Abstract
This paper presents macroeconomic forecasting by using a time-varying Bayesian compressed vector autoregression approach. We apply a random compression by using projection matrix to randomly select predictive variables in vector autoregression (VAR), and then perform true out-of-sample forecast where the forecast values are averaged across all estimated models, containing different in both explanatory variables and number of those variables by using Bayesian model averaging (BMA). In addition to this, we allow the parameters in Bayesian compressed VAR to be time-varying by implementing dynamic model averaging (DMA) algorithm that is applicable with VAR using forgetting factor to control the degree of time-varying in the estimating parameters. We validate the performance of the proposed method via real macroeconomic data including up to 53 variables. The empirical results demonstrate that the predictive performance of time-varying Bayesian compressed VAR can beat traditional VAR types which are considered to have a potentiality to deal with large size variables.
Highlights
Big data has become increasingly important in econometric field
Bayesian model averaging (BMA) method was applied to weight computation for each random compressed vector autoregression (VAR), this reduces the sensitivity of random projection matrix
The contribution of this work is that we present a way to select important variables via BMA and dynamic model averaging (DMA)
Summary
Econometricians, from macroeconomic point of view, are typically interested in working on how to implement an efficient forecasting method for large dimensional data as much as possible. Research work with other methods such as random projection, or compressing the data into a smaller matrix size instead of shrinking the priors on parameters has been proposed; see [25]. Bayesian model averaging (BMA) method was applied to weight computation for each random compressed VAR, this reduces the sensitivity of random projection matrix. We applied and implemented the model to predict key macroeconomic variables of Thailand and illustrated that by using Bayesian compressed VAR, we are able to improve the predictive performance relative to the traditional VAR such as Factor Augmented VAR, Dynamic Factor Model, Bayesian VAR with Minnesota prior, and Bayesian AR(1).
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.