Improvements for the Iowa Economic Forecasting Model
Is Version Of
The Iowa Department of Revenue (IDR) forecasts the state's personal income, wages/salaries, and employment for the state’s Revenue Estimating Conference (REC) three times a year. These forecasts are of great importance as they influence decisions regarding state policy and projects. Using the state’s original models and data input I make suggestions on possible improvements that could be adopted within the Bayesian Vector Autoregression forecasting models. My suggestions focused on changing the parameter values (theta being the constant ratio of the previous data’s variance used to decrease the priors’ influence on the predictions, and lambda, the standard deviation of the prior data) and number of lags in dependent and independent variables used in the Bayesian vector autoregression models. These suggestions were based on detailed statistical analyses on specific variables within the models and the input data that IDR uses in forecasting the state’s economy. The error measurement used is mean absolute percentage error (MAPE) since it is the error measurement that IDR uses when evaluating its forecasting errors. My suggestions improved IDR’s forecasting performance based on both in-sample and out-of-sample comparisons with their current model and should subsequently have a direct impact on the government’s policy and project decisions if they decide to implement them.
econometrics, bayesian, forecasting