Pricing the mortality tail: two series in one harvest record

A marine site does not produce one mortality series. It produces two. Baseline mortality runs between 0.4 and 0.9 per cent of biomass a month and behaves like a cost of goods. Event mortality removes between 8 and 38 per cent of a cohort in under three weeks and behaves like a credit loss. Fitting one distribution to the combined record gives a number that is too high for ordinary months and far too low for the months that decide whether a facility is repaid.

Quantitative Strategies separates the two series at a threshold of 2.5 per cent monthly mortality. The working set holds 9,400 site-months contributed by eleven operators across 26 marine sites, each dataset admitted under the provenance standard that applies to every input the division uses. Seventy-one site-months sit above the threshold. That is one event for every 132 site-months of operation, or roughly one event per site every eleven years. The median event removed 14 per cent of standing biomass and the worst removed 38 per cent.

Events are correlated inside a water body and close to uncorrelated across water bodies. Of the 71 events, 23 fell in clusters of two or more sites inside the same water body within 30 days. A lender holding three sites in one inlet therefore holds something near one risk. A lender holding three sites in three separated bodies of water holds something near three. Diversification in this sector is a question of hydrography, not a question of counting cages or counting licences.

That distinction sets the advance rate. Against standing biomass the division supports 55 per cent where a borrower concentrates its sites in one water body, and 70 per cent where the sites sit in three or more separated bodies of water. The price difference between the two structures is 180 basis points, which is close to the difference in modelled expected loss over a three-year facility. Neither figure is a negotiating position. Both are outputs, and a borrower that separates its sites moves from one to the other.

The series will not price husbandry. Records end at the loss and rarely record the response to it. Twelve of the 71 events carry no documented cause at all, and no dataset the division holds distinguishes an operator that moved stock early from one that waited. The model therefore stops where the site visit begins. Every marine position is inspected under the welfare audit programme, and the inspection report sits beside the model output when the Investment Committee prices a facility.

The mortality model is registered under the Model Risk Management policy with a named accountable person, a quarterly back-test and a written statement of the conditions in which it must not be used. It is model-assisted work and it is disclosed to clients as such. No facility is priced by the model alone. A person signs the price, and that person can be named to any client who asks how the number was reached.

Published 2025-04-08 by the Quantitative Strategies division. Research is prepared for eligible counterparties and does not constitute advice.