International Spillover Effects of Quantitative Easing Policy: A Case Study on the U.S. and Japan

ABSTRACT

This study analyses the influence of the changing nature of the U.S. quantitative easing (QE) policy’s international spillover on the Japanese economy following the global financial crisis. Based on existing studies, we construct a two-country vector autoregression model for the U.S. and Japan and estimate the time-varying parameters to comprehend the changing macroeconomic and financial structures of both countries. We not only confirmed the existence of spillover effects but also found significant changes in them over time. The spillover effects on Japan are larger for the real economy in the earlier phase and for financial markets in the later phase of the U.S. QE policy.

KEYWORDS:

• U.S. quantitative easing policies
• TVP-VAR
• international spillover effect
• Japan

JEL CLASSIFICATION:

• E44
• E52
• E58

Acknowledgments

We thank Osaka Bankers Association and JSPS KAKENHI Grant Numbers 16H03618 and 20K13533 for the financial support. This is a revised version of the paper presented at the Japan Society of Monetary Economics, Fifteenth Annual Conference of Asia-Pacific Economic Association and Fourth Annual Conference of the Japan Economy Network.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Ijiri and Matsubayashi (2019) show that the Japanese QE policy effects changed significantly.

² $A_t$ is a recursive restriction following prior studies, such as Nakajima (2011) and Ijiri and Matsubayashi (2019). The dimensions of $z_t$, ${\beta }_t$, $A_t$, and ${\mathrm{\Sigma }}_t$ are $(n\times 1)$, $(n_s^2\times 1)$, $(n\times n)$, and $(n\times n)$, respectively.

³ The dimensions of $w_{\beta },w_{\alpha }$, and $w_{\sigma }$ are $(n^{2}s\times n^{2}s),((n^2-n)/2\times (n^2-n)/2),$ and $(n\times n)$. This analysis was conducted following the program used by Nakajima (2011). We use the following initial states of TVP: ${\beta }_0\sim N(0,10I)$, $a_0\sim N(0,10I)$, and ${\sigma }_0\sim N(0,10I)$. Note that ${\tilde{\omega }}_{\beta ,j}^2,{\tilde{\omega }}_{a,j}^2,{\tilde{\omega }}_{\sigma ,j}^2$ are the $j$-th diagonal elements of the $j$-th diagonal elements of ${\omega }_{\beta },{\omega }_a,{\omega }_{\sigma }$: ${\tilde{\omega }}_{\beta ,j}^2\sim IG(100,0.001),{\tilde{\omega }}_{a,j}^2\sim IG(10,0.001),{\tilde{\omega }}_{\sigma ,j}^2\sim IG(10,0.001)$. Further, we check the autocorrelations and $p$-values of Geweke’s (1992) CD statistics for each model in Appendix.

⁴ In the U.S., the unemployment rate is an important indicator of the Federal Reserve’s monetary policy goals.

⁵ All data were sourced from Datastream. All variables except $U{R}^{US}$ are in their logarithmic forms. All data are demeaned. The two models set two lags.

⁶ This study uses a Bayesian estimation using the Markov chain Monte Carlo method. An initial sample of 30,000 observations is generated and discarded before generating another sample of 30,000 observations. Based on the final 30,000 observations, impulse responses are calculated with the parameters estimated for each point in time.

⁷ This does not accord with the result of Deckle and Hamada (2015), who estimated the fixed parameter model.

⁸ We check this result with the impulse response at 3 months; the figures are omitted due to space limitations.

Additional information

Funding

This work was supported by the Japan Society for the Promotion of Science [16H03618,20K13533], and Daiginkyo Forum [Special award in 2015] of The Osaka Bankers Association.