Heterogeneity in the volatility spillover of cryptocurrencies and exchanges

Financial Innovation

Table 6 Predicted power and relative contributions of explanatory variables

Panel 1: BTC’s total volatility connectedness as dependent variable
Actual	Predicted		Lagged RES_OE_BTC	Lagged RES_CB_BTC	Lagged VolBF	Lagged VolOE
4.11E-06	4.04E-06	Individual Contribution(%)	25.39%	23.34%	24.18%	27.10%
		Category	Exchanges-Return		Exchanges-Trading volume
		Aggregate contribution(%)	48.72%		51.28%

Panel 2: ETH’s total volatility connectedness as dependent variable
Actual	Predicted		Lagged GSCI_Energy	Lagged RV_BN_ETH	Lagged RV_OE_ETH	Lagged VolCB	Lagged RES_CB_ETH	Lagged RES_BN_ETH	Lagged bt_ETH
1.15E-06	1.13E-06	Individual Contribution(%)	10.59%	14.46%	11.98%	17.77%	17.83%	17.92%	9.44%
		Category	Macroeconomy	Exchanges-volatility		Exchanges-Trading volume	Exchanges-Return		ETH-Technical
		Aggregate contribution(%)	10.59%	26.45%		17.77%	35.75%		9.44%

Panel 3: LTC’s total volatility connectedness as dependent variable
Actual	Predicted		Lagged Total_LTC	Lagged PV_OE	Lagged PV_LTC
1.15E-06	1.13E-06	Individual Contribution(%)	46.48%	28.23%	25.29%
		Category	Lagged DV	Exchanges-Internet Concern	LTC-Internet Concern

(1) The variables in stepwise models are all in logarithm, except RES (the simple return). All independent variables are lagged in one order
(2) In three panels, some variables require first differences for their non-stationary. We use the Augmented Dickey-Fuller (ADF) tests to test a unit root in time series. Some variables are stationary in levels, including RES_BF_NTC, RES_BN_BTC, RES_CB_BTC, RES_OE_BTC, RV_BF_BTC, RV_BN_BTC, RV_CB_BTC, RV_OE_BTC, BTC_Volatility, bt_BTC, and tx_BTC in Panel 1. RES_BF_ETH, RES_BN_ETH, RES_CB_ETH, RES_OE_ETH, RV_BF_ETH, RV_CB_ETH, ETH_Volatility, bt_ETH in Panel 2. RES_BF_LTC, RES_BN_LTC, RES_CB_LTC, RES_OE_LTC, LTC_Volatility, bt_LTC in Panel 3. Other variables are in the first differences. After the appropriate transformation, all the dependent and independent variables are stationaries
(3) All variables are winsorized at 5th and 95th percentiles
(4) To control the heteroskedasticity, we use OLS and the robust standard error in the stepwise regression. We use HAC standard error to estimate the optimal model to overcome the heteroscedasticity and serial correlation, which is illustrated in Table 12
(5) The relative contributions of the optimal explanatory variables are estimated by the standardized coefficients proposed by Bring (1994)