【FinE】Generalized Hyperbolic(GH)函数/SGT函数及Toolbox的使用

Generalized Hyperbolic

This toolbox contains necessary tools related to the skewed generalized t(SGT) and the generalized hyperbolic (GH) distributions. Distribution fitting is also included.

Toolbox实现函数如下

函数文件	函数接口	附注
ghcdf.m	p=ghcdf(x, Param, m, sigma)	Generalized Hyperbolic cumulative density function (cdf)
ghfit.m	[parmhat, se, parmci, output]=ghfit(X, alpha, flag)	PARMHAT=ghfit(X)返回描述GH的最大似然估计，LAMBDA=PARMHAT(1)为尾部指数参数, ZETA=PARMHAT(2)为形状参数，RHO=PARMHAT(3)为偏度参数；SE返回估计的标准误差；PARMCI返回参数估计的95%置信区间；OUTPUT返回结构化的输出包括拟合的偏度和峰度以及MLE； […]=GITFIT(X, ALPHA) 返回100(1-ALPHA) percent 参数估计的C.I.；FLAG=1进一步给出估计的均值和标准差，PARMHAT(4)为均值，PARMHAT(5)为标准差
ghpdf.m	y=ghpdf(x, param, m, sigma)	GH函数的pdf

ghcdf

计算GH分布的累计概率密度函数(cdf)，根据传入参数Param, m, sigma计算在x处的值

%% GH的CDF计算
% 使用 Flexible distributions toolbox计算
X = returns(:, 1);
[parmhat, se, parmci, output] = ghfit(X, 0.05, 1);

p = ghcdf(0.001, parmhat(1:3));

RHO>0表示GHD为右偏，RHO<0表示GHD为左偏

ghfit

使用GH分布拟合序列，在设置FLAG=1的情况下可以得到拟合的偏度(skewness)和峰度(kurtosis)

[parmhat, se, parmci, output] = ghfit(X, 0.05, 1);

ghpdf

GH分布的pdf函数，当$\zeta \to \infty$且$\varrho =0$时，GH概率密度函数变为normal distribution，当$\varrho <0$，GH分布呈现左偏的形状，当$\varrho >0$，GH分布呈现右偏的形状

ghpdf(x, Param, m, sigma)给出在点x处的pdf值，拟合收益率序列如下图所示

% 拟合分布
sample_ret = returns(:, 1);
xmin = min(sample_ret)-0.01;
xmax = max(sample_ret)+0.01;

xp = xmin: 0.01:xmax;
[parmhat, se]=ghfit(sample_ret);
x_mean = mean(sample_ret); % 一阶矩信息
x_sigma = std(sample_ret); % 二阶矩信息

yp = ghpdf(xp, parmhat, x_mean, x_sigma);

hold on;
yyaxis left; % 激活左轴
plot(xp, yp, 'b--');

yyaxis right; % 激活右轴
histogram(X, 'Normalization', 'probability', 'FaceAlpha', 0.4);
legend('GH-fitted', 'returns');
hold off;

Skew Generalized t

SGT(skew generalized t)是一种五参数分布，pdf如下

形状参数$\eta >0,\psi >2$，偏度参数$|\lambda |<1$，$B(\cdot )$为$\beta$函数，$sgn(\cdot )$为符号函数，$\mu$为位置参数，$\theta$为正则化因子，使得$\mathbb{V}(x)={\sigma }^2$

sgtcdf

返回在x点处的累计概率密度

p = sgtcdf(x, Param, m, sigma)

sgtfit

% sgtfit parameter estimates and confidence intervals for skewed generalized t data
% 其余和GH类似

sgtinv

计算Inverse of skewed generalized T cumulative density function的CDF

sgtpdf

SGT分布对数据进行拟合，当$\eta =2$时，函数成为standardized student's t-distribution with d.o.f ψ. 当$\psi =\infty$函数变为GED(η). 当$\eta =2,\psi =\infty$时，SGT变为skew normal.

%% SGT的CDF计算
[sgt_p, se]=sgtfit(sample_ret);
yp = sgtpdf(xp, sgt_p, x_mean, x_sigma);

hold on;
yyaxis left; 
plot(xp, yp, 'b--');

yyaxis right;
histogram(sample_ret, 'Normalization', 'probability', 'FaceAlpha', 0.4);
legend('SGT-fitted', 'returns');

hold off;

sgtrnd

从Skewed Generalized T分布中进行抽样

skewtfit

Skewd Distributions：Fernandez and Steel(1998) proposed a quite general approach that allows the introduction of skewness in any continuous unimodal and symmetric distribution by changing the scale at each side of the mode

$$f(z\mid \xi )=\frac{2}{\xi +\frac{1}{\xi }}[f({\xi }_z)H(-z)+f(\frac{z}{\xi })H(z)]$$
其中$0<\xi <\infty$为形状参数，描述函数的非对称性程度，$\xi =1$为对称分布，$f(z\mid \xi =1)=f(z)$.
$H(z)=(1+sgn(z))/2$为Heaviside unit step funciton.

[parmhat, se, parmci, output]=skewtfit(X, alpha, flag)

References

Flexible distributions toolbox
Highly flexible distributions to fit multiple frequency financial returns