startval.m.zip_gamma_grid search_grid search matlab_grid-search

% Function to compute starting valus for a multiple-regime Smooth % Transition Regression (STR). The algorith is based on a grid search over % the parameters gamma and c. % Last update: October 4, 2006. function [alphaf,betaf,lambdaf,gammaf,cf,bestcost]=startval(y,X,W,q,m,gamma,c) % inputs: % ------ % y: dependent variable. % X: regressors. % W: dummy regressors. % q: transition variable. % m: number of nonlinear terms (number of regimes - 1) % gamma: gamma values for the previous nonlinear terms. % c: c values for the previous nonlinear terms. % outputs: % ------- % alphaf: starting value for alpha. % lambdaf: starting value for lambda. % gammaf: starting value for gamma. % cf: starting value for c. % bestcost: cost function evaluated at the starting point. bestcost=999999999999999999999999; [T,nX] = size(X); nW = size(W,2); % Maximum and minimum values for gamma % ------------------------------------ maxgamma = 50; mingamma = 10; rategamma = 1; % Maximum and minumum values for c % -------------------------------- minc=prctile(q,10); maxc=prctile(q,90); ratec=(maxc-minc)/200; gamma(m,1) = 0; c(m,1) = 0; for newgamma=mingamma:rategamma:maxgamma for newc=minc:ratec:maxc gamma(m,1) = newgamma; c(m,1) = newc; Z = [X W]; for i=1:m fX(:,i) = siglog(gamma(i)*(q-c(i))); Z = [Z repmat(fX(:,i),1,nX).*X]; end theta = pinv(Z'*Z)*Z'*y; alpha = theta(1:nX); if isempty(W)==1 beta=[]; else beta = theta(nX+1:nX+nW); end lambda = reshape(theta(nX+nW+1:end),nX,m); if isempty(W)==1 yhat = X*alpha + sum((fX*lambda').*X,2); else yhat = X*alpha + W*beta + sum((fX*lambda').*X,2); end e = y - yhat; cost = sum(e.^2)/T; if cost<=bestcost bestcost = cost; gammaf = gamma; cf = c; alphaf = alpha; betaf = beta; lambdaf = lambda; end end end