Market Dislocation and the Dollar Liquidity Shortage after Lehman: Intra-day Microstructure Evidence from the RMB. Non-deliverable Forward Market

Transcription

1 Market Dislocation and the Dollar Liquidity Shortage after Lehman: Intra-day Microstructure Evidence from the RMB Non-deliverable Forward Market Eric Girardin 1 and Michael M. Hutchison 2 Draft May 5, Aix-Marseille University (AMSE), CNRS and EHESS: eric.grardin@univ-amu.fr 2 University of California, Santa Cruz: hutch@ucsc.edu. This paper was written when Michael Hutchison was visiting Aix Marseille School of Economics, whose support is gratefully acknowledged. 1

2 1. Introduction Turmoil in global financial markets increased markedly with the bankruptcy of Lehman Brothers on September 15th, An immediate announcement effect was evident in some markets, as the failure led to revisions in expectations about the solvency of many similar financial institutions. Within a few days, however, many other financial markets in Europe and around the world, seemingly less connected to U.S. sub-prime mortgage and derivative problems, also experienced increasing volatility and price declines. Foreign exchange and money markets were not exceptions, with initial liquidity problems for U.S. banks spreading to European banks and institutions around the world. A global dollar liquidity shortage is often cited as the reason for high volatility and dysfunction in many financial markets at the time, confounded by declining creditworthiness of U.S. and European financial institutions (Baba and Packer, 2009b). This global dollar shortage problem ultimately resulted in the Federal Reserve System extending numerous swap arrangements with foreign central banks. The offshore non-deliverable forward (NDF) market in China was no exception to these disruptions (Hutchison et al., 2012). A substantial gap arose between the forward discount on the offshore NDF market compared to the onshore forward market during the trading days immediately following the failure of Lehman, shown graphically in figure 1 for a daily time series ranging from January 2006 through December 2009 and, in figure 2, a daily time series for the month of September Both the onshore and offshore forward markets traded at about a 1% premium (annualized) in the days prior to Lehman, indicating modest expected RMB appreciation against USD. A few days after Lehman s failure, however, market prices sharply diverged with the onshore premium declining by less than a 100 basis points while the offshore premium declined by 500 basis points (to a 4% discount). This gap between the onshore and offshore markets in China is unprecedented, either before or since the global financial crisis period. This study investigates what occurred in the offshore RMB NDF market during the hours and days immediately prior and following the collapse of Lehman, using a unique intra-day high frequency data set. We consider trading dynamics during this period, investigating fiveminute intervals of intraday data on prices, volumes and bid-ask spreads, allowing us to focus on how the global financial crisis (GFC) shock was transmitted differentially to the NDF market in Hong Kong and onshore forward market in Shanghai. We are able to identify two distinct events. First, the immediate Lehman announcement effect reduced the offshoreonshore forward price gap, as the domestic forward rate appreciated more than the appreciation in the offshore NDF rate. This was followed two days later by the USD liquidity shortage effect: a sharp depreciation in the NDF and market dislocation between the offshore and onshore markets, widening bid-ask spreads and erratic trading volume. We argue in this paper that the dollar liquidity shortage primarily hit the Hong Kong NDF market, where market makers are large international commercial banks whose treasury departments were facing dollar liquidity shortages in their global activities. By contrast, the onshore forward price was much less affected by the GFC, although bid-ask spreads did rise significantly, 2

3 presumably because Chinese (largely) state-owned banks operating in this market were not significantly liquidity constrained. Our descriptive analysis of market dynamics and our formal empirical analysis during these days support this interpretation of events. The NDF market had only a few trades and little price movement at the time of the Lehman bankruptcy news on the 15th, as the market was virtually closed due to the Mid-Autumn national holiday in China. The onshore forward market was completely closed that day (no trades and no price movement). The Lehman announcement was evident at market opening on the 16 th for about one-half of trading, resulting in appreciation of the RMB (domestic and offshore) and narrowing of the price gap, while the bid-ask spread on the domestic forward market jumped. Trading volume was normal. No significant change in the gap occurred on the 17 th and the bid-ask spread on the onshore market fell back to normal. By the far the largest transmission of the global shock, which we identify as the global liquidity shortage, emerged at the start of trading on the 18th September, when sharp dislocations between the NDF and forward markets emerged and continued until trading was virtually suspended at noon. Normal market trading continued on the late afternoon of the 18 th and on the 19 th. This period experienced a doubling of bid-ask spreads and high volatility in the NDF market. However, much less turmoil was evident in the onshore forward market. Our empirical results, employing a simple regression model framework, clearly isolate a significant Lehman announcement distinct form the dollar liquidity shortage effect. Model estimates also indicate important roles for market liquidity, proxied by the bid-ask spread and trading volume. To our knowledge, this is the first intra-day evidence linking dollar liquidity shortages to market dislocations between the onshore and offshore foreign exchange market. The next section of the paper discusses the literature on dollar liquidity shortages and financial market dysfunction at the time of the GFC. Section 3 considers the RMB NDF market and presents a detailed descriptive analysis of the events in the offshore and onshore markets around the time of the Lehman collapse. Section 4 discusses our data. Section 5 presents the empirical results. Section 6 concludes. 2. Literature Review: Offshore Markets, CIP and Dollar Liquidity Shortage during the GFC Foreign exchange (FX) markets were not insulated from the GFC. What had begun as deterioration in a relatively limited segment of the US subprime mortgage sector quickly spread to other markets, especially those of credit and securitized products (BIS, 2008; IMF, 2008). Uncertainty about losses increased the liquidity needs of financial institutions as well as their reluctance to lend to each other in money markets. Baba et al. (2008, 2009a,b), in a series of papers, document heightened volatility in the FX swap markets across several G10 currency pairs beginning in the summer of 2007, arguing that dollar funding shortages of European financial institutions, combined with increased counterparty risks, were largely responsible for dislocations in the FX swap market prior to September Facing unfavorable demand and supply conditions and the associated impairment of liquidity in interbank markets, they argue that many European financial institutions moved to actively convert euros into dollars through FX swaps, creating a one-sided market as US counterparties became more cautious about lending dollars. They consider the short-term covered interest parity (CIP) condition between the US dollar and the 3

4 euro through the foreign exchange (FX) swap market. Sharp and persistent deviations from the CIP condition observed during the turmoil are found to be significantly associated with differences in counterparty risk between European and US financial institutions. They conclude that understanding the dislocations in the FX swap market is all the more important given the rapidly growing role of FX swaps in foreign currency funding by financial institutions globally. Furthermore, they find evidence that US-dollar term funding auctions by the ECB, supported by US dollar swap lines with the Federal Reserve, alleviated the level of dislocations, as well as the instability of the FX swap market. Coffey et al. (2009) also find evidence of dislocations in foreign exchange markets during 2007 and onwards. They document deviations from the Covered Interest Parity (CIP) with respect to different dollar interest rates and exchange rate pairs of the dollar vis-à-vis other currencies. The study shows that proxies for margin conditions and cost of capital are significant determinants of these CIP deviations or basis. They also argue, similar to other authors, that uncertainty about counterparty risk became a significant determinant of CIP deviations after the collapse of Lehman and that supply of dollars by the Federal Reserve to foreign central banks via reciprocal currency arrangements (swap lines) reduced CIP deviations. In particular, they find that the announcement on October that the swap lines would become unlimited reduced CIP deviations substantially. They conclude that a breakdown of arbitrage transactions in the international capital markets during the crisis was partly due to lack of funding and to heightened counterparty credit risk, and that central bank interventions helped to reduce the funding liquidity risk of global institutions. Several papers have focused specifically on RMB markets around the time of the Lehman bankruptcy. Maziad and Kang (2012) argue that rapid expansion of RMB trade settlement and issuance of RMB-denominated bonds by the Chinese government and corporates in Hong Kong have created some feedback channels across onshore RMB markets. They employ a bivariate GARCH model to understand the inter-linkages between onshore and offshore markets and find that, while developments in the onshore spot market exert an influence on the offshore spot market, offshore forward rates have a predictive impact on onshore forward rates. They also find evidence of volatility spillovers between the two markets. In addition, they find that spillover patterns have been different under different market conditions; a period of offshore market dislocation until the end of 2011, and a period of normal market conditions since then. Chung et al. (2013) also investigate price disparity measured as the absolute value of the difference between the renminbi onshore deliverable forward and offshore non-deliverable forward exchange rates. They look at the determinants of these price disparities and argue that different assessments of the path of interest rates can lead to a different valuation of onshore and offshore forward prices. In particular, they argue that capital controls and the complexity of central bank policy make it difficult to extract information about the monetary stance from the Chinese interest rate term structure. They postulate that investors make use of a pricing model to form their own assessment about the monetary stance. They argue that parameter uncertainty, in addition to market-wide aggregate uncertainty and illicit capital flows in China s balance of payments help to explain price disparities. Much of the above literature considers gaps between onshore and offshore forward rates, and resulting covered interest rate parity differentials, especially those emerging after the 4

5 Lehman bankruptcy. Several papers attribute these gaps to a dollar liquidity shortage. However, no paper of which we are aware has considered the detailed intra-day trading and market dynamics of the onshore and offshore markets around the time of Lehman. This is the focus of our research. 3. Data and NDF Markets An NDF market develops when the onshore forward markets are either not developed or have restricted access such as in the Chinese domestic market. These markets, which are located offshore that is, in financial centers outside the country of the restricted currency and involve contract settlement without delivery in the restricted currency, allow offshore agents with the restricted-currency exposures to hedge their exposures and speculators to take a position on the expected changes in exchange rates or exchange rate regimes. Also active in the NDF markets are arbitrageurs who have access to both forward markets. Volumes in the NDF market increase with increasing interest or investment in the currency and with increasing restrictions on convertibility. When currencies are fully convertible, NDF markets are not observed. 3 The renminbi Non-Deliverable Forwards (NDF) for RMB have been mainly traded in the over-the-counter market in Hong Kong since Sources of arbitrage between offshore NDF markets and onshore forward markets include onshore corporates operating in both Mainland and Hong Kong, since they can trade either in onshore or offshore markets, though there are still wide-ranging restrictions on repatriation of funds. While onshore and offshore banks are not allowed to trade in offshore and onshore markets, respectively, both can trade in the Non-Deliverable Forward market. The NDF market therefore offers opportunities of arbitraging for both onshore and offshore banks. 4 The source of the second-by-second data over the trading hours for September employed in our study is Thomson-Reuters. The data includes second-by-second bid and ask quotes of the three month non-deliverable and onshore forward RMB dollar exchange rates and the number of trades. The NDF and forward rates are taken as the midpoint of the bid and ask prices. For our empirical analysis, we compute the 5-minute average of the NDF and forward prices and the bid-ask spreads in both markets. This is done to reduce the microstructure noise present in the second-by-second sample. We compute the 5- minute sum for the second-by-second number of trades in the NDF and forward markets to obtain our trading volume numbers. 4. Financial market dysfunction in the RMB Non-deliverable forward market Figures 1 and 2 plot market developments for RMB NDF and onshore forward rates as well as forward discounts using daily data for the period and the month of September 2008, respectively, described in the introduction. However, onshore markets in China were closed on Monday, 15 th September, due to a national holiday, so the Lehman 3 Lipscomb (2005) provides a useful overview of NDF markets. 4 See Morgan Stanley (2011) for more detailed discussion ( CNH: Understanding the CNH and CNY Relationship. ). 5

6 announcement effect was delayed. Figure 3 shows developments on the th September. Prices are averaged over five minute intervals. The dashed vertical line in the graph shows the exact timing of the Lehman bankruptcy announcement (1:15pm September 15 th, Hong Kong time) and the solid vertical line shows market opening on the 16 th at 9:15am. The first half-hour of trading on the 16 th absorbed the immediate announcement effect, as this is the first time that both the onshore and offshore markets were open following the Lehman bankruptcy filing. An immediate appreciation of the RMB forwards, both onshore and offshore, is evident. In addition, the onshore-offshore price gap, whereby the NDF rate was systematically stronger than the onshore forward market (by about 0.3 percent) in the days before Lehman, dropped to almost zero. This movement occurred because the onshore forward rate appreciated more than the NDF rate after the Lehman announcement. Substantial volatility was noted through the day on the 16 th, but the two forward rates moved closely in tandem and no substantial price gap opened. No market dysfunction is evident. However, wider market movements in the days following the Lehman announcement are seen in Figure 4. In particular, intraday movements on NDF and forward prices (RMB per USD), and the price gap (equal to 100 times the natural log of the NDF/forward ratio) is shown in the figure for September 11-12th and 15-19th. (Markets were closed on the weekend, 13-14th September.) Figure 5 shows the bid-ask spread for trading in the NDF market (left-hand-scale) and the number of trades (right-hand-scale). The precise timing of the Lehman bankruptcy filing is again shown in the figures by a dashed vertical line. The solid vertical lines in the figures delineate market opening (9:15am) on the 16th and 18 th September, respectively. Table 1 shows summary statistics for key variables in these markets. Three distinct periods are evident from Figures 4 and 5 and Tables 1 and 2. The first period, covering the 11-12th September, shows the NDF market was consistently selling for a basis point premium above the domestic forward market, the bid-ask spread was low and stable at about ½ basis point on average, and trading was active (5 trades on average per five-minute interval). The onshore forward market had about twice the average volume and half the spread of the NDF market. The second distinct period, discussed in part above in the context of Figure 3, is framed by the days of the 16-17th September. The Lehman effect, delayed by market closures, was felt at market opening on the 16th with the NDF premium over the onshore forward dropping to almost zero and a rise in volatility (measured by standard deviation in Table 1). The NDF bid-ask spread increased slightly, and volume was initially light before returning to normal on the 17th. Although remaining low, bid-ask spreads doubled in the domestic forward market. Spreads declined and trading volume rose on the 17th, at levels similar to the days prior to Lehman. In sum, the markets absorbed the Lehman news easily with a noticeable price effect on the NDF market but otherwise normal trading activity. The third distinct period is framed by the 18-19th and characterized by substantial price movements, high volatility, and a widening bid-ask spread. From market opening on the 18th, NDF and forward prices moved in opposite directions, depreciating and appreciating respectively, opening a wide wedge between the two markets (with the NDF at a discount relative to onshore forward). The price gap went from near zero to a median value of 70 basis points and reached a maximum value of over 100 basis points during the trading day. Volatility in the market doubled, with the standard deviation rising to about 0.29 on the 18th 6

7 September. The bid-ask spread also jumped. The market ceased to be operating normally in the afternoon, as there was no price movement and the bid-ask spread was at a constant (low) level, from 12pm to 4:45pm, although some trades were recorded. Volatility continued when usual market activity resumed, with the bid-ask spread having doubled on average on the 19th compared to the average levels seen over the 11-17th. Trading volumes fluctuated somewhat in the NDF market but normal trading continued throughout the 19th. By contrast with the sharp depreciation in the NDF rate during 18-19th September, the onshore forward rate was virtually constant. The bid-ask spread increased markedly, but was nonetheless half that of the NDF market on average. Overall, despite some volatility rise, a much more tranquil market environment in the domestic market is evident form the figures and summary statistics reported in Table 1. This delayed effect on the 18th-19th is not a Lehman announcement effect which was evident on the 16 th. Rather, the sharp rise in the NDF/forward gap (relative depreciation of NDF) appears related to other factors, likely the world-wide dollar liquidity shortage emphasized by other researchers. It also does not appear related to a sudden shift in expectations of a substantial RMB depreciation as was the case during the large NDF movement during the Asian financial crisis in October 1997 (Zhang, 2004; pp ). In addition, no major changes in Chinese capital controls occurred at the time. 5. Empirical results Our descriptive analysis of the NDF market around the time of Lehman suggests that the market responded to a Lehman announcement effect, beginning at market opening on the 16 th September (Open 16 ) and continuing to be operative in the first half-hour of the 16th September (Lehman) as markets absorbed the news. A few days later, on the th September, the effects of the dollar liquidity shortage (USDShort) became evident in Hong Kong markets. In addition, we documented in section 4 large movements in the bid-ask spread (Spread) and trading volume (Volume) in the NDF market at the time of the dollar liquuidty shortage. We test these effects more formally in the context of a simple multivariate regression model. The dependent variable of the model is the absolute value of the (log) ratio of the NDF and forward prices, which we term the price gap (Gap). 5 We employ the absolute value the price gap in order to capture the separation or dislocation between the onshore and offshore markets the offshore rate may be trading at a premium or discount relative to the onshore rate and be measuring a given price gap. Our measure is appropriate as an indicator of market dislocation, as opposed to a measure of price differentials or deviations from covered interest parity (CIP). For example, the size and direction of deviations from CIP are important when measuring the effects of controls on the inflows or outflows of capital (e.g. 5 The absolute value of Log(NDF/FWD)x100, i.e. the absolute value of the percentage difference in offshore and onshore prices, is our measure expressed in levels. We also estimate the model in first differences. Chung et al. (2013) label this absolute value measure a price disparity between the ndf and forward markets. Similarly, Ma and McCauley (2008) examine for China the absolute value of the onshore and offshore yield differential. 7

8 Hutchison et al. 2012). In our case, however, we are interested in the conditions and variables that may influence the absolute magnitude of the price gap in either direction. To investigate these factors more formally, we estimate models that explain the price gap (Gap) as a linear function of the NDF ask-bid spread (Spread) and NDF trading volume (number of transactions, Volume). We expect a rise in the ask-bid spread to reflect greater market risk in the NDF market, as well as liquidity shortages that would lead to a larger price gap. A rise in trading volume, by contrast, reflects greater liquidity in the NDF market and should lead to a fall in the forward price gap. Beyond these market factors that may reflect the price gap during the entire sample period, we direct our attention to the Lehman announcement effect and the USD liquidity shortage effect. The total Lehman effect is measured by sum of two time dummy variables (Sum_16 th _Lehman) reflecting the initial five-minute interval at market opening on the 16 th September (9:15am to 9:20am, Open 16 ) and the first half-hour of trading (9:15am to 9:45am, Lehman) on that day. The total dollar shortage liquidity effect (Sum_Lehman) is measured by the sum of three time dummy variables: the immediate market opening time dummies (first five minutes of trading from 9:15am to 9:20am) for the 18 th (Open 18 ) and 19 th (Open 19 ) September, plus a time dummy variable reflecting all of the trading hours for 18 th -19 th September (USDShort). We also control for the immediate market opening (5-minute period) effect on the other days of our sample, as is standard in intra-day empirical work. We expect a negative coefficient for the total Lehman effect, as the NDF premium over the onshore forward falls (reflecting a larger appreciation in NDF than in the onshore market forward). The US dollar initially (and temporarily) depreciated virtually worldwide with the Lehman announcement, and in China this paradoxically had the effect of a large appreciation (fall of the RMB price of USD) of the onshore forward and more modest appreciation of the NDF. These movements reduced the NDF premium that had been a characteristic of the market prior to Lehman. Our strongest prior, however, is the expectation that the dollar shortage liquidity effect on the th sharply increases the price gap. Unit root tests, reported in Table 2, indicate that the absolute price gap is stationary (a unit root is rejected using conventional ADF, and the Phillips-Perron, tests) for the first four days of our sample period but not on the 18 th and 19 th. For this reason, we estimate the model both in levels and first-differences. (Tests of a unit root in the first-difference of the price gap are always rejected). We include a lagged dependent variable in both level and firstdifference specifications of the model to take into account autoregressive properties in the data. We examine a model either for the level (equation 1) or for the change in the absolute value (equation 2) of the price gap (Gap): (1) Gap t = α 0 + α 1 Gap t 1 + α 2 open it + α 3 Lehman t + α 4 USDShort t + α 5 Spread t + α 6 (USDShort Spread) t + α 7 Volume t + α 8 (USDShort Volume) t + μ t (2) Gap t = α 0 + α 1 Gap t 1 + α 2 open it + α 3 Lehman t + α 4 USDShort t + α 5 Spread t + α 6 (USDShort Spread) t + α 7 Volume t + α 8 (USDShort Volume) t + ε t 8

9 where Spread is the ask-bid spread in the NDF market, Volume is number of trades in the NDF market, Open it is a dummy for the first five minutes of trading on day i (i = 12th, 16 th, 17 th, 18 th, 19 th September) and USDShort is a dummy equal to unity on 18th and 19th September (and zero otherwise) to proxy for the dollar liquidity shortage. We also include interactive terms, USDShort x Spread and USDShort x Volume, in order to capture potential shifts in the influence of these variables on the price gap between the tranquil and the dollar shortage period. is the first difference operator and t is the current 5-minute interval time subscript. Linear least squares estimates for various iterations of the model with the forward price gap in levels are reported in Table 3. The models estimated in first differences are reported in Table 4. We use HAC-corrected standard errors in order to account for autocorrelation and heteroskedasticity in the residuals, using the Bartlett kernel, Newey-West fixed band-with method. The presence of heteroskedasticity is indicated by the summary statistics (standard deviations), reported in Table 2, and is confirmed by the Breuch-Pagan-Gordfey (BPG) heteroskedasticity test. Similarly, the presence of autocorrelation is indicated by the Ljung Box Q statistic reported in Tables 3 and 4 with 12 lags, reflecting a full hour of returns. The first column of Table 3 reports our baseline model specification. The next columns report estimates where the model is augmented to include Lehman (column 2) and both the Lehman and USD liquidity shortage effects (column 3). The fourth and fifth columns further augment the model with the spread and volume variables as well as interactive terms involving these variables. The total Lehman effect (Sum_16 th _Lehman) and the total USD liquidity shortage effect (Sum_USDShort) is reported at the bottom of the table, together with net effect of spreads (Sum_spread) and net effect of volume (Sum_volume) during the USD shortage period. All of the model results indicate a high degree of persistence when the model is estimated in level form (coefficients on the lagged dependent variable ranging from 0.94 to 0.98). The baseline model results indicate a significant negative Lehman announcement effect, between to percent, as well as a very strong and significant positive dollar liquidity shortage effect ranging from 0.19 to 0.23 percent. These effects are robust and highly significant (at the 1 percent level of confidence). Both effects are as expected and indicate that markets responded to the Lehman announcement by initially narrowing the onshoreoffshore price gap, followed in a few days by a very sharp rise in the gap reflecting dislocations associated with the dollar liquidity shortage We find that larger bid-ask spreads in the NDF market significantly decrease the price gap during tranquil periods (columns 4 and 5 of Table 3), with an estimated response ranging from -3.9 to However, no price gap response to the bid-ask spread is found during the liquidity shortage episode (i.e., Sum_Spread, reported in the lower panel of the table, is not significantly different from zero). By contrast, order volume does not appear to have an effect on the price gap during normal or tranquil periods, but is highly correlated with the price gap during the USD shortage episode. In particular, the interactive term and Sum_volume are negative and statistically significant at the 1 percent level of confidence. 9

10 These results indicate that important structural shifts occurred during the dollar liquidity shortage period. The models estimated in first differences are reported in Table 4. The results are robust in that large impacts of Lehman and the USD dollar liquidity period are suggested. As expected, the explanatory power of the models is reduced when estimated in first-difference form, with adjusted R-squares ranging from 0.10 to 0.24, but all models show highly statistically significant coefficents. No autocorrelation is noted in the residuals, as indicated by the Q statistic. The BPG test does not indicate any evidence of heteroskedasticity. The lagged dependent variable is highly significant but with much less persistence than previously, with coefficients ranging from to The Lehman effect (Sum_16 th _Lehman) is again highly significant in all model specifications at the 1% level of confidence. Similarly, the USD shortage effect (Sum_18 th _19 th ) is again positive, ranging from 0.11 to 0. 21, and highly significant (at the 1 percent level of confidence). (Note that, in first difference form, only the two market opening dummies are included in the estimates to capture the USD dollar liquidity shortage). The effect of the bid-ask spread is clearly evident during the liquidity crisis period, in contrast with the previous estimates. In particular, higher spreads appear to lower the price gap during the tranquil period (-5.8) but sharply increase the price gap during the liquidity shortage episode (the interactive terms are positive and highly significant; Sum_spread is -32 and significant at the 1 percent level). By contrast, in this specification, trading volume does not appear to have an effect on the price gap either during tranquil or the dollar shortage periods. 6. Conclusion The disruptions following the announcement of the Lehman bankruptcy were not limited to advanced countries markets. This paper documents that capital controls and ample foreign exchange reserves enabled China to be partly insulated from global dollar liquidity shortages. However, the Chinese offshore NDF market was disrupted and became decoupled from the onshore market. We argue that this is because market makers in the offshore NDF market, where the settlement currency is the USD, were large international banks subject to severe liquidity constraints several days after the Lehman bankruptcy. By contrast, the Chinese banks operating on the onshore forward market, largely state owned, were less liquidity constrained. Our contribution to the literature is a detailed descriptive and econometric evaluation of this episode based on an original intraday database during the days immediately preceding and following the Lehman event. Using this data, we are able to track and evaluate the precise timing and nature of the differential impact of that event on the offshore (NDF) and onshore (forward) exchange rate markets for the RMB. While it is evident that the scissor s like movement in the two forward rates is a major stylized fact, this actually occurred several days after the Lehman announcement. The immediate effect of Lehman was to appreciate both onshore and offshore markets, and reduce the price gap between the two. Even at the time of the dollar liquidity shortage, the onshore forward rate reflected an expected (small) appreciation of the RMB. By contrast, the NDF rate depreciated sharply, from a point reflecting a modest expected appreciation to apparently anticipating a substantial 10

11 depreciation of the Chinese currency. We argue that, contrary to what happened during the East Asian crisis, it is unlikely that the NDF market started expecting substantial depreciation for the RMB. Rather, the stylized facts and econometric analysis support the view that the sudden shift was attributable to a global dollar shortage that effectively disconnected the onshore and offshore forward markets. Offshore prices at this time were dominated by high volatility, a rise in the onshore-offshore price gap, widening bid-ask spreads, and erratic trading volume.these results and others reported above suggest that separating the Lehman announcement effect from the dollar liquidity shortage effect is critical in understanding the market dynamics involved in the financial transmission of the GFC worldwide. 11

12 References Baba, Naohiko and Frank Packer (2009a), Interpreting Deviations from Covered Interest Parity during the Financial Market Turmoil of , Journal of Banking & Finance Baba, Naohiko and Frank Packer (2009b), From Turmoil to Crisis: Dislocations in the FX Swap Market Before and After the Failure of Lehman Brothers, Journal of International Money and Finance Bank for International Settlements (2008). Annual Report. Chung, Tsz-Kin, Cho-Hoi Hui and Ka-Fai Li (2012), Determinants and Dynamics of Price Disparity in Onshore and Offshore Renminbi Forward Exchange Markets, Hong Kong Institute for Monetary Research Working Paper No. 24/2012 (October). Dickey, David A. and Wayne A. Fuller (1979), Distribution of the Estimators for Autoregressive Time Series with a Unit Root, Journal of the American Statistical Association, 74, p Coffey, Niall, Warren Hrung and Asani Sarkar (2009), Capital Constraints, Counterparty Risk and Deviations from Covered Interest Rate Parity Federal Reserve Bank of New York Staff Report No. 393 (October). Elliott, Graham, Thomas J. Rothenberg, and James H. Stock (1996), Efficient Tests for an Autoregressive Unit Root, Econometrica 64, Griffoli, Tommaso Mancini and Angelo Ranaldo (2011), Limits to Arbitrage During the Crisis: Funding Liquidity Constraints and Covered Interest Parity, manuscript. Swiss National Bank (February). Hutchison, Michael, Gurnain Pasricha, and Nirvikar Singh (2012), Effectiveness of Capital Controls in India: Evidence from the Offshore NDF Market, IMF Economic Review 60, International Monetary Fund (2008). Financial Stability Report. Kwaitkowski, Dennis, Peter C.B. Phillips, Peter Schmidt, and Yongcheol Shin (1992), Testing the Null Hypothesis of Stationarity against the Alternative of a Unit Root, Journal of Econometrics, 54, Ma., G. and R. N. McCauley (2008), Efficacy of China s capital controls: Evidence from price and flow data, Pacific Economic Review, 13(1), Maziad, Samar and Joong Shik Kang (2012), RMB Internationalization: Onshore/Offshore Links, IMF Working Paper WP/12/133 (May). Nelson, Daniel B. (1991), Conditional Heteroskedasticity in Asset Returns: A New Approach, Econometrica 59:

13 Phillips, Peter C.B and Pierre Perron (1988), "Testing for a Unit Root in Time Series Regression", Biometrika, 75, Zhang, Peter G. (2004), Chinese Yuan (Renminbi) Derivative Products World Scientific Publishing, Singapore. 13

14 Table 1 Summary Statistics on NDF-Forward Gap, Spreads and Trading Volume Full Sample Sample 11-12th 11th 12th 16th 17th 18th 19th 16-19th Sept Sept Sept Sept Sept Sept Observations log(ndf/fwd)*100 Mean Median Maximum Std. Dev NDF_ SPREAD Mean Median Maximum Std. Dev VOLUME_NDF Mean Median Maximum Std. Dev FWD_ SPREAD Mean Median Maximum Std. Dev VOLUME_FWD Mean Median Maximum Std. Dev

15 Table 2: Unit-root and stationarity tests th 11-12th 16-17th 18-19th a. ABS_NDF- FWD ADF * -4.67** Phillips Perron ** -4.45** KPSS Elliott-Rothenberg ** Stock b. NDF Spread ADF -3.27* -7.20*** -7.53*** Phillips Perron -7.07*** -6.81*** -7.63*** -4.96*** KPSS *** Elliott-Rothenberg- 6.12* 1.89*** 1.57*** 17.9 Stock c. NDF Volume ADF -9.42*** -6.50*** -7.54*** -4.75*** Phillips Perron *** -9.73*** -7.72*** -7.41*** KPSS 0.15** Elliott-Rothenberg- Stock 1.00*** 2.33*** 1.53*** 3.59*** ***Unit root is rejected at the 1% level (** : 5%; *: 10%). Source of critical values: Dickey et al. (1979), Elliott et al. (1996), Phillips et al. (1988). Stationarity rejected at the 1% level ( : 5%; : 10%). Source of critical values: Kwaitkowski et al. (1992). In all cases series are in levels and constant plus trend are allowed.unit root and stationarity tests were also applied to each series in first-difference form. In every case unit roots (stationarity) were also rejected (not rejected). 15

16 Table 3: Level estimation for Absolute value of NDF-forward gap (Gap) (1) (2) (3) (4) (5) C (1.82)* (1.62)* (1.87)* (2.42)*** (2.92)*** Gap(-1) (68.1)*** (68.2)*** (57.6)*** (59.9)*** (55.9)*** Open 12 TH (19.3)*** (18.9)*** (19.9)*** (11.1)*** (11.0) Open 16TH (33.4)*** (6.37)*** (6.26)*** (6.26)*** (6.13)*** Open 17TH (4.21)*** (3.97)*** (3.97)*** (0.43) (0.06) Open18TH (172.1)*** (167.3)*** (37.6)*** (37.5)*** (11.4)*** Open19TH (1.80)* (1.81)* (1.41) (0.42) (1.98)** Lehman (2.77)*** (2.85)*** (2.89)*** (2.97)** USDShort (1.91)** (0.07) (0.98) Spread (2.17)** (2.74)*** USDShort*SPREAD (1.06) (1.79)* Volume (0.75) USDShort*Volume (2.59)*** Adj. R-squared AIC Log likelihood Q(12) [0.00] [0.00] [0.00] [0.00] [0.00] BPG [0.00] [0.00] [0.00] [0.00] [0.00] Tests Sum_16 th _Lehman (31.9)*** (30.9)*** (27.0)*** (21.7)*** Sum_USDShort (16.9)*** (5.78)*** (5.44)*** Sum_Spread (0.51) (0.09) Sum_Volume (2.53)*** Dependent variable: Gap, as defined in text. HAC standard errors & covariance (Bartlett kernel. Newey-West fixed bandwidth = 6.0). t-statistics between brackets, significance: ***1%; **5%, and *10%. Q is the Ljung-Box test of the null of no residual autocorrelation, and BPG is the Breusch-Pagan-Godfrey test of the null of homoskedasticity. The last panel reports the sum of the noted coefficients and the Wald tests testing whether their sum equals zero. 16

17 Table 4: First difference estimation for change in absolute value of NDF-forward gap ( Gap) (1) (2) (3) (4) C (0.18) (0.40) (0.68) (0.68) Gap(-1) (2.74)*** (2.85)*** (2.07)** (2.04)** Open 12 TH (13.5)*** (13.3)*** (5.90)*** (5.90)*** Open 16TH (13.5)*** (6.16)*** (5.80)*** (5.73)*** Open17TH (1.14) (0.89) (0.07) (0.10) Open18TH (95.3)*** (95.3)*** (2.67)*** (2.86)*** Open19TH (19.2)*** (18.4)*** (0.76) (0.03) Lehman (2.69)*** (2.79)*** (2.77)*** Spread (1.82)* (1.85)* USDShort* Spread (3.89)*** (4.00)*** Volume (0.91) USDShort* Volume (1.25) Adj. R-squared AIC Log likelihood Q(12) [0.39] [0.34] [0.12] [0.19] BPG [0.98] [0.99] [0.97] [0.89] Tests Sum_18 th _19 th (46.4)*** (45.9)*** (2.31)*** (2.48)*** Sum_16 th _Lehman (13.4)*** (9.64)*** (9.93)*** Sum_ Spread (3.47)*** (3.59)*** Sum_ Volume (1.04) Dependent variable: Gap, as defined in text. HAC standard errors & covariance (Bartlett kernel. Newey-West fixed bandwidth = 6.0). t-statistics between brackets, significance: ***1%; **5%, and *10%. Q is the Ljung-Box test of the null of no residual autocorrelation, and BPG is the Breusch-Pagan-Godfrey test of the null of homoskedasticity. The last panel reports the sum of the noted coefficients and the Wald tests testing whether their sum equals zero. 17

18 Figure 1 RMB Offshore (NDF) and Onshore Forward Markets, I II III IV I II III IV I II III IV I II III IV NDF_DISCOUNT_AR FWD_DISCOUNT_AR 18

19 Figure 2 RMB Offshore (NDF) and Onshore Forward Market Discount, September M9 FWD_DISCOUNT_AR NDF_DISCOUNT_AR Figure 3 The Lehman Announcement: NDF and Forward Rates on September 15 th and 16 th :00 14:00 16:00 10:00 14:00 16:00 9/15 9/16 Log NDF Log FWD *Intraday data, 5-minute intervals, September 15-16, The dashed vertical line is the time of Lehman bankruptcy filing (September 15 at 1:15pm Hong Kong time) when the markets were closed in Shanghai. The solid vertical line is market opening on the 16 th at 9:15am. 19

20 Figure 4 NDF and Forward gap : Lehman and Liquidity Effects M9 Log NDF Log FWD *Intraday data, 5-minute intervals, from September and The dashed vertical line is the time of Lehman bankruptcy filing (September 15 at 1:15pm Hong Kong time) when the markets were closed in Shanghai. First solid vertical line is market opening on the 16 th at 9:15am. The second solid vertical line is market opening on 18 th September at 9:15am, the beginning of the dollar liquidity shortage phase. Note inactivity in the NDF market from noon to late afternoon on 18 th. 20

21 Figure 5 NDF Spreads and Trading Volume: Lehman and Liquidity Effects M9 Bid-Ask Spread Trading Volume *Intraday data, 5-minute intervals, from September and The dashed vertical line is the time of Lehman bankruptcy filing (September 15 at 1:15pm Hong Kong time) when the markets were closed in Shanghai. First solid vertical line is market opening on the 16 th at 9:15am. The second solid vertical line is market opening on 18 th September at 9:15am, the beginning of the dollar liquidity shortage phase. Note inactivity in the NDF market from noon to late afternoon on 18 th. 21

22 22