Prev: How to combine cellfun and strfind?
Next: GUI, function and Workspace
From: Ash Ash on 27 Jul 2010 03:41
"John G" <johan.grahn(a)kolumbus.fi> wrote in message <go0t26$8kc$1(a)fred.mathworks.com>...
> Hello,
>
> i need to perfom a dynamic quantile regression (see for example Engle and Manganelli 2004) on a binary series to make sure there is no serial correlation.
>
> I managed to find a file that i think is the correct dynamic quantile regression (after all it was igned by the author), but it was part of a larger program and i could not work out how to use it on my binary series, i attached the code below the message.
>
> Could someone guide me in the right direction on how to do this?
>
> Any help would be appriciated!
>
> Johan
>
> function output = DQtest(OUT, MODEL, T, y, THETA, VaR, Hit, D, gradient)
>
> % ****************************************************************************************************************************************
> % * *
> % * Codes for the paper "CAViaR: Conditional Autoregressive Value at Risk by Regression Quantile" by Robert Engle and Simone Manganelli *
> % * *
> % * By SIMONE MANGANELLI, European Central Bank, Frankfurt am Main. *
> % * Created on 15 September, 2000. Last modified 20 February 2002. *
> % * *
> % ****************************************************************************************************************************************
> %
> %
> % Compute both the in sample and out of sample DQ test.
> % For the in sample DQ test, use the formulae of theorem 4:
> % Hit'X * inv(MM') * X'Hit / THETA*(1-THETA), where M = X' - inv(T)*(X'*H * grad(f)) * inv(D) * grad(f)'
> %
> % If OUT=1, the output is the p-value of the in sample DQ test.
> % If OUT=2, the output is the p-value of the out of sample DQ test.
> %
> %*****************************************************************************************
>
> % Compute the quantile residuals.
> residuals = y + VaR;
>
>
> % Set up the bandwidth for the k-Nearest neighbor estimator.
> SortedRes = sort(abs(residuals));
> if THETA == 0.01;
> k = 40;
> elseif THETA == 0.05
> k = 60;
> else error('This program considers only two confidence levels, 1% and 5%')
> end
>
> BANDWIDTH = SortedRes(k);
>
> LAGS = 4; % Number of lagged hits to be included in the test.
>
> %
> %********************************************************************************************
> % Compute the regressor matrix X.
> %
> Constant = ones(T-LAGS, 1);
> HIT = Hit(LAGS+1 : T);
> VaRforecast = VaR(LAGS+1 : T);
> ylag = y(LAGS : T-1);
> %VaRforecastlag = VaR(LAGS : T-1);
>
> Z = zeros(T-LAGS, LAGS);
> for s = 1:LAGS
> Z(: ,s) = Hit(s : T - (LAGS + 1 - s));
> %VaRforecastlag(: ,s) = VaR(s : T - (LAGS + 1 - s));
> %ylag(: ,s) = y (s : T - (LAGS + 1 - s));
> %ylagsquare(: ,s) = y(s : T - (LAGS + 1 - s)).^(2*(LAGS-s+1));
> end
>
> Xout = [Constant, VaRforecast, Z]; % Instruments for the out of sample test.
> Xin = [Z]; % Instruments for the in sample test.
>
> XHNABLA = zeros(size(Xin,2),size(gradient,2)); %, size(estimatedParameters,1));
> NABLA = gradient(LAGS+1:T,:); % Exclude the first LAGS rows from the gradient (for comformability reasons).
>
> %
> %********************************************************************************************
> % Estimate the matrices that enter the In Sample DQ test.
> if OUT == 1
> for i = 2:size(Xin,1)
> if abs(residuals(i)) <= BANDWIDTH
> XHNABLA = XHNABLA + (Xin(i,:))'*gradient(i,:);
> end
> end
> end
>
> XHNABLA = XHNABLA/(2*BANDWIDTH*T);
>
> M = Xin' - XHNABLA*inv(D)*NABLA';
>
> %
> %********************************************************************************************
> % Compute the DQ tests.
> DQstatIn = (HIT'*Xin * inv(M*M') * Xin'*HIT) / (THETA*(1-THETA));
> DQstatOut = (HIT'*Xout * inv(Xout'*Xout) * Xout'*HIT) / (THETA*(1-THETA));
>
> DQin = 1 - chi2cdf(DQstatIn, size(Xin, 2)); % Compute the P-value of the in sample DQ test.
> DQout = 1 - chi2cdf(DQstatOut, size(Xout, 2)); % Compute the P-value of the out of sample DQ test.
>
> %
> %**********************************************
> % Select the output of the program.
> if OUT == 1
> output = DQin;
> elseif OUT == 2
> output = DQout;
> else error('Wrong output selected. Choose OUT = 1 for in sample DQ test, or OUT = 2 for out of sample DQ test.')
> end
From: Ash Ash on 27 Jul 2010 03:43
Hello,
did you get an answer how to do this? I have the same problem now.
 | 
Pages: 1
Prev: How to combine cellfun and strfind?
Next: GUI, function and Workspace