Climate Risk Data Lineage and Proxies

By Jonas Osman Abdelghafour.

Climate-related financial risk analytics — the ECB thematic reviews, the TCFD recommendations, and IFRS S2 disclosures — require institutions to map their exposures to physical and transition climate hazards and to quantify the resulting risk. In practice most of the difficulty is not in the modelling but in the data: physical asset locations, sector classifications, emissions intensities, and hazard scenarios rarely align cleanly with the counterparty-level data institutions actually hold. A disciplined approach to lineage, proxies, and uncertainty is what makes the exercise defensible.

Exposure mapping is the first bottleneck

Climate risk analytics operate on exposures at the level of physical assets (real estate, infrastructure) or economic activities (sector, sub-sector, geography). Financial systems typically carry exposures at the level of counterparty, facility, or security. Mapping between the two is a data problem before it is a modelling problem.

A workable exposure map records, for each exposure: the counterparty identifier, the physical asset or activity identifier where available, the sector classification (NACE, GICS, or equivalent), the geography at the finest available level, the emissions intensity where relevant, and the source of each field. The source field is what makes the exercise auditable.

The proxy hierarchy

Almost no institution has actual, verified data for every exposure. A proxy hierarchy makes the fallback explicit and consistent. A typical hierarchy is:

1. Reported counterparty data (verified emissions, disclosed asset locations).
2. Third-party estimated data (commercial providers such as CDP, MSCI, S&P Trucost).
3. Sector-and-geography averages from national statistics or IEA/EDGAR datasets.
4. Sector average only.
5. Portfolio-wide fallback with an uplift for uncertainty.

Each exposure is tagged with the level of the hierarchy it uses. Analytics can then be reported both including and excluding proxy-heavy exposures, and users see how much of the answer depends on the least reliable inputs.

Uncertainty quantification

Climate scenarios themselves (NGFS reference scenarios, IEA WEO cases) carry substantial uncertainty in transition-policy pathways, technology assumptions, and hazard translation to financial impact. Layered on top, proxy-driven exposure data adds a further band of uncertainty that is often larger than scenario uncertainty for individual portfolios.

Uncertainty should be reported alongside point estimates. Common approaches include scenario ranges, sensitivity to proxy hierarchy level, and — where data supports it — Monte Carlo propagation of proxy uncertainty. A single-number climate loss estimate without uncertainty bands overstates the precision of what is actually a materially uncertain exercise.

Controls that make the data reliable

Four controls have disproportionate impact. First, data lineage documented from source system to analytical output, so any climate metric can be traced back to the fields it depends on. Second, refresh cadence tied to source data update frequency, so stale proxies do not silently persist. Third, coverage reporting that shows what fraction of exposures uses each hierarchy level, tracked over time. Fourth, independent review of the mapping and proxy logic by the second-line function, in line with actuarial model governance and the three lines of defence.

Governance

Climate risk metrics increasingly feed into risk appetite, capital planning, and disclosure. That elevates the data quality requirement. Board and risk committee reporting should include, alongside the headline exposure and loss metrics, an explicit statement of coverage, proxy dependence, and known limitations. This is consistent with the direction of ECB supervisory expectations and with IFRS S2's requirement to disclose material assumptions and uncertainties.

Common failure modes

Three patterns recur. Proxy usage is not tracked, so users of the analytics cannot tell where the numbers are anchored. Scenario outputs are compared across institutions as if they were comparable, when methodological differences dominate. And climate metrics are reported without uncertainty bands, giving false comfort about a genuinely uncertain exercise.

Limitations

Even a well-controlled climate risk data infrastructure remains subject to scenario uncertainty, model risk in the hazard-to-financial-loss translation, and the possibility that observed climate change diverges from all currently used scenarios. Data controls reduce the noise; they do not eliminate the underlying uncertainty.

Conclusion

Credible climate risk analytics are a data-engineering discipline as much as a modelling one. Lineage, an explicit proxy hierarchy, uncertainty reporting, and independent review are what let the outputs be used for decisions rather than treated as illustrative. Related notes: predictive modeling in insurance and ORSA scenario design and board use.

Written by Jonas Osman Abdelghafour, actuary and financial risk manager. Background and contact details are on the about page.