Portfolio optimization with CVaR budgets
∗ Kris Boudt - Peter Carl - Brian G. Peterson
R/Finance 2010
April 16th, 2010
∗ K.U.Leuven and Lessius 1/42 History
Outline Risk budgets: A primer on risk budgets
CVaR budgets as objective or constraint in Standard tool to quantify risk allocation; portfolio allocation
Dynamic portfolio allocation Previous research: non-normality return Conclusion series, CVaR: PerformanceAnalytics. Appendix Instrument to adjust marginally portfolios;
This research: Use of risk budgets as ob- jective and/or constraint in portfolio alloca- tion styles: PortfolioAnalytics;
Collaborative: Peter Carl & Brian Peter- son, David Ardia, Christophe Croux. 2/42 Motivation equity/bonds portfolio
Outline Bender, Briand, Nielsen, Stefek (JPM, Winter 2010): A primer on risk budgets
CVaR budgets as “Traditional approaches to structuring policy portfolios for strategic objective or constraint in portfolio allocation asset allocation have not provided the full potential of diversification.
Dynamic portfolio allocation Portfolios based on a 60/40 allocation between equities and bonds
Conclusion remain volatile and dominated by equity risk.”
Appendix Minimum risk portfolios tend to be dominated by bond risk and have lower expected returns.
3/42 Motivation equity/bonds portfolio
Outline Bender, Briand, Nielsen, Stefek (JPM, Winter 2010): A primer on risk budgets
CVaR budgets as “Traditional approaches to structuring policy portfolios for strategic objective or constraint in portfolio allocation asset allocation have not provided the full potential of diversification.
Dynamic portfolio allocation Portfolios based on a 60/40 allocation between equities and bonds
Conclusion remain volatile and dominated by equity risk.”
Appendix Minimum risk portfolios tend to be dominated by bond risk and have lower expected returns.
Optimize the risk allocation directly in the portfolio strategy. Examples: X A 60/40 risk allocation portfolio or an equal-risk portfolio
X The most risk diversified portfolio subject to return/risk targets. 3/42 Outline
Outline Risk budgets: A primer on risk budgets
CVaR budgets as objective or constraint in Review on risk budgets; portfolio allocation
Dynamic portfolio allocation Use of risk budgets as objective and/or Conclusion constraint in portfolio allocation styles. Appendix Illustrations:
X Static bond-equity portfolio, R code (DEoptim, see also Guy Yollin’s slides RFinance 2009);
X Dynamic 4 assets.
4/42 Outline
A primer on risk budgets
VaR budget
CVaR budget
Min CVaR portfolio
CVaR budgets as objective or constraint in A primer on risk budgets portfolio allocation
Dynamic portfolio allocation
Conclusion
Appendix
5/42 VaR budget on 60/40 portfolio
Outline > library(PortfolioAnalytics) A primer on risk budgets > data(indexes) VaR budget
CVaR budget > head(indexes[,1:2]) Min CVaR portfolio US Bonds US Equities CVaR budgets as objective or constraint in 1980-01-31 -0.027189977 0.0610 portfolio allocation
Dynamic portfolio 1980-02-29 -0.066876898 0.0031 allocation 1980-03-31 0.005275587 -0.0987 Conclusion 1980-04-30 0.099246261 0.0429 Appendix 1980-05-31 0.000000000 0.0562 1980-06-30 0.060511451 0.0296
6/42 VaR budget on 60/40 portfolio
Outline > apply(indexes[,1:2],2,’mean’) A primer on risk budgets US Bonds US Equities VaR budget
CVaR budget 0.006916187 0.008238420 Min CVaR portfolio > apply(indexes[,1:2],2,’sd’) CVaR budgets as objective or constraint in US Bonds US Equities portfolio allocation
Dynamic portfolio 0.01810161 0.04476569 allocation
Conclusion
Appendix
7/42 VaR budget on 60/40 portfolio
Outline > w6040 <- c(0.4,0.6) A primer on risk budgets > library(PerformanceAnalytics) VaR budget
CVaR budget > VaR(R=indexes[,1:2], weights=w6040, Min CVaR portfolio + portfolio_method="component") CVaR budgets as objective or constraint in $MVaR portfolio allocation
Dynamic portfolio [1,] 0.04336715 allocation $contribution Conclusion US Bonds US Equities Appendix -0.0002303964 0.0435975440 $pct_contrib_MVaR US Bonds US Equities -0.005312695 1.005312695
8/42 VaR budgets
Outline ∂VaR(w) VaR A primer on risk budgets Ci = wi ∂wi VaR budget CVaR budget Gourieroux,´ Laurent and Scaillet (2000): Min CVaR portfolio
CVaR budgets as objective or constraint in CiVaR = −E[wiri|rp = −VaR] portfolio allocation
Dynamic portfolio allocation
Conclusion Estimation:
Appendix X Simulation
X Explicit formulae Cornish-Fisher estimator (Boudt, Peterson and Croux, 2008; Peterson and Boudt, 2008).
9/42 Outline
Outline Pearson [2002, p.7]: “Value-at-risk has some well known limitations,
A primer on risk budgets and it may be that some other risk measures eventually supplants VaR budget
CVaR budget value-at-risk in the risk budgeting process.”
Min CVaR portfolio
CVaR: 95% Value at Risk and 95% Conditional VaR CVaR budgets as for a Skewed Student t objective or constraint in portfolio allocation Coherent risk mea- 95% VaR Dynamic portfolio allocation sure (most notably: Conclusion subadditive); Appendix Density
Less incentive to load 95% CVaR on the tail risk below
the VaR used. 0.0 0.1 0.2 0.3 0.4
−10 −5 0 5 10
Portfolio returns
10 / 42 CVaR budgets
Outline ∂CVaR(w) CVaR A primer on risk budgets Ci = wi ∂wi VaR budget CVaR budget Scaillet (2002): Min CVaR portfolio
CVaR budgets as objective or constraint in CiCVaR = −E[wiri|rp ≤−VaR] portfolio allocation
Dynamic portfolio allocation Estimation:
Conclusion X Appendix Simulation
X Explicit formulae Cornish-Fisher estimator (Boudt, Peterson and Croux, 2008).
11 / 42 CVaR budgets
Outline > ES(R=indexes[,1:2], weights=w6040, A primer on risk budgets + portfolio_method="component") VaR budget
CVaR budget $MES Min CVaR portfolio [1,] 0.07725177 CVaR budgets as objective or constraint in $contribution portfolio allocation
Dynamic portfolio US Bonds US Equities allocation -0.001066194 0.078317964 Conclusion $pct_contrib_MES Appendix US Bonds US Equities -0.01380155 1.01380155
12 / 42 CVaR in portfolio allocation
Outline As an objective: Minimum CVaR portfolio (E.g. Rockafellar A primer on risk budgets
VaR budget and Uryasev, 2000)
CVaR budget Min CVaR portfolio As a constraint: Min CVaR/SD portfolio under CVaR CVaR budgets as objective or constraint in constraints (E.g. Alexander and Baptista, 2004, Krokhmal, portfolio allocation
Dynamic portfolio Palmquist and Uryasev, 2002). allocation
Conclusion Why? Better risk measure + convex function of portfolio Appendix weights (easier to optimize).
13 / 42 Min CVaR portfolio
Outline > library(DEoptim) A primer on risk budgets > obj <- function(w) { VaR budget
CVaR budget + if (sum(w) == 0) { w <- w + 1e-2 } Min CVaR portfolio + w <- w / sum(w) CVaR budgets as objective or constraint in + ES(R=indexes[,1:2],weights = w)$MES portfolio allocation
Dynamic portfolio + } allocation > out <- DEoptim(fn = obj, lower = rep(0, 2), Conclusion + upper = rep(1, 2), DEoptim.control(itermax=50)) Appendix > wstar <- out$optim$bestmem > wMinCVaR <- wstar / sum(wstar) > print(wMinCVaR) US Bonds US Equities 0.96443348 0.03556652 14 / 42 CVaR budgets
Outline > ES(R=indexes[,1:2], weights=wMinCVaR, A primer on risk budgets + portfolio_method="component") VaR budget
CVaR budget $MES Min CVaR portfolio [1,] 0.01102894 CVaR budgets as objective or constraint in $contribution portfolio allocation
Dynamic portfolio US Bonds US Equities allocation 0.0106366796 0.0003922610 Conclusion $pct_contrib_MES Appendix US Bonds US Equities 0.96443349 0.03556651
15 / 42 CVaR budgets
Outline Weight allocation Risk allocation A primer on risk budgets style bond equity bond equity VaR budget
CVaR budget 60/40 weight 0.40 0.6 -0.01 1.01 Min CVaR portfolio 60/40 risk alloc 0.84 0.16 0.40 0.60 CVaR budgets as objective or constraint in Min CVaR Conc 0.86 0.14 0.50 0.50 portfolio allocation Min CVaR 0.96 0.04 0.96 0.04 Dynamic portfolio allocation
Conclusion
Appendix
16 / 42 Outline
A primer on risk budgets
CVaR budgets as objective or constraint in portfolio allocation CVaR budgets as objective or Risk budget constraints
Risk budget objective
Efficient Frontier constraint in portfolio
Dynamic portfolio allocation allocation Conclusion
Appendix
17 / 42 Traditional use
Outline ∂CVaR(w) Risk contribution CVaR A primer on risk budgets Ci (w)= wi ∂wi CVaR budgets as objective or constraint in portfolio allocation TM Risk budget constraints Litterman (1999): Hot Spots and hedges. Risk budgets as
Risk budget objective ex post instrument to adjust marginally portfolios. Efficient Frontier
Dynamic portfolio allocation Keel and Ardia (2009):
Conclusion
Appendix 1. Only precise for infinitesimal changes, poor approximations for realistic reallocations.
2. Assume changing a single position keeping fixed all other positions >< full investment constraint.
18 / 42 Proposal I: Risk budget constraints
Outline Avoid downside risk concentration by: A primer on risk budgets
CVaR budgets as CiCVaR objective or constraint in li ≤ %CiCVaR ≡ ≤ ui portfolio allocation CVaR Risk budget constraints
Risk budget objective Efficient Frontier Min CVaR portfolio with
Dynamic portfolio allocation X 60/40 risk allocation constraint Conclusion X Appendix equal risk allocation constraint.
19 / 42 60/40 risk allocation portfolio
Outline > obj <- function(w) { A primer on risk budgets + if (sum(w) == 0) { w <- w + 1e-2 } CVaR budgets as objective or constraint in + w <- w / sum(w) portfolio allocation
Risk budget constraints + CVaR <- ES(R=indexes[,1:2],weights = w) Risk budget objective + tmp1 <- CVaR$MES Efficient Frontier
Dynamic portfolio + tmp2 <- max(CVaR$pct_contrib_ES allocation + - c(0.405, 0.605) , 0) Conclusion + tmp3 <- max(c(0.395, 0.595) - Appendix + CVaR$pct_contrib_ES , 0) + out <- tmp1 + 1e3 * tmp2 + 1e3 * tmp3 + }
20 / 42 60/40 risk allocation portfolio
Outline > out <- DEoptim(fn = obj, lower = rep(0, 2), A primer on risk budgets + upper = rep(1, 2), DEoptim.control(itermax=50)) CVaR budgets as objective or constraint in > wstar <- out$optim$bestmem portfolio allocation
Risk budget constraints > w6040riskalloc <- wstar / sum(wstar) Risk budget objective > print(w6040riskalloc) Efficient Frontier
Dynamic portfolio US Bonds US Equities allocation 0.8382035 0.1617965 Conclusion
Appendix
21 / 42 60/40 risk allocation portfolio
Outline > ES(R=indexes[,1:2], weights=w6040riskalloc, A primer on risk budgets + portfolio_method="component") CVaR budgets as objective or constraint in $MES portfolio allocation
Risk budget constraints [1,] 0.01400341 Risk budget objective $contribution Efficient Frontier
Dynamic portfolio US Bonds US Equities allocation 0.005671224 0.008332185 Conclusion $pct_contrib_MES Appendix US Bonds US Equities 0.4049888 0.5950112
22 / 42 Special case: Equal-risk portfolio
Outline Min CVaR with equal-risk constraint A primer on risk budgets
CVaR budgets as objective or constraint in %CiCVaR(w)=1/N (i =1,...,N) portfolio allocation
Risk budget constraints
Risk budget objective
Efficient Frontier Dynamic portfolio In this portfolio: allocation wi ∂CVaR/∂wj Conclusion = . wj ∂CVaR/∂wi Appendix Downweights “hot spots”: positions for which a marginal
decrease in weight leads to a large reduction in CVaR.
23 / 42 Proposal II: Risk budget objective
Outline Avoid downside risk concentration by: A primer on risk budgets
CVaR budgets as objective or constraint in min max CiCVaR(w) portfolio allocation w i
Risk budget constraints
Risk budget objective
Efficient Frontier
Dynamic portfolio allocation
Conclusion
Appendix
24 / 42 Proposal II: Risk budget objective
Outline Avoid downside risk concentration by: A primer on risk budgets
CVaR budgets as objective or constraint in min max CiCVaR(w) portfolio allocation w i
Risk budget constraints
Risk budget objective Efficient Frontier Objective trades off Risk Minimization and Risk Diversification, Dynamic portfolio allocation since:
Conclusion
Appendix max CiCVaR = CVaR max{%C1CVaR,..., %CN CVaR} i
24 / 42 Relation with equal-risk portfolio
Outline If the set of equal-risk portfolios is non-empty and the A primer on risk budgets
CVaR budgets as minimum CVaR concentration portfolio has a unique optimum. objective or constraint in portfolio allocation Then the minimum CVaR concentration portfolio criterion is Risk budget constraints equivalent to: Risk budget objective Efficient Frontier min CVaR(w) w Dynamic portfolio allocation s.t. %C1CVaR = ... =%CN CVaR Conclusion
Appendix But computationally more simple and has also a solution if there is no equal-risk portfolio.
25 / 42 Min CVaR Concentration portfolio
Outline > obj <- function(w) { A primer on risk budgets + if (sum(w) == 0) { w <- w + 1e-2 } CVaR budgets as objective or constraint in + w <- w / sum(w) portfolio allocation
Risk budget constraints + CVaR <- ES(R=indexes[,1:2],weights = w) Risk budget objective + out <- max(CVaR$contribution) Efficient Frontier
Dynamic portfolio + } allocation > out <- DEoptim(fn = obj, lower = rep(0, 2), Conclusion + upper = rep(1, 2),DEoptim.control(itermax=50)) Appendix > wstar <- out$optim$bestmem > wMinCVaRConc <- wstar / sum(wstar) > print(wMinCVaRConc) US Bonds US Equities 0.8584465 0.1415535 26 / 42 Min CVaR Concentration portfolio
Outline > ES(R=indexes[,1:2], weights=wMinCVaRConc, A primer on risk budgets + portfolio_method="component") CVaR budgets as objective or constraint in $MES portfolio allocation
Risk budget constraints [1,] 0.01315665 Risk budget objective $contribution Efficient Frontier
Dynamic portfolio US Bonds US Equities allocation 0.006578325 0.006578323 Conclusion $pct_contrib_MES Appendix US Bonds US Equities 0.5000001 0.4999999
27 / 42 Overview portfolio allocations
Outline Weight allocation Risk allocation A primer on risk budgets style bond equity bond equity CVaR budgets as objective or constraint in 60/40 weight 0.40 0.6 -0.01 1.01 portfolio allocation
Risk budget constraints 60/40 risk alloc 0.84 0.16 0.40 0.60
Risk budget objective Min CVaR Conc 0.86 0.14 0.50 0.50 Efficient Frontier Min CVaR 0.96 0.04 0.96 0.04 Dynamic portfolio allocation
Conclusion
Appendix
28 / 42 Adding a return target: Efficient frontier
Outline Min CVaR Min CVaR Concentration Efficient frontier Efficient frontier A primer on risk budgets
0.0085 S&P 500 0.0085 S&P 500 CVaR budgets as objective or constraint in 0.0080 0.0080 portfolio allocation 0.0075 0.0075 Risk budget constraints
0.0070Monthly return 0.0070Monthly return Risk budget objective
0.0065 US bond 0.0065 US bond Efficient Frontier
Dynamic portfolio 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 allocation Monthly CVaR (full) Monthly CVaR (full) Monthly CVaR concentration(dashed) Monthly CVaR concentration(dashed)
Conclusion Weight allocation Weight allocation 1.0 1.0 Appendix 0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
0.0065 0.0068 0.0071 0.0074 0.0077 0.008 0.0083 0.0067 0.007 0.0073 0.0076 0.0079 0.0082
Monthly return Monthly return US bond S&P 500 29 / 42 Outline
A primer on risk budgets
CVaR budgets as objective or constraint in portfolio allocation
Dynamic portfolio allocation
Dynamic strategies Dynamic portfolio allocation
Data
Conclusion
Appendix
30 / 42 Dynamic investment strategies
We consider the following investment strategies, quarterly Outline
A primer on risk budgets rebalancing, 4 assets:
CVaR budgets as objective or constraint in 1. Benchmark strategies: portfolio allocation Dynamic portfolio X allocation “Equal Weight”
Dynamic strategies X “Min CVaR” Data
Conclusion X “Min CVaR + 40% Position Limit”
Appendix X “Min CVaR + EW return target”
2. Strategies that use CVaR budgets:
X “Min CVaR + 40% Perc CVaR Alloc Limit”
X “Min CVaR Conc”
X “Min CVaR Conc + EW return target”. 31 / 42 Summary statistics data:
4 assets: Merrill Lynch US bond, S&P 500, MSCI EAFE and S&P GSCI Real monthly returns Jan 1976-December 2009, total return indices
US bond S&P 500 MSCI EAFE S&P GSCI Mean (in %) 0.32 0.52 0.39 0.10 StdDev (in %) 1.86 4.46 4.98 5.50 Historical 95% CVaR (in %) 3.64 10.64 12.46 13.58
Quarterly rebalanced based on time-varying conditional moment estimates (EWMA mean, GARCH volatility, rolling 8 year correlation, coskewness and cokurtosis). Weight allocation:
Min CVaR Min CVaR + CVaR Alloc Limit Min CVaR + Return Target
1 1 1
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3 1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3 1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3
Min CVaR Conc Equal Weight Min CVaR Conc + Return Target
1 1 1
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3 1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3 1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3
US bond S&P 500 EAFE GSCI CVaR allocation:
Min CVaR Min CVaR + CVaR Alloc Limit Min CVaR + Return Target
1 1 1
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3 1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3 1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3
Min CVaR Conc Equal Weight Min CVaR Conc + Return Target
1 1 1
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3 1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3 1984Q1 1988Q3 1993Q1 1997Q3 2002Q1 2006Q3
US bond S&P 500 EAFE GSCI Out of sample performance:
Equal Min CVaR Min CVaR Conc Weight Position CVaR Alloc Return Return Limit Limit Target Target Mean (in %) 0.40 0.44 0.43 0.43 0.47 0.43 0.43 StdDev (in %) 2.85 1.38 2.41 1.64 1.74 1.85 2.02 Hist 95% CVaR (in %) 6.95 2.72 5.78 3.49 3.59 4.11 4.52 Portfolio turnover (in %) 1.27 2.30 3.10 2.70 4.50 1.82 4.45
T∗− 1 1 Portf. turnover = X w(i)t+1 − w(i)t+ . NT∗ t=1 Out of sample cum performance, relative to min CVaR:
Equal Weight Min CVaR + Position Limit Min CVaR + CVaR Alloc Limit
1.2 1.2 1.2
1.0 1.0 1.0
0.8 0.8 0.8
0.6 0.6 0.6
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
Min CVaR + Return Target Min CVaR Conc Min CVaR Conc + Return Target
1.2 1.2 1.2
1.0 1.0 1.0
0.8 0.8 0.8
0.6 0.6 0.6
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 Drawdowns higher than 10% on portfolio strategies over the period January 1984-December 2009:
Equal Min CVaR Min CVaR Conc Weight Position CVaR Alloc Return Return Limit Limit Target Target ∗ Credit crisis 0.47 0.10 0.37 0.18 0.14 0.24 0.24 ∗∗ Dot-com bubble burst 0.28 0.19 0.11 0.11 ∗∗∗ Asian-Russian crisis 0.13 0.12 0.11 ∗∗∗∗ Black Monday 0.11 0.13 0.12 0.12
∗ Dec 2007-Oct 2008 for the Min CVaR strategy, Nov 2007-Feb 2009 for all other styles.
∗∗ Start: Sept 2000 for all styles. End: Jan 2002 for Min CVaR Conc styles, July 2002 for the Min CVaR with position limit style, September 2002 for all other styles.
∗∗∗ Aug 1997-Aug 1998 for equal-weight strategy, Oct 1997-Aug 1998 for the Min CVaR + Return target and Nov 1997-Aug 1998 for the Min CVaR with position limit style.
∗∗∗∗ Black Monday: Sept-November 1987. Outline
A primer on risk budgets
CVaR budgets as objective or constraint in portfolio allocation
Dynamic portfolio allocation Conclusion Conclusion
Appendix
38 / 42 Conclusion
Outline CVaR budgets are useful for:
A primer on risk budgets CVaR budgets as Ex post analysis of the portfolio risk allocation; objective or constraint in portfolio allocation Dynamic portfolio And input in the portfolio allocation strategy through allocation
Conclusion X minimum CVaR Concentration objective Appendix X and/or risk allocation constraints.
39 / 42 References
Outline Software: R packages DEoptim, PerformanceAnalytics and A primer on risk budgets
CVaR budgets as PortfolioAnalytics objective or constraint in portfolio allocation Related research papers: Dynamic portfolio allocation 1. With B. Peterson and C. Croux: Estimation and decomposition
Conclusion of downside risk for portfolios with non-normal returns. Journal Appendix of Risk, Winter 2008.
2. With B. Peterson: Component VaR for a non-normal world. RISK, November 2008.
3. With D. Ardia, P.Carl, K. Mullen and B. Peterson. DEoptim for non-convex portfolio optimization. SSRN.
4. With P. Carl and B. Peterson. Portfolio optimization with CVaR
budgets. 40 / 42 Outline
A primer on risk budgets
CVaR budgets as objective or constraint in portfolio allocation
Dynamic portfolio allocation Appendix Conclusion
Appendix
41 / 42 %CV aR(1) = f(w(1)) for bivariate normal portfolio:
µ µ ρ − σ σ µ µ ρ − σ σ µ µ ρ σ σ 1= 2=0, = 0.5, 1= 2=1 1=1, 2=0, = 0.5, 1= 2=1 1= 2=0, =−0.5, 1=1, 2=2
1.0 1.0 1.0 0.8 0.8
0.6 0.5 0.6 0.4 0.4
0.2 0.0 0.2 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
µ µ ρ σ σ µ µ ρ σ σ µ µ ρ σ σ 1= 2=0, =0, 1= 2=1 1=1, 2=0, =0, 1= 2=1 1= 2=0, =0, 1=1, 2=2 1.0 1.0 1.0
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4 0.2 0.2 0.2 0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
µ µ ρ σ σ µ µ ρ σ σ µ µ ρ σ σ 1= 2=0, =0.5, 1= 2=1 1=1, 2=0, =0.5, 1= 2=1 1= 2=0, =0.5, 1=1, 2=2 1.0 1.0 1.0
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0