Is gold a safe haven for the dynamic risk of foreign exchange?

Future Business Journal

Table 5 The results of TVP-PVAR model

Parameter	Mean	SD	95% Interval	CD	Inefficiency
\({(}\sum \beta {)}_{{1}}\)	0.0023	0.0003	[0.0018, 0.0029]	0.366	5.20
\({(}\sum \beta {)}_{{2}}\)	0.0023	0.0003	[0.0018, 0.0029]	0.113	3.70
\({(}\sum a {)}_{{1}}\)	0.0056	0.0017	[0.0034, 0.0099]	0.796	15.53
\({(}\sum a {)}_{{2}}\)	0.0048	0.0024	[0.0031, 0.0079]	0.787	22.10
\({(}\sum h {)}_{{1}}\)	0.0054	0.0015	[0.0034, 0.0091]	0.516	13.45
\({(}\sum h {)}_{{2}}\)	0.0056	0.0016	[0.0034, 0.0097]	0.846	10.42

Done on EViews 12
In this study, we assume as a diagonal matrix for simplicity. Compared to the non-diagonal assumption, previous experiences of Primiceri [38] and Nakajima [34] indicate that this assumption is not sensitive to the results. The following priors are assumed for the i -th diagonals of the covariance matrices: \({(}\sum \beta {)}_{i}^{{ - 2}} \sim Gamma \, (40,0.02)\) \({(}\sum a {)}_{i}^{{ - 2}} \sim Gamma \, (40,0.02)\) \({(}\sum h {)}_{i}^{{ - 2}} \sim Gamma \, (40,0.02)\)
For the initial state of the time-varying parameter, rather flat priors are set as \(\mu_{{\beta_{0} }} = \mu_{{a_{0} }} = \mu_{{h_{0} }} = 0\), and \(\sum {\beta_{0} } = \sum {a_{0} } = \sum {h_{0} } = 10 \times I\). To compute the posterior estimates, we draw M = 10,000 observations after the initial 1000 observations are discarded. The estimates of \(\sum \beta\) and \(\sum \alpha\) are multiplied by 100. SD = standard deviation; CD = convergence diagnostics statistics