Inflation Target at Risk: A Time-varying Parameter Distributional Regression

Yunyun Wang, Tatsushi Oka, Dan ZhuarXiv 2024 · DOI · arXiv

TL;DR

This paper introduces a Time-varying Parameter Distributional Regression (TVPDR) model to estimate the entire probability distribution of inflation over time. The authors demonstrate that deflation and excessive inflation risks are driven by fundamentally different economic forces, and that the traditional Phillips Curve relationship—the trade-off between unemployment and inflation—largely breaks down during extreme economic episodes.

What Problem Does This Paper Address?

Most central banks aim for a specific inflation target (usually 2%). However, traditional "mean-based" econometric models focus on predicting the average outcome, often ignoring the "tails" of the distribution—the risks of deflation or runaway inflation.

Inflation dynamics are notoriously "messy": they are state-dependent, skewed, and prone to "fat tails" (extreme events happening more often than a normal distribution would suggest). While existing tools like Quantile Regression can look at specific points (like the 5th or 95th percentile), they often require awkward secondary steps to reconstruct a full predictive density. The authors address the need for a unified framework that can model the entire distribution of inflation while allowing the impact of macroeconomic drivers (like oil prices or unemployment) to change over time.

Methods

The authors propose the TVPDR model. It combines two powerful ideas:

  1. Distributional Regression (DR): Instead of modeling the mean, it models the Conditional Distribution Function (CDF) directly.
  2. Time-varying Parameters (TVP): It assumes the relationship between predictors and inflation isn't static; the regression coefficients evolve according to a "random walk."

Technical Innovation: * Monotonicity Constraint: A major challenge in distributional regression is ensuring the estimated CDF is "monotonic" (i.e., the probability of inflation being $\le 3\%$ cannot be lower than the probability of it being $\le 2\%$). The authors developed a novel Markov Chain Monte Carlo (MCMC) algorithm that enforces this constraint during the estimation process. * Shapley Value Decomposition: To make the "black box" of a distributional model interpretable, they use Shapley values to attribute changes in tail risks to specific macroeconomic drivers (e.g., how much of the 2022 inflation spike was due to energy vs. demand).

Datasets

The study focuses on U.S. inflation data from 1982:Q1 to 2024:Q4:

  • Target Variables: Headline CPI and Core CPI (excluding food and energy).
  • Predictors: A suite of macroeconomic variables sourced from the Federal Reserve Bank of St. Louis (FRED), including:
  • Lagged inflation and 5-year inflation expectations.
  • The unemployment rate (to test the Phillips Curve).
  • Energy and food price inflation.
  • Real GDP growth, Federal funds rate, and the National Financial Conditions Index.

Key Insights and Ideas

  • The Asymmetry of Risk: Deflationary risks and inflationary risks are not two sides of the same coin. Deflation risk is almost exclusively driven by demand-side weakness (low GDP growth) and "inflation inertia." Conversely, excessive inflation risk is primarily triggered by supply-side shocks, specifically energy and food prices.
  • The "Flattening" Phillips Curve is a Tail Phenomenon: The paper finds that the link between unemployment and inflation (the Phillips Curve) is much weaker at the edges of the distribution. This suggests that while unemployment might influence "average" inflation, it has very little power to predict or prevent extreme tail events.
  • Energy Shocks are Regime-Dependent: Energy price spikes have a massive, asymmetric impact on high-inflation risk but almost zero impact on deflation risk. Furthermore, this impact is time-sensitive; energy prices mattered much more during the post-pandemic recovery than in previous decades.

Key Results

  • Superior Forecasting: The TVPDR model outperformed several benchmarks—including standard Linear Regression, Quantile Regression, and Stochastic Volatility models—in terms of Root Mean Squared Error (RMSE) and Continuous Ranked Probability Score (CRPS), particularly during recessionary periods (2008 and 2020).
  • Capturing the 2022 Surge: The model successfully captured the dramatic shift in 2022, where the probability of "excessive inflation" (CPI > 4%) surged to near 0.9, while mean-based models were much slower to adjust.
  • Calibration: Using Probability Integral Transform (PIT) tests, the authors showed that their model provides a "statistically valid" description of the true underlying uncertainty, whereas simpler models often fail to account for the actual frequency of extreme outcomes.

Limitations and Future Work

  • Computational Complexity: Estimating a full distribution with time-varying parameters for every threshold is computationally intensive. The authors note that while their MCMC algorithm is efficient, it still involves a "substantial proliferation" of parameters.
  • Parameter Evolution: The model assumes parameters follow a Gaussian random walk. Future research could explore "jump" processes or regime-switching models to better capture sudden structural breaks in the economy.
  • Global Application: The current study is limited to U.S. data. Testing the TVPDR framework on emerging markets, where inflation distributions are often even more volatile and skewed, remains an open area for research.

Summary updated

Abstract

Inflation exhibits state-dependent, skewed, and fat-tailed dynamics that make risk a central concern for monetary policy. Accordingly, inflation risks are distributional and cannot be fully captured by mean-based models. We propose a flexible time-varying parameter distributional regression model that estimates the full conditional distribution of inflation, allowing macroeconomic drivers to have nonlinear and asymmetric effects across the distribution. Applied to U.S. inflation, the model captures major shifts in tail-risk probabilities. Analysis of risk drivers shows that deflationary pressures arise primarily from demand-side weakness and inflation persistence, whereas upside risks are driven mainly by supply-side shocks, particularly energy price inflation. Examining the impact of key drivers further reveals that the unemployment-inflation relationship weakens in the distributional tails. Energy price shocks, by contrast, have little effect on deflation risk but exhibit strongly time-varying and asymmetric effects on high-inflation risk.
BibTeX
@misc{wang2024inflation,
  author = {Yunyun Wang and Tatsushi Oka and Dan Zhu},
  title = {{Inflation Target at Risk: A Time-varying Parameter Distributional Regression}},
  year = {2024},
  eprint = {2403.12456},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2403.12456},
}
Figure