fourier unit root tests

I am assuming you are referring to the F-test statistics used to test for the null hypothesis of linearities as discussed in Enders and Lee, 2012. The computation of these are not yet included in the TSPDLIB library but are planned for the next version.

I've written some code that demonstrates how this can be after the Fourier_ADF procedure.

To use this code you should:

• Have the tspdlib library installed.
• Call the Fourier_ADF procedure first to find the optimal Fourier frequency and the number of lags.
• Copy and paste the code below to the program file being used to call the Fourier_ADF procedure.

The code to generate the F-stat following the Fourier_ADF call is below:

proc (2) = fourierADFFTest(y, model, k, p);
    local t, dy, ly, dc, dt, sink, cosk,
        dep, y1, ldy, sbt, trnd, sinp, cosp, z1, z2,
        b1, e1, sig21, se1, ssr1, b2, e2, sig22, se2, ssr2,
        k1, k2, lmat, j, cv_fstat;

    // Check number of columns of y
    if cols(y) > 1;
        errorlogat "Cannot run test, y must be single column.";
        end;
    endif;

    t = rows(y);

    dy = diff(y, 1);
    ly = lagn(y, 1);

    // Deterministic term=constant
    dc = ones(t, 1);

    // Deterministic term=trend
    dt = seqa(1, 1, t);
    sink = sin(2 * pi * k * seqa(1, 1, t)/t);
    cosk = cos(2 * pi * k * seqa(1, 1, t)/t);

    // Find lmat
    lmat = zeros(rows(dy), p + 1);
    j = 1;
    do while j <= p;
        local tmp;
        tmp = lagn(dy, j);
        lmat[., j] = lagn(dy, j);
        j = j + 1;
    endo;

    // Prepare data
    dep = trimr(dy, p + 1, 0);
    y1 = trimr(ly, p + 1, 0);
    ldy = trimr(lmat, p + 1, 0);
    sbt = trimr(dc, p + 1, 0);
    trnd = trimr(dt, p + 1, 0);
    sinp = trimr(sink, p + 1, 0);
    cosp = trimr(cosk, p + 1, 0);

    if p == 0;
        if model == 1;
            z1 = y1~sbt;
        endif;

        if model == 2;
            z1 = y1~sbt~trnd;
        endif;

    elseif p > 0;
        if model == 1;
            z1 = y1~sbt~ldy[., 1:p];
        endif;
        if model == 2;
            z1 = y1~sbt~trnd~ldy[., 1:p];
        endif;
    endif;

    // Unrestricted model
    z2 = z1~sinp~cosp;

    // Estimate model one
    // Restricted model
    {b1, e1, sig21, se1, ssr1} = myols(dep, z1);

    // Unrestricted model
    {b2, e2, sig22, se2, ssr2} = myols(dep, z2);

    // Count dimension of models
    k1 = cols(z1);
    k2 = cols(z2);

    // Generate F_stat
    F_stat = ((ssr1 - ssr2)/(k2 - k1))/(ssr2/(t-k2));

    // Find critical values
    cv_fstat = fstatcv(model, t);

     _printHeaderTSPDLIB(0, 0, 2, model, 4, 0, "Fourier ADF F-Stat", "linearity", 0);

    // Print results & Critical values
    local top, side, stats, dec, colwidth;
    side = "F-stat:";
    top = "";
    stats = F_stat; 
    dec = 3;
    colwidth = 15;

    _printStatsTable(top, side, stats, cv_fstat', dec, colwidth, 0, p, k);
    _printConclusion(F_stat, cv_fstat, 1, "linearity");

    retp(F_stat, cv_fstat);
endp;

proc (1) = fstatcv(model, T);
    local crit;

    // Critical Values (see, Enders & Lee, 2012, Table 1b)
    crit = zeros(5, 3);

    if model == 1;
        if T <= 150;
            crit = { 6.35 7.58 10.35 };
        endif;

        if 151 < T and T <= 349;
            crit = { 6.25 7.41 10.02 };
        endif;

        if 350 < T and T <= 500;
            crit = { 6.16 7.29 9.78 };
        endif;

        if 500 < T;
            crit = { 6.11 7.25 9.72 };
        endif;
    endif;

    if model == 2;
        if T <= 150;
            crit = { 7.78 9.14 12.21 };
        endif;

        if 151 < T and T <= 349;
            crit = { 7.62 8.88 11.70 };
        endif;

        if 350 < T and T <= 500;
            crit = { 7.53 8.76 11.52 };
        endif;

        if 500 < T;
            crit = { 7.50 8.71 11.35 };
        endif;
    endif;
    retp(crit);
endp;

To call the fourierADFFTest using the data vector y[., "Y"]:

// Find F-stat
{ f_stat, p_val } = fourierADFFTest(y[., "Y"], model, f, p);

Hi;

I added the above code to the fourier_adf (src) file. Then file fourier_adf (e) "// Find F-stat { f_stat, p_val } = fourierADFFTest(y[., "Y"], model, f, p);" I added the command. But when I run it, I get the error "G0025 : Undefined symbol: 'fourierADFFTest' [fourier_adf.e, line 25]".

Kara,

Try refreshing the TSPDLIB library after adding the code to the fourier_adf.src file. You can do this using the Refresh menu option from the TSPDLIB library tool.

I refreshed the TSPDLIB library after adding the code to the fourier_adf.src file. But it still gave this error:

G0025 : Undefined symbol: 'Fourier_ADF' [fourier_adf.e, line 25]

line 25 :

// Find F-stat
{ f_stat, p_val } = fourierADFFTest(y[., "Y"], model, f, p);

Can you confirm that you are loading the TSPDLIB library at the top of the fourier_adf.e file? It should include this line somewhere at the top of the file:

library tspdlib;

Yes, there is a library tspdlib line at the top of the fourier_adf.e file.

Kara,

Please email me directly [email protected]

Eric