Forecasting Industry Employment for a Resource - based Economy Using Bayesian Vector Autoregressive Models

Chang K Seung,Sung K Ahn

doi:10.52324/001c.8170

Abstract

Bayesian vector autoregressive (BVAR) models are developed to forecast industry employment for a resource-based economy. Two different types of input-output (I-O) information are used as priors: (i) a reduced-form I-O relationship and (ii) an economic-base version of the I-O information. Out-of-sample forecasts from these two I-O-based BVAR models are compared with forecasts from an autoregressive model, an unconstrained VAR model, and a BVAR model with a Minnesota prior. Results indicate most importantly that overall the model version with economic base information performs the best in the long run.

Highlights

Since Doan, Litterman, and Sims (1984) first used the Bayesian vector autoregressive (BVAR) approach to forecast macroeconomic variables, numerous studies have been conducted for national macroeconomic time series studies (e.g., Todd, 1984; Litterman, 1986; LeSage and Magura, 1991) or regional time-series studies (e.g., Amirizadeh and Todd, 1984; Magura, 1990; Partridge and Rickman, 1998; Puri and Soydemir, 2000; Rickman, 2001; Rickman, 2002)
Lags of 1 month and 12 months were chosen for unrestricted vector autoregressive model (UVAR) model because of (1) the prominent auto-correlations shown at those lags, (2) the partial autoregression matrices that are significant at those lags, (3) information criteria that support the model, and (4) parsimony
This means that the economic base information incorporated in the f (i, j) matrix used in IO2_VAR improves the forecasting capability in the long run in terms of the number of most accurate forecasts

Summary

INTRODUCTION

Since Doan, Litterman, and Sims (1984) first used the Bayesian vector autoregressive (BVAR) approach to forecast macroeconomic variables, numerous studies have been conducted for national macroeconomic time series studies (e.g., Todd, 1984; Litterman, 1986; LeSage and Magura, 1991) or regional time-series studies (e.g., Amirizadeh and Todd, 1984; Magura, 1990; Partridge and Rickman, 1998; Puri and Soydemir, 2000; Rickman, 2001; Rickman, 2002). Rickman, Miller, and McKenzie (2009) developed a model in which a prior of proportionality between basic and nonbasic sectors is imposed in estimating the equations None of these studies or other previous regional BVAR studies used a framework in which the economic base relationship is specified as Bayesian prior for a ―resource-based‖ economy. SEUNG & AHN: BVAR FORECASTING OF INDUSTRY EMPLOYMENT based I-O relationships [i.e., reduced-form employment relationships as in Partridge and Rickman (1998) and Rickman (2001)] (IO1_VAR), and (5) a BVAR which uses employmentbased I-O relationships with economic base relationships incorporated (IO2_VAR) This is the first example of BVAR application to a regional level analysis for an economy dependent on a natural resource economic base.

Bayesian VAR

X u2R R 1

Input-Output Information

MODEL SPECIFICATION AND DATA

MODEL IMPLEMENTATION AND RESULTS

CONCLUSIONS