Re: [Gretl-devel] again processor-detecting dependent behavior in arima

Generating strange arima output may be pc-dependent You may try open bad_data.gdt smpl 1 194 series sty=diff_series/sd(diff_series) list zli = y_one y_two matrix bad = mread("bad_matrix32") # or matrix bad = mread("bad_matrix36") # if paste bad_matrix32 and bad_matrix64 $workdir there's no need to input path y = sty+bad[1] arima 3 0 0; 1 0 0; y 0 zli --verbose If both models are ok : i.e. something like this (64) Model 3: ARMAX, using observations 1950:1-1998:2 (T = 194) Estimated using AS 197 (exact ML) Dependent variable: y Standard errors based on Hessian             coefficient   std. error      z       p-value  ---------------------------------------------------------  const       6.48832      0.0467055    138.9      0.0000   ***  phi_1       0.567313     0.159434       3.558    0.0004   ***  phi_2      −0.158255     0.107503      −1.472    0.1410    phi_3      −0.120526     0.130267      −0.9252   0.3549    Phi_1      −0.0597471    0.121899      −0.4901   0.6240    y_one       0.203394     0.233933       0.8695   0.3846    y_two      −2.07615      0.389196      −5.334    9.58e-08 *** or this (32) Model 4: ARMAX, using observations 1950:1-1998:2 (T = 194) Estimated using AS 197 (exact ML) Dependent variable: y Standard errors based on Hessian             coefficient   std. error      z       p-value  ---------------------------------------------------------  const       6.38832      0.0467055    136.8      0.0000   ***  phi_1       0.567313     0.159434       3.558    0.0004   ***  phi_2      −0.158255     0.107503      −1.472    0.1410    phi_3      −0.120526     0.130267      −0.9252   0.3549    Phi_1      −0.0597471    0.121899      −0.4901   0.6240    y_one       0.203394     0.233933       0.8695   0.3846    y_two      −2.07615      0.389196      −5.334    9.58e-08 *** you can try this open bad_data.gdt smpl 1 194 series sty=diff_series/sd(diff_series) list zli = y_one y_two series y = 0 matrix matr=seq(1,1000)/10^2 n = nelem(matr) ret = zeros(n,1) loop i = 1..n -q    y = sty+matr[i]    catch arima 3 0 0; 1 0 0; y 0 zli -q    err = $error    ret[i] = err? -$huge: $lnl endloop eval imaxc(ret) mad = selifr(seq(1,1000)',ret.<-200)/100 eval mad # note : mean(sty)~0 catch freq 1 --matrix=mad --plot=display # than run, for example y = sty+mad[1] arima 3 0 0; 1 0 0; y 0 zli --verbose y = sty+mad[2] arima 3 0 0; 1 0 0; y 0 zli --verbose y = sty+mad[3] arima 3 0 0; 1 0 0; y 0 zli --verbose ############# ? arima 3 0 0; 1 0 0; y 0 zli --verbose NLS: failed to converge after 1605 iterations Error executing script: halting ################### expected Oleh