CVaR 极端视图的熵池(Matlab代码实现)

 👨‍🎓个人主页：研学社的博客 

💥💥💞💞欢迎来到本博客❤️❤️💥💥

🏆博主优势：🌞🌞🌞博客内容尽量做到思维缜密，逻辑清晰，为了方便读者。

⛳️座右铭：行百里者，半于九十。

📋📋📋本文目录如下：🎁🎁🎁

目录

💥1 概述

📚2 运行结果

🎉3 参考文献

🌈4 Matlab代码实现

---

💥1 概述

本文Black - Litterman方法的完全灵活视图推广，以有效处理分布尾部的极端视图。首先，我们提供了一个递归算法来处理关于条件风险值的视图，这是完全灵活视图的原始实现无法直接处理的。其次，我们在一个网格上同时表示先验分布和后验分布，而不是通过蒙特卡洛场景：这样就可以简单地覆盖底层分布的远尾。

📚2 运行结果

 

 部分代码：

load pseudodata;

xi = 100 * cell2mat(data(:, 2));
n = length(xi);

% bandwith
bw = kernelbw(xi);

% weights
lambda = log(2) / (n / 2);
wi = exp(-lambda * (n - (n:-1:1)'));
wi = flipud(wi) / sum(wi);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Prior market model
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% kernel density
market = [];
market.mu  = mean(xi);
market.pdf = @(x) kernelpdf(x, xi, bw, wi);
market.cdf = @(x) kernelcdf(x, xi, bw, wi);
market.inv = @(x) kernelinv(x, xi, bw, wi);
market.VaR95 = market.inv(0.05);
market.CVaR95 = quadgk(@(x) (x .* market.pdf(x)), -100, market.VaR95) / 0.05;

% numerical (Gauss-Hermite grid) prior 
load ghq1000.mat % load mesh of GH zeros 
tmp = (ghqx-min(ghqx))/(max(ghqx)-min(ghqx)); % rescale GH zeros so they belong to [0,1]
epsilon = 1e-10;
Lower = market.inv(epsilon);
Upper = market.inv(1-epsilon);
X  = Lower + tmp * (Upper-Lower); % rescale mesh

p = integrateSubIntervals(X, market.cdf);
p = normalizeProb(p);
J = length(X); 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Entropy posterior from extreme view on mean and CVaR
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
view = [];
view.mu   = mean(xi) - 1.0; 
view.CVaR95 = market.CVaR95 - 1.0; 

% Netwton Raphson
s = []; 
idx = cumsum(p) <= 0.05 ;
s(1) = sum(idx);
[p_, KLdiv] = optimizeEntropy(p, idx', 0.05, [ones(1, J); X'; (idx .* X)'], [1; view.mu; 0.05 * view.CVaR95]);
%[p_, KLdiv] = optimizeEntropy(p, [idx'; (idx .* X)'], [0.05; 0.05 * view.CVaR95], [ones(1, J); X'], [1; view.mu]);

doStop = 0;
i = 1;
while ~doStop
   i = i + 1;

   idx = [ones(1, s(i-1)), zeros(1, J - s(i-1))]';
   [dummy, KLdiv_s]  = optimizeEntropy(p, idx', 0.05, [ones(1, J); X'; (idx .* X)'], [1; view.mu; 0.05 * view.CVaR95]);
   %[dummy, KLdiv_s]  = optimizeEntropy(p, [idx'; (idx .* X)'], [0.05; 0.05 * view.CVaR95], [ones(1, J); X'], [1; view.mu]);
   
   idx = [ones(1, s(i-1) + 1), zeros(1, J - s(i-1) - 1)]';
   [dummy, KLdiv_s1]  = optimizeEntropy(p, idx', 0.05, [ones(1, J); X'; (idx .* X)'], [1; view.mu; 0.05 * view.CVaR95]);
   %[dummy, KLdiv_s1] = optimizeEntropy(p, [idx'; (idx .* X)'], [0.05; 0.05 * view.CVaR95], [ones(1, J); X'], [1; view.mu]);
   
   idx = [ones(1, s(i-1) + 2), zeros(1, J - s(i-1) - 2)]';
   [dummy, KLdiv_s2]  = optimizeEntropy(p, idx', 0.05, [ones(1, J); X'; (idx .* X)'], [1; view.mu; 0.05 * view.CVaR95]);
   %[dummy, KLdiv_s2] = optimizeEntropy(p, [idx'; (idx .* X)'], [0.05; 0.05 * view.CVaR95], [ones(1, J); X'], [1; view.mu]);

   % first difference
   DE  = KLdiv_s1 - KLdiv_s;
   % second difference
   D2E = KLdiv_s2 - 2 * KLdiv_s1 + KLdiv_s;
   % optimal s
   s = [s; round( s(i-1) - (DE / D2E) )]; 

   tmp = [];
   idx  = [ones(1, s(i)), zeros(1, J - s(i))]';
   [tmp.p_, tmp.KLdiv]  = optimizeEntropy(p, idx', 0.05, [ones(1, J); X'; (idx .* X)'], [1; view.mu; 0.05 * view.CVaR95]);
   %[tmp.p_, tmp.KLdiv] = optimizeEntropy(p, [idx'; (idx .* X)'], [0.05; 0.05 * view.CVaR95], [ones(1, J); X'], [1; view.mu]);
   p_ = [p_, tmp.p_]; 
   KLdiv = [KLdiv; tmp.KLdiv]; %#ok<*AGROW>
   
   % if change in KLdiv less than one percent, stop
   if abs((KLdiv(i) - KLdiv(i-1)) / KLdiv(i-1)) < 0.01 
      doStop = 1;
   end
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Display results
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
figure;
plot(X, p, 'b', 'LineWidth', 1.5);
hold on;
plot(X, p_(:, end), 'r', 'LineWidth', 1.5);
hold on; plot(market.mu, 0.0, 'o', 'Color', 'Blue', 'MarkerSize', 8, 'MarkerFaceColor', 'Blue'); 
hold on; plot(market.CVaR95, 0.0, '^', 'Color', 'Blue', 'MarkerSize', 8, 'MarkerFaceColor', 'Blue'); 
hold on; plot(view.mu, 0.0, 'o', 'Color', 'Red', 'MarkerSize', 8, 'MarkerFaceColor', 'Red'); 
hold on; plot(view.CVaR95, 0.0, '^', 'Color', 'Red', 'MarkerSize', 8, 'MarkerFaceColor', 'Red'); 
grid on;
xlabel('Returns [%]');
h = legend('Prior', 'Posterior');
set(h, 'FontSize', 8);
set(gca, 'YTickLabel', '');

🎉3 参考文献

部分理论来源于网络，如有侵权请联系删除。

[1]A. Meucci, D. Ardia, S.Keel (2010) "Fully Flexible Extreme Views",

🌈4 Matlab代码实现