residual trading days tests in JDemetra+ 2.x.y

[AQLT]

In JDemetra+ 2.x.y, the residual trading days tests was perform with an AR(1) model estimated by OLS while in rjd3toolkit it is now estimated using another method (ML). In previous versions of rjd3toolkit there was a model called AR which was also using an ols model :
https://github.com/jdemetra/jdemetra-core/blob/2b12a60f0b7135f00fab7031233ddf93f33045b6/demetra-sa/demetra-sa-r/src/main/java/demetra/sa/r/TradingDaysTests.java#L33-L38
If think it would be usefull to keep this model in order to be able to reproduce the tests used in JDemetra+ 2.x.y

[palatej]

Hi Alain,

I suppressed the test because it was wrong (as the test in JD2). More specifically, the model was:
y(t)= a + by(t-1) + ctd(t) + e(t).

A more correct model should be:
y(t) - phiy(t-1)=a + b(td(t) - phi*td(t-1)) + e(t)

The new test is based on phi = 1.
In practice, setting phi = 1 doesn't change much the result of the test and it is more robust (and significantly faster).
To be noted that for stationary series, I put phi = 0