RDP 2019-03: Explaining Monetary Spillovers: The Matrix Reloaded 6. What Determines the Strength of Spillovers?

Jonathan Kearns, Andreas Schrimpf and Fan Dora Xia

April 2019

The primary goal of this section is to shed light on the different channels by examining which macro and financial variables determine the strength of spillover effects under the specification of Equation (2). The empirical results are reported in Tables 4–8. Our interpretation of the results presented below closely adheres to the framework of the three channels outlined above.

6.1 Domestic Economic Conditions

To test the domestic economic conditions channel, we interact monetary policy shocks with measures of economic linkages across economies. The main prediction of the domestic economic conditions channel is that economies with tighter economic linkages with shock originator economies should receive stronger spillovers. We first use trade variables to capture the direct economic linkages between economies. The trade-related variables we use are: bilateral export openness (exports from the recipient economy to the originator economy relative to GDP), bilateral import openness, as well as variables typically used in the trade gravity equation literature such as common language, weighted distance and time difference.

The results are presented in Table 4, pointing to a very limited explanatory power of the domestic economic conditions channel in determining spillover strength. Among all specifications, only the coefficient in front of the interaction term of bilateral trade with the ECB path shock is statistically significant. That said, this effect is no longer significant when removing euro area countries from the set of recipient economies, suggesting that among euro area countries trade openness may be a proxy for other factors. These results do not indicate there is a measurable role of the domestic economic conditions channel in determining spillovers.

Table 4: Spillovers and Bilateral Trade Linkages
    Target Path Premium R2 (%)
Exports ECB 0.01 0.10 0.03 2.5
(0.21) (2.73) (0.68)
ECB (excl EA) −0.01 0.01 −0.01 3.0
(−0.12) (0.15) (−0.27)
Fed 0.04 0.01 −0.01 5.0
(0.47) (0.22) (−0.15)
Imports ECB −0.03 0.08 0.00 2.4
(−0.74) (2.47) (−0.02)
ECB (excl EA) −0.01 0.02 −0.02 3.0
(−0.29) (0.49) (−0.31)
Fed 0.01 0.01 0.00 5.0
(0.20) (0.52) (0.02)
Common language Fed 0.08 0.05 −0.09 5.5
(0.59) (0.60) (−1.18)
Weighted distance Fed 0.00 0.00 0.00 5.5
(−0.50) (0.55) (0.09)
Time difference Fed −0.03 −0.02 −0.02 5.5
(−0.74) (−1.02) (−0.83)
Notes: The table reports the results of panel regressions as given by Equation (2) with various recipient-specific conditional variables Xi,t − 1 measuring bilateral trade linkages and other controls; the dependent variable is the daily change in 10-year bond yields in our set of 47 recipient economies; as regressors, besides the monetary shocks for the ECB and the Fed, some specifications also consider the daily change in the US Treasury yield and the VIX as global controls; the report coefficients correspond to γ ^ j in Equation (2); t-stat from PCSE are given in parentheses; cells coloured red (blue) indicate statistically significant positive (negative) coefficients at a 10 per cent confidence level; exports and imports (per cent of GDP) are measured in standard deviations from the mean

However, trade is only a small portion of the economic linkages between economies which also include the actions of multinational companies, information and investment flows and common global demand shocks. Hence, we also consider a measure of broader economic linkages, by looking at the commonality in macroeconomic conditions across economies. For this purpose we use long-term realised correlations in growth and inflation, without specifying the detailed mechanism underlying the correlation. Results using these measures as interaction terms are presented in Table 5.[24] None of the macro commonality measures robustly show up as significant when interacted with monetary policy shocks, however, further putting the validity of the domestic economic conditions channel in doubt.

Table 5: Spillovers and Commonality in Macro Conditions
    Target Path Premium R2 (%)
Inflation correlation ECB 0.25 0.32 0.51 2.6
(0.90) (1.35) (1.12)
ECB (excl EA) 0.47 −0.29 0.53 3.4
(1.07) (−0.77) (0.87)
Fed −0.05 −0.39 0.20 5.9
(−0.12) (−1.70) (0.81)
Growth correlation ECB −0.16 0.44 0.34 3.0
(−0.86) (2.68) (1.00)
ECB (excl EA) −0.04 0.18 0.39 5.4
(−0.19) (1.06) (0.98)
Fed 0.02 0.28 0.50 5.5
(0.05) (1.00) (1.68)
Notes: The table reports the results of panel regressions as given by Equation (2) with various recipient-specific conditional variable Xi,t −1 measuring common macroeconomic conditions; the dependent variable is the daily change in 10-year bond yields in our set of 47 recipient economies; as regressors, besides the monetary shocks for the ECB and the Fed, specifications also include the daily change in the US Treasury yield and the VIX as global controls; the reported coefficients correspond to γ ^ j in Equation (2); t-stat from PCSE are given in parentheses; cells coloured red (blue) indicate statistically significant positive (negative) coefficients at a 10 per cent confidence level; inflation correlation and growth correlation are measured as a 20-year rolling correlation of realised CPI inflation and realised real GDP growth, respectively

6.2 FX Regime Channel

To test the FX regime channel, we interact monetary policy shocks with measures of FX regimes in our panel regression framework. The FX channel predicts that economies ‘pegging’ their currencies to those of the shock originator should experience stronger spillovers. Rather than rely on ‘de jure’ measures of FX regimes, we construct de facto measures as in De Grauwe and Schnabl (2008), which essentially boils down to the realised bilateral exchange rate volatility between the originator and recipient economies.[25]

The results reported in Table 6 indicate that the FX channel yields greater power than the domestic economic conditions channel in explaining variation in spillover strength across economies. In the case of ECB shocks, the coefficient in front of the interaction term of the FX regime measure and the path shock is negative and significant. The more dampened FX volatility is, for example due to an explicit or implicit currency peg, the larger the spillover of interest rate shocks. FX volatility remains a robust variable in explaining cross-country differences in spillover strengths also when removing the euro area from the set of recipient economies. In the case of Fed policy shocks, the coefficient in front of the interaction term of our FX regime measure and the risk premium shock is marginally significant. Overall, these results suggest that spillover strengths are to some extent related to FX regimes, consistent with recent findings in Han and Wei (2016).

Table 6: Spillovers and the FX Channel
    Target Path Premium R2 (%)
FX volatility ECB 0.24 −0.49 −0.29 2.9
(1.27) (−2.82) (−0.96)
ECB (excl EA) −0.25 −0.48 −0.78 3.9
(−1.18) (−2.01) (−1.87)
Fed 0.02 0.24 −0.22 5.6
(0.04) (0.86) (−1.61)
Notes: The table reports the results of panel regressions as given by Equation (2) with the recipient-specific conditional variable Xi,t − 1 measuring FX volatility with respect to shock-originating economies; the dependent variable is the daily change in 10-year bond yields in our set of 47 recipient economies; as regressors, besides the monetary shocks for the ECB and the Fed, specifications also include the daily change in the US Treasury yield and the VIX as global controls; the reported coefficients correspond to γ ^ j in Equation (2); t-stat from PCSE are given in parentheses; cells coloured red (blue) indicate statistically significant positive (negative) coefficients at a 10 per cent confidence level; FX volatility is measured as a 1-year rolling realised volatility estimate, based on squared daily spot FX changes (%)

6.3 Risk Premium Channel

To assess the validity of the risk premium channel, we interact monetary policy shocks with measures of financial openness. The main idea is that the more financially open and interconnected an economy, the larger the impact of fluctuations in global risk appetite and financial conditions on bond yields. We explore a range of financial openness measures, including bilateral capital flows and the overall level of cross-border investments. Specifically, the bilateral variables used are: foreign currency debt denominated in the currency of the originator economy (i.e. either in US dollars or euro), and portfolio debt, portfolio equity, loans and FDI (all bilateral between the originator and recipient economies, assets and liabilities separately). We also use aggregate measures of financial openness: debt assets, portfolio assets, FDI assets and financial derivative assets (and separately, the equivalent liability measures) as well as the Chinn-Ito measure of financial openness.[26] Most of these variables are statistically significant in explaining the strengths of spillovers from the Fed and ECB.[27]

Given the correlation between these measures and to avoid any ensuing multicollinearity issues, we run separate regressions with each pair of these measures, checking which variables do not lose significance after controlling for other measures. This exercise helps us to determine which proxies are most powerful in capturing financial openness and in explaining spillover strengths. As can be gleaned from Table 7, two measures stand out, foreign currency debt and portfolio equity from originator economies.[28]

Table 7: Spillovers and Financial Interconnectedness
  FX debt Debt (from) Equity (from) Equity (to) Portfolio assets Portfolio liabilities FDI assets FDI liabilities
FX debt na yes yes yes yes yes yes yes
Debt (from) no na yes yes yes yes yes yes
Equity (from) yes yes na yes yes yes yes yes
Equity (to) no no no na no no no no
Portfolio assets yes yes no yes na yes yes yes
Portfolio liabilities yes no no yes no na no no
FDI assets yes yes no yes no no na no
FDI liabilities yes no no yes yes no no na
Notes: The table reports the results of regressions based on various financial openness indicators; it shows whether the row variable remains significant (t-stat > 1.69) after controlling for the column variable when both of them are included simultaneously in panel regression Equation (2); two variables, FX debt and equity investment from the originator economy remain significant even when included with all other controls

Our finding on the importance of financial openness in explaining spillovers is consistent with Rey (2013), as it points to important spillovers of major central banks' monetary policies to other countries' long-term rates and hence an impact on local financial conditions, regardless of whether the capital account is managed or not. To better differentiate between the risk premium and FX channel, we test whether FX regime and financial openness conditions present different channels. To this end, we include both FX volatility and our two financial openness measures as conditioning variables. Table 8 shows that FX volatility retains its significance in explaining cross-sectional variation in spillover strengths despite the addition of our financial openness measures. This result suggests that the FX regime represents a distinct and relevant channel, at least for explaining spillovers from ECB monetary policy shocks.

Table 8: Distinguishing FX and Financial Channels
  Foreign currency debt   Portfolio equity from originator   FX volatility
Target Path Premium Target Path Premium Target Path Premium
ECB 0.06 0.06 0.14   0.08 0.09 0.07   0.25 −0.39 −0.28
(1.41) (1.72) (2.42) (1.20) (1.76) (1.54) (1.10) (−1.92) (−0.86)
ECB −0.05 0.07 0.00   0.08 0.04 0.16   −0.44 −0.33 −0.92
(−1.36) (1.89) (0.06) (1.34) (0.84) (1.46) (−1.82) (−1.23) (−1.91)
Fed 0.16 0.13 −0.06   −0.03 −0.07 0.15   0.20 0.36 −0.18
(1.77) (2.57) (−1.80) (−0.38) (−1.20) (2.25) (0.29) (1.10) (−1.06)
Notes: The table reports the results of panel regressions as given by Equation (2) with recipient-specific conditional variable Xi,t − 1 including foreign currency debt, portfolio equity from shock-originating economies and FX volatility with respect to currencies in shock-originating economies; the dependent variable is the daily change in 10-year bond yields in our set of 47 recipient economies; as regressors, besides the monetary shocks for the ECB and the Fed, specifications also include the daily change in the US Treasury yield and the VIX as global controls; the reported coefficients correspond to γ ^ j in Equation (2); t-stat from PCSE are given in parentheses; cells coloured red (blue) indicate statistically significant positive (negative) coefficients at a 10 per cent confidence level

Footnotes

We estimate the commonality in economies' business cycle and inflation with a 20-quarter rolling regression. The results are robust to sensible variations of this set-up. [24]

FX volatility is calculated from the bilateral exchange rate between the originator and recipient economies. [25]

Ideally, we would like to have each economy's fixed income holdings in different currencies as a financial openness measure given its important role in portfolio choices of global fixed income investors. Unfortunately, such granular data does not exist for all economies we considered. [26]

Comprehensive results are provided in Tables A3–A6 in the Online Appendix. [27]

It is possible that recipient economies experiencing strong spillovers may take measures tightening financial openness to tame spillovers. This would result in negative relation between spillover strength and financial openness. The potential downward bias would actually make our evidence supporting the risk premium channel stronger. [28]