ABSTRACT
This paper examines macroeconomic effects of external shocks and their transmission mechanisms in one of the most commodity-abundant countries-Mongolia using a large Bayesian vector autoregression (BVAR) based on the approach proposed by Bańbura, Giannone, and Reichlin [(2010). Large Bayesian Vector Auto Regressions. Journal of Applied Econometrics, 25, 71–92]. Nine structural shocks (five external and four domestic shocks) are identified using a recursive ordering. Results show that external shocks are important sources of macroeconomic volatility in Mongolia. Commodity price shocks affect the economy through exchange rate and budget expenditure channels, while China’s growth and FDI shocks are primarily transmitted through the real sector and bank lending channels.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Gan-Ochir Doojav http://orcid.org/0000-0003-2416-9452
Notes
1 Standard VAR models may suffer from a loss of degrees of freedom, which decrease geometrically with the number of variables and proportionally with the number of lags included, resulting in inefficient estimates and increase in forecasting errors. In the classic econometrics, the curse of dimensionality can be resolved by including few numbers of variables and lags in VAR. Another alternative way is to use augmented vector autoregression (FAVAR) introduced by Forni, Hallin, Lippi, and Reichlin (Citation2000) and Stock and Watson (Citation2002).
2 De Mol et al. (Citation2008) studied the robustness of Bayesian regression to changes in number of variables and sample size. They concluded that when more variables are added (especially when double-exponential priors are used), the Bayesian regression choose fewer variables that explain most of the total variance of explanatory variables. This implies that macroeconomic variables have collinearity and carefully designed small-scale model can have the same results as a large-scale model.
3 Over-fitting issue arises when a model has too many parameters compared to the sample size. Over-fitted models tend to have poor predictive performance and overreact to slight changes in the sample.
4 In the case of Mongolia, there are fiscal rules such as restricting fiscal deficit and debt ratios; however implementation of the rules failed in practice.
5 Seasonal adjustment is made in only FDI data because it has strong seasonal fluctuations.
6 Log marginal likelihoods (log marginal data densities) for BVAR (2) and BVAR(4) are −3117.56 and −3109.43, respectively. The posterior odds ratio is 3394.8, computed by considering the BVAR(4) model as the null hypothesis (against the alternative hypothesis: BVAR(2) model), and the result offers ‘very strong’ evidence in favour of the BVAR(4) model.
7 This sample is called a training sample. It is the minimum length of time suited for econometric evaluations.
8 In 2014, FDI inflows was 4.25 billion US dollars (about 40 percent of GDP), then once the first phase of Oyu Tolgoi project was completed, it dropped to almost zero in 2015.
Additional information
Notes on contributors
Gan-Ochir Doojav
Gan-Ochir Doojav is Director General of Research and Statistics Department, Bank of Mongolia. He earned his master and PhD degrees in Economics from the Australian National University (ANU).
Davaajargal Luvsannyam
Davaajargal Luvsannyam is Director of Research Division, Research and Statistics Department, Bank of Mongolia. He earned his master and PhD degrees in Economics from the Australian National University (ANU).