Bio-risk modelling for finance

A biodiversity footprint for an investment portfolio looks like a measurement and is not one. Nobody visits the sites. The number is produced by a chain of models, and understanding where it can be trusted means understanding what each link in that chain actually does.

The chain

It starts with money. An input-output model of the economy converts a company’s revenue or spending into physical throughput by sector and country, including the upstream tiers a company never sees. Those quantities are then multiplied by pressure coefficients: hectares of land occupied, cubic metres of water abstracted, tonnes of nutrients released per unit of output. Finally a response function converts pressure into a biodiversity quantity — commonly mean species abundance relative to an undisturbed reference, or potentially disappeared fraction of species — which is summed across the portfolio and reported as a single figure.

The response step usually rests on the species-area relationship, the long-standing empirical observation that species richness scales with habitat area as a power law with an exponent typically between about 0.15 and 0.35. It is one of the more robust regularities in ecology. It is also silent about which species, and about timing: habitat loss commits species to extinction that persist for decades afterwards, the extinction debt, so the relationship describes a future equilibrium rather than a present state.

What the resolution really is

The binding limitation is spatial. Because the first link is an economic table indexed by sector and country, two companies in the same sector and country receive near-identical footprints per unit of turnover, whatever their actual sites. Yet siting is precisely what determines ecological consequence — a hectare converted inside an irreplaceable endemic range and a hectare of already-degraded pasture are not the same hectare, and the model as constructed cannot tell them apart. Increasing decimal places does not increase resolution.

The dependency side has a different defect. Frameworks that score how much a sector relies on pollination, water-flow regulation or coastal protection produce ordinal ratings — high, medium, low — derived from expert judgement. Ordinal labels can be ranked but not meaningfully averaged, and the common practice of weighting them into a portfolio-level score treats them as though they were quantities.

Ecosystem services also behave nonlinearly. Pollination falls off with distance from remaining habitat, aquifers draw down past recovery, fisheries collapse rather than decline smoothly. A linear footprint model cannot represent a threshold, and thresholds are where the financial loss occurs.

Why validation is unavailable

Credit risk models are disciplined by data: decades of default series exist, so a model that predicts badly is visibly wrong. Nature-risk models have no equivalent. There is no long history of financial losses cleanly attributable to biodiversity decline, because the losses arrive through weather, disease, yield and regulation, and are recorded under those names. Without a loss series, these models cannot be back-tested, only reviewed for plausibility.

That does not make them useless. A screen that identifies which holdings sit in water-stressed basins or depend on a single pollinated crop is genuinely informative about exposure. Read as an ordering of concerns it is defensible; read as a valuation of nature it is not.