.

I apply my general framework to U.S. data from 1959 to 2024. I use a set of four macroeconomic aggregates (i.e., real GDP growth, real personal income, real manufacturing and trade sales, and employment), which are the same four sectoral variables typically utilized in this literature (e.g., Stock and Watson, 1991). My evidence shows substantial cyclical variation in the broad-based balance of risks: macroeconomic skewness displays a procyclical pattern, with a tendency to rapidly decline to negative territory during downturns and to rise during expansions (see Figure below). In other words, macroeconomic risks are tilted more toward bad (good) outcomes during recessions (expansions). Interestingly, this measure is partially correlated with GDP growth skewness that conditions on past economic and financial conditions (e.g., Adrian, Boyarchenko, and Giannone, 2019), suggesting that financial conditions are not necessarily a driver of asymmetric macroeconomic risks.

To evaluate the practical relevance of the skewness measure produced by the model, I assess the model’s out-of-sample performance in predicting risks to GDP growth. The focus lies on whether incorporating a broad set of macroeconomic indicators via my framework improves predictive accuracy relative to univariate benchmark approaches: a univariate version of the model and the quantile regression framework of Adrian, Boyarchenko, and Giannone (2019). My framework delivers better signals of downside risk around U.S. recessions, particularly the 2001 and 2008 episodes, when univariate benchmark models either mis-measure downside risks or understate recession probabilities. These improvements reflect the value of leveraging information from multiple macroeconomic indicators beyond GDP alone. A formal forecast evaluation confirms the model’s relative gains.

Keywords: Dynamic Factor Models, Markov-Switching, Skewness
Codes JEL : C34, C38, C53, E37