B. Estimating integration of onshore and offshore renminbi markets This research differs from other papers in the literature in that it assesses the integration of renminbi offshore and onshore markets in that it applies a Threshold Autoregession model (TAR)model to estimate the no-arbitrage band and checks how much the basis is outside the band. Other studies, for example HKMA (2012); Ding, Tse, and williams(2012); Maziad and Kang(2012); and Wu and Pei(2012), assess renminbi onshore/offshore market linkages using Granger causality test or GARCH models. They focus on price discovery directions that within the band the basis follows a random walk as transaction costs make arbitrage and volatility spillovers between the two markets. Estimation in this paper exploits the fact unprofitable. In contrast, outside the band it follows an autoregressive process as arbitrage is profitable and moves the basis back towards the band. The estimated parameters of the autoregressive process indicate the speed of this convergence back to the band and will depend on the volume of the arbitraging capital flows between onshore and offshore markets possible under existing capital controls The data are differentials between the daily Cny and CNh(onshore and Figure 5. CNY and CNH Spot Exchange Rate Differentials offshore)dollar spot exchange rates The estimated width of the band - Spot exchange rate difference(CNY-CNH) 253 pips, roughly one quarter of a percentage point(Figure 5). Statistical tests confirm that the basis follows random walk within the band and an 500 autoregressive process outside it (Annex Table 3). The data start in nber 2010 when CNH first began to trade actively. Despite over Sources: Bloomberg LP, and IMF staff estimations. 600 observations the sample is 1 Band is estimated with the TAR model on the sample of 1 September 2010-31 January 2013. Around 56 percent of observations are within the band. relatively short and the estimated width of the band could be different for different sample periods, either for statistical easons or because transaction costs can change with institutional market reforms. An example of how differences in market infrastructure can lead to different band widths comes rom applying the TAr model to other onshore-off-shore exchange rate Table 1. tar model estimation results s. as is done for the differentials ependent Variable: CNY-CNH Results between the onshore three-month CNY weaker than CNH (pos basis) L5 percent of time forward rate and (i) the offshore Autoregressive coefficient three-month nondeliverable forward Implied "half life 25 days rate(NDF)and (ii) three-month CNY-CNH basis trades within band 56 percent of time deliverable forward rate in hong CNY stronger than CNH(neg basis) 29 percent of time Kong SAr, respectively, as shown in the annex Implied"half life Note: *** indicates significance at the l percent level. The technical ar The estimation results find limited provides full estimation results.6 B. Estimating Integration of Onshore and Offshore Renminbi Markets This research differs from other papers in the literature in that it assesses the integration of renminbi offshore and onshore markets in that it applies a Threshold Autoregession model (TAR) model to estimate the no-arbitrage band and checks how much the basis is outside the band. Other studies, for example HKMA (2012); Ding, Tse, and Williams (2012); Maziad and Kang (2012); and Wu and Pei (2012), assess renminbi onshore/offshore market linkages using Granger causality test or GARCH models. They focus on price discovery directions and volatility spillovers between the two markets. Estimation in this paper exploits the fact that within the band the basis follows a random walk as transaction costs make arbitrage unprofitable. In contrast, outside the band it follows an autoregressive process as arbitrage is profitable and moves the basis back towards the band. The estimated parameters of the autoregressive process indicate the speed of this convergence back to the band and will depend on the volume of the arbitraging capital flows between onshore and offshore markets possible under existing capital controls. The data are differentials between the daily CNY and CNH (onshore and offshore) dollar spot exchange rates. The estimated width of the band is 253 pips, roughly one quarter of a percentage point (Figure 5). Statistical tests confirm that the basis follows a random walk within the band and an autoregressive process outside it (Annex Table 3). The data start in September 2010 when CNH first began to trade actively. Despite over 600 observations, the sample is relatively short and the estimated width of the band could be different for different sample periods, either for statistical reasons or because transaction costs can change with institutional market reforms. An example of how differences in market infrastructure can lead to different band widths comes from applying the TAR model to other onshore-off-shore exchange rate pairs; as is done for the differentials between the onshore three-month forward rate and (i) the offshore three-month nondeliverable forward rate (NDF) and (ii) three-month deliverable forward rate in Hong Kong SAR, respectively, as shown in the Annex. The estimation results find limited -1500 -1000 -500 0 500 1000 1500 2000 -1500 -1000 -500 0 500 1000 1500 2000 Sep-10 Nov-10 Jan-11 Mar-11 May-11 Jul-11 Sep-11 Nov-11 Jan-12 Mar-12 May-12 Jul-12 Sep-12 Nov-12 Jan-13 Spot exchange rate difference (CNY-CNH) Band: (-32, 221) 1/ Sources: Bloomberg L.P.; and IMF staff estimations. 1/ Band is estimated with the TAR model on the sample of 1 September 2010‒31 January 2013. Around 56 percent of observations are within the band. Figure 5. CNY and CNH Spot Exchange Rate Differentials Dependent Variable: CNY-CNH Results CNY weaker than CNH (pos. basis) 15 percent of time Autoregressive coefficient 0.97*** Implied "half life" 25 days CNY-CNH basis trades within band 56 percent of time CNY stronger than CNH (neg. basis) 29 percent of time Autoregressive coefficient 0.88*** Implied "half life" 6 days Table 1. TAR Model Estimation Results Note: *** indicates significance at the 1 percent level. The technical annex provides full estimation results