Kalman filter question

Question

Hello,

I am modeling the influence of systemic risk on a portfolio formed of equity indices. In the first case I am considering that stock returns are following a pure-diffusion process and every return is influenced by a marked-wide risk factor and k region-specific factors.(every international equity index belongs to a country and each country belongs to a region and there are k regions). In order to estimate the expected returns and sensitivities towards this risk I am performing a Kalman filter. This are the equations:
x(t+1) = b+F*x(t) +  e(t), e(t) ~ N(0,S); 
z(t) = a + A*x(t) + w(t), w(t)~ N(0,W)
x k+1xT represents the matrix containing the brownian motion increments
F-k+1xk+1; Z-nxT contains the observed returns of indices

I have tried to implement a code,but I only get 0 values or Nan for all arrays. Is definitely something wrong with the way I initialized the variables:
function [xfilt, Vfilt, Vmfilt, loglik] = kalmanFilter( Z, A, W, F, S, a, b)
T = size(Z,2);
n = size(Z,1);
k=3
xfilt = ones( k, T);
Vfilt = zeros( k, k, T);
Vmfilt = zeros( k, k, T);
V=zeros(k,k,T)
Vm=zeros(k,k,T)
xhat=ones(k,T)
xhatm=ones(k,T)
if n~=size(A,1)
	error('Dimension error - A & Z must agree.');
end

if (n~=size(W,1) || n~=size(W,2))
	error('Dimension error - W is not valid covariance matrix.');
end

if (k~=size(F,2))
	error('Dimension error - F not square.');
end

if (k~=size(A,2))
	error('Dimension error - F & A do not agree.');
end

if (k~=size(S,1) || k~=size(S,2))
	error('Dimension error - S is not valid.');
end

if nargin < 6, a=0;
else a=zeros(n,T);
end
if nargin < 7, b=0; 
else b= ones(k,T);
end


% Initialize log-likelihood
loglik = 0;


for i=1:T
	% Time update
		xhatm = F*xhat + b;
		Vm(:,:,i) = F*V(:,:,i)*F'+ S;
  	end
	
	% Correction
	H = A*Vm(:,:,i)*A' + W;
	K = Vm(:,:,i) * A'*inv(H);
	xhat= xhatm+ K*(Z - a - A*xhatm);
	V= (eye(k) - K*A)*Vm(:,:,i)*(eye(k) - K*A)';
	
	xfilt(:,i) = xhat(:,i);
	Vfilt = V;
	Vmfilt(:,:,i) = Vm(:,:,i);
	
	% Update log likelihood
	loglik = loglik -(n/2)*log(2*pi) - .5 * log( abs(det(H)) ) -.5*(( Z - a - A*xhatm )'*inv(H)*( Z - a - A*xhatm ));
    end


A call this funtion from another.m file:
load DAT1.dat
Z=DAT1
T = size(Z,2);
n = size(Z,1);
k = 3;
xfilt = ones( k, T);
Vfilt = zeros( k, k, T);
Vmfilt = zeros( k, k, T);
A=eye(n,k)
W=zeros(n,n)
F=zeros(k,k)
S=zeros(k,k)
a=zeros(n,T)
b=ones(k,T);
[xfilt, Vfilt, Vmfilt, loglik] = kalmanFilter( Z, A, W, F, S, a, b)

Could you please help me, any suggestions will be greatly appreciated?

Thank you in advance,
Isabela

Kalman filter question

0 Comments
Show -2 older comments Hide -2 older comments

Answers (0)

Tags

Community Treasure Hunt

Kalman filter question

0 Comments Show -2 older comments Hide -2 older comments

Answers (0)

Tags

See Also

Community Treasure Hunt

0 Comments
Show -2 older comments Hide -2 older comments