Motivation Empirical Study Summary Liquidity and the Value-at-Risk Lidan, Li 1 University 2 of Konstanz Webex Presentation, 10 Dec 2010 Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Outline 1 Motivation Fundamental Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies 2 Empirical Study Our Model Data Estimations Causality between liquidity and volatility Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Basic Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies Motivation Why care about liquidity? aects prices even when fundamental value remains constant (Amihud, Mendelson & Wood (1990) ) high liquidity improves market eciency (Amihud, Mendelson & Lauterbach (1997)) aggregate measures predictable, exhibit commonality (Chordia, Roll & Subrahmanyam (2001), Hasbrouck & Seppi (2001), Amihud (2002), Jones (2002), Huberman & Halka (2001), Korajczyk & Sadka (2007)) it is priced (Pastor & Stambaugh (2003), Acharya & Pedersen (2004, 2005), Sadka (2005)) experiences shocks (Huang (2003)) Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Basic Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies Problem Liquidity is not observable Transmission mechanism into price process is unclear. Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Basic Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies Market Liquidity Wikipedia: market liquidity is an asset's ability to be sold without causing a signicant movement in the price and with minimum loss of value Common denition: ability to buy or sell signicant quantities of a security quickly, anonymously, and with minimal or no price impact Price of immediacy Kyle (1985) 3 dimensions- tightness, depth, resiliency Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Basic Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies Reasons for market illiquidity asynchronous arrivals of buyers and sellers, role for market-makers bid-ask spread - inventory holding (Garman (1976), Stoll (1978), Amihud & Mendelson (1980), Ho & Stoll (1978)). But costs is found to be small! (Stoll (1978), George et. al. (1991), Madhavan & Smidt (1991)) Chacko et al. (2008) assumes zero inventory costs, shows that the asynchronous arrivals of buyers and sellers give market makers transitory pricing power Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Basic Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies Reasons for market illiquidity (II) adverse selection due to information asymmetry: Bagehot (1971) Kyle (1985), Glosten (1989), Easley & O'Hara (1987), Glosten & Harris (1988) show that the eect of asymmetric information is most likely captured by price impact of trades PIN type measures: No of speculators in the market: Grossman & Miller (1988), Brunnermeier & Pedersen (2008) Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Basic Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies Relationship between liquidity and volatility rst order eects of liquidity: expected stock returns ∝ illiquidity proxies (Amihud (1986), Brennan & Subrahmanyan (1996), Datat et. al. (1998), Easley et. al. ,Pastor & Stambaugh (2003),etc ) second order eects of liquidity Trading activity ∝liquidity, ∝-volatility (Tinic (1972), Benston & Hagerman (1974) ) Constantinides (1986) equilibrium model: nds that higher volatility corresponds to higher trading costs, and hence a wider no-trade zone. Modern automated auction markets: state of limit order book describes liquidity supply of asset (Biais, et al (1995), Handa & Schwartz (1996), Ahn et. al. (2001), Beltran et. al. (2005) ) Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Basic Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies Liquidity adjusted VaR VaR :Quantile forecast: model for forecasting volatility + method of computing quantiles Bangia et. al. (1999): LVaRt ,q = σt ,return Zq,return + σt ,spread Zq,spread Strong assumptions. Berkowitz (2000) : impact of transaction sizes on prices Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Basic Problem Denition of Liquidity Eects of Illiquidity Liquidity Proxies Liquidity Proxies Bid-ask spread type measures: inside bid ask spread, quote slope QS = log BS Depth Volume related measures: volume, number of transactions Price impact measures: Kyle (1985), Admanti & Peiderer (1988): ∆pt = λ qt + ψ[Dt − Dt −1 ] + yt , Hasbrouck (1991), Foster & Viswanathan (1993) : qt ∆pt n n = αp + ∑ βj ∆pt −j + ∑ γj qt −j + τt j =1 j =1 = αp + ψ[Dt − Dt −1 ] + λ τt + vt Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Joint distribution of returns and liquidity Angelidis & Benos (2006): decompose components of spread. LVaR has intraday U shape pattern LVaR should incorporate instantaneous eects of liquidity on returns: f (rt , lt |Ft −1 ) = f (rt |lt , Ft −1 )f (lt |Ft −1 ) p = −∞ fq (rt , lt |Ft −1 )dldr Estimating f (rt |liqt , Ft −1 )−Incorporating liquidity in conditional R LVaR R 1. density of returns Linear: Garch (1,1) : ht = ω + αεt2−1 + β ht −1 + γ liqt quantile garch: Qut (τ|Ft −1 ) = β0∗ + ∑pi=1 βi∗ Qut−i (τ|Ft −i −1 ) + ∑qj=1 γj∗ |ut −j | nonlinear : exponential garch: ln 2. Estimating ht = ω + αεt2−1 + β ln ht −1 + γ ln liqt f (liq |F Lidan, Li, Liquidity and the of Value at Risk states )-Dynamic modeling liquidity Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Data NYSE TAQ data ltered by rules adapted from Barndor-Nielsen et. al. (2008) 27 stocks 2005-2008: 1007 days. Table: descriptive statistics mean sdev skewness kurtosis JB stat aa ret -0.002219 0.023831 -1.132675 12.742556 4197 c ret -0.002926 0.030443 -2.705678 38.880062 55244 Lidan, Li, ibm ret 0.000744 0.013055 -0.443328 7.383169 839 aa qs 0.015206 0.006741 1.337592 5.824882 635 c qs 0.009254 0.007832 5.148196 39.97825 61821 Liquidity and the Value at Risk ibm qs 0.050768 0.049192 4.895691 39.052015 58557 Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Line graph Lidan, Li, Liquidity and the Value at Risk Our Model Data Estimations Causality between liquidity and volatility Motivation Empirical Study Summary standard Garch tests 25 out of 27 reject for homoskedasticity, 16 out of 27 reject test for linear garch Table: linear GARCH test pvalues aa aig axp ba bac c 1 0.11596215 4.6121941e-011 4.3572475e-005 0.21836067 5.1321147e-006 4.3712979e-005 5 4.6864754e-005 2.9456266e-015 1.8173076e-008 0.0063437401 1.4207275e-010 9.6559111e-007 10 3.3693466e-007 1.1832488e-034 1.0403632e-006 0.020689865 4.8448195e-012 1.405496e-006 Table: nonlinear GARCH test aa aig axp ba bac c SB test -1.179 0.8214 2.519 1.446 2.06 0.3814 pvals 0.238 0.411 0.011 0.148 0.039 0.7028 NSB test 1.96 -2.469 -2.820 -1.338 -4.877 6.081 pval 0.049 0.0135 0.0048 0.1806 1.07e-006 1.19e-009 Lidan, Li, PSB test 0.299 -7.562 -3.699 -0.653 -2.60 -0.8981 pval 0.764 3.963e-014 0.0002 0.513 0.009 0.369 Liquidity and the Value at Risk General 4.11 60.954 17.395 2.51 27.28 43.00 pval 0.249 3.675e-013 0.0005 0.473 5.13e-006 2.454e-009 Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Garch(1,1) estimations aa aig axp c dd gm hpq hon ibm jpm ω1 variance regressor-daily ave quote slope 0.000063 (0.334954) 0.001592 ( 1.959817) 0.003653 ( 3.068729) 0.004808 (4.295362) -0.001142 (-1.040142) 0.008403 (0.997375) 0.0019726 (1.038327) 0.004711 ( 1.540569) 0.003999 (3.991991) 0.003861 (3.889822) α1 0.077991 (3.562869) 0.182233 (4.026930) 0.111610 (4.091173) 0.183844 (5.665177) 0.082773 (2.406288) 0.079120 (1.764832) 0.039546 ( 2.397216) 0.076807 ( 2.663340) 0.101512 (4.030168) 0.129481 (4.010669) Lidan, Li, β1 0.920312 (36.081008) 0.832144 (24.215470) 0.882698 (29.541562) 0.816131 (27.672662) 0.881455 (13.315929) 0.921642 (21.122267) 0.898113 (24.128927) 0.864799 (16.760280) 0.842225 (19.832386) 0.867526 (26.715588) γ1 0.000478 (0.538177) 0.000078 (0.251349) -0.000226 (-0.408095) -0.001528 (-1.227071) 0.001141 ( 1.065728) -0.000614 (-0.405460) 0.002905 (1.409540) 0.001372 (1.311829) 0.000415 (1.506518) -0.000351 (-0.475982) Liquidity and the Value at Risk Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Shifting focus to extreme quantiles Longin (2000) and Bali (2000) : volatility measures based on asset return distributions cannot produce accurate measures of market risk during volatile periods. Allow for regressors to impact conditional tail quantiles dierently from central quantiles Koenker & Basset(1978) regression quantiles: least asymmetric absolute deviation problem Lidan, Li, Liquidity and the Value at Risk Our Model Data Estimations Causality between liquidity and volatility Motivation Empirical Study Summary Quantile Garch(1,1) Table: aa- quantile regression results τ 0.01 0.05 0.1 0.2 0.3 ω β α γ -0.032949683 -0.3133369 -2.2400412 -0.01384044 -0.013332437 -0.68619769 -0.99403384 -0.0073056359 -0.016070251 -0.18213752 -0.41068334 -0.0065702043 -0.011544173 -0.10697263 -0.26433129 -0.0044813663 -0.0059909927 -0.077650486 -0.19348258 -0.0025501655 0.4 qgarch (90 % -0.0046672071 -0.011825408 -0.026522968 -0.0017563458 Table: c- quantile regression results τ 0.01 0.05 0.1 0.2 0.3 ω β α γ -0.032607317 -1.2018977 -0.85429112 -0.016805871 -0.0067585568 -1.1193023 -1.0093918 -0.011114935 -0.0036311977 -0.74001279 -0.80873688 -0.0065269578 -0.0027424452 -0.49995051 -0.4553683 -0.0045542156 -0.0019580268 -0.29255192 -0.21614284 -0.0010507862 Lidan, Li, Liquidity and the Value at Risk 0.4 qgarch (90 % -0.0009154247 -0.17398816 -0.16251356 -0.0016848939 Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Exponential Garch (1,1) aig aa axp ba c ibm hpq c 1.081412 (0.724338) 1.210381 (2.386101) -0.053551 (-0.156410) 2.645763 (2.417373) 0.580475 (0.592333) 0.905157 (1.697931) 2.223891 (1.983677) ω α β γ -0.774965 (-3.336173) -0.176737 (-2.787408) 0.053638 (1.192938) -0.141037 (-2.178208) -0.075331 (-4.754929) -0.116812 (-3.560465) -0.119265 (-1.486515) 0.070650 (1.274415) 0.032776 (2.366915) 0.079636 (2.512507) 0.017978 (2.304863) 0.013632 (5.060816) 0.046812 (2.250848) 0.026403 (2.340134) 0.741077 (6.312925) 0.916418 (28.404282) 0.847086 (13.629401) 0.959246 (41.845684) 0.985397 (409.584020) 0.893339 (16.699990) 0.914802 (33.813713) -0.374399 (-2.245243) -0.076850 (-2.316936) 0.424063 (2.311409) 0.045636 (0.896446) -0.045602 (-8.114259) 0.018581 (0.185414) 0.080283 (1.927078) Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Granger Causalities between liquidity and volatility out of 27 stocks, 22 reject for nongranger causality bet liq and realized vol liq to realized vol realized vol to liq lags 5 10 15 20 5 10 15 20 aa 25.7680 18.4836 14.5109 12.7633 33.8057 25.8839 21.0978 18.0936 aig 56.2347 42.6982 34.5208 30.3883 52.9909 38.7367 31.8736 27.1626 axp 10.2775 6.4562 4.83922 3.9788 11.0912 10.6735 10.9238 10.3448 ba 14.4012 10.8708 9.78064 8.1009 14.2308 10.6756 10.2098 8.48768 c 9.8891 6.7312 5.6338 6.8044 14.2877 11.5201 10.2247 10.1749 cat 0.8948 0.5821 0.2083 0.62697 0.0044 -0.4372 -0.1587 -0.3014 dd 9.9097 8.4811 7.5559 7.4770 16.5684 14.5606 12.6370 11.5005 dis 11.3784 7.92746 7.3078 6.1666 12.7966 11.3234 10.9793 9.4725 ge 2.2663 1.6180 1.4740 4.2036 4.8376 3.1967 2.2948 1.2522 gm 0.5730 1.2118 1.3306 2.4816 1.8345 1.0424 0.2204 -0.0610 Lidan, Li, Liquidity and the Value at Risk Our Model Data Estimations Causality between liquidity and volatility Motivation Empirical Study Summary Granger causalities in tails Hong et. al. (2009): tests for Granger causality in risk. H10 : P (Y1t H1A : P (Y1t Risk indicator: < −V1t |I1,t −1 ) = P (Y1t < −V1t |I1,t −1 , I2,t −1 ) < −V1t |I1,t −1 ) 6= P (Y1t < −V1t |I1,t −1 , I2,t −1 ) Z1t ≡ 1(Y1t < −V1t ). Investigate cross-spectrum between Z1t , Z2t ,f (ω) ≡ 21π ∑∞j =−∞ ρ(j )e −ij ω , ω ∈ [−π, π], i = H10 : f10 (ω) ≡ 0 1 2π ∑ j =−∞ Lidan, Li, √ −1 ρ(j )e −ij ω , ω ∈ [−π, π] Liquidity and the Value at Risk Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Granger Causality in tails (2) Using kernels for downweighting lags: fˆ(ω) ≡ 21π ∑jT=−11−T k (j /M )ρ̂(j )e −ij ω Comparing: L2 (fˆ, fˆ10 ) ≡ 2π Rπ fˆ(ω) − fˆ10 (ω)|2 d ω = ∑Tj =−11 k 2 (j /M )ρ̂ 2 (j ). −π | Test statistic: Q1 (M ) ≡ [T ∑Tj =−11 k 2 (j /M )ρ̂ 2 (j ) − C1T (M )]/D1T (M )1/2 Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Our Model Data Estimations Causality between liquidity and volatility Granger causality in tails Qtrun Qreg lags 15 20 5 10 15 20 aa 0.01 -2.7385391 -3.1622076 2.2121 8.2582 10.347 10.234 aa 0.05 -2.7385424 -3.162197 2.9476 3.3150 4.1031 3.6476 aa 0.10 -2.7385339 -3.1621925 1.5139 1.7369 1.1432 1.3408 aa 0.15 -2.7385242 -3.1621837 4.2837 2.8853 1.9985 1.3405 aa 0.20 -2.7385572 -3.1622173 2.1726 2.7303 1.4026 1.0893 aa 0.25 -2.738533 -3.1621903 2.9063 2.5707 2.0531 1.7652 c 0.01 -2.7385973 -3.1622637 0.9503 4.9678 3.3964 2.3592 c 0.05 -2.7385925 -3.1622525 1.1715 -0.2205 1.5387 0.8250 c 0.10 -2.7386058 -3.1622708 -0.8000 -1.296 -1.2263 -0.5628 c 0.15 -2.7386072 -3.1622727 2.3162 1.388 1.3304 0.5815 c 0.20 -2.7386036 -3.1622693 1.1267 0.9242 0.4232 0.3453 c 0.25 -2.738607 -3.1622721 0.7047 0.0710 -0.3740 -0.6736 ibm 0.01 -2.7386114 -3.1622761 -1.500 -2.146 -2.645 -3.060 ibm 0.05 -2.7385247 -3.1621989 6.158 4.489 3.546 3.466 ibm 0.10 -2.7385326 -3.1622048 3.267 2.611 2.406 3.554 ibm 0.15 -2.7385238 -3.1621834 3.064 3.327 2.156 2.931 ibm 0.20 -2.7385331 -3.1621874 3.879 3.821 3.359 2.92 ibm 0.25 -2.7385434 -3.1622023 1.143 3.959 3.406 2.763 Lidan, Li, Liquidity and the Value at Risk Motivation Empirical Study Summary Summary Causality of liquidity in the means and tail dier Direct incorporation of liquidity measure into tails is more robust (no distributional assumption) Outlook testing for instantaneous causality dynamic model for liquidity states dierent liquidity proxies Lidan, Li, Liquidity and the Value at Risk