Market Segmentation, Fundamentals or Contagion? Assessing Competing Explanations for CNH-CNY Pricing Differentials

Michael Funke, Chang Shu, Xiaoqiang Cheng and Sercan Eraslan

WP442 A parsimonious approach to incorporating economic information in measures of potential output iii

Market Segmentation, Fundamentals or Contagion? Assessing Competing Explanations for CNH-CNY Pricing Differentials 1

Michael Funke, Chang Shu, Xiaoqiang Cheng and Sercan Eraslan †

Abstract

Renminbi internationalisation has added complexity in understanding the currency movement as the offshore exchange rate diverges from the onshore rate. Using extended GARCH models, this paper attributes the onshore-offshore exchange rate differential to underlying driving forces, including market segmentation, fundamentals and global factors. Policy dummies for changes in constraints on currency trading and cross border renminbi flows are used to measure market segmentation. These are found to have important influence on both the level and volatility of the differential. Fundamentals and global factors matter too, although the impact of global factors fades away when considered jointly with policy variables.

Keywords: Renminbi Exchange Rates, China, onshore and offshore markets, GARCH Models

JEL Classification: F31, C22, C52

1 The views expressed in this paper are entirely those of the authors. They do not necessarily

represent the views of the Bank for International Settlements and/or the Hong Kong Monetary Authority.

† SHU:, Bank for International Settlements, chang.shu@bis.org; FUNKE, Hamburg University, michael.funke@uni-hamburg.de; CHENG: Hong Kong Monetary Authority, axcheng@hkma.gov.hk; ERASLAN, Hamburg University, sercan.eraslan@wiso.uni-hamburg.de

WP442 A parsimonious approach to incorporating economic information in measures of potential output v

Table of Contents

Introduction ............................................................................................................................................... 7

Institutional Background ...................................................................................................................... 8

Renminbi internationalisation and the rise of offshore renminbi market ............... 8

Onshore and offshore rennminbi foreign exchange markets: a brief comparison .............................................................................................................................. 9

Onshore and offshore renminbi differential: a snapshot ............................................. 11

Measuring market segmentation, fundamentals and contagion ....................................... 12

Measuring market segmentation using policy variables .............................................. 12

Macroeconomic and market fundamentals ....................................................................... 15

Global market contagions ......................................................................................................... 16

Modelling the CNH-CNY differential: Design ............................................................................ 16

Empirical results ..................................................................................................................................... 18

Basic GARCH(1,1) model............................................................................................................ 18

Extended GARCH(1,1) models ................................................................................................. 20

Concluding remarks and policy discussions ............................................................................... 25

References ................................................................................................................................................ 26

WP 7

Introduction

With the growing Chinese economy and its external linkages, the renminbi’s influence has been rising and understanding renminbi movements has become increasingly important. In the latest Triennial Central Bank Survey on the foreign exchange market, the Bank for International Settlements (BIS) found that the renminbi now ranks the ninth most traded currency in the world, and the most traded in Asia. Shu, Chow and Chan (2007) first note that the renminbi’s impact in Asia became discernible after the exchange rate reform in July 2005 when China abandoned the fixed exchange rate and moved to a managed float regime. Studies using more recent periods show that such regional impact has grown stronger (Fratzscher and Mehl, 2011, Henning, 2012 and Subramanian and Kessler, 2012). There is also evidence that influence of the renminbi has gone beyond Asia (Fratzscher and Mehl, 2011, Balasubramaniam, Patnail and Shab, 2011, and Subramanian and Kessler, 2012). To the extent renminbi movements could affect other economies’ growth and potentially require policy adjustment, understanding and modelling the renminbi exchange rate is of great importance to many central banks. It is also of great interest for market participants as changed economic prospects in China and related economies will likely entail adjustment in their investment strategies.

Yet understanding renminbi movements has been made more important and more challenging at the same time by the process of renminbi internationalisation as an offshore renminbi foreign exchange market has developed. The Chinese authorities have promoted the use of the renminbi outside mainland China since 2009, quickly removing restrictions on renminbi use under current account and gradually widening renminbi use under capital account. The market has responded enthusiastically to these policies, and a wide range of offshore renminbi products in the fixed income, equities and foreign exchanges have been grown rapidly. Among these, the renminbi foreign exchange market, dubbed the CNH market, has recorded particularly strong growth, and is rapidly gaining in on the offshore non-deliverable forwards (NDF) market which has had a much longer history. As the only foreign exchange market with officially sanctioned deliverable forwards offshore, a second set of renminbi spot and forward rates have emerged (McCauley, Shu and Ma, 2014).

Unlike the NDF market, the CNH market represents essentially the same financial product as its onshore counterpart.2 Yet unlike major currencies such as the US dollar, the euro and the Japanese yen which have no material difference between onshore and offshore rates, deviations exist between the onshore and offshore renminbi rates and can be substantial from time to time. The deviations may occur for a number of reasons. Compared to the onshore market (known as the CNY market), there is no presence of the central bank in the offshore market or trading restrictions. As such, the two markets may respond differently to the same set of economic and financial fundamentals. The offshore market is also more closely linked with the global financial markets, and thus more vulnerable to changes in global liquidity conditions and risk appetite of international investors. And the limited integration of two markets means that the deviations cannot be arbitraged away. Questions arise as to which factors, market segmentation fundamentals or global factors, are important in driving the onshore-offshore price differential. These questions are of significant interest to policymakers in China that seek to promote the international use of the renminbi, but do not wish to see increases in exchange rate volatility at this initial stage of the process. They are relevant for other central banks and market participants who wish to understand the significance of different signals from the two renminbi rates (Shu, He and Cheng, 2014).

This study aims to examine drivers of the onshore-onshore pricing differential in the renminbi foreign exchange market, an issue that has been little studied previously. There have been some studies investigating the pricing differential of China’s onshore and offshore equities, eg. Peng, Miao

2 An overview of the offshore renminbi market development can be found in Minikin and Lau (2013).

8 WP

and Chow (2007). Studies on China’s onshore and offshore foreign exchange markets tend to focus on causality between the two, eg. Cheung and Rime (2013), Wu and Fei (2012) and Maziad and Kang (2012). One study that looks at the factors causing onshore and offshore markets compares the CNY forward rates with NDF rates (Li, Hui and Chung, 2012). The study closest to ours is undertaken by Craig, Hua, Ng and Yuen (2013) which suggests that capital controls and shifts in global market sentiment explain much of the divergence in onshore and offshore spot rates.3

Our study differs from these earlier analyses in both data selection and econometric methodology, and will enrich understanding on the pricing differential on several important fronts. A large sample afforded by higher frequency data (daily) enables us to explore a much wider set of potential factors of influence, unlike the VAR model in Craig, Hua, Ng and Yuen (2013) being constrained by the number of monthly observations. Also critically, we compile a series of policy variables that capture the change in the degree of market segmentation. These variables are included in extended GARCH models with a view of assessing how policy changes might affect the level and volatility of the price differential, ie. the first and second moments of the differential.

The remainder of the paper is organised as follows. Section 2 lays out the institutional background which gives rise to the segmentation of the onshore and offshore markets. Section 3 discusses the factors that can drive the deviations between the two markets. In particular, based on policy measures that aim at relaxing restrictions on foreign exchange trading and on cross border renminbi flows, we will compile a set of variables that help gauge the reduction on market segmentation. GARCH and extended GARCH models employed for assessing the determinants of the CNY-CNH differential will be outlined in Section 4, and empirical results reported in Section 5. Section 6 considers policy implications from our work.

Institutional Background

Renminbi internationalisation and the rise of offshore renminbi market

The past decade has witnessed an evolution of Mainland China’s exchange rate regime from fixed to managed-floating rates with greater flexibility. In 1994, the multiple exchange rates for the renminbi were unified, and the currency became fully convertible under current account in 1996. A more significant move came when the People’s Bank of China (PBoC) announced in July 2005 that China would implement a managed floating exchange rate system based on market supply and demand and in reference to a basket of currencies, instead of pegging to the US dollar. The daily trading band against the US dollar has been progressively widened, from ±0.3% to ±0.5% in May 2007 and then to ±1% in April 2010. In March 2014, this trading band was once again widened to ±2%. As the clearest indication yet of the future direction, Governor Zhou of the PBoC pledged in November 2013 that the central bank will [over time] largely withdraw from regular intervention in the market.

China has also embarked on promoting RMB internationalisation in recent years. Chinese authorities have facilitated use of the renminbi outside mainland China since July 2009, moving quickly to remove restrictions on the use of the renminbi in current account transactions. Cross border renminbi settlement, now a major source of offshore renminbi liquidity, was introduced for trade on a trial basis in July 2009, and broadened over the course of the next three years to cover all current account transactions in China. Conscious of the risks associated with currency internationalisation, the opening for the use of the

3 See also IMF (2013, Box 2, p. 16) for an attempt at synthesising the debate on onshore and offshore RMB markets.

WP 9

renminbi in capital account transactions has taken a more gradual pace. Sources of offshore funds have been broadened, eg. overseas direct investment in the renminbi by Chinese enterprises from 2011 and more recent rules to make offshore lending easier for by mainland banks. The PBoC has also set up bilateral local currency swap facilities with overseas central banks and monetary authorities, with a view to supporting the international use of the renminbi and to provide a contingent source of liquidity. By 2013, 20 agreements for such facilities have been signed, with the total amount reaching RMB 1.6 trillion.

In the meantime, offshore renminbi is permitted to be used onshore in a wider variety of ways and on a growing scale, such as in the onshore interbank bond market by offshore financial institutions, foreign direct investment by foreign firms, and the renminbi Qualified Foreign Institutional Investor (R-QFII) scheme (under which Hong Kong-based brokerage firms could offer renminbi investment products to non-Chinese residents that are invested in onshore bond and stock markets).

Apart from opening channels of cross border renminbi flows, measures on the technical front have facilitated expansion of renminbi liquidity and market development. In Hong Kong, the spot fixing for the offshore renminbi exchange rate was launched in June 2011 by the Treasury Markets Association (TMA) – an industry association supported by the Hong Kong Monetary Authority (HKMA), and the interbank interest rate fixing for the renminbi (CNH HIBOR), the first offshore renminbi interest rate benchmark, was introduced in June 2013 and expected to facilitate the development of renminbi products such as syndicated loans and cross-currency swaps.

Market forces have responded quickly to official policies, and vibrant offshore renminbi markets are taking shape. For instance, renminbi deposits and CDs in Hong Kong had reached more than RMB 1 trillion by November 2013, rising from RMB 322 billion at end-2010. According to BIS international securities data, renminbi offshore bonds outstanding at end-2013 amounted to around RMB 490 billion, rising from around RMB 100 billion at end-2010. Since July 2010, Hong Kong has produced a second set of spot and forward exchange rates for the renminbi – the CNH rates – for delivery of the renminbi against the US dollar outside Mainland China. A second set of renminbi yield curves has also been formed with the active bond issuance by the Chinese Ministry of Finance and firms from both inside and outside China. The renminbi offshore centres have now also spread to other regions such as London, Singapore and Taiwan.

Onshore and offshore rennminbi foreign exchange markets: a brief comparison

Among the different segments of offshore renminbi markets, the foreign exchange market, referred to as the CNH market, has grown particularly rapidly. It has a number of distinctive features in terms of the participant base and regulation, as compared to the onshore foreign exchange market, ie. the CNY market.

The CNY market, with its comparatively long history and deep liquidity, remains highly regulated. Established in 1994, the CNY market developed forward and derivatives trading from 1997, and has seen rapid growth in recent years. According to the latest BIS Triennial Survey, its average daily turnover surged from USD 0.6 billion in 2004 to USD 20 billion in 2013 (Table 1). Among different types of trading, spot contracts account for the majority of transactions, while forwards and derivatives have a smaller share. The access to the wholesale market is restricted to domestic entities including banks, finance companies (subsidiaries of

10 WP

large SOEs), and subsidiaries of foreign banks. Importantly, foreign exchange transactions between banks and their customers are required to be backed by underlying real demand (i.e. current account transactions), and pure speculative trades are prohibited. The daily fixing, announced each morning at 9.30 am by China Foreign Exchange Trade System (CFEFS) in Shanghai, is based on quotations from a panel of banks. The PBoC has a presence in the market in order to maintain exchange rate stability, notwithstanding increased flexibility in the rate in recent years. The continuing constraints on exchange rate movements may have dampened the development of the onshore market. Despite the phenomenal growth, the scale of the CNY market is small in relation to China’s economic links with the rest of the world. By comparison, for example, Japan’s external trade is half the size of China’s, yet the onshore trading of the Japanese yen is over eight times of onshore renminbi trading.

By contrast, the CNH market, short in history and growing exponentially, is a free market and has a more diversified range of products. The exchange rate is determined freely by market forces; neither the PBoC nor the HKMA intervenes in the CNH market. Spot trading in the CNH market became active in August 2010 following the launch of cross-border renminbi trade settlement, and deliverable forwards and derivatives have been developed subsequently. As shown by the BIS Triennial Survey, its daily turnover has grown to USD 7.3 billion in less than three years since its inception in late 2010 (Table 1).4 This size of turnover is remarkable in view of the much smaller renminbi liquidity pool offshore than onshore. A spot fixing for the CNH rate was launched in June 2011 by the Treasury Markets Association (TMA), providing a reference rate for derivatives instruments. This market is accessible by all entities outside Mainland China for purposes other than trade and personal use, eg.

4 The figures in the table refer to the renminbi trading in Hong Kong. The respective figures for offshore renminbi trading

globally are USD 12.8 billion for spot and USD 7.1 billion for deliverable forwards.

Onshore and offshore RMB markets Table 1

CNY markets1 CNH markets1

Products Spot, forward, swap and options Spot, forward, swap, and options

Market participants Central bank, domestic banks, finance companies (subsidiaries of large

SOEs) and domestic subsidiaries of foreign banks

Exporters, importers, offshore financial institutions, hedge funds and Hong Kong

residents

Price formation mechanism Managed float Free float

Central bank intervention Yes No

Trading band ±0.5% - ±2%4 No

Regulatory authorities PBoC HKMA

Daily turnover in April 20132

Spot US$ 17.6 bn US$ 5.1 bn

Deliverable forward US$ 2.4 bn US$ 2.2 bn

Bid-ask spread3 17 pips 33 pips 1 Against the US dollars. 2 Adjusted for local and cross-border inter-dealer doubling counting. 3 Average of 2013 Q4. 4 Trading bands widened from ±0.3% to ±0.5% on 21 May 2007, then to ±1% on 16 April 2012, and then to ±2% on 17 March 2014.

Sources: BIS Triennial Survey and HKMA Quarterly Bulletin, December 2013.

WP 11

investment, hedging, and so on.5 Reportedly, corporates, banks, asset managers and even hedge funds, which traditionally use the NDF market, have increasingly switched to the CNH market as a result of this market’s rapid improvement in liquidity. Bid-ask spreads have narrowed. The bid-ask spread of spot trading has compressed to the range of 20-40 pips, significantly lower than the volatile range of 30 to over 300 pips in late 2010 and much closer to the onshore spread. The greater latitude in the market may have facilitated rapid expansion and proliferation of products. Notably, forwards and derivatives account for around 85% of the CNH market, in contrast to the smaller share of this segment onshore.

The distinctive features of the CNY and CNH markets and continuing segmentation of the onshore and offshore renminbi markets in general (including money, bond and foreign exchange markets) result in deviations of the two exchange rates. The different pricing mechanisms give rise to potentially different movements of the two rates. For example, even when both respond to the same set of economic news, eg. on macroeconomic data and policy changes, the CNH rate may react more strongly as there are no trading limits or concerns over central bank actions. At the same time, the offshore markets are more exposed to global factors. As fundamentally the same financial product, the two foreign exchange markets should move close to each other if they are completely free of trade barriers. Arbitrage between the onshore and offshore markets does take place. Arbitrage activities can help facilitate similar pricing of the onshore and offshore markets (money market and foreign exchange market alike). Yet, given still relatively narrow channels of cross-border flows and transaction cost, pricing in the CNH/CNY markets can diverge.

Onshore and offshore renminbi differential: a snapshot

Indeed, both the CNY and CNH rates follow the broad trend, appreciating by around 10% since September 2010 (Graph 1, left-hand panel). But the CNH rate displayed greater volatility in daily movements, with a wider trading range and a bigger standard deviation (Graph 1, centre panel). To examine closely the deviations between the two, let CNYt and CNHt denote the daily onshore and offshore FX rate against the US dollar taken the end of the trading day. The exchange rate premium/discount Dt is defined as 𝐷𝑡 = 100 ∙ 𝑙𝑛 �𝐶𝑁𝐻

𝐶𝐻𝑌�. Between September 2010 and September 2013,

two periods stood out when the deviations between the CNY and CNH rates were particularly wide. The first period was at the very early stage of the CNH operations in late 2010 when the conversion quota for trade-settlement related renminbi transactions placed severe restrictions on liquidity in the offshore renminbi market. The quota ceiling was breached in October 2010 due to overwhelming offshore demand for the renminbi, leading to a much stronger CNH than the CNY. The reverse situation occurred in autumn 2011 against the backdrop of a sharp increase in risk aversion globally due to the intensified European debt crisis. The surge in demand for the US dollar led to a breach of the quota ceiling for renminbi conversion in Hong Kong, and the CNH weakened much more than the CNY as a result.6 In addition, the volatility of the CNY rate during the episodes of the breach of the conversion quota also intensified significantly, echoing the findings by Moziad and Kang (2012) that there exist volatility spillovers from offshore to onshore renminbi market. Since then, liquidity in the

5 Renminbi foreign exchange transactions for trade settlement and personal use can be undertaken through the Bank of

China (HK), a designated Clearing Bank by the PBoC for renminbi business in Hong Kong, at rates close to onshore rates. However, there exists an (undisclosed) net position quota for trade settlement and a conversion quota for personal transactions (ie. RMB 20,000 per day per account).

6 Some observers, eg. Craig, Hua, Ng and Yuen (2013), have discarded those observations surrounding October 2010 and September 2011 in the empirical work on the ground that the variation cannot be encompassed in a general modelling framework. On the contrary, we believe that a satisfactory model should be able to capture these episodes.

12 WP

CNH market has improved significantly, with lower bid-ask spreads and narrower deviations from the CNY market (Graph 1, right-hand panel).

From a time series point of view, Dt shows clear evidence of volatility clustering: there were extended periods where Dt volatility was low, and periods when Dt volatility was high. In order to test for potential autoregressive conditional heteroscedastic (ARCH) effects, we conduct Engle´s (1982) ARCH test. A large critical value indicates rejection of the null hypothesis in favour of the alternative. The ARCH test result (4,503.72, probability value = 0.000) strongly points to the presence of ARCH effect in the series.

Reminbi exchange rate Graph 1

CNY and CNH differentials Market volatility2 Bid-ask spreads Lhs: vis-à-vis USD Rhs : in basis points In per cent In pips

1 A positive number indicates implied depreciation. 2 It shows the maximum, minimum and mean plus/minus 1 standard deviation of the sample of daily changes of exchange rate vis-à-vis the US dollar between 1 September 2010 and 30 November 2013.

Sources: Bloomberg; authors’ calculations.

Measuring market segmentation, fundamentals and contagion

Given the fact that the CNY and CNH rates are essentially the same financial product, their pricing differentials may well be driven by the segmentation between onshore and offshore markets due to existing controls on capital account in China. Meanwhile, different fundamentals such as economic or market conditions in Mainland China and Hong Kong can affect the pricing of the CNH and CNY rates differently. On top of these factors, global market conditions may also play a role in causing the CNH-CNY differential, as the offshore market is more prone to risk contagion in the financial markets across the world. These different sets of factors and their measurement will be discussed in turn in details. The emphasis is placed on compiling measures of market segmentation.

Measuring market segmentation using policy variables

Policies that may lead to the CNH-CNY differential may include measures that place restrictions on onshore and offshore trading and those on cross border renminbi flows. We code the policy changes into dummy variables to measure the degree of market segmentation. This approach has widely adopted to gauge the extent of capital controls, institutional changes and so on.

WP 13

One example of direct trading restrictions is the daily trading band in the onshore market. It constrains the responsiveness of the CNY rate to changes in economic and market conditions. Greater exchange rate flexibility would allow the CNY market to be priced more in line with economic fundamentals and market sentiment. This would allow the onshore and offshore rates to move in a more consistent way, thus reducing the differential between the two. A policy dummy reflecting the gradually widening of the trading band is considered to see whether it might lower the volatility of the CNH-CNY differential (Table 2).

For the CNH market, the conversion quota under the cross-border renminbi trade settlement scheme constituted a constraint at the earlier stages of market development. Depletion of the quota would sharply dry up market liquidity, and thus increase the volatility of the CNH-CNY differential. The pricing of the CNH rate would exhibit substantial deviations from that of the CNY, as seen in the previous section. We introduce two dummies to capture the two periods when the quota was breached in late 2010 and late 2011 respectively (Table 2).

A range of policies aimed at expanding cross border renminbi flows and facilitating development of offshore renminbi markets may also have affected the CNH-CNY differential. These measures include the following.

• Current account measures: cross-border renminbi trade settlement

The cross-border renminbi trade settlement scheme was introduced in July 2009 on a trial basis, allowing selected importers and exporters in five mainland cities to settle trade transactions with counterparts in Hong Kong, Macau and ASEAN countries. The scheme was gradually widened to cover all external trade, signifying the complete removal of restrictions on the use of the renminbi for current account transactions. On the surface, the trade settlement scheme merely allows Chinese firms to use the renminbi to settle their cross-border trade transactions. In reality many firms have used the scheme to channel funds across the border between mainland China and Hong Kong, and conduct exchange rate arbitrage.

• Capital account: inward and outward renminbi flows

A series of measures have been announced since the second half of 2010 which encourage cross border renminbi flows under capital account. The major channels opened for offshore renminbi flows back to mainland China include: a) eligible offshore financial institutions invest offshore renminbi funds in the mainland interbank bond market; b) foreigners undertake direct investment in mainland China using offshore renminbi; c) qualified foreign institutional investors invest offshore renminbi in mainland stock and bond markets; and d) onshore non-financial corporations to raise renminbi funds offshore. There are fewer channels for renminbi outflows. These include mainland firms using the renminbi to undertake overseas direct investment and mainland banks extending renminbi loans to domestic enterprises for operating overseas.

Three dummies are compiled based on the different sets of measures for expanding cross border renminbi flows: trade settlement, renminbi inward flows and renminbi outflow flows. Table 2 lists the timeline of changes in the policies and associated dummy variable values, with higher values representing more liberal regimes for flows.

Among the institutional variables compiled in this sub-section, while breaching the conversion quota might have immediate effects on the CNH-CNY differential, others may have longer term impact. Most notably, the policies for cross border renminbi flows should increase efficiency of the offshore renminbi market, and thus reduce volatility of the CNH-CNY differential over the long term. The renminbi trade settlement scheme initially increased the offshore renminbi supply as there were more participating mainland importers than exporters, but renminbi inflows and outflows under the

14 WP

scheme became more balanced over time. Policies for renminbi outflows should help expand the pool of offshore renminbi. While policies for renminbi inflows to the mainland may take funds out of offshore renminbi pool, they may encourage the use of the CNH market. The widening channels for renminbi flows also increase arbitrage opportunities. For example, a merchant can choose to settle trade either onshore or offshore depending on which exchange rate is more favourable. Similar opportunities are also becoming possible under capital account transactions. Over a longer horizon, both markets become more efficient as larger cross-border renminbi flows facilitate price discovery in the two markets. Taking into account all these effects, it is not certain that these policies may not induce a change in the relative pricing of the CNH to CNY rate in a certain direction, yet they are likely to lead to lower volatility of the differential.

Policy dummies between September 2010 and September 2013 Table 2

From To Definition

Cross-border renminbi trade settlement (TS) 0 23 August 2010 23 August 2011 An early stage of the scheme in which selected

companies from 20 Mainland provinces can settle trade with the rest of the world in renminbi

1 24 August 2011 1 March 2012 Expanded the scheme to selected companies in all Mainland provinces

2 2 March 2012 30 September 2013 Pilot scheme closed. All trading companies in China became eligible for cross-border renminbi trade settlement

Capital account liberation i) Inward renminbi capital flows (IF) 0 23 August 2010 13 October 2011 Foreign central banks, offshore renminbi clearing

banks and Participating Banks allowed to invest RMB raised offshore in the Mainland interbank bond market

1 14 October 2011 15 December 2011 Approved foreigners allowed to invest renminbi raised offshore in Mainland firms directly, including through the provision of renminbi cross-border loans

2 16 December 2011 2 April 2012 Qualified foreign institutional investors allowed to invest renminbi raised offshore in listed Mainland bonds and equities

3 3 April 2012 7 May 2012 Investment quotas of qualified foreign institutional investors expanded

4 8 May 2012 30 September 2013 Rules formalised for onshore non-financial corporations to issue offshore renminbi bonds

ii) Outward renminbi capital flows (OF) 0 23 August 2010 12 January 2011 No renminbi outflows permitted under capital

account 1 13 January 2011 30 March 2012 Mainland firms allowed to take renminbi offshore

for overseas direct investment (ODI) in foreign firms

2 31 March 2012 30 September 2013 Mainland banks allowed to extend renminbi loans to “going-out” domestic enterprises.

WP 15

Policy dummies for various dimensions (continued) Table 2

From To Definition

Market liquidity condition improvements i) Liquidity facility (OL) 0 23 August 2010 14 June 2012 No offshore renminbi liquidity support 1 15 June 2012 30 September 2013 HKMA provides CNH liquidity facility to

participating banks in Hong Kong ii) Trading band (TB)3 0 23 August 2010 15 April 2012 Daily trading band for the USD/CNY rate at ±0.5% 1 16 April 2012 30 September 2013 Daily trading band widened to ±1 % Depletion of conversion quota i) Quota1

1 0 23 August 2010 26 October 2010 Offshore renminbi conversion (at onshore rates)

quota under trade settlement arrangement in place 1 27 October 2010 4 November 2010 Quota breached for the first time 0 5 November 2010 30 September 2013 New quota imposed ii) Quota2

2 0 23 August 2010 23 September 2011 Quota for offshore renminbi conversion (at

onshore rates) under trade settlement arrangement in place

1 24 September 2011 3 October 2011 Quota breached for the second time 0 4 October 2011 30 September 2013 New quota imposed.

Sources: HKMA, PBoC.

Macroeconomic and market fundamentals

Macroeconomic developments are among the fundamental determinants of exchange rates. Announcements of macroeconomic data can trigger immediate adjustment in exchange rates movements, as they lead market participants to revise expectations on an economy’s outlook and thus adjust their portfolios. One stand of literature in recent years studies the effects of macroeconomic announcements on exchange rates. This literature typically finds that exchange rates appreciate in response to unexpected declines in inflation and stronger-than-expected growth (eg. Andersen, Bollerslev, Diebold and Bega, 2004 and Faust, Rogers, Wang and Wright, 2007). In the case of the renminbi, different participant bases in the onshore and offshore markets may react to the same set of macroeconomic news differently due to different interpretations of the news. The absence of currency intervention and trade band restrictions may also permit the CNH rate to react more strongly to the same set of information.

To capture the impact of macroeconomic fundamentals, two types of variables will be considered in this study. The first type is based on macroeconomic forecast surveys, taking the surprise elements in key macroeconomic indicators, ie. deviations of outturns from projections obtained from Bloomberg. The indicators considered are GDP growth, industrial production growth and the Purchasing Manager Index. The advantage of measuring surprise in macroeconomic fundamentals in this way is that they do not rely upon an estimated model and therefore the estimates aren´t contaminated by the so-called generated regressor problem (Pagan, 1984).

16 WP

The second type is stock indices as indirect measures of impact of macroeconomic data on sentiment of different markets. 7 The Shanghai stock exchange composite index is used to proxy onshore and offshore market sentiment respectively. For the Hong Kong Market, the Hang Seng sub-index covering dual listed companies (ie. those listed both in the mainland and Hong Kong) is used.

Differences of onshore and offshore renminbi market conditions are also important factors driving the CNH-CNY pricing differentials. As discussed earlier, the onshore spot market remains deeper and more liquid than the offshore market, but the latter is growing rapidly. To capture the impact of the evolution of the offshore market deepening and liquidity conditions, we employ two measures: a) the bid-ask spread of the CNH rate; and b) the ratio between the CNH and CNY bid-ask spreads. Higher spreads or higher ratio between the two markets indicate worsening liquidity conditions in the CNH market. It is recognised that poor liquidity can lead to discounts the price of a financial asset (Amihud and Mendelson, 1986). In the meantime, it may also be associated with greater volatility of the CNH-CNY differentials.

To improve the liquidity condition in the offshore renminbi market, the HKMA launched the renminbi liquidity facility in mid-2012. We compile a dummy variable for this facility to gauge its effectiveness in improving liquidity and thus reducing volatility of the CNH-CNY pricing differentials (Table 2).

Global market contagions

Apart from domestic conditions, exchange rate movements are also under the influence of global conditions. Tighter global liquidity or a reduction in risk appetite might lead to fund outflows from emerging markets, resulting in weaker currencies. The influence is likely to be bigger for the CNH market which is more connected with global financial markets, while largely effective, albeit leaky, capital controls insulate the CNY market more from external shocks. Thus, global financial shocks might also drive wedge between the CNH and CNY rates. We consider two types of indicators – global liquidity and investor risk appetite. There are several price-based or quantity-based measures of global liquidity in the literature.8 High frequency price-based indicators, namely 10-year US Treasury bond yields and US 5-year swap rates, are used in this study. Investor risk appetite is proxied by the VIX, an implied volatility index that measures the market's expectation of 30-day S&P 500 volatility priced in S&P 500 options. The VIX is considered to be the world´s premier barometer of investor sentiment and market volatility.

Modelling the CNH-CNY differential: Design

As a benchmark for modelling the CNH-CNY differential, we start from a parsimonious generalised autoregressive conditional heteroscedasticity [GARCH(p,q)]. As noted earlier, the CNH-CNY differential series, Dt, shows some evidence of volatility clustering. Engle (1982) and Bollerslev (1986) showed that volatility clustering, or conditional heteroscedasticity, can be modelled using a simple generalised autoregressive conditional heteroscedasticity [GARCH(p,q)] model of the form. A GARCH(p,q) for Dt is given as:

7 Nonetheless, it should be noted that equity prices may also be moved by fundamentals not related to macroeconomic

conditions, such as corporate profits and market liquidity. 8 See Committee on the Global Financial System (2011) for a summary discussion on concepts and measurements of global

liquidity.

WP 17

(1) 𝐷𝑡 = 𝜇 +�𝜙𝑖𝐷𝑡−𝑖 +𝑟

𝑖=1

𝜀𝑡

(2) 𝜀𝑡 = �ℎ𝑡𝑧𝑡

(3) ℎ𝑡 = 𝜔 + ∑ 𝛼𝑖𝜀𝑡−𝑖2𝑞𝑖=1 + ∑ 𝛽𝑖ℎ𝑡−𝑖

𝑝𝑖=1 ,

where zt is assumed to be an iid N(0,1) random variable, ht is the conditional variance of εt given {𝜀𝑠, 𝑠 < 𝑡}, ω > 0, and the αi and βi parameters are assumed to be positive to ensure that the conditional variance ht is positive. The lagged-dependent variables typically capture autocorrelation caused by market microstructure or non-trading day effects.9

In the next step, we extend the basic GARCH(p,q) model by adding explanatory variables in the mean and conditional variance equations. Simple GARCH(p,q) models are unlikely to be the true data generation process, and a more flexible modelling of the mean and conditional variance dynamics will undoubtedly improve the explanatory power of the model. More importantly, GARCH(p,q) provides a mechanical way to describe the behaviour of a heteroscedastic time series, but gives no indication about which factors have caused such behaviour to occur. As such, it does not offer insight for understanding the determinants of the 𝐷𝑡 series and its time-varying volatility. Adding further explanatory variables to the mean and conditional variance equations sets our study apart from in the literature on modelling exchange rates, as it allows us to explore ‘deeper’ drivers of exchange rates and of the pricing differential in renminbi onshore and offshore markets.

In our extended GARCH model, the mean equation takes the following form:

(4) 𝐷𝑡 = 𝜇 + ∑ 𝜙𝑖𝐷𝑡−𝑖 + ∑ 𝜿𝑙′𝒙𝑡−𝑙 +𝐿𝑙=0

𝑟𝑖=1 𝜀𝑡,

where x is a 𝑘 × 1 vector of (weakly) exogenous explanatory variables. The variables we consider to enter the conditional mean equation include the trade settlement conversion quota dummies QUOTA1 and QUOTA2, the bid-ask spread (SPREAD) in the CNH market (or the ratio of the CNH and CNY bid-ask spread as an alternative), the share price ratio of Hong Kong to Shanghai (SHARE), macroeconomic surprises, and US interest rates. The depletion of the conversion quota is expected to lead to sharp, yet temporary widening of the differential. The dummy for the quota for late 2010 is expected to have a negative sign as the CNH rate carried a premium over the CNY rate at the time, i.e. the CNH rate is smaller than the CNY rate meaning the former is stronger than the latter vis-à-vis the US dollar. The opposite is true for the period in late 2011, and the coefficient on the quota dummy should be positive. A positive sign is expected for the coefficient on the US interest rates, ie. tighter global liquidity conditions, as reflected in higher US interest rates, might lead to a sharper weakening of the CNH rate compared to the CNY rate. The liquidity indicators may carry a positive sign, reflecting the premium placed on an illiquid asset.

9 Omitting relevant explanatory variables from the mean equation increases the variance of the error term which potentially

biases the GARCH results. Misspecification of the conditional mean equation may also cause the estimated residuals to be correlated which, in turn, causes the squared residuals be to correlated. Therefore, care should be taken when specifying the conditional mean equation of GARCH models.

18 WP

Just as further explanatory variables may be added to the conditional mean equation, weakly exogenous explanatory variables may also be added to the conditional variance equation in a straightforward way giving:

(5) ℎ𝑡 = 𝜔 + ∑ 𝛼𝑖𝜀𝑡−𝑖2𝑞𝑖=1 + ∑ 𝛽𝑖ℎ𝑡−𝑖 + ∑ 𝝍𝑘

′ 𝒘𝑡−𝑘𝐾𝑘=1

𝑝𝑖=1 ,

where w is an 𝑚 × 1 vector of (weakly) exogenous variables that may account for the heteroscedastic nature of the disturbances. The variables we consider for the conditional variance equation include the policy dummies for the cross border renminbi settlement scheme (TS), inward renminbi flows (IF) and outward renminbi flows (OF), offshore renminbi liquidity facility (OL), trading band relaxation of the CNY market (TB) and the global risk measure VIX. For the policy variables, we consider them in two forms: a) one variable for each policy area with different steps combined into one dummy variable; b) individual steps in each policy area, if there are different stages, are broken into separate dummies, eg. TS1, TS2 and TS3 for the three stages in the cross border renminbi settlement scheme. The reason to consider separate stages in one policy area is that the extent of each step is often different, and thus their impact might differ.

For the estimation, daily data from September 2010 (when quotes for the CNH rate became regular) to September 2013 are used, excluding weekends and other holiday non-trading periods.

It is worth noting that estimating GARCH models with additional dummy variables entails some non-trivial risks and poses challenges in the computation. Doornik and Ooms (2008) have demonstrated that regression-GARCH models with dummy variables in the conditional mean equation may lead to multimodality likelihood functions. Since reaching a global maximum of the log-likelihood function is not guaranteed using standard optimisation techniques like the BFGS algorithm, estimation has to be treated with care. In light of this problem, we have explored the surface of the log-likelihood by perturbating the starting values and re-estimating the GARCH parameters. In order to avoid similar pitfalls in modelling the variance, we have also followed Doornik and Ooms (2008) to add a corresponding dummy in the conditional variance equation as a robustness check.10

Empirical results

Basic GARCH(1,1) model

In estimating a basic GARCH(p,q) specification, model selection information criteria point to the simple GARCH(1,1) model. The best-fitting basic GARCH(1,1) models and the associated diagnostics are presented in Table 3.

10 The estimation methodology looks straightforward, though it is in fact complex because of the large number of

parameters to be estimated. Computational tractability requires appropriate starting values in order to achieve convergence to the global maximum. To achieve this, appropriate starting values were obtained using the simplex algorithm. The preliminary iterations avoid problems with the multimodality and/or discontinuity of the likelihood function.

WP 19

Benchmark GARCH model estimates1 Table 3

(I) (II)

Mean equation

µ -0.044*** -0.003

(8.3) (0.9)

φ1 - 0.918***

- (44.2)

Variance equation

ω 0.005*** 0.001***

(5.6) (3.1)

α1 0.848*** 0.248***

(8.2) (4.7)

β1 0.198*** 0.737***

(3.0) (15.6)

Diagnostic tests

LogL 221.64 595.74

LB(15) 2052.7 14.21

(0.00) (-0.51)

ARCH(1) 0.2 0.008

(-0.65) (-0.93)

WC(15) 12.39 18.19

(-0.65) (-0.25) 1 Sample from Septempber 2010 – 20 September 2013. ***, **, * indicator significance at 1%, 5% and 10% levels respectively. For the parameters t-values robust to heteroscedasticity are given in parentheses. For the residual tests prob-values are given in parentheses. LB(15) is the Ljung-Box Q statistic for 15 lags. ARCH(1) is the LM-test for 1st order ARCH effects. WC(15) is the modified Ljung-Box Q-statistic for serial dependence of squared residuals. The test statistic is robust to heteroscedasticity and resported for autocorrelations up to lag 15 (West and Cho, 1995). Diagnostic tests are carried out on the standardised residuals. The normal error distribution is utilised. All models are estimated using the BFGS algorithm, using numerical derivatives.

All explanatory variables are statistically significant at the 5 percent level and the models do a good job of capturing the observed volatility clustering in Dt. Model I in Table 3 points to high nonstationary volatility in the fitted GARCH(1,1) model, as 𝛼1 + 𝛽1 > 1. In this case, the GARCH(1,1) model becomes an integrated GARCH(1,1) (IGARCH) model. Adding the lagged dependent variable to the model solves this problem (Table 3, Model II).11 The parameter φ1 shows high persistence in Dt, implying a rather smooth evolution of the premium/discount through time. The implied half-life of a volatility shock 𝑙𝑛(0.5)

𝑙𝑛(𝛼1+𝛽1) in Model II is 𝑙𝑛(0.5)

𝑙𝑛(0.248+0.737)= 45.9 days. So this model implies that the

conditional volatility is very persistent.

11 Lumsdaine and Ng (1999) and Mikosch and Starcia (2004) have suggested that observed IGARCH(1,1) behaviour may result

from misspecification of the conditional mean and the conditional variance equation.

20 WP

Extended GARCH(1,1) models

The estimation results for the extended GARCH models proxies for segmentation (ie. policy variables), fundamentals, and global market contagion introduced are presented in Table 4. In this set of specifications, the fundamentals are represented by the bid-ask spread in the CNH market and share price ratio of the Shanghai and Hong Kong equity markets. The GARCH(1,1) specification is retained for the extended GARCH models. Hansen and Lunde (2004) have provided compelling evidence that it is difficult to find volatility models that beat the plain GARCH(1,1).12

12 All models have been estimated assuming a Gaussian error distribution. Weiss (1986) and Bollerslev and Wooldridge

(1992) have shown that even when normality is inappropriately assumed, maximising the Gaussian log-likelihood results in quasi-maximum-likelihood estimates that are consistent and asymptotically normally distributed.

WP 21

GARCH model estimates including policy segmentation variables1 Table 4

(I) (II) (III) (IV) (V) (VI) (VII)

Mean equation µ 0.026*** 0.014*** 0.018*** 0.02*** 0.019*** 0.019*** 0.015***

(11.60) (5.10) (2.70) (7.50) (2.70) (7.40) (2.60)

φ1 0.74*** 0.754*** 0.748*** 0.748*** 0.747*** 0.75*** 0.762***

(45.50) (49.40) (32.40) (46.20) (45.40) (47.60) (30.85) Segmentation – policy variables

QUOTA1 -0.31*** -0.292*** -0.291*** -0.292*** -0.302*** -0.289*** -0.307***

(3.00) (3.40) (3.20) (3.60) (3.10) (3.50) (4.60)

QUOTA2 0.118 0.138 0.11 0.101 0.123 0.105 0.236

(0.50) (0.60) (0.50) (0.50) (0.70) (0.50) (1.40) Fundamental variables

SPREADt 0.078*** 0.076*** 0.077*** 0.077*** 0.077*** 0.076*** 0.062*** (17.70) (19.70) (9.40) (18.20) (16.60) (19.40) (7.60)

SHAREt -0.227*** -0.136*** -0.176*** -0.188*** -0.183*** -0.192*** -0.149***

(8.70) (5.70) (2.80) (7.20) (2.70) (7.70) (2.80) Variance equation ω 0.001*** 0.091*** 0.002*** 0.002*** 0.001*** 0.001*** 0.103***

(36.30) (329.50) (2.60) (52.20) (33.10) (28.90) (13.80)

α1 0.158*** 0.111*** 0.147*** 0.148*** 0.125*** 0.157*** 0.292***

(29.50) (19.10) (3.70) (22.50) (19.70) (23.70) (3.20)

β1 0.818*** 0.841*** 0.8*** 0.797*** 0.833*** 0.798*** 0.509***

(254.30) (237.80) (15.30) (206.40) (252.20) (202.70) (3.10)

Segmentation – policy variables

TS1 -0.09*** -0.089***

(137.60) (46.50)

TS2 -0.088*** -0.077***

(513.30) (18.60)

TS3 -0.091*** -0.062***

(292.20) (5.90)

IF -0.0002** -0.005**

(2.30) (2.20)

OF -0.0006*** -0.006*

(30.60) (1.80)

TB -0.0005** (2.40) OL -0.0006*** (11.70) Contagion variables ∆lnVIXt 0.014*** 0.006** 0.011*** 0.014*** 0.012*** 0.013*** 0.004 (11.40) (2.50) (3.00) (9.40) (7.40) (9.30) (1.20) Diagnostic tests LogL 655.89 678.6 662.61 662.26 661.54 663.32 699.62 LB(15) 24.19 23.79 23.62 3.91 24.25 24.19 22.08 (0.06) (0.07) (0.07) (0.07) (0.06) (0.06) (0.11) ARCH(1) 0.89 2.54 1.12 0.89 1.65 0.98 0.03 (0.35) (0.11) (0.29) (0.34) (0.19) (0.32) (0.87) WC(15) 15.99 19.77 14.88 15.81 16.42 15.03 11.88 (0.38) (0.18) (0.46) (0.39) (0.36) (0.45) (0.69)

22 WP

1 Sample from Septempber 2010 – 20 September 2013. ***, **, * indicator significance at 1%, 5% and 10% levels respectively. For the parameters t-values robust to heteroscedasticity are given in parentheses. For the residual tests prob-values are given in parentheses. LB(15) is the Ljung-Box Q statistic for 15 lags. ARCH(1) is the LM-test for 1st order ARCH effects. WC(15) is the modified Ljung-Box Q-statistic for serial dependence of squared residuals. The test statistic is robust to heteroscedasticity and resported for autocorrelations up to lag 15 (West and Cho, 1995). Diagnostic tests are carried out on the standardised residuals. The normal error distribution is utilised. All models are estimated using the BFGS algorithm, using numerical derivatives.

In these extended GARCH models, the impacts of segmentation, fundamentals, and global market contagions on both the mean and volatility of CNH-CNY pricing differentials are largely in line with our expectations.

Among the policy variables, depletion of renminbi conversion quota, has both statistically and economically significant impact on the daily CNH-CNY pricing differentials in the conditional mean equation. In particularly, the CNY was priced around 30% more expensive than the CNH during the first episode of quota depletion in late 2010 compared to normal market conditions. In comparison, the impact of the second quota depletion seems to have little impact on the pricing differentials in most specifications. Exploring with different dynamic structures, it is found that QUOTA2 has an impact on the CNH-CNY differential with a 5 day lead. It widened the gap by making the CNH notably weaker than the CNY amidst the sharp rise in global reversion upon deepening of the European sovereign debt crisis. The lead in QUOTA2 may reflect that the strain in the CNH market had already intensified before the formal breach of the conversion quota.

The policy measures rolled out by the authorities aiming for increasing improving cross border renminbi flows are shown to significantly lower the volatility of the CNH-CNY pricing differentials when entering the conditional variance equations individually. For the cross border renminbi trade settlement, the dummies for different stages of the scheme (TS1, TS2, and TS3) are individually significant, but the cumulative dummy for this scheme is not significant. For the other liberalisation measures, the collective or cumulative dummy variables are important. These variables all carry a negative sign. This suggests that the enlargement of cross-border renminbi trade settlement scheme, gradual liberalisation of capital accounts by allowing greater inward and outward capital flows (IF and OF), offering short-term renminbi liquidity support to offshore banks (OL) are negatively associated with the volatility of onshore-offshore renminbi pricing differentials (Table 4, Models II-VI). Among all policy measures, the enlargement of cross-border settlements has the biggest impact in reducing the volatility between the CNH-CNY differential, with their coefficients more than ten times the size of IF and OF. This is in line with the fact that the introduction of cross-board trade settlement is the cornerstone of renminbi internationalisation, and has made the biggest contribution to the offshore renminbi pool. The significance of the dummies for individual phases of the trade settlement scheme shows that the different stages have different impacts on the differential. Indeed, TS1 led to a greater reduction in the differential volatility than TS2 and TS3, suggesting that the first expansion of the scheme (from trade between 5 Mainland metropolitan cities/provinces and a few selected regions outside the Mainland to that between 20 provinces and the rest of world) had the biggest effect in expanding offshore renminbi liquidity.

Among the other policy variables, relaxation of the trading band in the CNY market (TB) is found to reduce volatility, but the magnitude is relatively small (Table 4, model V). Our sample does not cover the period following the latest widening of the band in March 2014. Inferring from the estimation results, the further band widening should continue to compress the volatility of the CNH-CNY differential. The access to short-term liquidity instruments provided by the HKMA has limited impact on the market liquidity conditions (OL). The variable’s coefficient is statistically significant, but very small in magnitude (Table 4, Model VI). This may be due to the fact that the facility has not been frequently used, possibly owing to a stigma attached to using such facilities. Resorting to the central bank’s help may be viewed as a negative signal to the market, and thus banks are less prone to use the instrument.

WP 23

In the final model of Table 4, all policy variables enter the estimation simultaneously (Table 4, Model VII). The impacts of policy measures such as improving offshore market liquidity (OL), and gradual relaxation of CNY trading band (TB), however, seem to be less robust and drop out of the specification. One possible reason is that the timing captured by these variables may coincide with the introduction of other policies.

Turning to fundamental variables, market liquidity (SPREAD) and market sentiment (SHARE) also play an important role in explaining the onshore-offshore renminbi pricing differentials in the conditional mean equation. For instance, greater bid-ask spreads in the CNH market (either measured as the CNH bid-ask spread, or the ratio between the CNH and CNY bid-ask spreads) tend to result a discount of the offshore renminbi relative to its onshore counterpart, in line with the empirical evidence that illiquidity of a market may toll the players for holding the underlying financial asset. Improvement in offshore market sentiment, proxied by the rising of dual-listed H-shares relative to A-shares, on the other hand, leads to a more appreciated renminbi in the CNH relative to the CNY market.

The global factors seem to have weaker impact than we expected. The US interest rates, measured either as the 10-year US treasury bond yield or 5-year swap rates, are not significant in any specification.So we omit them in the interests of parsimony. By comparison, the VIX variable, the risk aversion indicator, is highly significant in specifications with individual policy variables, but less so when entering the conditional variance equation jointly with all policy variables. It carries a positive sign. The estimates for the VIX may have a number of interesting implications. Risk appetite of the global investor community can affect the renminbi market. With the CNH market more linked with the markets in the rest of the world, a rise in risk aversion globally will increase the volatility of the differential between the CNH and CNY rates. Yet, the effects of global factors may be dampened by policies that encourage cross border renminbi flows. When cross border flows are sufficiently large, the CNH-CNY differentials triggered by global capital movements will be quickly eroded.

On the model properties, the GARCH-specific variables remain largely robust in these extended GARCH models and point to persistence in both level and volatility of CNH-CNY differentials. The parameters ω, α and β remain significant. Thus, all previous results related to market segmentation, fundamentals and contagion do not preclude the existence of GARCH effects. The greater log-likelihood values indicate that the extended models are superior statistical characterisations of the CNH-CNY premium/discount compared to the basic GARCH(1,1) models.

Table 5 reports additional specifications in which macroeconomic surprise variables for mainland China are added.13 The estimation results in Table 3 encompass both the modelling approaches in Table 1 and Table 2, and include variables that are significant in the previous specifications. The macroeconomic surprise variables enter the estimation individually initially (Table 5, Models I-III), and simultaneously in the final model (Table 5, Model IV).

13 The hypothesis that discontinuities and jumps in Dt correspond to Chinese macroeconomic surprise releases is consistent

with Andersen et al. (2003, 2007) and Lahaye et al. (2011).

24 WP

GARCH Model Estimates Including Macroeconomic Surprise Variables Table 5

(I) (II) (III) (IV)

Mean equation µ 0.015** 0.016** 0.016*** 0.017***

(2.30) (2.50) (2.70) (2.90) φ1 0.758*** 0.765*** 0.773*** 0.764***

(24.40) (33.60) (32.20) (33.80) Segmentation – policy variables

QUOTA1 -0.302*** -0.309*** -0.309*** -0.31***

(4.50) (4.90) (4.90) (5.00) QUOTA2 0.212 0.242 0.241 0.241

(1.00) (1.30) (1.40) (1.30) Fundamental variables

SPREADt 0.063*** 0.06*** 0.061*** 0.061***

(5.90) (8.50) (7.80) (8.90) SHAREt -0.154** -0.156*** -0.156*** -0.167***

(2.50) (2.70) (2.80) (3.10) IP-S 0.024 0.016

(1.60) (1.10) GDP-S 0.224** 0.191*

(2.20) (1.80) PMI-S 0.017 0.017

(0.90) (0.90) Variance equation

ω 0.083*** 0.058*** 0.061*** 0.063*** (7.50) (10.70) (9.30) (165.80)

α1 0.265** 0.307*** 0.3*** 0.309*** (2.00) (4.20) (3.90) (5.00)

β1 0.57** 0.487*** 0.493*** 0.487*** (2.10) (4.10) (3.40) (6.50) Segmentation – policy variables

TS1 -0.072*** -0.045*** -0.048*** -0.05*** (34.80) (30.60) (27.90) (162.20)

TS2 -0.06*** -0.032*** -0.034*** -0.036*** (8.80) (9.50) (9.30) (23.6)

TS3 -0.048*** -0.017** -0.019** -0.022*** (2.90) (2.00) (2.00) (38.20)

IF -0.005 -0.005*** -0.006** -0.006*** (1.30) (2.80) (2.50) (60.80)

OF -0.005 -0.006** -0.006** -0.006*** (1.00) (2.30) (2.10) (24.20)

Contagions variables

∆lnVIXt 0.004 0.004 0.004 0.003 (1.10) (1.10) (1.30) (1.10) Diagnostic tests

LogL 700.95 701.73 699.93 702.65

LB(15) 22.29 22.26 21.94 21.54 (0.10) (0.10) (0.11) (0.12)

ARCH(1) 0 0.07 0.06 0.08 (0.97) (0.79) (0.80) (0.77)

WC(15) 12.37 11.67 12.02 12.17 (0.65) (0.70) (0.68) (0.67)

WP 25

1 Sample from September 2010 – 20 September 2013. ***, **, * indicator significance at 1%, 5% and 10% levels respectively. For the parameters t-values robust to heteroscedasticity are given in parentheses. For the residual tests prob-values are given in parentheses. LB(15) is the Ljung-Box Q statistic for 15 lags. ARCH(1) is the LM-test for 1st order ARCH effects. WC(15) is the modified Ljung-Box Q-statistic for serial dependence of squared residuals. The test statistic is robust to heteroscedasticity and resported for autocorrelations up to lag 15 (West and Cho, 1995). Diagnostic tests are carried out on the standardised residuals. The normal error distribution is utilised. All models are estimated using the BFGS algorithm, using numerical derivatives.

Most macroeconomic variables are not significant in these models. The only macroeconomic

surprise variable that has some significance is GDP growth. Some other key macroeconomic indicators (eg. inflation, export growth, both in the form of surprises, and a summary surprise index compiled by the CitiBank) have also been tested, and found not to be significant. Rather than concluding that macroeconomic fundamentals do not matter for the CNH-CNY differential, this might reflect the erratic market response to macroeconomic releases. Sometimes stronger-than-expected growth is welcomed, taken as avoiding a sharp economic slowdown at this juncture. Yet, at different points of a business cycle, it may stoke fear of overheating and concerns over potential monetary tightening. As such, it might be difficult to get a consistent sign for macroeconomic variables in the model. To some extent, the SHARE variable may have captured the market response to macroeconomic news, although it may also reflect impact of other information such as corporate profits and liquidity conditions.

Concluding remarks and policy discussions

The renminbi exchange rate has been closely watched both domestically and internationally. Renminbi internationalisation has created another dimension of complexity in understanding renminbi movements as the offshore foreign exchange market has generated another set of spot and (deliverable) forward rates. Yet limited research has been undertaken to understand the deviations between the onshore and offshore renminbi rates and their implications. In filling in lacuna, this study examines to what extent market segmentations, fundamentals and global factors drive the differential. An extended GARCH model is used in modelling the differential, which allows analysis on these forces’ impact on the level and volatility of the differential. A great deal of attention is devoted to measuring segmentation. For this purpose, a series of policy dummies are compiled in order to gauge changes in institutional factors or the degree of restrictions on onshore trading, offshore trading and cross border renminbi flows.

The analysis suggests that among the three types of driving forces, market segmentation, or institutional factors play an important role. The conversion quota for renminbi trade settlement placed constraints at the early stage of the CNH market developments, and the incidences of breaching the quota caused sharp temporary widening of the CNH-CNY differential. On the positive side, the relaxation in policy constraints, be it the trading band in the onshore market, liquidity support to the offshore market, or cross border transactions in the renminbi, improves the market efficiency by reducing volatility of the differential. Fundamental factors, some financial and some economic, appear to mainly affect the level of the differential. Global factors have comparatively smaller impact, particularly when considered jointly with policy variables.

The empirical findings of this study underscore the role of policy in facilitating development of renminbi markets and the importance of further capital liberalisation. The pricing difference in onshore and offshore renminbi exchange rates is informative, given restrictions on onshore foreign exchange trading rules and barriers to cross border renminbi movements. The onshore rate is more reflective of policy intentions, while the offshore rate may capture better underlying demand and supply for the currency and global forces. Thus the differential of the two is the market solution for the current stage of capital account opening. Nonetheless, it is a solution for a less than optimal situation where inefficiency arises in the two markets. The inefficiency can take different forms: mis-reflection of underlying economic and market conditions in the onshore market, and greater vulnerability of the

26 WP

offshore market to liquidity shocks due to continuing liquidity constraints. Given the notable impact policy relaxation has already had, further capital account is expected to be a key driver to improve efficiency in both markets and narrow their pricing differential. Conceivably, over the longer term when two markets are fully integrated, the renminbi markets will evolve like those of the US dollar whereby the location of trading is immaterial, and the onshore and offshore rates will give consistent pricing signals.

The concern over the offshore market driving up onshore market volatility should not be overplayed, as liberalisation can enable the much deeper onshore market to anchor the exchange rate and better absorb volatility in both markets. Some increases in CNY volatility, particularly if triggered by changes in fundamentals, are desirable to the extent that this helps break expectations of one-way movement in the renminbi and dampen speculative flows based on these expectations. Yet, unwarranted rises in exchange rate volatility due to, for example, liquidity constraint in the offshore market, are not desirable. One such case was the more volatile movements in the CNY rate during the episode when the breach of the conversion quota in late 2010 led to a jump in the CNH volatility and a sharp widening of the CNH-CNY differential. Segmenting the two markets does not necessarily contain volatility in the offshore market. On the contrary, our results suggest that liberalisation measures can counter other factors, such as global investor sentiment, in reducing volatility of the CNH-CNY differential. The lower pricing uncertainty, as reflected by the narrower differential, could help decrease volatility in the renminbi foreign exchange markets. This is beneficial particularly when the uncertainties arise from market constraints and sharp swings in global factors not related to domestic economic fundamentals.

From the modelling perspective, alternative frameworks may be considered for enriching understanding of the CNH-CNY differential. It is suggested that the adjustment of the differential may have certain regularities, and it can differ depending on whether the CNH is at a premium or discount to the CNY. More generalised threshold GARCH models may be employed to capture these phenomena. Complex dynamics in onshore and offshore interaction may even call for further extensions of the GARCH framework such as the Markov switching GARCH model (see, eg. Chen, So and Lin, 2009) and double-threshold GARCH model (see, eg. Brooks, 2001, and Chen and So, 2006). These analytical alternatives may be considered in our future research.

References

Amihud Y., and H. Mendelson (1986) “Asset Pricing and the Bid-ask Spread”, Journal of Financial Economics 17, 223-249.

Andersen, T.G., Bollerslev, T., Diebold, F.X. and C. Vega (2003) “Micro Effects of Macro Announcements: Real-Time Price Discovery in Foreign Exchange”, American Economic Review 93, 38-62.

Andersen, T.G., Bollerslev, T., Diebold, F.X. and C. Vega (2007) “Real-Time Price Discovery in Stock, Bond and Foreign Exchange Markets”, Journal of International Economics 73, 251-77.

Balasubramaniam, V., Patnaik, I. and A. Shah (2011). “Who Cries about the Chinese Yuan?” National Institute of Public Finance and Policy, New Delhi, Working Paper 2011-89, May.

WP 27

Bollerslev, T. (1986) “Generalized Autoregressive Conditional Heteroscedasticity”, Journal of Econometrics 31, 307-327.

Bollerslev, T. and T. Wooldridge (1992) “Quasi-Maximum Likelihood Estimation and Inference in Dynamic Models with Time-Varying Covariances”, Econometric Reviews 11, 143-172.

Brockwell, P.J and R.A. Davis (2002) Introduction to Time Series and Forecasting, New York and Berlin (Springer).

Brooks, C. (2001) “A Double-Threshold GARCH Model for the French Franc/Deutschmark Exchange Rate”, Journal of Forecasting 20, 135-143.

Chen, C.W.S. and M.K.P. So (2006) “On a Threshold Heteroscedastic Model”, International Journal of Forecasting 22, 73-89.

Chen, C.W.S., So, M.K.P. and E.M.H. Lin (2009) “Volatility Forecasting with Double Markov Switching GARCH Models”, Journal of Forecasting 28, 681-697.

Chen, S.-W. and C.-H. Shen (2004) “GARCH, Jumps and Permanent and Transitory Components of Volatility: The Case of the Taiwan Exchange Rate”, Mathematics and Computers in Simulation 67, 201-216.

Cheung, Y. W. and D. Rime (2013) “The offshore renminbi exchange rate: microstructure and links with to the onshore market”. Presentation at the International Conference on Pacific Rim Economies and the Evolution of the International Monetary Architecture by the City University of Hong Kong and Journal of International Money and Finance.

Committee on the Global Financial System (2011) “Global liquidity – concept, measurement and policy implications”. CGFS Papers No. 45.

Craig, R.S., Hua, C., Ng, P. and R. Yuen, (2013) “Chinese Capital Account Liberalization and the Internationalization of the Renminbi”, IMF Working Paper WP/13/268.

Doornik, J.A. and M. Ooms (2008) “Multimodality in GARCH Regression Models”, International Journal of Forecasting 24, 432-448.

Engle, R.F. (1982) “Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of UK Inflation”, Econometrica, 50, 987-1007.

Faust, J., Rogers, J.H., Wang, S.-Y. and J.H. Wright (2007) “The high-Frequency Response of Exchange Rates and Interest Rates to Macroeconomic Announcements”, Journal of Monetary Economics 54, 1051-1068.

Fratzscher, M. and A. Mehl (2011)“China’s Dominance Hypothesis and the Emergence of a Tri-polar Global Currency System”, CEPR Discussion Paper No. 8671, November 2011.

Goyal, R., Jain, R. and T. Soumasree (2013) “Non deliverable forward and onshore India rupee market: a study on inter-linkeages”, Reserve Bank of India Working Paper No. 11/2013.

Hansen, P. and A. Lunde (2004) “A Forecast Comparison of Volatility Models: Does Anything Beat a GARCH(1,1) Model?”, Journal of Applied Econometrics 20, 873-889.

Henning, C. R. (2012) “Choice and Coercion in East Asian Exchange Rate Regimes”, Peterson Institute for International Economics Working Paper 12-15.

Huang, B.-Y., Chiou, J.-S. and P.S. Wu (2007) “Abnormal Profitability and Foreign Investment Based on the Investigation of Covered Interest Parity”, Physica A 384, 475-484.

IMF (2013) People´s Republic of China – Hong Kong Special Administrative Region, 2012 Article IV Consultation Discussions, Washington.

Lahaye, J., Laurent, S. and C. Neely (2011) “Jumps, Cojumps and Macro Announcements”, Journal of Applied Econometrics 26, 893-921.

28 WP

Levy, D., Bergen, M., Dutta, S. and R. Venables (1997) “The Magnitude of Menu Costs: Direct Evidence From Large U.S. Supermarket Chains”, Quarterly Journal of Economics 112, 791–825.

Li, K. F., Hui, C. H. and T.K. Chung (2012) “Determinants and dynamics of price disparity in onshore and offshore renminbi forward exchange rate markets”, Hong Kong Institute for Monetary Research Working Paper No. 24/2012.

Lumsdaine, R.L. and S. Ng (1999) “Testing for ARCH in the presence of a possibly misspecified conditional mean”, Journal of Econometrics 93, 257-279.

Maziad, S. and J.S. Kang (2012) “RMB Internationalization: Onshore/Offshore Links”, IMF Working Paper WP/12/133, Washington.

McCauley, R., Shu, C. and G.N. Ma (2014). “Non-deliverable forwards: 2013 and beyond.” BIS Quarterly Review, 2014 (March), 75-88.

Mikosch, T. and C. Starcia (2004) “Non-Stationarities in Financial Time Series, the Long-Range Dependence and the EGARCH Effects”, Review of Economics and Statistics 86, 378-384.

Minikin, R. and K. Lau (2013) The Offshore Renminbi, Singapore, John Wiley.

Pagan, A. (1984) “Econometric Issues in the Analysis of Regressions with Generated Regressors”, International Economic Review 25, 221-47.

Peng, W.S. Miao, H. and N. Chow (2007) “Price convergence between dual listed A and H shares”, Hong Kong Monetary Authority China Economic Issues, 6/07.

Shiller, R.J. (1981) “The Use of Volatility Measures in Assessing Market Efficiency” The Journal of Finance, 36, 291-304.

Shu, C., Chow, N. and J.-Y. Chan (2007) “Impact of the Renminbi Exchange Rate on Asian Currencies”, China Economic Issues No. 3/07, Hong Kong Monetary Authority January.

Subramanian, A. and M. Kessler (2012) “The Renminbi Bloc is here: Asia Down, Rest of the World to go?”, Peterson Institute for International Economics Working Paper 12-19.

Vlaar, P.J.G. and F.C. Palm (1993) “The Message in Weekly Exchange Rates in the European Monetary System: Mean, Reversion, Conditional Heteroscedasticity, and Jumps”, Journal of Business and Economic Statistics 11, 351-359.

Weiss, A.A. (1986) “Asymptotic Theory for ARCH Models: Estimation and Testing”, Econometric Theory 2, 107-131.

West, K.D. and D. Cho (1995) “The Predictive Ability of Several Models of Exchange Rate Volatility”, Journal of Econometrics 69, 367-391.

Wu, G. and C. Pei (2012) “Quantitative analysis of the pricing relationship between onshore and offshore renminbi exchange rates”. Fin