请教:HURST指数计算问题
小弟为MATLAB的小菜菜鸟,请假各位前辈。我在计算HURST指数的时候,发现使用同样数据我的方法计算出来的与其他人有很大差距,别人大致在0.6左右,而我的只是在0.3左右。望前辈指教问题之所在。
coef=polyfit(log(20:length(x))',RSana(x,20:length(x),'Hurst',1),1)
相关RS程序如下:
function =RSana(x,n,method,q)
if nargin<1 | isempty(x)==1
error('You should provide a time series.');
else
% x must be a vector
if min(size(x))>1
error('Invalid time series.');
end
x=x(:);
% N is the time series length
N=length(x);
end
if nargin<2 | isempty(n)==1
n=1;
else
% n must be either a scalar or a vector
if min(size(n))>1
error('n must be either a scalar or a vector.');
end
% n must be integer
if n-round(n)~=0
error('n must be integer.');
end
% n must be positive
if n<=0
error('n must be positive.');
end
end
if nargin<4 | isempty(q)==1
q=0;
else
if q=='auto'
t=autocorr(x,1);
t=t(2);
q=((3*N/2)^(1/3))*(2*t/(1-t^2))^(2/3);
else
% q must be a scalar
if sum(size(q))>2
error('q must be scalar.');
end
% q must be integer
if q-round(q)~=0
error('q must be integer.');
end
% q must be positive
if q<0
error('q must be positive.');
end
end
end
for i=1:length(n)
% Calculate the sub-periods
a=floor(N/n(i));
% Make the sub-periods matrix
X=reshape(x(1:a*n(i)),n(i),a);
% Estimate the mean of each sub-period
ave=mean(X);
% Remove the mean from each sub-period
cumdev=X-ones(n(i),1)*ave;
% Estimate the cumulative deviation from the mean
cumdev=cumsum(cumdev);
% Estimate the standard deviation
switch method
case 'Hurst'
% Hurst-Mandelbrot variation
stdev=std(X);
case 'Lo'
% Lo variation
for j=1:a
sq=0;
for k=0:q
v(k+1)=sum(X(k+1:n(i),j)'*X(1:n(i)-k,j))/(n(i)-1);
if k>0
sq=sq+(1-k/(q+1))*v(k+1);
end
end
stdev(j)=sqrt(v(1)+2*sq);
end
case 'MW'
% Moody-Wu variation
for j=1:a
sq1=0;
sq2=0;
for k=0:q
v(k+1)=sum(X(k+1:n(i),j)'*X(1:n(i)-k,j))/(n(i)-1);
if k>0
sq1=sq1+(1-k/(q+1))*(n(i)-k)/n(i)/n(i);
sq2=sq2+(1-k/(q+1))*v(k+1);
end
end
stdev(j)=sqrt((1+2*sq1)*v(1)+2*sq2);
end
case 'Parzen'
% Parzen variation
if mod(q,2)~=0
error('For the "Parzen" variation q must be dived by 2.');
end
for j=1:a
sq1=0;
sq2=0;
for k=0:q
v(k+1)=sum(X(k+1:n(i),j)'*X(1:n(i)-k,j))/(n(i)-1);
if k>0 & k<=q/2
sq1=sq1+(1-6*(k/q)^2+6*(k/q)^3)*v(k+1);
elseif k>0 & k>q/2
sq2=sq2+(1-(k/q)^3)*v(k+1);
end
end
stdev(j)=sqrt(v(1)+2*sq1+2*sq2);
end
otherwise
error('You should provide another value for "method".');
end
% Estiamte the rescaled range
rs=(max(cumdev)-min(cumdev))./stdev;
clear stdev
% Take the logarithm of the mean R/S
logRS(i,1)=log10(mean(rs));
if nargout>1
% Initial calculations fro the log(E(R/S))
j=1:n(i)-1;
s=sqrt((n(i)-j)./j);
s=sum(s);
% The estimation of log(E(R/S))
logERS(i,1)=log10(s/sqrt(n(i)*pi/2));
% Other estimations of log(E(R/S))
%logERS(i,1)=log10((n(i)-0.5)/n(i)*s/sqrt(n(i)*pi/2));
%logERS(i,1)=log10(sqrt(n(i)*pi/2));
end
if nargout>2
% Estimate V
V(i,1)=mean(rs)/sqrt(n(i));
end
end 在使用数据时,我使用过指数的对数收益率,也使用过对数收益率AR(1)的残差,但都是跟别人差距很大,望前辈指教。 求RS代码过程的解释? yejian111222 发表于 2009-8-12 11:37
在使用数据时,我使用过指数的对数收益率,也使用过对数收益率AR(1)的残差,但都是跟别人差距很大,望前辈 ...
你比没比较过你的算法和别人的有什么区别?
还是你没有拿到别人的程序,只有别人的结果?
你的程序比较长,估计很少有人会帮你逐步分析
最好找一个现成的程序,逐步比较过程量,看问题出在哪里
页:
[1]