Trading Imbalances and the Relative Prices of Stock Pairs April 2011 Mark S. Seasholes HKUST AH Xsec Clark Liu HKUST © MSS 2011 Page 1 AH Premium Index • Clark and I recently had a short paper accepted at Economics Letters – There are 42 companies in the AH Index as of April 2009 – Pa is a share price in China (mainland) converted to US$ – Ph is a share price in Hong Kong converted to US$ WgtAvg Pt a 100 h – Shares have same dividend and voting rights WgtAvg Pt 250 200 150 100 50 AH Xsec AH Premium Index © MSS 2011 Parity Apr-2009 Jan-2009 Oct-2008 Jul-2008 Apr-2008 Jan-2008 Oct-2007 Jul-2007 Apr-2007 Jan-2007 Oct-2006 Jul-2006 Apr-2006 Jan-2006 0 Page 2 Jiangxi Copper • Below is a graph of the ratio of Jiangxi Copper’s relative stock prices – Shares have same dividend and voting rights • Two interesting points: a) High level of difference Pt a 100 b) High level of volatility Pt h 400 350 300 250 200 150 100 50 AH Xsec © MSS 2011 Apr-2009 Jan-2009 Oct-2008 Jul-2008 Apr-2008 Jan-2008 Oct-2007 Jul-2007 Apr-2007 Jan-2007 Oct-2006 Jul-2006 Apr-2006 Jan-2006 0 Page 3 Economic Questions AH Xsec © MSS 2011 Page 4 Economic Questions • Main questions: – Why do a given company’s shares, trading at two different locations, have different prices? – Can we changes explain changes in relative prices? – Can we link trading imbalances to these changes? – Can we estimate economic magnitudes that tell us about the amount of noise in a developed stock market (Hong Kong)? – Ultimately (not today) I hope to learn something about the relative importance of studying sentiment vs. frictions in markets? • Frictions: – This paper continues my work on market frictions, investor trading behavior, and asset prices. – In this paper, we study a short-sale constraint in one market combined with limited risk-bearing capacity in two different markets. AH Xsec © MSS 2011 Page 5 Institutional Background AH Xsec © MSS 2011 Page 6 China (Mainland) Companies • Chinese companies can have different share classes – “a” shares trade in China (mainland) and are originally for domestic shareholders • Listed on Shanghai or Shenzhen stock exchanges – “h” shares trade in Hong Kong • There are additional share classes: – “b” shares trade in the PRC and are originally for foreigners – ADRs trade in the USA – “n” shares trade on the NYSE or Nasdaq – “s” shares trade in Singapore • While the shares are technically different classes, many have the same dividend and voting rights. AH Xsec © MSS 2011 Page 7 Why Study a- and h-shares? • Nice institutional setting – Two classes of shares from the same company with same dividend and voting rights. – Shares are not fungible. – Relative share prices are tracked by investors and the financial media on a daily basis (Hang Seng publishes the “AH Premium Index”) – Shanghai and Hong Kong are in the same time zone. – No off-exchange trading on either exchange. – Investors read/write using Chinese characters (simple and traditional). – Many investors can trade on both exchanges. – Both a-shares and h-shares have high turnover. There are few worries about liquidity and illiquidity in this study. AH Xsec © MSS 2011 Page 8 Theoretical Framework AH Xsec © MSS 2011 Page 9 Theoretical Model • We model an economy with a single firm that has two claims to its dividends (i.e., two share classes). – The shares are equal in all respects except they trade on two separate markets and are not fungible. • There are four dates t = { 0, 1, 2, 3 } • There are two markets { a , h } generally shown as super-scripts • Risky assets pay the same dividend (i.e. for each share class): t N 0, t D3 D 1 2 3 • We denote the prices of the risky assets at date t as: P , P a t AH Xsec P3a P3h D3 h t © MSS 2011 Page 10 Theoretical Model (2) We use “X” to denote the holdings of a group on a given date t: • Market a has two groups of investors that trade only in that market: – { Xti(a) , Xth(a) } i(a) = informed; h(a) = noise traders • Market h has two groups of investors that trade only in that market: – {Xti(h) , Xth(h) } i(h) = informed; h(h) = noise traders • Final group consists of arbitrageurs who may hold shares in both markets – {Xta(a) , Xta(h) } • Short-sale constraint – The arbitrageurs holdings in market a are constrained to be nonnegative on all dates: Xta(a) ≥ 0 – Assumption captures: one of the markets has a short sale constraint AH Xsec © MSS 2011 Page 11 Theoretical Model (3) • Part of the dividend (t) is revealed to all investors at each date t • Agent’s maximize expected CARA utility: U[ W(3) ] = – elW(3) – To simplify expressions, we need to make an assumption about risk aversion coefficients: l = lia = 0.5lih = 0.5la • We can solve for prices and holdings of each group at each date using backwards induction – At each date, we need to consider whether or not the short-sale constraint is binding – We then need to consider the possibility that the short-sale constraint might bind in the future • Key point: Arbitrageurs drive prices so that expected returns (not prices) are equal across markets AH Xsec © MSS 2011 Page 12 Theoretical Model (4) • We can solve for prices and holdings of each group at t={0, 1, 2, 3} – Therefore, we solve for changes in prices (returns) – We also solve for changes in holdings (“order imbalances” or “OIB”) • The AH Premium at date t is defined as: AH Premt = Pta – Pth • Date AH Premium in Non-Binding Case t=1 l 12 32 X1h ( a ) X1h ( h) l 32 a2 2 a2 h2 h.o.t. AH Premium in Binding Case l 22 12 32 X1h ( a ) X1h ( h) l 32 a2 2 a2 h2 h.o.t. Arbitrageurs care about the possibility that the short-sale constraint might bind in the coming periods AH Xsec © MSS 2011 Page 13 Theoretical Model (5) • A disclaimer… • This is a CARA-normal model and we define returns and the AH Premium using price differences. Market conventions define both using ratios and we will too in our empirical analysis AH Premt = Pta – Pth AH Xsec © MSS 2011 Pi ,at AH Premi ,t h 100 Pi ,t Page 14 Numerical Analysis AH Xsec © MSS 2011 Page 15 Numerical Analysis • • • • • We draw one set of random numbers and then calculate prices and holdings at each date: t = { 0, 1, 2, 3 } Calc. returns, OIBs, and auto-corr. over intervals t = 0→1 and t = 1→2 We repeat the exercise 170 times to simulate having 170 weeks of data Numerical parameters are chosen to match empirical moments For example: DXh(a) ~N[0,0.0050] and DXh(h) ~N[0,0.0020] • Same assumption about risk tolerance: l = lia = 0.5lih = 0.5la • Holdings chosen to match steady-state equilibrium values: Group Stock a Stock h Noise Traders 50% 50% Informed Investors 50% 25% Arbitrageurs 0% 25% 100% 100% Total AH Xsec © MSS 2011 Page 16 Numerical Analysis (2) The model generates a number of predictions: #1 #2 #3 #4 #5 #6 #7 AH Xsec Mean Reversion: The AH Premium is mean reverting Skewness: The AH Premium is positively skewed Binding of Short Sale Constraint: - This implication is not testable (quantities are not observable) Volume and Order Imbalances - Volume in mkt a is more than double that of mkt h Cross-Market Return Correlations - Correlation of ra and rh is 0.64 Return-OIB Correlations - Correlation is stronger in mkt a than in mkt h Company-Level AH Premiums - Average level is proportional to fundamental volatility © MSS 2011 Page 17 Empirical Data AH Xsec © MSS 2011 Page 18 Sample Selection & Table I, Panel A • We choose the 43 companies that have been included in the AH Premium Index between Jan-2006 and Apr-2009 • Average market capitalization is USD 12.6 billion (median USD3.6 bn) Name Air China Anhui Conch Anhui Expressway Bank of China Bankcomm Beijing N Star CCB CHALCO China Coal China COSCO Average Median AH Xsec A Ticker H Ticker 601111 0753 600585 0914 600012 0995 601988 3988 601328 3328 601588 0588 601939 0939 601600 2600 601898 1898 601919 1919 Mkt Cap (US$ mil) 3,611 8,784 549 31,116 34,094 1,147 133,111 8,833 5,781 5,954 12,639 3,611 © MSS 2011 # of Wks Indus (GICS Codes) of Data Airlines 140 Construction Materials 173 Transp. Infrastructure 165 Commercial Banks 147 Commercial Banks 102 Real Estate Mgmt 132 Commercial Banks 82 Metals & Mining 104 Oil, Gas & Fuels 60 Marine 96 118 122 Page 19 Empirical Data • Paper shows weekly results (defined Wednesday-to-Wednesday) – 03-Jan-2006 to 30-Apr-2009 • Price and return data – Prices and returns from Datastream – All values (prices) are converted to USD • FX rates from Datastream Pi ,at AH Premi ,t h 100 Pi ,t • Volume and turnover data – Number of tradeable shares (free floats) are from Hang Seng – Volume data from Datastream AH Xsec © MSS 2011 Page 20 Figure 1: Cross-Section of AH Premiums • Below is a graph of the 25th, 50th, and 75th highest AH Premiums Pt a 100 Pt h 400 350 300 250 200 150 100 50 25th AH Xsec 50th © MSS 2011 75th Apr-2009 Jan-2009 Oct-2008 Jul-2008 Apr-2008 Jan-2008 Oct-2007 Jul-2007 Apr-2007 Jan-2007 Oct-2006 Jul-2006 Apr-2006 Jan-2006 0 Parity Page 21 Figure 2: AH Premiums of Three Companies • Below is a graph of the AH Premium for three companies Pt a 100 Pt h 400 350 300 250 200 150 100 50 China Ship Jiangxi Copper AH Xsec Apr-2009 Jan-2009 Oct-2008 Jul-2008 Apr-2008 Jan-2008 Oct-2007 Jul-2007 Apr-2007 Jan-2007 Oct-2006 Jul-2006 Apr-2006 Jan-2006 0 Shenzhen Expressway Parity © MSS 2011 Page 22 Table I, Panel B • Overview stats of price related variables Name Air China Anhui Conch Anhui Expressway Bank of China Bankcomm Beijing N Star CCB CHALCO China Coal China COSCO Average Median AH Xsec Avg AH Premi,t 206.6 104.9 126.0 142.6 125.0 291.0 124.3 219.0 146.7 184.0 Stdev AH Premi,t 65.1 16.8 26.3 28.3 22.5 70.5 19.8 45.1 29.1 46.2 AR(1) Coef 0.94 0.87 0.93 0.94 0.92 0.89 0.91 0.86 0.78 0.84 Frac w/ Pa ≤ Ph 4% 43% 24% 15% 24% 0% 9% 0% 2% 0% Corr (ra, rh) 0.56 0.60 0.15 0.43 0.57 0.54 0.61 0.51 0.52 0.50 182.1 153.6 39.5 36.6 0.84 0.88 8% 2% 0.47 0.50 © MSS 2011 Page 23 Empirical Data: OIB Order imbalance data • From Thomson Reuters Tick History (TRTH) database • 563 million plus 61 million trades Step 1: For each stock “i” and each day “k” and each market {a or h}, we calculate order imbalances. Winsorized at the 0.5% and 99.5% levels OIB a* i ,k Step 2: Standardize daily, order imbalances. Step 4: Calculate a weekly, companylevel measure AH Xsec Buyia,k Sellia,k OIB Shrsia,k a i ,k © MSS 2011 OIBia,t OIBia,k mean OIBia,k 11:k 70 stdev OIBia,k 11:k 70 kweek "t " OIBia,k* Page 24 Table I, Panel C • Overview stats of trading related variables Name Air China Anhui Conch Anhui Expressway Bank of China Bankcomm Beijing N Star CCB CHALCO China Coal China COSCO Average Median AH Xsec Free Float Mkt a 0.16 0.26 0.33 0.03 0.25 0.43 0.92 0.21 0.18 0.20 Free Float Mkt h 0.54 1.00 1.00 0.39 0.62 1.00 0.20 1.00 0.98 0.96 Avg Turn(a) 0.17 0.07 0.10 0.11 0.10 0.24 0.07 0.11 0.13 0.15 Avg Turn(h) 0.07 0.04 0.02 0.07 0.03 0.05 0.04 0.07 0.05 0.10 Corr (OIBa, OIBh) 0.20 0.12 0.09 0.07 -0.09 0.11 -0.25 0.12 -0.11 0.09 0.24 0.22 0.89 1.00 0.13 0.12 0.05 0.05 0.01 0.04 © MSS 2011 Page 25 State Space (Statistical) Model AH Xsec © MSS 2011 Page 26 State Space Model with Trading Variables • Assume observable prices are composed of two unobservable components – Component #1: “Efficient price” ( mi,t ) follows a process of uncorrelated increments with a non-zero drift ( di,t ) – Component #2: “Transitory price pressure” ( si,t ) assumed stationary • In this paper, a given company has two observable stock prices and one efficient (fundament) price. Think about Hong Kong prices only pia,t mi ,t sia,t c pih,t mi ,t sih,t mi ,t mi ,t 1 d i ,t wi ,t wi ,t aOIBia,t hOIBih,t ui ,t sia,t a sia,t 1 aOIBia,t ia,t sih,t h sih,t 1 hOIBih,t ih,t AH Xsec © MSS 2011 Page 27 State Space Model (2) • Estimated on a stock-by-stock basis – We don’t estimate coefs for two companies due to lack of data – Price of stock i is in logs – The “c” is used to estimate the average amount that Pa is above Ph – The ( di,t ) is the required return on the stock – Assume ( ui,t ) and ( i,t ) are normally distributed and independent • We use Kalman filter to get estimates of mi,t and si,t : – Estimation is implemented in Ox ; Kalman filter w/ ssfpack add-on AH Xsec © MSS 2011 Page 28 Parameter Estimates • • Results related to the efficient price equation a h u Param 0.0012 -0.0001 0.0288 Stderr (0.0005) (0.0004) (0.0021) Fundamental information appears to be incorporated into prices via trading in the a-share market AH Xsec © MSS 2011 Page 29 Parameter Estimates (2) • • • Results related to the transitory price equations a h a h (a) (h) Param 0.8561 0.8469 0.0056 0.0040 0.0580 0.0584 Stderr (0.0055) (0.0116) (0.0006) (0.0006) (0.0023) (0.0023) A first-order auto-correlation coefficient of 0.85 implies shocks have halflives of 4.27 weeks. The -coefficients indicate that order imbalances affect transitory prices in both the a- and h-markets. AH Xsec © MSS 2011 Page 30 Economic Magnitudes • • We multiply coefficients by the standard deviations of our trading variables to better understand economic magnitudes Effic. Eq b.p. a (OIBa) h (OIBh) (w) 34 -4 203 Trading explains only a small about of efficient price changes AH Xsec © MSS 2011 Page 31 Economic Magnitudes (2) • We multiply coefficients by the standard deviations of our trading variables to better understand economic magnitudes Trans Eq b.p. Trans Eq b.p. a (OIBa) (Dsa) 166 h (OIBh) (Dsh) 122 797 576 • A one standard deviation shock to OIBa is associated to a 166 bp move in transitory a-share prices at a weekly frequency • A one standard deviation shock to OIBh is associated to a 122 bp move in transitory h-share prices at a weekly frequency AH Xsec © MSS 2011 Page 32 Variance Decomposition • We decompose total variance to get more economic results pih,t mi ,t sih,t pih,t 1 mi ,t 1 sih,t 1 Dpih,t Dmi ,t Dsih,t 2 ri h,t 2 mi ,t 2 Dsih,t 2Cov mi ,t , Dsih,t 2(rh) 2(Dm) 2(Dsh) 2Cov(m,Ds) Average 100% 39.6% 45.6% 14.8% Stderr n.a. (6.2%) (7.1%) (2.3%) AH Xsec © MSS 2011 Page 33 Cross-Sectional Analysis • We test Implication #7, by regressing a company’s avg(AH Premi,t) over the sample period on the standard deviation of efficient price changes which we can calculate as (wi,t) – This makes use of the Kalman filter as we extract the time series of mi,t and use it to calculate (wi,t) Avg AH Premi,t b0 b1 (wi,t ) ei,t • The coefficient b1 is estimated to be 0.26 with a 4.92 t-statistic. AH Xsec © MSS 2011 Page 34 Conclusions AH Xsec © MSS 2011 Page 35 Our Conclusions • Trading imbalances are associated with cross-border price deviations. • A one standard deviation shock to trading imbalances in mainland China (or Hong Kong) is associated with 166 bp (or 122 bp) movement in transitory prices at a weekly frequency. • We estimate 45.6% of a Hong Kong stock’s total variance is due to transitory price movements. • The half-live of these movements is around 4.2 weeks. • This research helps focus future efforts on understanding where the trading imbalances are coming from. AH Xsec © MSS 2011 Page 36 Comments on Investor Sentiment AH Xsec © MSS 2011 Page 37 Investor Sentiment • Currently, a “hot” topic or theme in financial economics – What is an exact definition of investor sentiment? • Scenario #1: Investors in both markets have relatively homogeneous beliefs about a company’s prospects. Good news is released. – Stock prices in both markets rise – AH Premium is constant – Trading volume is low (everyone wants to buy at old prices) • Scenario #2: Investors in both markets have relatively homogeneous beliefs about a company’s prospects. A “wave of positive sentiment” hits investors in Shanghai. – Prices in Shanghai rise – AH Premium increases – Trading volume in Shanghai is low (everyone wants to buy) AH Xsec © MSS 2011 Page 38 Investor Sentiment (2) • Scenario #3: Investors in both markets have similar dispersion in beliefs about a company’s prospects. Good news is released. – Prices in both Shanghai and Hong Kong rise – AH Premium is constant to a first order – Trading volume in Shanghai and Hong Kong can be large as investors rebalance portfolios based on the new information. • Scenario #4: A “wave of positive sentiment” hits a subset of investors in Shanghai – Prices in Shanghai can rise – AH Premium can go up – Trading volume in Shanghai can be large as the sentiment prone investors buy shares from the other investors in Shanghai – Trading volume in Shanghai and Hong Kong should be uncorrelated AH Xsec © MSS 2011 Page 39 Extra Slides AH Xsec © MSS 2011 Page 40 AH Xsec AH Premium Index © MSS 2011 Jan-2011 Sep-2010 May-2010 Jan-2010 Sep-2009 May-2009 Jan-2009 Sep-2008 May-2008 Jan-2008 Sep-2007 May-2007 Jan-2007 Sep-2006 May-2006 Jan-2006 AH Premium Index • Updated graph through last week 250 200 150 100 50 0 Parity Page 41