Core Path · L6.3.08

LGD models

14 minLGDWorkout LGDCure modelBeta regressionFractional logitDownturn LGD

By the end you will be able to

  1. 1Compute a workout LGD from dated recoveries and costs, and show how the discount rate, the weighting and open workouts move the long-run average.
  2. 2Model LGD in two stages, a cure logit and then a beta regression or a fractional logit, fitted by maximum likelihood (M-D08: Maximum likelihood estimation).
  3. 3Estimate downturn LGD from the defaults of a downturn and set it beside the booked add-on and the EBA's guidelines.

Why it matters

L3.3.08: Loss given default and recovery averaged twenty booked workouts and added a flat 8 points for a downturn. Now there are 598: every default of the corporate panel, ten years of them. Their average LGD is 50.4% or 43.9% depending on whether you discount, 47.4% if you weight by size, and the individual LGDs spread from zero to above 100%, with few near the average. A capital number that rests on LGD needs a model that respects that shape, and a downturn estimate the regulator will accept.