1
$\begingroup$

If a Treasury issue is trading special (100bp repo advantage vs GC in overnight repo), and it's worth -10bp in carry (assuming duration is 10) and the error to the Treasury spline is -5bp, would this mean the issue is cheap given the following :

1) The average error to spline is -2bp over the past 3 months

2) If we adjust the spline error (-5bp) to remove the effects of the repo advantage (-10bp), it would be -5 - (-10) = 5bp cheap to the Treasury spline.

I am expecting convergence from 5bp to -2bp. Am I missing something?

Improve this question
$\endgroup$
0
3
$\begingroup$

A bond like that should trade through the spline curve by an amount that reflects how long it is expected to remain special in the repo market. For example, 1 year at 100bp special is worth about 10bp on a 10yr note. However, 6 months at 100bp special is only worth 5bp of yield. Try to estimate how long it will remain special. You might want to look at the repo behavior of most recently issued notes as further auctions occur. Typically, as bonds get "old" they get less special.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thanks, but I'm trying to adjust (remove) the effects of specialness to get the true RV picture. Let's say I'm comparing two Treasury issues and one is more special than the other. It's difficult to make apples/apples comparison so I am removing the specialness effect to calculate the adjusted spread to spline. $\endgroup$ – VanillaCall Aug 4 '18 at 23:20
  • $\begingroup$ Yes I know. You can't do it unless you know how long the specialness will last. $\endgroup$ – dm63 Aug 5 '18 at 4:35
  • 1
    $\begingroup$ @dm63 is right. For example if the specialness only lasts 1day it is effectively not special. To estimate the duration of specialness, you can consider the relative adjustments to term repo. Its quite complicated though since the amount of data you have (some term repo prices) is not necessarily enough to parameterise a probability distribution of the amount of specialness on each future date, but you can formulate a model, +1 $\endgroup$ – Attack68♦ Aug 5 '18 at 15:26

Not the answer you're looking for? Browse other questions tagged or ask your own question.