InterpolaL (StdLib) works wrong for a Series with negative changes only

[Alfredo Torre]

InterpolaL (from StdLib) works wrong when the original Series only has negative changes and, at least, one no change. Please, see the next example:

Serie Ser.Qua = SubSer(
100*Pulse(y2009m01d01, Quarterly) +
 90*Pulse(y2009m04d01, Quarterly) +
 90*Pulse(y2009m07d01, Quarterly) +
 80*Pulse(y2009m10d01, Quarterly),
y2009m01d01, y2010m01d01);
Serie Ser.Mon = InterpolaL(Ser.Qua, Monthly);


The problem is in InvChEx function. I suggest this new code for InterpolaL:

Serie InterpolaL(Serie ser, TimeSet dtn)
{
  Serie dser   = ((F-1):ser)>>CalInd(W,Dating(ser));
  Serie iper   = SubSer(CalVar(dtn,Dating(ser)),First(ser),Last(ser));
  Serie inc    = dser/iper;
  Serie dNser1 = InvCh(inc, Log(CalInd(W, dtn2)));
  Serie dNser2 = ExpandOmit(dNser1);
  Serie dNser  = (B):dNser2;
  Date  ini    = Succ(First(dNser),dtn,-1);
  Serie sIniP  = FirstS(ser)*CalInd(C,dtn);
  Serie sIni   = SubSer(sIniP,ini,ini);
  SubSer(DifEq(1/(1-B),dNser,sIni),ini,Last(ser))
};

[Víctor de Buen Remiro]

Interpolation functions

[Víctor de Buen Remiro]

I dont't know who's the author of this function nor what is its target nor nothing about it.

I'm using my own interpolation attached functions.

Serie Ser.Qua = SubSer(
100*Pulse(y2009m01d01, Quarterly) +
 90*Pulse(y2009m04d01, Quarterly) +
 90*Pulse(y2009m07d01, Quarterly) +
 80*Pulse(y2009m10d01, Quarterly),
y2009m01d01, y2010m01d01);
Serie Ser.Mon = InterpolaL(Ser.Qua, Monthly);

//Linear interpolation
Serie Ser.Mon_1 = InvChInterp(Ser.Qua, Monthly, "LinearSpline", 1);
//Akima (smooth cubic) interpolation
Serie Ser.Mon_3 = InvChInterp(Ser.Qua, Monthly, "AkimaSpline", 3);

If you want to test them I can to add them to StdLib.

[Alfredo Torre]

That's great!

Just two more things to check if I've understood:

.- You suggest to create those functions in StdLib so they can be available in TOL. My doubt now is: aren't available now? Did you built them for this ticket?

.- Should InterpolaL be changed as

Serie InterpolaL(Serie ser, TimeSet dtn)
{ InvChInterp(ser, dtn, "LinearSpline", 1) };

?

[Víctor de Buen Remiro]

It will be changed in next release (tomorrow)

[Alfredo Torre]

Ok, thank you very much.

Looking forward to use it!

[Víctor de Buen Remiro]

(In [1141]) New interpolation methods in NameBlock StdLib::interpolate
Fixes #707

[Víctor de Buen Remiro]

Finally I've created a NameBlock StdLib::Interpolate for all interpolation related methods.

Serie Ser.Qua = SubSer(
100*Pulse(y2009m01d01, Quarterly) +
 90*Pulse(y2009m04d01, Quarterly) +
 90*Pulse(y2009m07d01, Quarterly) +
 80*Pulse(y2009m10d01, Quarterly),
y2009m01d01, y2010m01d01);
Serie Ser.Mon = InterpolaL(Ser.Qua, Monthly);

//Inverse dating change by means of linear interpolation
Serie Ser.Mon_1 = Interpolate::Scalar::invCh(Ser.Qua, Monthly, "LinearSpline", 1);
//Inverse dating change by means of Akima interpolation (smooth polynomial spline)
Serie Ser.Mon_3 = Interpolate::Scalar::invCh(Ser.Qua, Monthly, "AkimaSpline", 3);

[Víctor de Buen Remiro]

(In [1146]) New interpolation methods in NameBlock StdLib::interpolate
Fixes #707