Government Budget Predictions with Mixed Frequency Analysis Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 19, 2019
Creator
  • Ozkan, Nazire
    • Affiliation: College of Arts and Sciences, Department of Economics
Abstract
  • Based on the growing literature of Mixed Data Sampling (MIDAS) analysis, this dissertation proposes forecasting procedures for the U.S. federal and state government budgets and output growth. Mixed frequency analysis elucidates the information content of data sampled at different frequencies and, hence, enables more accurate forecasts than the conventional approach that aggregates all time series into the lowest common frequency. This dissertation consists of three essays, each of which is examined in a separate chapter. The first chapter proposes a real-time forecasting procedure involving a combination of MIDAS-type regression models constructed with predictors of different sampling frequencies to predict the annual U.S. federal government current expenditures and receipts. Evidence shows that forecast combinations of MIDAS regression models provide forecast gains over the traditional models, suggesting the use of mixed frequency data consisting of fiscal series and macroeconomic indicators in forecasting the annual federal budget. It is also shown that, although not statistically significant, MIDAS regressions with quarterly leads that are employed to have real-time forecast updates of the current year federal expenditures and receipts are found to have improved forecast performance compared to MIDAS regressions without leads. Using a sample of 48 mainland U.S. states, the second chapter considers the problem of forecasting state and local governments' expenditures and revenues. It first proposes a forecasting procedure that involves a simple mixed frequency data regression approach, namely combinations of Augmented Distributed Lag--Mixed Data Sampling (ADL-MIDAS) regression models. With this approach, for almost all states, it is found that the use of high frequency state-specific and national variables combined with a low frequency budget series provides forecast performance gains over the traditional models where all data are of the same low/annual sampling frequency. This chapter then proposes a procedure with a multiple equation regression model, specifically a Mixed Frequency--Bayesian Vector Autoregressive (MF-BVAR) model. The predictive ability of the proposed model is assessed against the forecast performance of a traditional, low frequency Bayesian Vector Autoregressive (BVAR) model. Although the forecast performance varies at the state level, the overall empirical forecast performance of the MF-BVAR is better than that of the traditional BVAR model. Finally, predictive abilities of the two proposed forecasting procedures are empirically examined and the results suggest that one cannot be chosen over the other. While the ADL-MIDAS model provides better forecasts for expenditure series across states, forecasts for revenue series are more accurately obtained via the MF-BVAR model. The third chapter proposes a method for producing current-quarter forecasts of the U.S. real Gross Domestic Income (GDI) growth with a range of available within the quarter monthly/ weekly/daily observations of macroeconomic and financial indicators, such as employment, industrial production, and stock prices. The real-time forecasting procedure involves a combination of MIDAS-type regression models constructed with predictors of different sampling frequencies. Evidence shows that forecast combinations of MIDAS regression models with monthly leads that are employed to have real-time forecast updates of the current-quarter GDI growth provide forecast gains over the traditional models, suggesting the use of readily available within-quarter data in forecasting current-quarter output growth.
Date of publication
Keyword
Subject
DOI
Identifier
Resource type
Rights statement
  • In Copyright
Advisor
  • Hill, Jonathan
  • Ghysels, Eric
  • Aguilar, Michael
  • Francis, Neville
  • Chaudhuri, Saraswata
Degree
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2014
Language
Publisher
Place of publication
  • Chapel Hill, NC
Access
  • There are no restrictions to this item.
Parents:

This work has no parents.

Items