Invoice currency choice, nonlinearities and exchange rate pass-through

ABSTRACT

Estimating time-varying thresholds as a proxy for exporter’s predicted exchange rates, this study proposes a new approach to analyse possible asymmetric behaviour of exchange rate pass-through (ERPT) or pricing-to-market (PTM) in Japanese exports between yen appreciation and depreciation periods. Constructing the industry-specific nominal effective exchange rate on a contract (invoice) currency basis, we perform the multivariate threshold near-vector autoregressive (near-MTVAR) estimation and reveal a strong tendency of symmetric ERPT in the short-run, between yen appreciation and depreciation periods. From the 2000s, however, Japanese machinery exporters increased the degree of PTM even in the long-run, while other industries raised the degree of long-run ERPT, reflecting the difference of product differentiation across industries. This evidence has significant implications for the recent unresponsiveness of the Japanese trade balance to the large depreciation of the yen.

KEYWORDS:

• Exchange rate pass-through
• pricing-to-market
• time-varying threshold
• multivariate threshold near-vector autoregressive (near-MTVAR) model
• yen appreciation and depreciation
• invoice currency choice

JEL CLASSIFICATION:

• C22
• F31
• E31

Acknowledgments

This study is conducted as a part of the Project “Exchange Rates and International Currency” undertaken at Research Institute of Economy, Trade and Industry (RIETI), and financially supported by the JSPS (Japan Society for the Promotion of Science), KAKENHI Grant Numbers JP16H03638, JP16H03627, JP16J11988, JP17KT0032, and JP19H01504. We wish to thank Etsuro Shioji, Takatoshi Ito, Eiji Ogawa, Yushi Yoshida, Junko Shimizu, Kentaro Kawasaki, Mi Dai, Jianwei Xu, and anonymous referees for their valuable comments. The views expressed in this paper are those of the authors and do not necessarily represent the view of institutions the authors belong to.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Abenomics typically denotes an economic-stimulus package that consists of ‘three arrows’ – bold monetary easing, flexible fiscal policy, and structural reforms, advocated and led by Prime Minister Shinzo Abe. Among the three arrows, we focus on the excessive monetary easing policies such as quantitative and qualitative easing (QQE) and the negative interest rate policy as a key factor in driving the yen to depreciate sharply from the end of 2012.

² See, for example, Marston (1990), Parsons and Sato (2008), and Yoshida (2010).

³ See Sato (2003) and Ito et al. (2012).

⁴ Fedoseeva and Werner (2016) use the one standard deviation of exchange rate changes as a threshold to divide large appreciation, depreciation, and ‘inaction’ bands. Florian (2013) also employs the two thresholds approach in estimating the nonlinear autoregressive distributed lag (NARDL) model, though analysing not the ERPT but the exchange rate impact on EMU exports to the United States. While this inaction band approach is useful, it is intrinsically based on the zero-threshold model.

⁵ Murase (2013) conducts a nonlinear estimation of ERPT by defining exchange rate volatility as a threshold variable. Although it is a useful approach, we assume the exchange rate in levels to be a threshold variable in the present research.

⁶ Another useful approach is to employ a time-varying parameter VAR approach to analyse possible changes in ERPT behaviour. See Shioji (2014, 2015) who investigated whether impulse responses of Japanese import prices to exchange rate shocks had changed over the sample period. See also Shimizu and Sato (2015) who applied the Kalman filter technique to a single-equation model of ERPT in Japanese exports.

⁷ Contract-NEER was first developed by Ceglowski (2010) and then extended by Ito et al. (2018).

⁸ This definition differs from the conventional definition of effective exchange rates. In the next section, we show how to construct the contract-NEER.

⁹ For example, in the four-year window from January 2000 to December 2003, the estimated threshold is treated as the threshold value (exporter’s predicted rate) for June 2003. The choice of a ‘four-year’ window is somewhat arbitrary. We tried various windows with different lengths to obtain an appropriate number of trials. Using a graphical investigation such as in Figure 3, we finally chose the four-year window.

¹⁰ We thank an anonymous referee for the suggestion of using the multivariate threshold (MT) VAR model.

¹¹ See Cushman and Zha (1997), Zha (1999), and Maćkowiak (2007) for an analysis using a near-VAR model with block exogeneity.

¹² Besides using a threshold variable, a Markov-switching VAR model can be employed to consider nonlinearities of ERPT/PTM. However, as the objective of this study is to investigate possible nonlinearities of ERPT/PTM in regard to different currency periods, especially taking into account firm exchange rate prediction. Using the threshold variable would be a more intuitive and straightforward approach to distinguish different currency periods. Therefore, the Markov-switching VAR model is beyond the scope of this study. We would like to thank an anonymous referee for suggesting the Markov-switching VAR model to us, which will be considered in our future research.

¹³ The Cholesky decomposition, typically employed in the VAR literature, deals with contemporaneous zero restrictions. This indicates that if including a real output variable that is exogenous to other endogenous variables in a VAR, the real output variable is treated exogenously to other variables only contemporaneously. In contrast, a near-VAR model can include the real output variable as a block exogenous one (i.e. in a foreign block of Equation (3)), where the real output variable is strictly exogenous to other variables. Indeed, Blanchard and Quah (1989) developed long-run zero restrictions under a VAR framework. But, the block exogeneity restriction is stricter than the Blanchard and Quah (1989) restrictions in that the block exogeneity does not allow any effect of the domestic block on the foreign block at all.

¹⁴ The programing codes and data set for our near-MTVAR model are available upon request.

¹⁵ As will be discussed below, we tried estimation with a different order of variables in the domestic block, and the result is very similar to the benchmark result.

¹⁶ For reference, the yen-based export prices of All Manufacturing and seven industries are plotted in Appendix Figure B1. As for All Manufacturing and three machinery industries (Figure B1(a)), export price series are less likely to be affected by any structural breaks or external shocks such as the global financial crisis in 2008–2009. In the case of other industries (Figure B1(b)), export price series are likely to be affected somewhat by the global financial crisis. But, overall, we do not have to control for any exogenous shocks when using the export price series for an empirical analysis.

¹⁷ The data on Japan’s trade invoice currency is published by the Ministry of Finance.

¹⁸ The BOJ collects export price data when cargo is loaded in Japan at the customs clearance stage, and when free-on-board (FOB) prices at a Japanese port of exports are surveyed. As long as they are traded in foreign currencies, sample prices are recorded in the original contract currency and finally compiled as the ‘export price index on the contract currency basis.’ To compile the ‘export price index on the yen basis,’ sample prices in the contract currency are converted into yen equivalents by using the monthly average exchange rate of the yen vis-à-vis the contract currency. See the BOJ website (https://www.boj.or.jp/en/statistics/pi/cgpi_2010/index.htm/) for further details.

¹⁹ This assumption is relaxed when constructing the contract-NEER in practice.

²⁰ By definition, the sum of the weights in Equations (4) and (5) is assumed to be unity.

²¹ Weights are based on the input values of goods (i.e. raw and intermediate materials, fuel, and energy) and services for the manufacturing industry, using purchasers’ prices in the IO Table during the 2005 base year, published by the Ministry of Internal Affairs and Communications.

²² The 20 countries and areas are France, Germany, the Netherlands, Australia, Canada, Hong Kong, Korea, Singapore, the United States, the United Kingdom, mainland China, India, Indonesia, Malaysia, the Philippines, Thailand, Vietnam, Russia, Mexico, and Taiwan.

²³ Although not reported in this paper, the accumulated impulse responses are statistically significant for at least the first five periods (the results are available upon request). As shown below, we present the results of both impulse responses and error-confidence bands for a sub-sample analysis.

²⁴ The results of both impulse responses and error-confidence bands for sub-sample periods are presented in Appendix Figure B2. All impulse responses are statistically significant at least for the first five periods.

²⁵ We would like to thank an anonymous referee for encouraging us to conduct the formal test of symmetry in impulse responses presented in Appendix 1.

²⁶ For a good survey of ERPT in developing and emerging economies, see Aron, MacDonald, and Muellbauer (2014).

²⁷ As of 2010, three major machinery industries account for 66.6% of Japanese total exports, while Metals and Chemicals account for 11.8% and 9.5%, respectively. The share of exports by industry is taken from the BOJ website (http://www.boj.or.jp/statistics/pi/cgpi_2010/index.htm/).

²⁸ By estimating the time-varying parameter model, Shimizu and Sato (2015) found that the coefficient of the contract NEER (i.e. PTM elasticity) declined sharply, but only temporarily, during the yen appreciation period from 2011 to 2012. They also found that Japanese exporters returned to the strong PTM behaviour after the rapid depreciation of the yen from the end of 2012.

²⁹ Ito et al. (2012, 2018) investigated a relationship between product differentiation and invoice currency choice using the firm-level information obtained from interviews with Japanese major machinery firms.

³⁰ The equation is based on Paternoster et al. (1998) who proposed the test for the equality of regression coefficients.

Additional information

Funding

This work was supported by the Japan Society for the Promotion of Science [KAKENHI JP16H03627, KAKENHI JP16H03638, KAKENHI JP16J11988, KAKENHI JP17KT0032, KAKENHI JP19H01504].