Economic Forecasting under Structural Changes and Heterogeneity

Taeyoung Doh Chair
Federal Reserve Bank of Kansas City
 
Taeyoung Doh Organizer
Federal Reserve Bank of Kansas City
 
Padma Sharma Organizer
Federal Reserve Bank of Kansas City
 
Tuesday, Aug 4: 2:00 PM - 3:50 PM
1765 
Topic-Contributed Paper Session 
Thomas M. Menino Convention & Exhibition Center 
Room: CC-156B 

Applied

Yes

Main Sponsor

Business and Economic Statistics Section

Co Sponsors

Government Statistics Section
Section on Bayesian Statistical Science

Presentations

A Bayesian Framework for Detecting Structural Changes in Time Varying Parameters of Panel Models

We develop a Bayesian framework for detecting structural breaks in panel data models. Our method offers greater flexibility in detecting structural breaks relative to existing tests by allowing for breaks to occur at different times for distinct parameters. By jointly detecting the presence of structural breaks and estimating parameters, our framework overcomes the problems of inaccurate inference associated with pre-testing in standard test-based approaches to addressing structural breaks. We incorporate unobserved heterogeneity specific to panel settings by allowing for fixed effects or interactive fixed effects in the model. A dynamic extension of the spike-and-slab prior provides the primary mechanism for change point detection and parameter estimation in the sub-periods. We show that we recover structural breaks in a wide range of simulation settings even when standard methods do not identify such breaks. We apply our method on data from Ditzen et al. (2024) to detect whether unconventional monetary policies resulted in structural breaks in bank loan growth. 

Keywords

Structural breaks

Bayesian estimation

Panel data 

Speaker

Padma Sharma, Federal Reserve Bank of Kansas City

A Large-Scale Bayesian VAR for Disaggregated GDP Components

We introduce a large scale Bayesian VAR for monthly indicators of disaggregated GDP
components and estimate it using Bayesian methods. By linking monthly indicators with relevant
GDP subcomponents and aggregating them following the national accounting convention,
we generate GDP forecasts as well as a few key macroeconomic variables. By accommodating
the heterogeneous release schedule of monthly indicators, our model can sequentially update
GDP forecasts at higher (daily or weekly) frequency than feasible based on only aggregate
variables. Our model's nowcasts are comparable to the widely used reference GDP tracking
model for most GDP components. The model's predictive density of GDP growth across data
vintages suggests that surprises in macroeconomic indicators can generate frequent changes
in downside tail risk, which has often attributed to changes in financial market conditions.
The departure from normality is pronounced when the predictive density is tilted to match the
previous quarter's headline GDP growth.
 

Keywords

Large-Scale Bayesian VAR

Disaggregated GDP Components

Density Forecasting 

Speaker

Taeyoung Doh, Federal Reserve Bank of Kansas City

Inflation Factors

We propose a novel framework to identify inflation factors that provide
timely estimates of supply and demand conditions shaping goods- and services-related
inflationary pressures. These factors are estimated from disaggregated Personal Consumption
Expenditures (PCE) data on prices and quantities, using a new Sign-Restricted
Dynamic Factor Model (SiR-DFM). Aggregate PCE inflation is then broken down
into the contributions of common factors—encompassing goods demand, goods supply,
services demand, services supply, and inflation expectations—and consumption
category-specific—that is, idiosyncratic—components. By implementing a series of
validation exercises, we show that the estimated inflation factors provide an informative
and coherent narrative of inflation dynamics and can be used for forecasting and
policy analyses.
 

Keywords

Inflation Expectation

Sign-Restricted Dynamic Factor Model 

Speaker

Danilo Leiva-Leon

Latent Worker Dynamics and Wage Changes

This paper investigates the effects of latent worker heterogeneity on wage changes among
new hires. We uncover latent labor market states and unobserved worker types by estimating
a multi-channel latent Markov model, using individual histories of labor force status and job
switches under various circumstances from the Survey of Income and Program Participation.
We identify latent labor-market states that are highly transient, typically experienced by
workers with unstable employment who frequently switch jobs. Individuals who encounter
these transient states within a one-year interval are classified as the vulnerable type, while
all others are classified as the stable type. The vulnerable type, representing 20 percent
of the civilian population, does not experience wage gains when switching jobs and faces
countercyclical wages at new jobs, in contrast to typical job switchers who experience wage
gains and acyclical wage changes. Long tenure among vulnerable workers further reduces
wage gains at new jobs and amplifies the countercyclicality of their wages, likely reflecting
the destruction of job-specific human capital and greater mismatch. We interpret these
empirical findings through a search-and-matching model in which heterogeneity among
workers and jobs accounts for the excess cyclicality in the wages of new hires.
 

Keywords

Latent Markov Model

Worker Heterogeneity

Survey of Income and Program Participation 

Speaker

Hie Joo Ahn, Federal Reserve Board

Mixed-Frequency Panel Regressions with Sparse and Heterogeneous Structures

This paper develops a mixed frequency panel regression framework for nowcasting and
forecasting a low frequency outcome using a large set of high frequency predictors. We
propose a method that captures both sparsity in distributed lag predictors and heterogeneity
across cross sectional units through latent group structures. In this setting, slope coefficients
are homogeneous within groups but heterogeneous across them. To estimate the model, we
introduce a doubly penalized least squares estimator that simultaneously selects the relevant
high frequency predictors and uncovers the underlying group structure without prior knowledge
of the number of groups or sparsity patterns. We establish oracle properties for the estimator
and show that it consistently identifies both the relevant predictors and the group memberships
in large samples. Monte Carlo experiments demonstrate strong finite sample performance. An
empirical application to U.S. metropolitan statistical area housing prices illustrates the gains
in mixed frequency nowcasting and forecasting achieved by incorporating sparsity and group
heterogeneity.
 

Keywords

High dimensionality

Mixed data sampling (MIDAS)

Parameter Heterogeneity

Penalized regression

Sparsity

Oracle property 

Speaker

Shahnaz Parsaeian