Ramsey reset specification

[bparksb]

The Ramsey Reset test uses predicteds in a auxiliary test equation.

The problem is numerical - if the dependent variable is large or small, Y_hat^2, Y_hat^3, Y_hat^4, Y_hat^5 can be very large or very small and will cause a 'near singular matrix' message when Eviews attempts to calculate the auxiliary equation.

One 'fix' is to use the standardized predicteds rather than the raw predicteds, i.e., (Y_hat-@mean(Y_hat))/@stdev(Y_hat). The standardized variables will be numerically better behaved and the log likelihood is identical (as it must be).

An example follows in which hours of labor is regressed on wages and a Ramsey 3 term test will not compute.
eq01.reset(3) produces
Near singular matrix error. Regressors may be perfectly collinear in "DO_ EQ01.RESET(3)".
while
equation eq02.ls hrs c wage hrsf^2 hrsf^3 hrsf^4 produces
Near singular matrix error. Regressors may be perfectly collinear in "EQUATION EQ02.LS HRS C WAGE HRSF^2 HRSF^3 HRSF^4".

I hope that you implement better numerical in the Ramsey test.

wfcreate(wf=testramsey) u 39
series hrs
series wage
HRS.fill 2157,2174,2062,2111,2134,2185,2210,2105,2267,2205,2121,2109,2108,2047,2174,2067,2159,2257,1985,2184,2084,2051,2127,2102,2098,2042,2181,2186,2108,2188,2203,2077,2196,2093,2173,2179,2200,2052,2197
WAGE.fill 2.905,2.97,2.3,2.511,2.791,3.04,3.222,2.493,2.8,2.356,2.922,2.499,2.796,2.453,3.582,2.909,2.511,2.516,1.423,3.636,2.983,2.573,3.262,3.234,2.28,2.304,2.912,3.015,2.786,3.01,3.273,1.901,3.009,1.899,2.959,2.971,2.98,2.63,3.413

equation eq01.ls hrs c wage
eq01.fit hrsf
eq01.reset(3)
genr hrsfstd = (hrsf-@mean(hrsf))/@stdev(hrsf)
equation eq02.ls hrs c wage hrsf^2 hrsf^3 hrsf^4
equation eq03.ls hrs c wage hrsfstd^2 hrsfstd^3 hrsfstd^4
equation eq04.ls hrs c wage hrsfstd^2 hrsfstd^3 hrsfstd^4 hrsfstd^5
equation eq05.ls hrs c wage hrsfstd^2 hrsfstd^3 hrsfstd^4 hrsfstd^5 hrsfstd^6