basic swap curve question

[thetrystero]

i'm trying to compute zero rates for malaysian klibor which pays quarterly coupon. i'm using the forward substitution method, as per formula below, but when i compare the zero rates to murex, the numbers get gradually worse with increasing tenor -- reaching differences of several percent for zero rates in the 15 yr tenor. what am i doing wrong here?

(D_n = [1-(r/4)\times\sum_{i=1/4}^{i=n-1/4}D_i] / (1+r/4))

E, F, G
rates, t, DF, DF Formula
3.673129893,0.25,0.9909254,'=1/(1+F43*E43/100)
3.695923196,0.50,0.9820021,'=1/(1+F44*E44/100)
3.71782616,0.75,0.9729450,'=1/(1+F45*E45/100)
0.916911484,1.00,0.9641486,'=(100-E46*SUM(G43:G45))/(100+E46)
0.958632695,1.25,0.9533779,'=(100-E47*SUM(G43:G46))/(100+E47)
0.958632695,1.50,0.9443253,'=(100-E48*SUM(G43:G47))/(100+E48)
0.958632695,1.75,0.9353586,'=(100-E49*SUM(G43:G48))/(100+E49)
0.958632695,2.00,0.9264771,'=(100-E50*SUM($G$43:G49))/(100+E50)


here are the rates:

DEPOSIT,o/n,3.504831723,
DEPOSIT,t/n,3.502331963,
MYR KLIBOR,1m,3.604654913,
MYR KLIBOR,2m,3.628793086,
MYR KLIBOR,3m,3.673129893,
MYR KLIBOR,6m,3.695923196,
MYR KLIBOR,9m,3.71782616,
MYR KLIBOR 3M,1Y,3.667645934,
MYR KLIBOR 3M,2Y,3.834530781,

 

[PatM]

day count convention?

Could be day count convention?
For the first three, you are using the exact fractions 0.25, 0.5, and 0.75 of a year.
It could be 91/365, 181/365, 273/365 of a year?
If you kept this tiny inaccuracy going over many periods out to 15 years, the accumulated error could be substantial.
I don't know the day count convention in Malaysia. I would send an email to Murex asking for their methodology paper for Malaysia.

Additionally, some switch between discrete rates and continuously compounded rates in one step for interpolation when building a zero curve. I don't know if that would make any difference in this case, but the method is slightly different from what you are saying in this post. Very small differences in these methods end up making bigger differences when you bootstrap repeatedly all the way out to 15 years. Google "zero curve methodology". I don't know exactly what Murex does for this.