Tactical Asset Allocation 2 session 6 Andrei Simonov Tactical Asset Allocation 1 3/22/2016 Agenda Statistical properties of volatility. – Persistence – Clustering – Fat tails Is covariance matrix constant? Predictive methodologies – Macroecon variables – Modelling volatility process: GARCH process and related methodologies – Volume – Chaos Skewness Tactical Asset Allocation 2 3/22/2016 Volatility is persistent Returns2 are MORE autocorrelated than returns themselves. Volatility is indeed persistent. Akgiray, JB89 Tactical Asset Allocation 3 3/22/2016 It is persistent for different holding periods and asset classes LAG FT All Share Daily r Daily |r| Weekly r Weekly |r| GB£/US$ Daily r Daily |r| 1 0.193 0.347 2 0.020 0.292 3 0.031 0.256 4 0.045 0.260 5 0.018 0.222 10 0.070 0.225 0.086 0.255 0.144 0.192 0.021 0.195 0.076 0.210 -0.052 0.207 -0.037 0.137 -0.022 0.133 -0.003 0.120 -0.008 0.068 -0.006 0.052 0.024 0.114 0.002 0.076 Sources: Hsien JBES(1989), Taylor&Poon, JFB92 Tactical Asset Allocation 4 3/22/2016 Volatility Clustering, rt=ln(St/St-1). US$/SEK ContCompRate 0.015 0.01 0.005 0 -0.005 -0.01 -0.015 198702-02 198802-02 198902-02 199002-02 Tactical Asset Allocation 199102-02 199202-02 199302-02 199402-02 199502-02 199602-02 199702-02 199802-02 199902-02 200002-02 200102-02 5 3/22/2016 Volatility clustering S&P100 ContComp Returns 0.04 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 feb87 feb88 feb89 feb90 Tactical Asset Allocation feb91 feb92 feb93 feb94 feb95 feb96 feb97 feb98 feb99 feb00 feb01 6 3/22/2016 Kurtosis & Normal distribution 4 2 x j x n(n 1) 3 ( n 1 ) KURTOSIS x) (n 2)( n 3) (n 1)( n 2)( n 3) std (Kurtosis Kurtosis= 0 for normal dist. If it is positive, there are so-called FAT TAILS 3.46 3.5 2.65 3 2.5 1.96 2 1.93 1.68 1.5 1 0.5 Tactical Asset Allocation 0 US $TO UK £ NOON NY US $ TO SWEDISH KRONA (JPM) US $ TO SWISS FRANC (JPM) S&P 100 MSCI SWEDEN 7 3/22/2016 us t A ralia us Be tri l a Ca gium D na en da m Fi ark n Fr land G an H erm ce on a g ny K Ire ong la n Ita d N J ly N ethe apa ew rl n Ze and a s N l and o Po rwa rtu y g Sp al Sw Sw ain it z ede er n la nd U K W U or W S ld or ex ld EAUS FE A Higher Moments & Expected Returns Average Excess Kurtosis in Developed Markets 6 5 4 3 2 1 0 -1 Data through June 2002 Tactical Asset Allocation 8 3/22/2016 rg e Banti n hr a Br a in a Chz il Cz i e c Co Chi l e l h om na Re b pu ia b Eg l ic G yp H ree t un c e ga In In ry do di ne a Is sia Jo rael rd M Ko an ala re a M ysi a M ex or ic o o N cco ig er Pa Om i a ki an Ph sta n ili Pe pp ru i Po nes Sa R lan ud u d i A ssi So Sl rab a ut ov i a h a Sr Afr ki a i L ica Ta a nk Th iw a a an Tuila n V rk d e Zi nez e y Comba uela m bw po e sit e A Higher Moments & Expected Returns Average Excess Kurtosis in Emerging Markets 6 5 4 3 2 1 0 -1 Tactical Asset Allocation Data through June 2002 9 3/22/2016 Extreme events Tactical Asset Allocation 10 3/22/2016 Normal distribution: Only 1 observation in 15800 should be outside of 4 standard deviations band from the mean. Historicaly observed: – 1 in 293 for stock returns (S&P) – 1 in 138 for metals – 1 in 156 for agricultural futures Tactical Asset Allocation 11 3/22/2016 What do we know about returns? Returns are NOT predictable (martingale property) Absolute value of returns and squared returns are strongly serially correlated and not iid. Kurtosis>0, thus,returns are not normally distributed and have fat tails -’ve skewness is observed for asset returns Tactical Asset Allocation 12 3/22/2016 ARCH(1) volatility at time t is a function of volatility at time t-1 and the square of of the unexpected change of security price at t-1. Ret(t)=f Ret(t-1)+et e(t)= s(t) z(t) s2(t)=a0+a1e2(t-1), z~N(0, 1) If volatility at t is high(low), volatility at t+1 will be high(low) as well Greater a1 corresponds to more persistency Tactical Asset Allocation 13 3/22/2016 Simulating ARCH vs Normal 1 ARCH(4) 0.8 0.6 0.4 0.2 ARCH(1) -0.2 493 481 469 457 445 433 421 409 397 385 373 361 349 337 325 313 301 289 277 265 253 241 229 217 205 193 181 169 157 145 133 121 109 97 85 73 61 49 37 25 13 1 0 Normal -0.4 Tactical Asset Allocation Normal ARCH(1) ARCH(4) 14 3/22/2016 GARCH=Generalized Autoregressive Heteroskedasticity volatility at time t is a function of volatility at time t-1 and the square of of the unexpected change of security price at t-1. s2(t)=a0+b1 s2(t-1)+ a1e2(t-1), e~N(0, s2t) If volatility at t is high(low), volatility at t+1 will be high(low) as well The greater b, the more gradual the fluctuations of volatility are over time Greater a1 corresponds to more rapid changes in volatility Tactical Asset Allocation 15 3/22/2016 Conditional and Unconditional Swedish 360days T-Bill Volatility 1.4 Standard Deviation 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1992-02-10 1994-01-10 1995-12-11 1997-11-10 1999-10-11 2001-09-10 Tactical Asset Allocation 16 3/22/2016 S&P return volatility 0.16 0.14 0.12 0.10 Conditional Unconditional 0.08 0.06 0.04 0.02 Tactical Asset Allocation feb-01 feb-00 feb-99 feb-98 feb-97 feb-96 feb-95 feb-94 feb-93 feb-92 feb-91 feb-90 feb-89 feb-88 feb-87 feb-86 feb-85 feb-84 feb-83 feb-82 0.00 Moving Av 17 3/22/2016 Persistence If (a1+b)>1, then the shock is persistent (i.e., they accumulate). If (a1+b)<1, then the shock is transitory and will decay over time For S&P500 (a1+b)=0.841, then in 1 month only 0.8414=0.5 of volatility shock will remain, in 6 month only 0.01 will remain Those estimates went down from 1980-es (in 1988 Chow estimated (a1+b)=0.986 Tactical Asset Allocation 18 3/22/2016 Forecasting power GARCH forecast is far better then other forecasts Difference is larger over high volatility periods Still, all forecasts are not very precise (MAPE>30%) xGARCH industry Tactical Asset Allocation 19 3/22/2016 Options’ implied volatilities Are option implicit volatilities informative on future realized volatilities? YES If so, are they an unbiased estimate of future volatilities? NO Can they be beaten by statistical models of volatility behavior (such as GARCH)? I.e. does one provide information on top of the information provided by the other? – Lamoureux and Lastrapes: ht = w + ae2t-1 + bh t-1 + gsimplied – They find g significant. Tactical Asset Allocation 20 3/22/2016 Which method is better? (credit due: Poon & Granger, JEL 2003) Number % of of studies studies HISTVOL> GARCH GARCH> HISTVOL 22 17 56% 44% HISVOL> ImpVol ImpVol > HISVOL 8 26 24% 76% GARCH> ImpVol ImpVol > GARCH 1 17 6% 94% Tactical Asset Allocation 21 3/22/2016 Straddles: a way to trade on volatility forecast Straddles delivers profit if stock price is moving outside the normal range If model predicts higher volatility, buy straddle. If model predicts lower volatility, sell straddle Tactical Asset Allocation Profit X ST 22 3/22/2016 Volatility and Trade Lamoureux and Lastrapes: Putting volume in the GARCH equiation, makes ARCH effects disappear. ht = w + ae2t-1 + bh t-1 + gVolume GARCH11 GARCH11+Vol Mean Median Mean Median Alfa1 0.102 0.066 0.057 0.037 Beta 0.626 0.716 0.016 0.000 Gamma 0.671 0.668 Beta+Alfa1 0.728 0.782 0.073 0.037 Heteroscedastisity is (at least, partially) due to the information arrival and incorporation of this information into prices. Processing of information matters! Tactical Asset Allocation 23 3/22/2016 What else matters? Macroeconomy US Div Yield (t-1) Unconditional 0.08 Conditional-"less than mean" 0.07 Conditional-"more than mean" 0.06 0.05 0.04 0.03 0.02 0.01 0 USA AUSTRALIA Tactical Asset Allocation CANADA GERMANY JAPAN U.K. 24 3/22/2016 Macroeconomic variables (2) Volatility condit on US Ret(t-1) and US yield curve(t-1) 0.1 Unconditional Conditional-"less than mean" 0.09 Conditional-"more than mean" 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 USA AUSTRALIA Tactical Asset Allocation CANADA GERMANY JAPAN U.K. 25 3/22/2016 Stock returns and the business cycle: Expansion Tactical Asset Allocation USA UK Sw itzerland Sw eden Spain Singapore Norw ay New Zealand Netherlands Japan Italy Ireland Hong Kong Germ any France Finland Denm ark Canada Belgium Austria Australia EAFE 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 AC World Volatility NBER Expansions and Contractions January 1970-March 1997 Contraction 26 3/22/2016 Predicting Correlations (1) Crucial for VaR Crucial for Portfolio Management – Stock markets crash together in 87 (Roll) and again in 98... – Correlations varies widely with time, thus, opportunities for diversification (Harvey et al., FAJ 94) Tactical Asset Allocation 27 3/22/2016 Predicting Correlations (2) Correlation M atrix --- Conditioned on: US Return (t-1) & US Yield Curve (t-1) Use “usual suspects” to predict correlations Simple approach “upup” vs. “downdown” Remove Black Monday ? yes USA AUSTRALIA CANADA GERMANYJAPAN U.K. 1) Unconditional 0.40 0.69 0.31 0.24 0.47 2) Conditional-"less than mean" 0.61 0.79 0.37 0.35 0.56 3) Conditional-"more than mean" 0.19 0.60 0.28 0.18 0.26 Australia CANADA GERMANYJAPAN U.K. 1) Unconditional 0.55 0.24 0.27 0.44 2) Conditional-"less than mean" 0.65 0.40 0.40 0.63 3) Conditional-"more than mean" 0.46 0.06 0.21 0.41 CANADA GERMANYJAPAN U.K. 1) Unconditional 0.27 0.26 0.49 2) Conditional-"less than mean" 0.37 0.40 0.60 3) Conditional-"more than mean" 0.29 0.24 0.41 GERMANY JAPAN U.K. 1) Unconditional 0.36 0.40 2) Conditional-"less than mean" 0.44 0.53 3) Conditional-"more than mean" 0.33 0.26 JAPAN U.K. 1) Unconditional 0.35 2) Conditional-"less than mean" 0.44 3) Conditional-"more than mean" 0.32 Tactical Asset Allocation 28 3/22/2016 Predicting Correlations (3) Tactical Asset Allocation 29 3/22/2016 Chaos as alternative to stochastic modeling Chaos in deterministic non-linear dynamic system that can produce random-looking results Feedback systems, x(t)=f(x(t-1), x(t-2)...) Critical levels: if x(t) exceeds x0, the system can start behaving differently (line 1929, 1987, 1989, etc.) The attractiveness of chaotic dynamics is in its ability to generate large movements which appear to be random with greater frequency than linear models (Noah effect) Long memory of the process (Joseph effect) Tactical Asset Allocation 30 3/22/2016 Example: logistic eq. X(t+1)=4ax(t)(1-x(t)) 0.8 0.7 0.6 0.5 0.4 0.3 a=0.5 a=0.75 0.2 0.1 0 1 6 11 Tactical Asset Allocation 16 21 26 31 36 41 46 31 3/22/2016 1 0.9 0.8 0.7 0.6 A=0.9 0.5 0.4 0.3 0.2 0.1 0 1 1 6 11 16 21 26 31 36 41 46 1 6 11 16 21 26 31 36 41 46 0.9 0.8 0.7 0.6 A=0.95 0.5 0.4 0.3 0.2 0.1 0 Tactical Asset Allocation 32 3/22/2016 Hurst Exponent Var(X(t)-X(0)) t2H H=1/2 corresponds to “normal” Brownian motion H<(>)1/2 – indicates negative (positive) correlations of increments For financial markets (Jan 63-Dec89, monthly returns): IBM 0.72 Coca-Cola 0.70 Texas State Utility 0.54 S&P500 0.78 MSCI UK 0.68 Japanese Yen 0.64 UK £ 0.50 Tactical Asset Allocation 33 3/22/2016 Long Memory Memory cannot last forever. Length of memory is finite. For financial markets (Jan 63-Dec89, monthly returns): IBM 18 month Coca-Cola 42 Texas State Utility 90 S&P500 48 MSCI UK 30 Industries with high level of innovation have short cycle (but high H) “Boring” industries have long cycle (but H close to 0.5) Cycle length matches the one for US industrial production Most of predictions of chaos models can be generated by stochastic models. It is econometrically impossible to distinguish between the two. Tactical Asset Allocation 34 3/22/2016 Correlations and Volatility: Predictable. Important in asset management Can be used in building dynamic trading strategy (“vol trading”) Correlation forecasting is of somewhat limited importance in “classical TAA”, difference with static returns is rather small. Pecking order: expected returns, volatility, everything else… Good model: EGARCH with a lot of dummies Tactical Asset Allocation 35 3/22/2016 Smile please! Black- Scholes implied volatilities (01.04.92) Tactical Asset Allocation 36 3/22/2016 us t A ralia u Be stri lg a Ca ium D na en da m Fi ark nl Fr and G a H erm nce on a g ny K Ire ong la n Ita d N J ly N ethe apa ew rl n Ze and a s N l an or d Po wa rtu y g Sp al Sw Swe ain it z de er n la n Ud K W U or W S ld or ex ld EAUS FE A Skewness & Expected Returns Average Skewness in Developed Markets 2 1.5 1 0.5 0 -0.5 -1 -1.5 -2 -2.5 Data through June 2002 Tactical Asset Allocation 37 3/22/2016 rg e Banti n hr a Br a in a Chz il Cz i e c Co Chi l e h lo na Re m pu bia b Eg l ic G yp H ree t un c e ga In In ry do di ne a Is sia Jo rae rd l M Ko an ala re a M ysi a M ex or ic o o N cco ig er PaOm i a ki an Ph sta n ili Pe pp ru i Po nes Sa la ud Ru nd i A ss So Sl rabi a ut ov i a h a Sr Afr ki a i L ica Ta a nk Th iw a a i an la T V ur nd e Zi nez ke y Comba uela m bw po e sit e A Skewness & Expected Returns Average Skewness in Emerging Markets 2 1.5 1 0.5 0 -0.5 -1 -1.5 -2 -2.5 Data through June 2002 Tactical Asset Allocation 38 3/22/2016 Skewness or ”crash” premia (1) Skewness premium =Price of calls at strike 4% above forward price/ price of puts at strike 4% below forward price 1 The two diagrams following show: That fears of crash exist mostly since the 1987 crash This shows also in the volume of transactions on puts compared to calls Tactical Asset Allocation 39 3/22/2016 Skewness or ”crash” premia (2) Tactical Asset Allocation 40 3/22/2016 Tactical Asset Allocation 41 3/22/2016 Skewness Skewness 2 1 0 Variance -1 -20 5 10 15 12.5 10 RF Expected Return 7.5 5 See also movie from Cam Harvey web site. Tactical Asset Allocation 42 3/22/2016 Where skewness is coming from? Log-normal distribution Behavioral preferences (non-equivalence between gains and losses) Experiments: People like +’ve skewness and hate negative skewness. Tactical Asset Allocation 43 3/22/2016 Conditional Skewness, Bakshi, Harvey and Siddique (2002) For 1996 f 5skew 7. 00 5. 44 3. 89 2. 33 0. 78 - 0. 78 - 2. 33 - 3. 89 - 5. 44 - 7. 00 - 0. 55 0. 31 1. 16 2. 02 2. 88 3. 73 0. 000 4. 59 0. 053 0. 106 5. 45 0. 158 l ogsi ze 6. 31 0. 211 0. 264 7. 16 0. 317 0. 370 8. 02 0. 422book_m kt 0. 475 8. 88 0. 528 9. 74 0. 581 0. 634 10. 59 0. 686 11. 45 0. 739 Tactical Asset Allocation 44 3/22/2016 What can explain skewness? Stein-Hong-Chen: imperfections of the market cause delays in incorporation of the information into prices. Measure of info flows – turnover or volume. Tactical Asset Allocation 45 3/22/2016 Co-skewness Describe the probability of the assets to run-up or crash together. Examples: ”Asian flu” of 98,” crashes in Eastern Europe after Russian Default. Can be partially explained by the flows. Important: Try to avoid assets with +’ve co-skewness. Especially important for hedge funds Difficult to measure. Tactical Asset Allocation 46 3/22/2016 Three-Dimensional Analysis Tactical Asset Allocation 47 3/22/2016 Alternative Vehicles Alternate Asset Classes Often Involve Implicit or Explicit Options 7 6 5 4 3 2 1 0 -1 -2 -3 -4 S&P 500 Global Macro 1 2 3 4 Source: Naik (2002) Tactical Asset Allocation 5 48 3/22/2016 Alternative Vehicles Alternate Asset Classes Often Involve Implicit or Explicit Options 8 6 4 2 S&P 500 Trend Followers 0 -2 -4 -6 -8 1 2 3 4 Source: Naik (2002) Tactical Asset Allocation 5 49 3/22/2016 Alternative Vehicles Alternate Asset Classes Often Involve Implicit or Explicit Options 7 6 5 4 3 2 1 0 -1 -2 -3 -4 S&P 500 FI Arb 1 2 3 4 5 Source: Naik (2002) Tactical Asset Allocation 50 3/22/2016 Alternative Vehicles Alternate Asset Classes Often Involve Implicit or Explicit Options 2 1.5 1 0.5 Delta(BAA-10yTBond) x10 FI Arb 0 -0.5 -1 -1.5 -2 1 2 Source: Naik (2002) Tactical Asset Allocation 3 4 5 51 3/22/2016 Alternative Vehicles Alternate Asset Classes Often Involve Implicit or Explicit Options .1 Panel B: PRAM Returns, 1990 - 1998 Risk Arb Return - Risk-free Rate .08 .06 9002 9811 9602 9504 9706 9011 9410 9704 9107 9010 9004 9304 9607 9411 9407 9712 9310 9311 9312 9105 9308 9710 9402 9801 9804 9610 9006 9606 9101 9508 9201 9207 9604 9301 9408 9611 9005 9208 9306 9106 9807 9608 9303 9409 9702 9605 9507 9404 9509 9505 9812 9203 9412 9405 9506 9502 9806 9701 9112 9802 9210 9307 9708 9206 9501 9603 9512 9510 9205 9209 9202 9511 9612 9109 9108 9805 9104 9810 9003 9609 9705 9711 9110 9302 9103 9211 9305 9707 9403 9406 9212 9309 9401 9204 9503 9709 9102 9001 9703 9007 9601 9803 9111 .04 .02 0 -.02 9008 9012 9809 9808 -.04 -.06 9009 -.08 -.1 -.2 -.16 -.12 -.08 -.04 0 .04 .08 .12 .16 .2 Market Return minus Risk-free Rate Tactical Asset Allocation Source: Figure 5 from Mitchell & Pulvino (2000) 52 3/22/2016 Alternative Vehicles Alternate Asset Classes Often Involve Implicit or Explicit Options 6 4 Event Driven Index Returns 2 0 -15 -10 -5 0 5 10 -2 -4 LOWESS fit -6 Source: Naik (2002) -8 Russell 3000 Index Returns Tactical Asset Allocation 53 3/22/2016 Co-skewness for hedge funds Co-Skewness Measure (Definition 2) (Total of 42 Funds, over Jan 1997 - Feb 2001) 4 Mean Returns (Geometric) 3.5 3 2.5 2 1.5 1 0.5 0 -0.4 -0.3 -0.2 -0.1 -0.5 0 0.1 0.2 0.3 Coskew ness Source: Lu and Mulvey (2001) Tactical Asset Allocation 54 3/22/2016