Yen internationalization and Japan's international reserves

Economic Modelling 52 (2016) 452–466

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/ecmod

Yen internationalization and Japan's international reserves Zhiwen Zhang a,b, Anthony J. Makin c,d,⁎, Qinxian Bai e a

School of International Studies, Sun Yat-sen University, Institute of Advanced International Studies, Guangzhou, China Asia-Pacific Department, The International Monetary Fund, Washington, DC, USA c Economics, Griffith Business School, Griffith University, Gold Coast Australia d Lee Kuan Yew School of Public Policy, National University of Singapore, Singapore e School of Economics, Liaoning University, China b

a r t i c l e

i n f o

Article history: Accepted 16 September 2015 Available online 18 October 2015 Keywords: International reserves Yen internationalization Foreign exchange Gold SDR

a b s t r a c t The relationship between major East Asian economies' international reserves and internationalization of their currencies presents a seeming paradox in international finance. While large international reserves may be expected to foster more widespread global use of a currency, strong growth of international reserves has been associated with very low Asian currency internationalization. Using the generalized method of moments (GMM) estimator we show the overall size of the Bank of Japan's international reserves, as well as the ratio of its foreign exchange reserves to international reserves, are negatively related to yen internationalization, while gold reserves and special drawing rights are positively related. © 2015 Elsevier B.V. All rights reserved.

1. Introduction A large international finance literature examines the asset and transactions use of currencies beyond their country of issue. One stream addresses what factors determine the choice of particular currencies used as foreign exchange reserves, while another explores factors contributing to the internationalization of major economies' currencies. Most studies have focused on the US dollar, which has been the predominant reserve currency, having played the pivotal role in the Bretton Woods fixed exchange rate system that prevailed post-war until 1971. It is also by far the most internationalized currency, despite the creation of the euro and the post-Bretton Woods emergence of first Japan, and then China, as economic superpowers. A country's international reserves comprised mainly of foreign exchange reserves, gold reserves, and special drawing rights (SDR) are the external assets held by its monetary authority to maintain exchange rate stability and provide a buffer against future currency crises. Currency internationalization on the other hand refers to the international extension of a national currency's basic functions of unit of account, medium of exchange and store of value. Successful currency internationalization results in widespread use of a national currency in international trade, international financial transactions and reserve assets held by foreign central banks and monetary authorities (see Chinn and Frankel, 2005, 2008; Cohen, 1971; Hartman and Issing, 2002; Krugman, 1984; Prasad, 2014). ⁎ Corresponding author. E-mail addresses: [email protected], [email protected] (Z. Zhang), t.makin@griffith.edu.au (A.J. Makin).

http://dx.doi.org/10.1016/j.econmod.2015.09.026 0264-9993/© 2015 Elsevier B.V. All rights reserved.

Whether a national currency becomes widely used internationally depends largely on the confidence that international market participants have in the currency-issuing country and since international reserves are intended to improve market confidence, large international reserve holdings may be expected to foster currency internationalization. Hence a question of particular relevance to international policy makers in this context is why the euro, renminbi and yen have not become more internationalized relative to the US dollar in light of their economies' increased prominence in the world economy and, especially in the cases of China and Japan, why increased currency internationalization has not increased along with very strong growth in their central banks' foreign exchange assets. Given the large foreign exchange reserves held by the People's Bank of China and Bank of Japan, which grew very rapidly from the turn of the century onwards, it may seem paradoxical that internationalization of the renminbi and yen has been so minimal. The paradox arises because the very existence of East Asia's huge foreign exchange asset holdings implies economies in the region have been heavily engaged in international monetary and trade transactions. What then is the nexus between the internationalization of an economy's currency and the quantum and composition of its international reserves? Addressing that question is the main aim of this paper, focusing specifically on the case of internationalization of the yen and Japan's international reserves as renminbi internationalization was only officially promoted from 2009. A key feature of Japan's total international reserves is the dominance of foreign exchange reserves in total reserves. During the first decade of the century Japanese foreign exchange reserves averaged 97% of total international reserves, while

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

gold reserves averaged only 2% of the total. In contrast, figures for the United States were 21% and 64% respectively, while the euro area shares were 50% and 43%.1 Earlier studies on the determinants of the currency composition of foreign exchange reserves include Dooley et al. (1989), Eichengreen and Mathieson (2000). More recently papers by Chinn and Frankel (2005, 2008) in this vein predicted an enhanced role for the euro while Subramanian (2011) finds that economic size is the primary determinant of reserve currency status. Meanwhile, Huang et al. (2014) examine the case of the renminbi as a possible reserve currency predicting it will be increasingly used internationally in the future. The literature on yen internationalization has two focal points. The first examines the potential international use of the yen (Ito et al., 2012; Sato, 1999; Taguchi, 1994; Tavlas and Ozeki, 1992). The second focuses on the successes and failures of yen internationalization, as well as possible renminbi internationalization (Chen, 2004; Frankel, 2011, 2012; Kawai and Takagi, 2011; Xu, 2005). However, within the existing empirical research on this topic, the focus is on investigating the yen as an invoicing currency in international trade (Fukuda and Ono, 2006; Ito et al., 2012; Sato, 1999, 2003), while this paper focuses on the influence of international reserves on yen internationalization. In what follows we systematically explore theoretical and empirical relationships between international reserve holdings and local currency internationalization. In preview, we find that the composition and size of international reserves play key roles in determining how international reserves contribute to currency internationalization. Specifically, the ratio of foreign exchange reserves to international reserves is negatively related to local currency internationalization, while gold reserves and SDR holdings are positively related; the size of international reserves relative to GDP also proves to be negatively related. The rest of the paper is organized as follows. Section 2 describes the key features of Japan's international reserves and a short history of yen internationalization. Section 3 presents the conceptual foundations and hypotheses for examining the relationship between international reserves and yen internationalization. Section 4 introduces the data sources, model specification and methodology to be used, before Section 5 presents our econometric results. Section 6 concludes the paper.

2. Evolution of Japan's international reserves and yen internationalization In this section, we first present the evolution of Japan's international reserves focusing on the size and composition of its reserves. We then trace the process of yen internationalization. Our sample period in what follows avoids the global financial crisis of 2009–10 and its aftermath which was a period characterized by exceptionally wide swings in financial variables, including capital flows, exchange rates and changes in foreign reserves. On the contrary, there was considerably less volatility in these variables during our 1976–2009 sample period which fairly evenly spans times when Japan was growing strongly and when it subsequently experienced lower growth prior to the global financial crisis.

2.1. International reserves Fig. 1 depicts the evolution of Japan's international reserves which changed dramatically from the early 1990s due to a strong increase in the foreign exchange reserves share and sharp decline in gold reserves with little change in SDR holdings and IMF reserve positions. On average, the foreign exchange reserves share has exceeded 96% since 1996 with gold reserves less than 3%, and SDR holdings 2%. 1

Data source: International Monetary Fund, International Financial Statistics (IFS).

453

Fig. 2 combines the composition and size of Japan's reserves showing the size of reserves being driven by the foreign exchange component, especially from 1994 onward. 2.2. Yen internationalization In 1964, Japan accepted the obligations of Article VIII of the IMF Articles of Agreement to allow free convertibility of the yen under its current account although there were government concerns that this could limit money supply control and add to exchange rate volatility (Frankel, 1984). As a result, Japan's monetary authorities were cautious about internationalization of the yen in the 1960s–1970s. However, this changed in the mid-1980s when new policy measures were introduced to promote yen internationalization (see Takagi, 2009). The international use of the yen subsequently peaked then spiraled downward during the so called “lost decades” of low economic growth in the 1990s and 2000s following the collapse of the asset bubble fuelled by easy bank credit during the 1980s (See Amyx, 2004 and Koo, 2009 for related discussion). Since then, not only has economic growth been minimal, mild deflation has occurred at times, and public debt has increased dramatically to be the highest in the OECD proportionate to GDP due to successive bouts of fiscal stimulus. Fig. 3 presents the trajectory of the yen's international use as a reserve currency. The “Mount Fuji” shape gray line intuitively displays the rise and fall of the yen's internationalization from 1976 to 2009. We could observe that, before 1991, the internationalization of the yen went upward with a gradual progress; in 1991, it reached its historic peak of 8.5%, expressed by the yen share in identified official holdings of foreign exchange. Again this de-internationalization broadly coincides with Japan's “lost decades” of the 1990s and 2000s. In sum, Figs. 1 and 3 show that yen internationalization appears to be closely and inversely related to foreign exchange reserves, rising when foreign exchange reserve holdings were low, and falling when they were high. In addition, the size of international reserves as a share of GDP is negatively associated with yen internationalization. 3. Linkages between international reserves and currency internationalization There are several theoretical reasons for an inverse relationship to exist between the scale of a central bank's foreign reserves and the extent to which that nation's currency is internationalized. First, the notion of currency competition suggests that when the economies of dominant reserve currencies are weak, due for instance to financial crises, the scope for internationalizing other currencies rises, and, vice versa, falls, if economic and financial conditions in the dominant reserve economy strengthen. Relatedly, the supremacy of the US dollar is explicable with reference to a currency inertia effect (Chinn and Frankel, 2005, 2008), the strength of which ultimately depends on the soundness of the reserve issuing country's macroeconomic and financial conditions. In this context, the dominance of the US dollar over the yen, renminbi and euro may seem anomalous in light of domestic monetary shocks the United States has experienced since the 2008–10 Wall Street banking crisis with the rise in public debt to historically high levels and the flooding of money markets with US dollars via quantitative easing. However, this reflects the underlying strength of United States' public institutions, its broad and deep financial markets, appropriately regulated, and robust and transparent legal framework (Prasad, 2014). Second, currency internationalization can be directly related to the operations of the central bank. Foreign exchange reserves accumulate in the first instance due to the central bank using domestic currency to buy other currencies in the foreign exchange market in order to prevent currency appreciation. In Japan's case, during the “lost decades” era a major reason for curbing yen appreciation in real terms against the dollar was to avoid Japan's loss of international competitiveness,

454

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

100

Percent of International Reserves

90 80

Foreign Exchange Reserves

70 60 50 40 30

Reserve Postion in the IMF

Gold Reserves

20

SDR holdings

10

2008

2006

2002

2004

2000

1998

1996

1994

1992

1990

1988

1986

1982

1984

1980

1978

1974

1976

1972

1968

1970

1966

1964

1962

1960

1956

1958

1954

1950

1952

1948

0

Fig. 1. The evolution of the Japanese international reserves composition. Sources: Authors' calculation based on raw data from International Financial Statistics (IFS); gold reserves in troy ounces have been converted into dollar value through the gold prices in London (Line 11276KRZZF… in IFS) (accessed in May–July 2010).

understood as the capacity of Japanese producers of tradable goods and services to compete on price grounds with foreign suppliers. Yen appreciation worsens Japanese competitiveness, other things equal, by making its exports of goods and services more expensive from foreign buyers' perspective. Meanwhile, appreciation makes imported goods and services cheaper than domestically produced products which negatively affects Japan's GDP and employment levels. Not only would this have been especially important over a lengthy interval when Japan's economic growth was subdued, yen appreciation would also have made all traded goods cheaper, thereby compounding the deflation problem Japan experienced over this time. Pressure on the nominal exchange rate to appreciate can arise from either the current or capital account sides of the external accounts, resulting in an increase (decrease) in the supply (demand) of foreign exchange relative to domestic demand (supply) of that currency, as illustrated in Fig. 4. For instance, if there is an increase in the supply of US dollars from S0 to S1 relative to demand for them in the dollar-yen foreign exchange market due to US monetary easing, the yen would

tend to appreciate from e0 to e1 (defined as yen per dollar with a fall implying appreciation). To prevent this, the Bank of Japan intervenes buying US dollars, which implies foreign exchange reserves accumulate by the amount FX shown on the horizontal axis in Fig. 4. Alternatively, a fall in demand for US dollars due to relatively lower US interest rates from D0 to D1 would also appreciate the yen. Under these circumstances, the Bank of Japan can similarly intervene buying US dollars, thereby accumulating an additional amount of US dollar reserves measured by distance YF in the figure. In sum, intervention by the Bank of Japan buying US dollars in exchange for yen to prevent real appreciation necessarily increases US dollar reserves, whereas sale of US dollars would have depleted foreign exchange reserves and strengthened the yen. Accordingly, the scale of foreign exchange reserves at any time is indicative of past central bank action to prevent exchange rate appreciation in order to avoid losses of international competitiveness and prevent deflation. To that extent, large foreign exchange reserve holdings signify

25

90 80

20

70 60

Foreign Exchange Reserves (Left Scale)

15

50 40 30 20 10

10

Gold Reserves (Left Scale) The Reserves size (Right Scale)

5

0 1948 1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

0

Percent of GDP

Percent of International Reserves

100

Fig. 2. The relations between the reserves composition and size in Japan. Source: Author's calculation based on raw data from IFS (accessed in May–July 2010).

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

9%

455

25%

8% 20%

7% 6% 5%

The Degree of Yen Internationalization (Left Scale)

15%

4%

10%

3% 2% The Reserves Size of Japan (Right Scale)

1%

0% 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

0%

5%

Fig. 3. The evolutions of the Japanese reserves size and yen internationalization. Notes: 1. The degree of yen internationalization is the share of the yen in total identified official holdings of foreign exchange. Data from 1976–2008 are from IMF Annual Report (1983–2009). Data for 2009 and 2010 are calculated by authors based on raw data taken from COFER (Currency Composition of Official Foreign Exchange Reserves). Among them, datum on 2010 refers to that of 2010Q1 which was downloaded in May 2010. 2. The reserves size of Japan is the Japanese international reserves as share of its GDP. Source: the same as that of Fig. 2.

that a currency may be ‘undervalued’, and hence out of line with market fundamentals.2 From the perspective of international users of the currency for transactions and invoicing purposes, this in effect means the currency has sub-optimal international purchasing power which would deter yen internationalization. On the other hand, if investors expect the Bank of Japan to allow future yen appreciation they could benefit, though this would run counter to past policy, be difficult to predict, and reduce the yen value of the central bank's reserves. Nonetheless, international investors are more likely to acquire yen to hold in the event of further expected appreciation when there is sustained upward pressure on the exchange rate, despite Bank of Japan intervention, as occurred for instance during the 1970s when the economy experienced very strong growth. Hence, the relationship between expectations, yen internationalization and reserves accumulation complex and time variant. Third, there are money supply and central bank balance sheet effects. When the Bank of Japan buys US dollars it expands the Japanese money supply, other things equal. In the longer term, intervention that persistently increases foreign exchange reserve holdings can therefore generate inflationary pressures which acts to offset any competitiveness gains achieved via reserve accumulation. For this reason, foreign exchange intervention is normally sterilized in most economies via central bank sales of domestic bonds to soak up excess domestic liquidity, though in Japan's case sterilization may have been less than complete due to underlying deflationary rather than inflationary pressures. However, to the extent the Bank of Japan's intervention was sterilized by domestic bond market intervention, a fall in domestic bonds on the assets side of its balance sheet would have offset the rise in foreign reserve assets backing the money supply on the liabilities side. Hence the sharp rise in the share of foreign assets in the Bank of Japan's balance sheet implies Japan's money supply in effect became more reliant on foreign fiat currency as backing. This weakens the yen's internationalization appeal from a balance sheet perspective. What then of external reserve holdings not in currency form? Gold, for instance, has long acted as an international store of value, medium of exchange, and unit of account and, unlike fiat money, adds intrinsic value on the asset side of the central bank's balance sheet. Hence, the higher the ratio of gold reserves in international reserves, the greater the confidence international investors should have in the currency the

2

See Makin (2007) for related discussion on China's exchange rate and reserves.

country issues leading to a higher level of currency internationalization. In contrast to the yen, a relatively high stock of gold with its high intrinsic international value is a significant item in the US government's balance sheet (see Greenspan, 2014 for related discussion). This, of itself, makes dollar internationalization more appealing from a prudential perspective, as gold has long been accepted as the ultimate form of payment in times of crisis. The SDR is another form of reserve asset and unit of account that was created in 1969 to ease the dollar crisis and add liquidity to the international monetary system. The allocation of the SDR is based on member economies' reserve positions with the IMF, its value currently determined by a currency basket composed of the U.S. dollar, euro, pound sterling and Japanese yen. As the IMF has ceased issuing SDR, its value is inherently more stable than ordinary sovereign currencies (Lago et al., 2009). Therefore, other things equal, a rise in a central bank's SDR holdings should favor currency internationalization from a balance sheet perspective. Similarly, reserve positions with the IMF or quotas affecting SDR allocations are also positive for currency internationalization. Empirically, whether the size of reserves is negatively or positively associated with yen internationalization depends on their composition. If the composition is heavily dominated by foreign exchange reserves, the negative effect of foreign exchange reserves is expected to prevail over the positive effects of gold and SDR. The above

Fig. 4. Foreign exchange market intervention and reserve accumulation. Source: this study.

456

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

graphical evidence and our theoretical reasoning therefore suggest two testable hypotheses. Hypothesis 1. The very high share of foreign exchange reserves has a negative effect on yen internationalization, while gold reserves and SDR holdings have a positive effect despite their excessively low shares.3 Hypothesis 2. The size of Japan's international reserves relative to GDP negatively influences yen internationalization. In what follows we formally test these hypotheses using advanced econometric techniques. 4. Data, model specification and methodology This section defines our variables and data sources and specifies the econometric methodology to be used in what follows. 4.1. Variables and data sources Based on data availability, this study uses annual data to investigate the influence of Japan's international reserves via composition and size measures on yen internationalization over the period 1976–2009.4 4.1.1. Yen internationalization (SHARE) A currency's internationalization can be measured by its uses in international trade, international financial transactions, and central banks' international reserves. The IMF regularly publishes data on the currency composition of foreign exchange reserves in its annual report which provides us with important data to measure the degree of the yen's international use as a reserve currency. In accordance with the existing literature (Chinn and Frankel, 2005, 2008), we use the share of the yen in total identified official holdings of foreign exchange to measure the degree of yen internationalization. Variables of international reserves: composition (FOREXR, GOLDR, and SDRR) and size (TRSHARE). As discussed in Section 2, the impacts of Japanese international reserves' composition on yen internationalization are complicated. Specifically, the extremely high share of foreign exchange reserves (FOREXR), has a negative effect on the yen's internationalization, but both ratios of gold reserves and SDR holdings to international reserves (GOLDR and SDRR), have positive effects. In addition, the size of Japanese international reserves (TRSHARE), which is the ratio of Japan's international reserves to its GDP, has an adverse impact on yen internationalization. All data are taken from International Financial Statistics (IFS).5 The existing literature (Bai and Zhang, 2011; Chinn and Frankel, 2005, 2008) has identified several long-run determinants of currency internationalization. They document that the economic power, financial market development, exchange rate appreciation, and currency inertial contribute to, while inflation is negatively associated with the local currency's international use. We follow them to measure these variables. Specifically, economic power (GSHARE) is measured by the share of a country's GDP in the aggregate world GDP. Data on the aggregate world GDP are retrieved from World Economic Outlook Database (WEO); data on Japanese GDP in the local currency, as well as the exchange rate, are taken from IFS and converted into GDP in USD through the market exchange rate (period average).6 3 Since the link between the quota and local currency internationalization is very weak, we would not empirically examine the effect of the Japanese quota on yen internationalization. 4 All data used in this study are publicly available. 5 Gold reserves (fine troy ounces) have been converted into values in the US dollar through the gold prices in London (Line 11276KRZZF…in IFS). 6 Since this study uses exchange rate in different variables, if there are no additional notes, we specifically refer to it as the market exchange rate (period average).

As for financial market development (STOCKT), it is constructed as the stock value traded at the Tokyo Stock Exchange divided by GDP in Japan. Data on stock value and Japanese GDP are taken from Global Financial Database (GFD) and IFS, respectively. Among them, data on Japanese GDP in yen have been converted into those in U.S. dollar through the market exchange rate. STOCKCAP, the ratio of Japan's stock market capitalization to Japanese GDP, is also used as an alternative indicator of STOCKT for robustness checks. Data sources for STOCKCAP are the same as those for STOCKT. Considering that the yen has appreciated in the sample period, we follow Cohen (2005) and choose the market exchange rate of the yen against the U.S. dollar to control for the currency strength (level exchange rate of the yen) (FOREX). The data on the market exchange rate are obtained from IFS. Currency Inertia (CSHARE (−1)) is captured by one-period lag of dependent variable. As well, Inflation (INFDIF) is measured by the differential between Japanese CPI inflation rate and that of advanced economies. Data on both time series are retrieved from IFS. In addition, we also control for variables like the once-in-acentury global financial crisis of 2007–2009 dummy (GFC), exchange rate volatility (VOLT) of the yen against the US dollar, and interest rate differential (TRDIFL) between Japan and the UK to complement information in our econometric regressions. Specifically, we construct GFC as taking 0 before 2007, and taking 1 otherwise. We measure VOLT as the first difference of annual logarithm of the market exchange rate of the yen to the U.S. dollar.7 Data are taken from IFS. As well, we measure TRDIFL as the difference between 13-week financing bill rate in Japan and the UK treasury-bill rate, and the data are also taken from IFS. 4.2. Model specification and methodology Chinn and Frankel (2005, 2008) pointed out that currency share cannot be linearly associated with economic power because the value of SHARE is bounded between 0 and 1. They suggest a logistic transformation of SHARE to establish a linear function. We follow their procedure to construct a measure of yen internationalization (CSHARE)8 and check a scatter plot to see if the linear relationship between CSHARE and economic power holds. By looking at Fig. 5, we find that LOGITSHARE (or CSHARE) is linearly associated with LNGSHARE (economic power). So we specify the estimation equation as follows: CSHAREt ¼ C þ β1 At þ β2 IN RESERVEt þ β3 CONTROLt þ ϵt among these variables: CSHARE is the dependent variable; t is time; C is the constant; βi, i = 1, 2, 3, are parameters to be estimated; A is a conditioning information set which includes the long-run determinants of yen internationalization, with GSHARE, FOREX, STOCKT or STOCKCAP, INFDIF and CSHARE (− 1); IN_RESERVE represents proxies of both composition and size of Japanese international reserves, with FOREXR, GOLDR, SDRR and TRSHARE, respectively; CONTROL refers to control variables consisting of GFC and VOLT as well as TRDIFL; and ϵt is the error term. In choosing methodologies, one of key concerns is the potential endogeneity issue in estimation which makes the ordinary least squares (OLS) estimator biased and inconsistent and requires us to find a proper method to overcome it. 7 As to the measure of exchange rate volatility, there is currently no agreement in the academia. Some scholars measure it as the first-difference of logarithm of the market exchange rate (see, for example, Dominguez, 1993, see pp. 15; Dominguez, 1998, see pp. 169 and pp. 170; Bonser-Neal and Tanner, 1996, see pp. 861), and others measure it as the standard deviation of the first difference of monthly logarithms of the exchange rate centering the year investigated. See, for example, Clark et al. (2004). 8 CSHARE = LOGIT (SHARE) = LOG(SHARE / (1 − SHARE)).

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

-2.0

LOGITSHARE

-2.4

-2.8

-3.2

-3.6

-4.0 2.0

2.2

2.4

2.6

2.8

3.0

LNGSHARE Fig. 5. LOGITSHARE vs. LNGSHARE. Source: this study.

In our study, the international reserves variables could be endogenous with different degrees: gold reserves and SDR holdings could be more exogenous than foreign exchange reserves. This is because foreign exchange reserves, gold reserves and SDR holdings are three of the four components of international reserves, and they must be highly correlated with each other. If we introduce them together into the same regression, there must be a severe multi-collinearity, making the estimates inaccurate. However, if we separately introduce each of these variables into the regressions, the presence of omitted variables implies that the individual international reserves variable correlates with the error term, leading to an endogeneity problem. More importantly, it is possible that there exists a bidirectional causal relationship between the currency share and the high ratio of foreign exchange reserves. That is, the lower degree of yen internationalization leads to a situation that the Bank of Japan must hold more foreign exchange reserves, while this holding of more foreign exchange reserves is not conducive to the process of yen internationalization as clarified in Section 2. However, gold reserves and SDR holdings do not face this reversal causality from yen internationalization because a high yen internationalization does not necessarily promote the Bank of Japan to buy gold and increase its SDR holdings. Such distinction makes foreign exchange reserves more endogenous than gold reserves and SDR holdings when estimation. In addition, economic power and financial development could also be correlated with the error term. In particular, CSHARE(− 1), the lagged endogenous dependent variable, appears on the right hand of the equation, implying there is a problem of endogeneity in the model, which has been theoretically proved by Davidson and MacKinnon(1999, pp. 91–93). Under these circumstances, the OLS estimator will be biased and inconsistent. Therefore, we employ the instrumental variable (IV) estimator to resolve this endogeneity problem. Since we have 34 observations, which is a large sample (n ≥ 30) and more than sufficient to conduct a reliable time series study, and do not assume that the error term is a spherical disturbance, we use the GMM estimator rather than the two-stage least squares (2SLS) estimator to overcome the endogeneity problem. One of challenges we are facing when using the GMM estimator is to find proper instruments. As valid instrumental variables, they must satisfy with two conditions: (1) they are correlated with endogenous variables based on economic reasons; (2) they are uncorrelated with the error term in the equation. Given that both gold reserves and foreign exchange reserves are two of key components of international reserves,

457

and that central banks and monetary authorities often use their foreign exchange reserves to buy gold, so we argue that gold reserves are highly, but negatively correlated with foreign exchange reserves. In this case, gold reserves are naturally good instruments for foreign exchange reserves. However, to avoid the correlation between gold reserves ratio and the error term, we use one and two-period lags of gold reserves ratio (GOLDR(−1) and GOLDR(−2)) as proper instrumental variables to instrument foreign exchange reserves ratio (Zhang, 2015). The potential high negative correlation coefficients between GOLDR, GOLDR(− 1), GOLDR(−2) and FOREXR will be reported in Table 2 as part evidence showing the validity of instrument selection. In addition, as suggested in Fair (1970), when estimating equations with lagged endogenous variables, lagged dependent and independent variables must be used as instruments. Therefore, except the abovementioned specific instrumental variables for foreign exchange reserves, instruments in all regressions of our paper include one and two-period-lagged dependent and independent variables as internal instruments.9 Finally, we use Hansen's (1982) J test of over-identifying restrictions to examine the validity of the instruments. The instruments are valid if we fail to reject the null hypothesis.10 As well, all standard errors of estimates are corrected by using the Newey and West (1987) procedure (or HAC method) and hence are heteroskedasticity and autocorrelation consistent. 5. Empirical results This section describes the data, examines correlations between variables of interest and tests for the stationarity of time series before analyzing the robustness of the benchmark regression results. 5.1. Data properties It is necessary for us to look at the data properties before estimation by presenting correlations and unit root tests of the relevant variables to confirm whether the data supports our linear specification. Correlation matrix of variables investigated is shown in Table 1. A highly positive correlation between GSHARE and CSHARE has been found, and its correlation coefficient is 0.840; STOCKT is moderately yet positively correlated to CSHARE, and its coefficient is 0.335; both GOLDR and SDRR are positively correlated, although FOREXR and TRSHARE are negatively correlated to CSHARE, and their correlation coefficients are 0.341, 0.379, − 0.148 and − 0.726, respectively. The correlation coefficients between GOLDR(− 1), GOLDR(− 2) and FOREXR are − 0.896 and − 0.831, respectively. For other variables, see Table 1. The above correlations imply, first, that the dependent variable is related to the independent variables as expected, second, that the data support our linear benchmark specification of long-run determinants of yen internationalization (although there may be a non-linear relationship between yen internationalization, foreign exchange reserves, gold reserves and SDR holdings respectively due to lower correlations that require robustness checks) and, third, that instrumental variables are highly correlated with endogenous variables as required for the choice of instruments. To ensure that the variables in question are stationary before estimation, we use the Augmented Dickey–Fuller (ADF) unit root test to investigate whether or not variables of interest have unit roots. The results show that, of the 17 variables in question, all variables are stationary 9 Lagged dependent and independent variables are proper internal instruments for endogenous independent variables because these lagged variables are highly correlated with their respective current values but uncorrelated with the error term in the present. For similar use, please refer to Liviatan (1963), Fair(1970) and Bond(2002). 10 The software used in this paper is Eviews8.0.

1.000 1.000 −0.507 1.000 −0.118 0.118 1.000 0.896 −0.241 0.411 0.424 −0.344 0.057 0.128 0.245 0.335 0.174 0.352 0.421 −0.559 −0.027 −0.488 0.285

1.000 −0.384 0.338 0.367 −0.353 0.211 0.001 0.117 0.227 0.062 0.212 0.305 −0.492 −0.146 −0.523 0.192

1.000 −0.544 −0.584 0.707 −0.910 0.829 0.783 0.734 0.777 0.773 0.714 −0.553 0.070 −0.308 0.437

1.000 0.981 −0.455 0.543 −0.384 −0.318 −0.256 −0.357 −0.272 −0.213 0.199 −0.185 0.082 0.042

1.000 −0.444 0.569 −0.413 −0.352 −0.284 −0.380 −0.304 −0.242 0.211 −0.122 0.056 0.029

1.000 −0.678 0.589 0.536 0.498 0.503 0.507 0.464 −0.298 0.031 −0.223 0.381

1.000 −0.944 −0.896 −0.831 −0.869 −0.866 −0.789 0.713 −0.033 0.303 −0.399

1.000 0.980 0.942 0.942 0.941 0.897 −0.854 −0.089 −0.321 0.471

1.000 0.979 0.934 0.962 0.948 −0.887 −0.153 −0.370 0.466

1.000 0.922 0.958 0.966 −0.914 −0.159 −0.438 0.480

1.000 0.911 0.871 −0.839 −0.095 −0.320 0.364

1.000 0.968 −0.918 −0.109 −0.515 0.590

1.000 −0.918 −0.155 −0.555 0.550

1.000 0.104 0.559 −0.519

GFC VOLT TRSHARE SDRR(−2) SDRR(−1) SDRR GOLDR(−2) GOLDR(−1) GOLDR FOREXR INFDIF STOCKCAP STOCKT FOREX

1.000 0.942 0.840 −0.028 0.335 0.316 −0.218 −0.148 0.341 0.445 0.527 0.379 0.543 0.595 −0.726 −0.071 −0.518 0.388

GSHARE CSHARE(−1) CSHARE

CSHARE CSHARE(−1) GSHARE FOREX STOCKT STOCKCAP INFDIF FOREXR GOLDR GOLDR(−1) GOLDR(−2) SDRR SDRR(−1) SDRR(−2) TRSHARE VOLT GFC TRDIFL

Table 1 Variable correlation matrix.

Notes: 1. All variables take logs except VOLT and GFC, INFDIF and TRDIFL plus 1 before taking log, respectively; 2. Data on variables are in percentage before taking logs. Among them, CSHARE = LOGIT(SHARE) = log(SHARE / (1 − SHARE)), and data on SHARE are in decimal.

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466 TRDIFL

458

except CSHARE, CSHARE(−1), SDRR and TRSHARE (see Table 2). In the presence of structural breaks of time series, the ADF test may lead to the false acceptance of the null hypothesis of a unit root. Therefore, we further use a new ADF-type test for unit roots with two structural breaks developed by Narayan and Popp (2010). These results show that CSHARE, CSHARE(−1), GSHARE, STOCKCAP, GOLDR, GOLDR(−1) and VOLT are non-stationary at the conventional significance levels (see Table 3). However, as Engle and Granger(1987) noted that, for two or more non-stationary time series, their linear combinations could be stationary. This implies that it is not necessary for us to make the individual time series stationary before estimation in this case.11 5.2. Benchmark regression results analysis To estimate the quantitative effects of Japanese international reserves' size and imbalanced composition on yen internationalization, we first employ the long-run determinants of the yen's internationalization in the conditioning information set to specify the regression benchmark. We then investigate the impacts of Japan's international reserves on the internationalization of the yen by introducing individual variables of the composition and size of Japan's international reserves into the regression. We subsequently extend the regression benchmark by individually introducing control variables of VOLT, GFC and TRDIFL into the benchmark. Based on this, we investigate the effects of the various indicators of Japan's international reserves on yen internationalization. We finally conclude with further robustness checks by using STOCKCAP, an alternative measure of financial market development, while examining the dynamic effects of GOLDR and SDRR on the yen's internationalization, respectively. Table 4a presents the empirical results of the effects of Japanese international reserves on yen internationalization. Column (1) reports the impacts of major variables in the benchmark on the internationalization of the yen. We find that GSHARE, the Japanese economic power, has a large and positive effect on yen internationalization, with an estimation coefficient of 0.812. The effect of GSHARE is the largest one in the benchmark regression12 and statistically significant at the level of 1%. FOREX, the level exchange rate reflecting the strength of the yen against the U.S. dollar, greatly contributes to yen internationalization, having an estimation coefficient of 0.564 and being statistically significant at the level of 1%. In the process of yen internationalization, CSHARE(− 1), the proxy of currency inertia, also has a statistically significant and positive impact, and its estimation coefficient is 0.416. The effect of financial market development (STOCKT) has a small, positive coefficient that is statistically significant at the level of 1%. Inflation (INFDIF) has a significantly negative impact on yen internationalization with an estimation coefficient of − 0.164. In addition, statistically significant J-Statistic (with a large P-value of 0.511) shows that the instruments are valid, while a high adjusted R2 (0.928) value means a very good fit. Moreover, the Q-statistic that tests for the residual serial correlation significantly supports the null hypothesis of no serial correlation in residual series at the conventional confidence level, and the Normality test shows that the residual series follow a normally distributed process, whereas stationary regression residual series at the 1%

11 To check if there is a “spurious regression” problem in the regression (Granger and Newbold, 1974), we can first test if there is a co-integration relationship in linear combinations of variables interested and then further confirm if the specification is correct. That is, we conduct the ADF unit root test on the regression residual series. If we can reject the null hypothesis of a unit root at the conventional statistical significance levels, it implies that the regression residual series are stationary, and the co-integration relationship noted above is confirmed. This shows that there is no spurious regression issue (Zhang, 2009, pp. 360–367). 12 When comparing these regression coefficients, they must first be standardized.

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

459

Table 2 Augmented Dickey–Fuller (ADF) unit root test results. Null hypothesis: a time series has a unit root. Models

Intercept

Variables

t-Statistic

P-value

Trend and intercept

CSHARE CSHARE(−1) GSHARE FOREX STOCKT STOCKCAP INFDIF FOREXR TRSHARE GOLDR GOLDR(−1) GOLDR(−2) SDRR SDRR(−1) SDRR(−2) VOLT TRDIFL

−2.117 −1.792 −3.889***

(0.240) (0.378) (0.008)

−3.174** −3.293** −3.856*** −4.498*** −2.062

(0.036) (0.027) (0.006) (0.002) (0.261)

t-Statistic

None P-value

−4.528***

(0.009)

−3.261* −3.537* −3.999** −1.823 −3.948** −3.850**

(0.091) (0.053) (0.020) (0.669) (0.022) (0.028)

−3.247*

(0.094)

t-Statistic

−4.126***

Lag length P-value

(0.000)

AIC 0 1 11 13 11 11 0 7 7 1 1 1 2 2 3 0 1

Notes: 1. All variables take logs except VOLT, INFDIF and TRDIFL plus 1 before taking log, respectively; 2. Data on variables are in percentage before taking log. Among them, CSHARE = LOGIT(SHARE) = log(SHARE / (1 − SHARE)), and data on SHARE are in decimal. 3. P values are reported in parentheses; 4. ***, ** and * represent the statistically significant levels of 1%, 5% and 10% respectively.

level signifies that the regression is not a spurious one.13 The regression results presented above are as expected, which shows that the estimation specification of the benchmark is very good and can fully reflect the major effects of the long-run determinants of yen internationalization. Now, we investigate the effects of Japanese international reserves on yen internationalization by introducing the proxies of the composition and size of Japanese international reserves into the benchmark. Column (2) displays the marginal effect of FOREXR on yen internationalization. The regression results show that the signs of determinants that affect local currency internationalization, in the long run, are completely consistent with those in the benchmark regression. In this context, FOREXR, the share of foreign exchange reserves in international reserves, has a negative impact on yen internationalization. The estimating coefficient is −0.869 and is statistically significant at the level of 1%. This implies, other things being equal, that as the ratio of foreign exchange reserves to international reserves increases by 1%, SHARE / (1 − SHARE) will decline by 0.869%, and thus SHARE, which is the degree of yen internationalization, will decrease by 0.465%. We can find this negative effect is large, around 0.5%, which is from the currency competition of existing reserve currencies, particularly the US dollar. This estimation result is highly consistent with our theoretical analysis in Section 2. Column (3) shows the effect of GOLDR on yen internationalization. It shows that the signs of the determinants of local currency internationalization are consistent with those in the benchmark regression, and GOLDR, the share of gold reserves in international reserves, contributes

13 1. We perform the ADF unit root test on residual series by automatically selecting lag length based on SIC. 2. Based on Table 2 in Mackinnon (2010), we calculate the critical values for the unit root test on the residual series based on co-integration regressions (with constant, no trend): 1) for 3 time series with I(1), the calculated critical values at the 1% level for sample sizes of 26, 30, 31 and 33 are − 4.895, −4.810, − 4.792 and −4.760, respectively. 2) for 4 time series with I(1), the calculated critical values at the 1% level for sample sizes of 26, 30, 31 and 33 are − 5.396, − 5.289, − 5.267, and − 5.227; while for the sizes of 30 and 31 at the levels of 5% and 10% are − 4.483, −4.095, −4.470, and −4.085 respectively. 3) for 5 time series with I(1), the calculated critical values at the 1% level for sample sizes of 30, and 31 are −5.737, and −5.711, while for the sample sizes of 30 and 31 at the 5% and 10% levels are −4.897, −4.494, as well as −4.881 and −4.482 respectively; also for the size of 27 at the 5% level is −4.953. 4) for 6 time series with I(1), the calculated critical values at the 1% level for sample sizes of 30 and 31 are −6.162 and −6.131. The null hypothesis is that residual series has a unit root; 4. ***, ** and * represent the statistical significance level of 1%, 5% and 10%, respectively.

to yen internationalization. The estimation coefficient is 0.08114 and is statistically significant at the level of 1%. In other words, other things being equal, as the ratio of gold reserves to international reserves increases by 1%, SHARE / (1 − SHARE) will increase by 0.081%, and thus SHARE will be enhanced by 0.075%. An increase in gold reserves implies an improvement of the composition of Japanese international reserves and a more reliable sovereign credit of the yen, which greatly contributes to the degree of yen internationalization. This is also consistent with our expectation. Column (4) displays the effect of SDRR on yen internationalization. We find that the signs of the long-run determinants of local currency internationalization remain the same as those in the benchmark. SDRR, the ratio of SDR holdings to international reserves, has a significantly positive effect on yen internationalization with an estimation coefficient of 0.074. In other words, all other things being equal, a 1% increase in the share of SDR holdings will raise SHARE / (1 − SHARE) by 0.074%, thus SHARE increases by 0.069%. An increase in SDRR implies an improvement of the composition of international reserves in Japan. This positive effect on yen internationalization is also consistent with our expectation. Column (5) of Table 4a presents the effect of TRSHARE, the size of Japan's international reserves, on yen internationalization. The results show that the signs of major determinants of local currency internationalization continue in line with those in the previous regressions. However TRSHARE, the ratio of international reserves to GDP, has a negative impact on yen internationalization with an estimation coefficient of − 0.151 and statistically significance at 1% level. In other words, other things being equal, a 1 percent increase in the share of international reserves accounted for GDP in Japan will reduce SHARE / (1 − SHARE) by 0.151%, thus SHARE decreases by 0.131%. As noted in Section 2, this negative effect is caused by the highly imbalanced composition with the extremely high share of foreign exchange reserves which is negatively associated with yen internationalization shown in Column (2). This result is also consistent with our expectation. In addition, we also report results by OLS to check endogeneity effects. These results are presented in Columns (6) to (10) of Table 4a Comparing the estimated coefficients with those by GMM, OLS 14 After standardization, the coefficient of 0.081 is 0.2214, which is much larger than the un-standardized one.

460

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

Table 3 Results of ADF-type unit root test with two structural breaks. Null hypothesis: a time series has a unit root Nr.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

Series

Sample

CSHARE CSHARE(−1) GSHARE FOREX STOCKT STOCKCAP INFDIF FOREXR TRSHARE GOLDR GOLDR(−1) GOLDR(−2) SDRR SDRR(−1) SDRR(−2) VOLT TRDIFL

1976–2009 1977–2009 1976–2009 1976–2009 1976–2009 1976–2009 1976–2009 1976–2009 1976–2009 1976–2009 1977–2009 1978–2009 1976–2009 1977–2009 1978–2009 1976–2009 1976–2009

T

34 33 34 34 34 34 34 34 34 34 33 32 34 33 32 34 34

M1

M2

Test statistic

TB1

TB2

K

Test statistic

TB1

TB2

K

−2.133 −2.551 −2.716 −1.756 −1.857 −0.298 −4.372** −5.296*** −0.954 −0.702 −0.050 −3.956* −4.496** −4.609** −3.233 −3.465 −3.978*

1984 1985 1985 1985 1989 1989 1994 1985 1986 1986 1986 1983 1989 1990 1991 1985 1985

2000 2001 1991 2000 1999 1998 1996 1992 1992 1994 1993 1988 1991 1992 1993 1995 1992

0 0 0 5 0 3 0 5 0 0 0 1 4 1 4 5 3

−2.537 −3.313 −1.983 −5.326** −4.597* −3.901 −5.229** −4.405* −5.67*** −2.398 0.326 −2.47 −3.888 −3.076 −4.678* −3.657 −4.241*

1991 1986 1985 1985 1989 1989 1994 1982 1986 1982 1988 1983 1985 1987 1987 1985 1986

1994 1997 1995 1996 1998 1998 1996 1985 1998 1998 1999 1991 1991 1992 1993 1995 1992

0 1 5 2 5 5 2 1 3 1 5 0 4 0 0 5 3

Notes: 1. The results are produced by Gauss 14.0 using the Narayan and Popp (2010) method; 2. M1 and M2 denote Model 1 and Model 2 respectively; TB1 and TB2 denote the true break years; 3. K is the optimal lag; 4. ***, ** and * represent the statistically significant levels of 1%, 5% and 10% respectively.

estimators underestimate the effects of international reserves and longrun determinants like economic power, exchange rate appreciation, financial market development as well as inflation, while overestimating

the effect of currency inertia on yen internationalization. These biased estimates are obvious evidence of the presence of an endogeneity problem addressed subsequently.

Table 4a The effects of international reserves on yen internationalization (regression results: GMM and OLS, STOCKT). Dependent variable: CSHARE = log(SHARE / (1 − SHARE)) GMM

C GSHARE FOREX STOCKT INFDIF CSHARE(−1)

OLS

Benchmark

FOREXR

GOLDR

SDRR

TRSHARE

Benchmark

FOREXR

GOLDR

SDRR

TRSHARE

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

−6.842*** (0.000) 0.812*** (0.000) 0.564*** (0.000) 0.135*** (0.006) −0.164*** (0.000) 0.416*** (0.003)

–

−5.873*** (0.000) 0.765*** (0.001) 0.349*** (0.001) 0.125** (0.017) −0.229*** (0.001) 0.353** (0.030)

−6.097*** (0.000) 0.758*** (0.001) 0.412*** (0.002) 0.144** (0.019) −0.223*** (0.000) 0.367*** (0.007)

−4.893*** (0.000) 0.532*** (0.004) 0.286*** (0.003) 0.160*** (0.000) −0.181*** (0.004) 0.317** (0.028)

−3.466*** (0.001) 0.304** (0.031) 0.348*** (0.000) 0.091* (0.097) −0.032 (0.472) 0.775*** (0.000)

–

−2.645*** (0.005) 0.260** (0.031) 0.153 (0.200) 0.094* (0.061) −0.059 (0.137) 0.721*** (0.000)

−2.800*** (0.001) 0.263** (0.024) 0.203** (0.025) 0.095** (0.050) −0.044 (0.284) 0.732*** (0.000)

−1.139 (0.299) −0.017 (0.908) 0.036 (0.766) 0.107** (0.019) −0.028 (0.431) 0.708*** (0.000)

FOREXR

0.561*** (0.000) 0.096*** (0.003) 0.181*** (0.000) −0.205*** (0.004) 0.497*** (0.000) −0.869*** (0.000)

GOLDR

0.081*** (0.000)

SDRR

0.065** (0.021) 0.074*** (0.010)

TRSHARE R2 Adj. R2 Obs. adj. Bandwidth J-statistic P-value Q-stat.(1) P-value Norm. test (P) LM test (F:P) No. of I(1) R.S. ADF test

0.401*** (0.002) 0.079 (0.127) 0.136*** (0.001) −0.101*** (0.004) 0.629*** (0.000) −0.658*** (0.000)

0.940 0.928 31 Fixed 5.258 (0.511) 0.003 (0.959) (0.281) – 3 −5.344***

0.945 0.934 31 Fixed 5.967 (0.743) 1.328 (0.249) (0.549) – 3 −6.406***

0.943 0.929 31 Fixed 5.007 (0.659) 1.058 (0.304) (0.810) – 4 −6.353***

0.938 0.923 31 Fixed 5.617 (0.585) 2.425 (0.119) (0.148) – 4 −6.892***

0.054* (0.088) −0.151*** (0.000) 0.950 0.938 31 Fixed 4.415 (0.731) 1.441 (0.230) (0.612) – 4 −6.687***

0.949 0.940 33 Fixed – – 0.071 (0.789) (0.578) (0.739) 3 −5.810***

0.957 0.949 33 Fixed – – 10.217 (0.521) (0.354) (0.348) 3 −6.738***

0.956 0.946 33 Fixed – – 1.511 (0.219) (0.433) (0.258) 4 −6.823***

0.955 0.944 33 Fixed – – 2.301 (0.129) (0.360) (0.165) 4 −7.098***

−0.148*** (0.001) 0.960 0.951 33 Fixed – – 2.580 (0.108) (0.442) (0.221) 4 −7.330***

Notes: 1. All variables take logs, and INFDIF plus 1 before taking log; 2. P-values are reported in the parentheses; 3. *, ** and *** represent the statistically significant levels of 10%, 5% and 1%, respectively; 4. Instrumental variables include one and two-period lags of dependent and all independent variables. As for regressions on foreign exchange reserves ratio, we also use one and two-period lags of gold reserves ratio to instrument it; the null hypothesis for the Hansen's J-test is that the instrument is valid; 5. When there is a lagged dependent variable in the right hand of the equation, the D–W statistic is invalid, so we report one lag of the Q-statistics and its P-value, respectively, to test if the residual series are serially correlated. The null hypothesis is that there is no serial correlation. 6. No. of I(1) denotes the number of time series with I(1) in the regression. 7. Re. ADF Test refers to t-statistic of the ADF unit root test on residual series. For specific explanation, please refer to Footnote #12 of this paper.

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

In sum, controlling for the major determinants of yen internationalization, the highly imbalanced composition of Japan's international reserves has complex effects on yen internationalization. A higher ratio of foreign exchange reserves is associated with lower yen internationalization whereas higher ratios of both gold reserves and SDR holdings result in a higher yen internationalization. In addition, as the size of Japan's international reserves increases, yen internationalization decreases because of the reserve's highly imbalanced composition dominated by foreign exchange reserves. 5.3. Non-linear specifications Since three international reserves variables (FOREXR, GOLDR and SDRR) are weakly correlated with yen internationalization, we introduce their respective squared terms (LFOREXR, LGOLDR and LSDRR) to the specifications of Columns (2), (3) and (4) in Table 4a, to check if this potential non-linearity affects our approach. The results are reported in Table 4b. Column (2a) reports the effect of foreign exchange reserves share on yen internationalization by adding LFOREXR, the squared term of FOREXR before centering. It shows the coefficient of FOREXR is positive, but highly insignificant. Meanwhile, LFOREXR is also highly insignificant at the conventional significance level. To avoid the potential effect of multicollinearity caused by high correlation between FOREXR and its squared term, we introduce the variable MLF by centering FOREXR using FOREXR minus its mean, and create a new squared term CLFOREXR using MLF. We then use MLF and CLFOREXR to repeat the regression of Column (2a) by replacing FOREXR and LFOREXR. The results are presented in Column (2b). We find that MLF has a negative and highly significant coefficient while CLFOREXR, its squared term, remains highly insignificant at the

461

conventional statistical level. This confirms that a non-linear relationship between yen internationalization and foreign exchange reserves share does not hold, even though their correlation is weak. Additionally, following the same procedure, we check the potential non-linear effects of GOLDR and SDRR by adding their respective squared terms (LGOLDR, LSDRR, CLGOLDR, and CLSDRR) into their associated linear specifications in Columns (3) and (4) of Table 4a respectively, before and after centering the two variables. The results show all squared terms are insignificant at the conventional statistical level (see Columns (3a), (4a), (3b), and (4b) in Table 4b). This also implies non-linear specification for the relationships between gold reserves and SDR holdings and yen internationalization does not fit the data. In sum, these tests confirm that our linear approach is appropriate. 5.4. Robustness checks To get more robust estimates, we first introduce VOLT, exchange rate volatility, to the benchmark specification, and then investigate the effects of the composition and size of Japan's international reserves on yen internationalization. The results are reported in Columns (1) to (4) of Table 5. Columns (1) to (4) show that after introducing VOLT, as compared to the previous regressions, there is no change in signs of the major determinants of local currency internationalization in the benchmark. Additionally, the signs of the proxies of the composition and size of Japanese international reserves are fully consistent with those in previous regressions. Quantitatively, there are some changes in estimation coefficients. Estimating coefficients of FOREXR, GOLDR, SDRR, and TRSHARE are − 0.784, 0.080, 0.066, and − 0.146 respectively and the corresponding effects on SHARE become − 0.440, 0.074, 0.062 and − 0.127, respectively. Compared to the regression results without

Table 4b The effects of international reserves on yen internationalization (squared terms: GMM, STOCKT). Dependent variable: CSHARE = log(SHARE / (1 − SHARE)) Squared terms before centering LFOREXR

C GSHARE FOREX STOCKT INFDIF CSHARE(−1) FOREXR GOLDR SDRR LFOREXR MLF CLFOREXR LGOLDR MLG CLGOLDR LSDRR MLS CLSDRR AR(7) R2 ADj. R2 Obs. adj. Bandwidth J-statistic P-value Q-stat.(1) P-value Norm. test (P) No. of I(1) R.S. ADF test

LGOLDR

Squared terms after centering LSDRR

CLFOREXR

CLGOLDR

CLSDRR

(2a)

(3a)

(4a)

(2b)

(3b)

(4b)

−7.798 (0.654) 0.574*** (0.000) 0.270*** (0.010) 0.129*** (0.001) −0.159*** (0.001) 0.524*** (0.000) 2.017 (0.800)

−6.170*** (0.000) 0.714*** (0.000) 0.446*** (0.000) 0.105*** (0.003) −0.120** (0.031) 0.390*** (0.001)

−3.554*** (0.000) 0.246** (0.016) 0.398*** (0.000) 0.126*** (0.001) −0.102*** (0.000) 0.806*** (0.000)

−4.463*** (0.000) 0.574*** (0.000) 0.270*** (0.010) 0.129*** (0.001) −0.159*** (0.001) 0.524*** (0.000)

−5.973*** (0.000) 0.714*** (0.000) 0.446*** (0.000) 0.105*** (0.003) −0.120** (0.031) 0.390*** (0.001)

−3.548*** (0.000) 0.246** (0.016) 0.398*** (0.000) 0.126*** (0.001) −0.102*** (0.000) 0.806*** (0.000)

0.184* (0.046) 0.009 (0.418) −0.286 (0.754) −0.501*** (0.003) −0.286 (0.754) −0.042 (0.163) 0.029 (0.260) −0.042 (0.163) 0.006 (0.530)

0.950 0.934 31 Fixed 7.124 (0.849) 1.447 (0.229) (0.361) 3 −6.545***

0.949 0.934 31 Fixed 6.180 (0.627) 0.938 (0.333) (0.549) 5 −5.359**

−0.571*** (0.000) 0.961 0.943 26 Fixed 8.492 (0.903) 2.150 (0.143) (0.400) 3 −6.331***

0.950 0.934 31 Fixed 7.124 (0.849) 1.447 (0.229) (0.361) 3 −6.545***

0.949 0.934 31 Fixed 6.180 0.627 1.511 (0.219) (0.433) 5 −5.359**

0.015 (0.282) 0.006 (0.530) −0.571*** (0.000) 0.961 0.943 26 Fixed 8.492 (0.903) 2.150 (0.143) (0.400) 4 −6.331***

Notes: 1. LFOREXR and CLFOREXR at the 1% level, and LSDRR as well as MLS at the 5% level are stationary in M1, while MLF is stationary at the 10% level in M2 tested using the Narayan and Popp (2010) method. However, LGOLDR, CLGOLDR, MLG and CLSDRR are non-stationary at the conventional statistical significant level. 2. For others, please refer to the notes of Table 4a.

462

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

Table 5 The effects of international reserves on yen internationalization (robustness check I: GMM and STOCKT). Dependent variable: CSHARE = log(SHARE / (1 − SHARE))

C GSHARE FOREX STOCKT INFDIF CSHARE(−1) FOREXR GOLDR SDRR TRSHARE VOLT GFC TRDIFL R2 ADj. R2 Obs. adj. Bandwidth J-statistic P-value Q-stat.(1) P-value No. of I(1) Re. ADF test

FOREXR

GOLDR

SDRR

TRSHARE

FOREXR

GOLDR

SDRR

TRSHARE

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

– 0.499*** (0.000) 0.100*** (0.001) 0.165*** (0.000) −0.168*** (0.005) 0.573*** (0.000) −0.784*** (0.000)

−5.190*** (0.000) 0.662*** (0.001) 0.313*** (0.002) 0.112** (0.012) −0.210*** (0.000) 0.429*** (0.001)

−5.946*** (0.000) 0.731*** (0.000) 0.422*** (0.000) 0.139*** (0.006) −0.213*** (0.000) 0.409*** (0.001)

−3.906*** (0.003) 0.397** (0.043) 0.232** (0.031) 0.131*** (0.002) −0.153*** (0.008) 0.423*** (0.001)

– 0.433*** (0.000) 0.086*** (0.002) 0.110** (0.017) −0.144*** (0.008) 0.658*** (0.000) −0.660*** (0.000)

−6.296*** (0.002) 0.826*** (0.004) 0.423*** (0.005) 0.114** (0.019) −0.208*** (0.000) 0.389** (0.021)

−10.978*** (0.004) 1.572*** (0.006) 0.946** (0.013) 0.049 (0.353) −0.149*** (0.001) 0.270 (0.234)

−6.099*** (0.003) 0.818*** (0.006) 0.335** (0.040) 0.099*** (0.005) −0.119*** (0.000) 0.232** (0.032)

0.080*** (0.000)

0.057*** (0.006) −0.068 (0.301)

0.066** (0.012) 0.118 (0.210)

0.052 (0.722)

0.109 (0.286)

−0.146*** (0.000) −0.049 (0.733)

0.949 0.936 31 Fixed 5.583 (0.849) 0.989 (0.320) 4 −6.271***

0.945 0.929 31 Fixed 4.903 (0.672) 1.115 (0.291) 5 −6.382***

0.939 0.921 31 Fixed 5.216 (0.634) 1.554 (0.212) 4 −6.546***

0.953 0.938 31 Fixed 4.763 (0.689) 2.388 (0.122) 4 −7.102***

0.027 (0.808) 0.092 (0.165) 0.068 (0.140) 0.954 0.937 31 Fixed 5.710 (0.839) 2.473 (0.116) 4 −6.993***

0.022 (0.831) 0.097 (0.144) 0.036 (0.366) 0.949 0.928 31 Fixed 5.364 (0.718) 1.774 (0.183) 5 −6.735***

−0.062 (0.665) 0.593** (0.022) 0.140** (0.044) 0.907 0.867 31 Fixed 2.656 (0.915) 0.394 (0.530) 4 −5.983***

−0.119*** (0.003) −0.041 (0.560) 0.154** (0.042) 0.071*** (0.010) 0.955 0.935 30 Fixed 7.021 (0.934) 0.854 (0.356) 4 −6.124***

Notes: The same as the notes of Table 4a.

VOLT (see Columns (2) to (5) of Table 4a), the new estimates show that both the negative effects of both FOREXR and TRSHARE and the positive effects of both GOLDR and SDRR decrease slightly. This implies that, controlling for exchange rate volatility of the yen, the effects of international reserves only slightly declined, showing that their impacts are relatively stable and robust. In what follows, we introduce GFC and TRDIFL, the once-in-acentury global financial crisis dummy variable and interest rate differential, to add more information to our model. After that, we continue to test to determine if the effects of various proxies of Japan's international reserves on yen internationalization result in any new change. The estimation results are presented in Columns (5) to (8) in Table 5. Columns (5) to (8) show that, controlling for the effects of the oncein-a-century global financial crisis and interest rate differential, the signs of the major determinants of yen internationalization in the benchmark and various proxies of the composition and size of Japan's international reserves remain the same, except that SDRR has a negative, but insignificant coefficient (−0.068). Quantitatively, the estimation coefficients have a slight decline. Specifically, estimates of FOREXR, GOLDR and TRSHARE are −0.660, 0.057 and −0.119 respectively. The corresponding effects on SHARE become − 0.398, 0.054 and − 0.106, respectively. These regressions characterize that, the more independent variables, the slightly smaller estimation coefficients. However, these results are highly consistent with those in the previous regressions. To further check the robustness of the estimation results reported above, we use an alternative measure of financial market development, STOCKCAP, to replace STOCKT, to re-specify the regression benchmark. We then repeat the regressions presented in Columns (2) to (5) of Table 4a and those in Table 5. The regression results are displayed in Tables 6 and 7. According to these results, we find that the signs of all of variables in question are highly consistent with those in Tables 4a, 4b and 5. This implies that the results remain robust. In addition, because the supply of gold and SDR is relatively fixed worldwide while the US dollar assets have been increasing over time, the above regressions only examine the contemporaneous relationships between yen internationalization and gold reserves and SDR holdings. This may not capture the whole picture of the much complicated

relationship between them. To produce more robust empirical evidence, we also investigate the respective dynamic relationship between yen internationalization and gold reserves and SDR holdings by

Table 6 The effects of international reserves on yen internationalization (robustness check II: GMM and STOCKCAP). Dependent variable: CSHARE = log(SHARE / (1 − SHARE))

C GSHARE FOREX STOCKCAP INFDIF CSHARE(−1)

Benchmark

FOREXR

GOLDR

SDRR

TRSHARE

(1)

(2)

(3)

(4)

(5)

−6.676*** (0.000) 0.759*** (0.002) 0.572*** (0.000) 0.143*** (0.004) −0.162*** (0.006) 0.453*** (0.004)

–

−6.115*** (0.001) 0.776*** (0.002) 0.372*** (0.002) 0.144*** (0.008) −0.246*** (0.003) 0.342* (0.065)

−6.269*** (0.000) 0.742*** (0.001) 0.459*** (0.000) 0.158*** (0.001) −0.242*** (0.000) 0.385*** (0.005)

−5.230*** (0.001) 0.555*** (0.010) 0.319*** (0.006) 0.178*** (0.001) −0.206*** (0.005) 0.285 (0.107)

FOREXR

0.552*** (0.000) 0.125*** (0.001) 0.189*** (0.001) −0.230*** (0.005) 0.507*** (0.000) −0.890*** (0.000)

GOLDR

0.086*** (0.000)

SDRR

0.069*** (0.007)

TRSHARE R2 ADj. R2 Obs. adj. Bandwidth J-statistic P-value Q-stat.(1) P-value No. of I(1) R.S. ADF test

0.933 0.920 31 Fixed 4.975 (0.547) 0.101 (0.751) 4 −5.599***

0.934 0.921 31 Fixed 5.257 (0.811) 1.860 (0.173) 4 −6.685***

Notes: The same as the notes of Table 4a.

0.935 0.918 31 Fixed 4.769 (0.688) 1.312 (0.252) 5 −6.503***

0.929 0.912 31 Fixed 5.671 (0.579) 2.539 (0.111) 4 −6.971***

−0.160*** (0.001) 0.939 0.924 31 Fixed 4.236 (0.752) 1.759 (0.185) 4 −6.883***

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

463

Table 7 The effects of international reserves on yen internationalization (robustness check III: GMM and STOCKCAP). Dependent variable: CSHARE = log(SHARE / (1 − SHARE))

C GSHARE FOREX STOCKCAP INFDIF CSHARE(−1) FOREXR GOLDR SDRR TRSHARE VOLT GFC TRDIFL R2 ADj. R2 Obs. adj. Bandwidth J-statistic P-value Q-stat.(1) P-value No. of I(1) R.S. ADF test

FOREXR

GOLDR

SDRR

TRSHARE

FOREXR

GOLDR

SDRR

TRSHARE

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

– 0.464*** (0.000) 0.127*** (0.001) 0.172*** (0.000) −0.182*** (0.006) 0.591*** (0.000) −0.785*** (0.000)

−5.138*** (0.001) 0.632*** (0.006) 0.316*** (0.007) 0.125*** (0.008) −0.210*** (0.002) 0.446*** (0.004)

−5.953*** (0.000) 0.700*** (0.001) 0.452*** (0.000) 0.144*** (0.001) −0.222*** (0.000) 0.433*** (0.001)

−4.040** (0.012) 0.396* (0.091) 0.246** (0.044) 0.144*** (0.004) −0.161** (0.015) 0.410** (0.014)

– 0.354*** (0.008) 0.059 (0.116) 0.063 (0.238) −0.116** (0.032) 0.714*** (0.000) −0.541*** (0.001)

−6.537*** (0.004) 0.841*** (0.009) 0.450*** (0.006) 0.119** (0.032) −0.204*** (0.000) 0.388** (0.047)

−9.094** (0.012) 1.221** (0.018) 0.805** (0.021) 0.063 (0.358) −0.133*** (0.001) 0.421* (0.083)

−4.734** (0.011) 0.647** (0.016) 0.200 (0.180) 0.052* (0.053) −0.107*** (0.002) 0.278*** (0.008)

0.084*** (0.000)

0.050*** (0.008) −0.055 (0.372)

0.061*** (0.010) 0.065 (0.503)

0.030 (0.835)

0.086 (0.398)

−0.153*** (0.000) −0.081 (0.585)

0.940 0.925 31 Fixed 5.210 (0.877) 1.620 (0.203) 5 −6.603***

0.940 0.921 31 Fixed 4.838 (0.680) 1.546 (0.214) 6 −6.612***

0.932 0.911 31 Fixed 5.075 (0.651) 1.867 (0.172) 5 −6.712***

0.945 0.928 31 Fixed 4.886 (0.674) 3.050 (0.081) 5 −7.413***

−0.093 (0.425) 0.143** (0.039) 0.137** (0.020) 0.944 0.924 31 Fixed 6.101 (0.807) 2.482 (0.115) 5 −7.054***

−0.037 (0.729) 0.122* (0.080) 0.063* (0.077) 0.944 0.920 31 Fixed 5.532 (0.700) 2.447 (0.118) 6 −7.077***

−0.058 (0.700) 0.510** (0.016) 0.165** (0.030) 0.915 0.878 31 Fixed 3.945 (0.786) 0.751 (0.386) 5 −6.191***

−0.140*** (0.000) −0.171*** (0.001) 0.131** (0.050) 0.111*** (0.000) 0.948 0.924 30 Fixed 6.718 (0.945) 1.085 (0.297) 5 −6.243***

Notes: The same as the notes of Table 4a.

introducing one and two-period lags of GOLDR and SDRR, respectively. The estimation results are reported in Tables 8 and 9. We find that, even considering the complicated effects that the relatively fixed gold reserves and SDR holdings might have on yen internationalization, the dynamic effects of gold reserves and SDR holdings on yen internationalization are significantly positive. The estimating results are highly consistent with the contemporaneous results in previous regressions. In sum, after controlling for the effects of exchange rate volatility of the yen, the once-in-a-century global financial crisis and interest rate differential, the negative and positive effects of various proxies of the Japanese international reserves' composition and size remain the same as those in the previous regressions without them. Even if we use an alternative measure of financial market development, the estimation results have no substantial changes. In addition, when we carefully investigate the dynamic relationships between yen internationalization and gold reserves and SDR holdings, our conclusions still hold. This shows that the estimating results in this study are robust and reliable, and can offer powerful explanations of the lower degree of yen internationalization.

6. Conclusions and policy implications In this study, we analyze the relationship between Japan's international reserves and the internationalization of the yen, finding that, overall, Japan's international reserves significantly negatively influence yen internationalization. Closer scrutiny of the composition of reserves shows that Japan's high share of foreign exchange reserves has a large negative effect on yen internationalization, while both gold reserves and SDR holdings have a positive effect despite their relatively small shares and after controlling for the yen's exchange rate volatility, the global financial crisis of 2007–09 and interest rate differentials. These empirical results are also robust to an alternative measure of financial market development and the dynamic effects of gold reserves and SDR holdings on the yen's internationalization.

There are two key policy implications of this study. The first is that if greater internationalization of the yen remains a long term policy goal, then persistent foreign exchange market intervention to prevent yen appreciation is inadvisable since this leads to ever greater foreign exchange reserves accumulation, predominantly in US dollars, as the Bank of Japan has to purchase US dollars with yen to reduce upward pressure on the dollar–yen exchange rate. This is at odds with achieving higher yen internationalization on theoretical and empirical grounds. Second, to further internationalize the yen, the Bank of Japan needs to reweight the components of its international reserves by raising gold reserves and SDR holdings relative to its foreign exchange reserves. This is not to suggest that the Bank of Japan should begin purchasing significant quantities of gold at current market prices as soon as possible, but when market conditions are propitious, for example when the US dollar is very strong, given gold prices tend to trough at such times. Apart from improving our understanding as to why the yen's internationalization is so low, the study provides a point of reference for China which has actively promoted renminbi internationalization though has a similar international reserves composition to Japan's. Insufficient data limits testing our hypotheses for China however as renminbi internationalization only began in 2009. Future research could further examine this important issue using recent cross-sectional data for all major currencies.

Acknowledgments This study was funded by the Humanities and Social Sciences Research Foundation of the Chinese Ministry of Education under grant 14YJA790087 and the Fundamental Research Funds for the Central Universities under grant 1109043-13200-1137103. This is a revised version of a paper entitled “International reserves and local currency internationalization: An empirical study on the Japanese yen” circulating for comments, which was presented at Harvard Fairbank Center for Chinese Studies and the 87th Annual Conference of Western Economic Association International. The authors thank Richard Cooper, Koichi Hamada, Taiji Furusawa, Menzie Chinn, Rawi Abdelal, Stijn Claessens,

464

Table 8 Dynamic effects of gold reserves on yen internationalization (regression results: GMM). Dependent variable: CSHARE = log(SHARE / (1 − SHARE)) STOCKT

C GSHARE FOREX

GOLDR(−1)

GOLDR(−2)

GOLDR(−1)

GOLDR(−2)

GOLDR(−1)

GOLDR(−2)

GOLDR(−1)

GOLDR(−2)

GOLDR(−1)

GOLDR(−2)

GOLDR(−1)

GOLDR(−2)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

−5.786*** (0.000) 0.760*** (0.000) 0.302*** (0.010) 0.124*** (0.008)

−5.737*** (0.000) 0.665*** (0.003) 0.363*** (0.001) 0.156*** (0.000)

−5.595*** (0.000) 0.730*** (0.000) 0.304*** (0.009) 0.116** (0.012)

−5.047*** (0.000) 0.543*** (0.009) 0.312*** (0.005) 0.162*** (0.000)

−6.956*** (0.001) 0.924*** (0.002) 0.447*** (0.002) 0.126** (0.032)

−11.322*** (0.000) 1.470*** (0.000) 0.954*** (0.000) 0.223*** (0.005)

−5.436*** (0.000) 0.689*** (0.001) 0.295** (0.011)

−6.399*** (0.000) 0.685** (0.011) 0.471*** (0.000)

−5.314*** (0.000) 0.669*** (0.002) 0.302** (0.011)

−5.323*** (0.001) 0.531** (0.029) 0.391*** (0.001)

−6.493*** (0.004) 0.842** (0.011) 0.427*** (0.010)

−12.174*** (0.001) 1.603*** (0.002) 0.984*** (0.000)

0.125*** (0.004) −0.172*** (0.006) 0.375*** (0.006) 0.085*** (0.000)

0.198*** (0.002) −0.179*** (0.002) 0.413** (0.042)

0.116*** (0.005) −0.173*** (0.005) 0.396*** (0.004) 0.081*** (0.001)

0.176*** (0.002) −0.148*** (0.008) 0.504*** (0.008)

0.111* (0.068) −0.185*** (0.000) 0.369* (0.064) 0.049* (0.057)

0.230** (0.016) −0.132*** (0.010) 0.117 (0.661)

STOCKCAP INFDIF CSHARE(−1) GOLDR(−1)

−0.174*** (0.002) 0.324*** (0.008) 0.0803*** (0.002)

GOLDR(−2)

−0.137*** (0.003) 0.410*** (0.010)

−0.174*** (0.001) 0.354*** (0.004) 0.081*** (0.004)

0.059** (0.023)

VOLT

−0.138*** (0.003) 0.474*** (0.003)

0.017 (0.894)

0.071** (0.018) 0.165* (0.065)

0.945 0.928 31 Fixed 5.633 0.583 0.781 (0.377) (0.496) 5 −6.156***

0.942 0.924 30 Fixed 7.165 (0.519) 0.005 (0.942) 0.451 4 −5.520***

GFC TRDIFL R2 ADj. R2 Obs. adj. Bandwidth J-statistic P-value Q-stat.(1) P-value Nor. test (P) No. of I(1) Re. ADF test

0.944 0.930 31 Fixed 5.623 0.584 0.700 (0.403) (0.446) 4 −6.101***

0.942 0.927 30 Fixed 7.131 0.523 0.036 (0.850) (0.398) 3 −5.609***

−0.181*** (0.000) 0.317* (0.077) 0.054** (0.040)

0.013 (0.908) 0.115* (0.075) 0.038 (0.420) 0.949 0.928 31 Fixed 5.807 (0.669) 1.496 (0.221) 0.723 5 −6.575***

−0.114*** (0.009) 0.226 (0.199)

−0.045 (0.246) 0.087 (0.388) 0.304*** (0.001) 0.001 (0.990) 0.945 0.921 30 Fixed 6.988 (0.538) 0.974 (0.324) 0.994 4 −6.365***

0.042** (0.049)

0.939 0.924 31 Fixed 5.491 0.600 1.256 (0.262) (0.483) 5 −6.423***

0.928 0.909 30 Fixed 6.796 0.450 0.146 (0.703) (0.492) 4 −5.890***

−0.005 (0.963)

0.050** (0.043) 0.040 (0.702)

0.940 0.922 31 Fixed 5.546 0.594 1.378 (0.240) (0.524) 6 −6.486***

0.933 0.911 30 Fixed 7.041 0.425 0.214 (0.644) (0.495) 5 −5.954***

−0.049 (0.685) 0.128* (0.078) 0.079* (0.070) 0.944 0.920 31 Fixed 6.038 0.643 2.270 0.132 0.847 6 −6.954***

−0.040 (0.241) −0.009 (0.936) 0.326*** (0.002) 0.026 (0.611) 0.934 0.904 30 Fixed 6.754 0.563 1.399 0.237 0.945 5 −6.575***

Notes: 1. All variables take logs, and INFDIF plus 1 before taking log; 2. P-values are reported in the parentheses; 3. *, ** and *** represent the statistically significant levels of 10%, 5% and 1%, respectively; 4. Instrument variables include one and twoperiod lags of dependent variable and all independent variables. Of them, instruments for Columns (2) and (4) also include GOLDR(−2), while for Columns (5), (6), (11) and (12) include GFC itself; the null hypothesis for the Hansen's J-test is that the instrument is valid; 5. When there is a lag dependent variable in the right hand of the equation, the D–W statistic is invalid, so we report one lag of the Q-statistics and its P-value, respectively, to test if the residual series are serially correlated. The null hypothesis is that there is no serial correlation. 6. No. of I(1) denotes the number of time series with I(1) in the regression. 7. Re. ADF Test refers to t-statistic of the ADF unit root test on residual series. For specific explanation, please refer to Footnote #12 of this paper.

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

STOCKT

STOCKCAP

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

465

Table 9 Dynamic effects of SDR holdings on yen internationalization (regression results: GMM). Dependent variable: CSHARE = log(SHARE / (1 − SHARE)) STOCKT SDRR(−1)

C GSHARE FOREX STOCKT STOCKCAP INFDIF CSHARE(−1) SDRR(−1)

STOCKCAP SDRR(−2)

SDRR(−1)

SDRR(−2)

SDRR(−2)

SDRR(−1)

SDRR(−2)

SDRR(−1)

SDRR(−2)

SDRR(−1)

SDRR(−2)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

−4.989*** (0.000) 0.646*** (0.001) 0.233* (0.076) 0.120*** (0.001) –

−5.716*** (0.002) 0.623** (0.032) 0.409** (0.017) 0.170*** (0.001) –

−4.672*** (0.000) 0.595*** (0.000) 0.208* (0.073) 0.122*** (0.001) –

−5.645*** (0.003) 0.595** (0.046) 0.417*** (0.010) 0.173*** (0.001) –

−6.812*** (0.000) 0.891*** (0.000) 0.446*** (0.001) 0.144** (0.019)

−10.304*** (0.000) 1.328*** (0.000) 0.867*** (0.000) 0.185*** (0.009) –

−5.024*** (0.003) 0.626*** (0.006) 0.236 (0.103) –

−5.356*** (0.004) 0.573* (0.051) 0.376** (0.021) –

−4.678*** (0.001) 0.577*** (0.003) 0.212 (0.108) –

−5.420*** (0.010) 0.571* (0.089) 0.405** (0.015) –

−7.213*** (0.000) 0.986*** (0.000) 0.464*** (0.004) –

−10.115*** (0.000) 1.315*** (0.001) 0.841*** (0.000) –

−0.220*** (0.000) 0.322*** (0.003) 0.117*** (0.001)

−0.145*** (0.001) 0.446** (0.039)

−0.227*** (0.000) 0.344*** (0.000) 0.126*** (0.000)

−0.141*** (0.002) 0.467** (0.037)

−0.220*** (0.000) 0.308** (0.036) 0.078** (0.040)

0.137*** (0.003) −0.239*** (0.000) 0.315** (0.021) 0.126*** (0.000)

0.158*** (0.003) −0.151*** (0.010) 0.453** (0.044)

0.132*** (0.003) −0.242*** (0.000) 0.342*** (0.002) 0.131*** (0.000)

0.154*** (0.002) −0.130** (0.019) 0.481* (0.060)

0.055 (0.189) −0.051* (0.096) 0.213*** (0.009) 0.128*** (0.000)

0.154** (0.029) −0.127*** (0.003) 0.277 (0.220)

SDRR(−2)

0.047 (0.430)

VOLT

0.131 (0.123)

0.045 (0.433) 0.129 (0.200)

0.946 0.929 31 Fixed 6.434 (0.599) 0.722 (0.396) 4 −4.462*

0.937 0.917 30 Fixed 7.420 (0.492) 0.043 (0.836) 4 −5.168**

GFC TRDIFL R2 ADj. R2 Obs. adj. Bandwidth J-statistic P-value Q-stat.(1) P-value No. of I(1) Re. ADF Test

SDRR(−1)

0.945 0.931 31 Fixed 6.542 (0.587) 1.130 (0.288) 3 −6.374***

0.937 0.921 30 Fixed 7.012 (0.535) 0.002 (0.964) 3 −5.436***

0.170** (0.043) 0.147** (0.026) 0.017 (0.749) 0.950 0.929 31 Fixed 6.300 0.710 1.466 (0.226) 4 −4.692**

−0.120*** (0.002) 0.290* (0.099)

−0.041 (0.187) 0.112 (0.268) 0.275*** (0.000) 0.029 (0.545) 0.950 0.927 30 Fixed 7.106 (0.626) 0.980 (0.322) 4 −6.372***

0.059 (0.310)

0.938 0.923 31 Fixed 6.537 (0.587) 1.614 (0.204) 4 −6.618***

0.931 0.913 30 Fixed 7.243 (0.511) 0.201 (0.654) 4 −5.826***

0.119 (0.174)

0.044 (0.427) 0.033 (0.768)

0.939 0.920 31 Fixed 6.329 (0.610) 1.125 (0.289) 5 −4.639*

0.932 0.911 30 Fixed 7.498 (0.484) 0.054 (0.816) 5 −5.579**

0.117 (0.172) 0.221*** (0.000) 0.033 (0.377) 0.960 0.935 27 Fixed 9.155 (0.935) 2.065 (0.151) 5 −5.725**

−0.041 (0.248) −0.005 (0.964) 0.284*** (0.000) 0.068 (0.145) 0.944 0.919 30 Fixed 7.202 (0.616) 1.536 (0.215) 4 −6.610***

Notes: 1. All variables take logs, and INFDIF plus 1 before taking log; 2. P-values are reported in the parentheses; 3. *, ** and *** represent the statistically significant levels of 10%, 5% and 1%, respectively; 4. Instrument variables include one and two-period lags of dependent variable and all independent variables. Of them, instruments for SDRR(−1) are SDRR(−1), SDRR(−2) and SDRR(−3), instruments for SDRR(−2) are SDRR(−2), SDRR(−3) and SDRR(−4), while instruments for Columns (5), (6), (11) and (12) also include GFC itself; the null hypothesis for the Hansen's J-test is that the instrument is valid; as well, Column (11) includes AR(6) to reduce residual serial correlation. 5. When there is a lag dependent variable in the right hand of the equation, the D–W statistic is invalid, so we report one lag of the Q-statistics and its P-value, respectively, to test if the residual series are serially correlated. The null hypothesis is that there is no serial correlation. 6. No. of I(1) denotes the number of time series with I(1) in the regression. 7. Re. ADF Test refers to t-statistic of the ADF unit root test on residual series. For specific explanation, please refer to Footnote #12 of this paper.

Gian Luca Clementi, Takeshi Hamashita, Daniel Walker, Guohua Peng, Jiang Cheng, Zhenxi Chen, Linan Huang and Tingting Qu for their helpful comments on earlier versions of this paper, and Chikahisa Sumi and Yongzheng Yang for their helpful discussions, as well as Regina M. Abrami and Lydia Chen for their kind help. We also thank Paresh Kumar Narayan for sharing Gauss code on unit root test, and Li Zhao for advising the use of Gauss software. Of course, the authors are solely responsible for all errors and misinterpretations in the paper. References Amyx, J., 2004. Japan's Financial Crisis: Institutional Rigidity and Reluctant Change. Princeton University Press. Bai, Q.X., Zhang, Z.W., 2011. The size of foreign exchange reserve and local currency internationalization: an empirical study on the Japanese yen. Econ. Res. J. 10, 137–149 (In Chinese). Bond, S., 2002. Dynamic panel data models: a guide to micro data methods and practice. Port. Econ. J. 1, 141–162. Bonser-Neal, C., Tanner, G., 1996. Central bank intervention and the volatility of foreign exchange rates: evidence from the options market. J. Int. Money Financ. 15 (6), 853–878. Chen, H., 2004. The road toward the internationalization of the Japanese yen. World Econ. Polit. 5, 65–70 (In Chinese). Chinn, M., Frankel, J., 2005. Will the euro eventually surpass the dollar as leading international reserve currency? NBER Working Paper No. 11510 Chinn, M., Frankel, J., 2008. The euro may over the next 15 years surpass the dollar as leading international currency. NBER Working Paper No. 13909.

Clark, P., Tamirisa, N., Wei, S.J., Sadikov, A., Zeng, L., 2004. Exchange rate volatility and trade flows—some new evidence. IMF Occasional Paper No. 235. Cohen, B.J., 1971. The Future of Sterling as an International Currency. Macmillan, London. Cohen, B., 2005. Currency choice in international bond issuance. BIS Q. Rev. 2, 53–66. Davidson, R., MacKinnon, J.G., 1999. Econometric Theory and Methods. Oxford University Press, New York, pp. 91–93. Dominguez, K.M., 1993. Does central bank intervention increase the volatility of foreign exchange rates? NBER Working Paper No. 4532 Dominguez, K.M., 1998. Central bank intervention and exchange rate volatility. J. Int. Money Financ. 17 (1), 161–190. Dooley, M.P., Lizondo, J.S., Mathieson, D.J., 1989. The currency composition of foreign exchange reserves. IMF Staff Papers Vol. 36, No. 2 (June). Eichengreen, B., Mathieson, D.J., 2000. The currency composition of foreign exchange reserves: retrospect and prospect. IMF Working Paper WP/00/131. Engle, R.F., Granger, C.W.J., 1987. Co-integration and error correction: representation, estimation, and testing. Econometrica 55 (2), 251–276. Fair, R.C., 1970. The estimation of simultaneous equation models with lagged endogenous variables and first order serially correlated errors. Econometrica 38 (3), 507–516. Frankel, J.A., 1984. The yen/dollar agreement: liberalizing Japanese capital markets. Policy Analysis in International, Economics No. 9. Frankel, J.A., 2011. Historical precedents for internationalization of the RMB. CGS/IIGG Working Paper (October). Frankel, J.A., 2012. Internationalization of the RMB and historical precedents. J. Econ. Integration 27 (3), 329–365 (September). Fukuda, S.I., Ono, M., 2006. On the determinants of exporters' currency pricing: history vs. expectations. J. Jpn. Int. Econ. 20, 548–568. Granger, C., Newbold, P., 1974. Spurious regressions in econometrics. J. Econ. 2 (2), 111–120. Greenspan, A., 2014. “Golden rule: why Beijing is buying” foreign affairs. Snapshot (September 29).

466

Z. Zhang et al. / Economic Modelling 52 (2016) 452–466

Hansen, L.P., 1982. Large sample properties of generalized method of moments estimators. Econometrica 50 (4), 1029–1054. Hartman, P., Issing, O., 2002. The international role of the euro. Journal of Policy Modeling 315–345. Huang, Y.P., Wang, D.l., Fan, G., 2014. Paths to a reserve currency: internationalization of the Renminbi and its implications. ADBI Working Paper Series No. 482 (May). Ito, T., Koibuchi, S., Sato, K., Shimizu, J., April 2012. The choice of an invoicing currency by globally operating firms: a firm-level analysis of Japanese exporters. Int. J. Finance Econ. 17 (4), 305–320. Kawai, M., Takagi, S.J., 2011. The RMB as a key international currency: lessons from the Japanese experience. Notes Prepared for the Asia–Europe Economic Forum (January 10–11, Paris). Koo, R., 2009. The Holy Grail of Macroeconomics—Lessons From Japan's Great Recession. Wiley & Sons, New York. Krugman, P.A., 1984. The International Role of the Dollar: Theory and Prospect. In: Bilson, John F.O., Marston, Richard C. (Eds.), University of Chicago Press, Chicago 261-27. Lago, I.M.y., Duttagupta, R., Goyal, R., 2009. The debate on the international monetary system. The IMF Staff Position Note, SPN/09/26 (November). Liviatan, N., 1963. Consistent estimation of distributed lags. Int. Econ. Rev. 4 (1), 44–52. MacKinnon, J.G., 2010. Critical values for cointegration tests. Queen's Economics Department Working Paper No. 1227. Makin, A.J., 2007. Does China's huge external surplus imply an undervalued Yuan? China World Econ. 15 (3), 89–102 (2007). Narayan, P.K., Popp, S., 2010. A new unit root test with two structural breaks in level and slope at unknown time. J. Appl. Stat. 37 (9), 1425–1438.

Newey, W., West, K., 1987. A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55 (3), 703–708. Prasad, 2014. The Dollar Trap: How the US Dollar Tightened Its Grip on Global Finance. Princeton University Press, New Jersey. Sato, K., 1999. The international use of the Japanese yen: the case of Japan's trade with East Asia. World Econ. 4 (06), 547–584. Sato, K., 2003. Currency invoicing in Japanese exports to East Asia: implication of the yen internationalization. Asian Econ. J. 17 (2), 129–154. Subramanian, A., 2011. Renminbi rules: the conditional imminence of the reserve currency transition. Peterson Institute for International Economics Working Paper Series WP 11-14 (September). Taguchi, H., 1994. On the internationalization of the Japanese yen. In: Ito, Takatoshi, Anne, Krueger, Volume, N.B.E.R.-E.A.S.E., 3, (Eds.), Macroeconomic Linkage: Savings, Exchange Rates and Capital Flows. University of Chicago Press, Chicago. Takagi, S., 2009. Internationalizing the yen, 1984–2003: unfinished agenda or mission impossible? BIS Research Papers Tavlas, S.G., Ozeki, Y., 1992. The internationalization of currencies: an appraisal of the Japanese yen. IMF Occasional Paper. Xu, M.Q., 2005. Considerations on RMB internationalization and regionalization based on the yen's success and failure. World Econ. Study 12, 39–44 (In Chinese). Zhang, X.D., 2009. Applied Quantitative Economics. China Machine Press, Beijing, pp. 360–367 (in Chinese). Zhang, Z.W., 2015. The size of foreign exchange reserves and local currency internationalization: cross-country evidence. Singap. Econ. Rev. 60 (4).