Who should interpolate Libor: submitter or administrator?

ICE Benchmark Administration (IBA), the new administrator of Libor has published a position paper on the future evolution of Libor. The core of the paper is a shift to “a more transaction-based approach for determining LIBOR submissions” and a “more prescriptive calculation methodology”. In this post, I discuss the following IBA proposals regarding interpolation and extrapolation:

Interpolation and extrapolation techniques are currently used where appropriate by benchmark submitters according to formulas they have adopted individually.

We propose that inter/extrapolation should be used:

1. When a benchmark submitter has no available transactions on which to base its submission for a particular tenor but it does have transaction-derived anchor points for other tenors of that currency, and
2. If the submitter’s aggregate volume of eligible transactions is less than a minimum level specified by IBA.

To ensure consistency, IBA will issue interpolation formula guidelines

Para 5.7.8

In my view, it does not make sense for the submitter to perform interpolations in situations that are sufficiently standardized for the administrator to provide interpolation formulas. It is econometrically much more efficient for the administrator to perform the interpolation. For example, the administrator can compute a weighted average with lower weights on interpolated submission – ideally the weights would be a declining function of the width of the interpolation interval. Thus where many non interpolated submissions are available, the data from other tenors would be virtually ignored (because of low weights). But where there are no non-interpolated submissions, the data from other tenors would drive the computed value. The administrator can also use non linear (spline) interpolation across the full range of tenors. If submitters are allowed to interpolate, perverse outcomes are possible. For example, where the yield curve has a strong curvature but only a few submitters provide data on the correct tenor, these will differ sharply from the incorrect (interpolated) submissions of the majority of the submitters. The standard procedure of ignoring extreme submissions would discard all the correct data and average all the incorrect submissions!

Many people tend to forget that even the computation of an average is an econometric problem that can benefit from the full panoply of econometric techniques. For example, an econometrician might suggest interpolating across submission dates using a Kalman filter. Similarly, covered interest parity considerations would suggest that submissions for Libor in other currencies should be allowed to influence the estimation of Libor in each currency (simultaneous equation rather than single equation estimation). So long as the entire estimation process is defined in open source computer code, I do not see why Libor estimates should not be based on a complex econometric procedure – a Bayesian Vector Auto Regression (VAR) with Garch errors for example.

Posted at 12:08 pm IST on Thu, 23 Oct 2014         permanent link

Categories: benchmarks, statistics

Comments