David Walker compares how firms describe internal models in their solvency and financial condition reports and considers how the practice may evolve
Describing an internal model in simple language is no mean feat. To win approval to use the models under Solvency II, EU insurers in the first place had to pen tens of thousands of words. But the publicly disclosed explanations, demanded in solvency and financial condition reports (SFCRs), have typically run to the length of a school essay.
One UK consultant said distilling years of work and complex models into just 1,900 words – the average length of descriptions published by Europe’s largest groups – “was certainly a challenge”.
But he added that what emerged was pleasing to read, and seemed largely to satisfy the requirement that even a layperson could grasp content of SFCRs, including on internal models.
"It is about describing how you use your model, how it works - at a high, but meaningful, level"
Catherine Drummond, a partner at consultancy Lane Clark & Peacock (LCP) in London, who has analysed SFCRs from 2016, counselled insurers discussing their internal models in this year’s reports to keep who reads them in mind.
This might prove challenging, given most insurers did not plan tracking readership, according to conversations last year with InsuranceERM.
Drummond said: “The SFCR is not for the regulator, that’s what the regulatory supervisory report (RSR) is for. So it’s not about giving away commercially sensitive information, but it is about describing how you use your model, how it works - at a high, but meaningful, level - and the main differences compared to the standard formula.”
More than words
She said some of the more successful reports filed last year, to her mind, had “high-level diagrams of how the modelling process works and concise summaries describing the key differences versus the standard formula”.
Some went beyond the minimum information required, to reveal whether risk model outputs from internal models over or undershot those from the standard formula, “giving readers of these reports much needed context and additional clarity”.
"Instead of talking about the '99.5th VaR', describe the process of ordering losses by size and picking the hundredth highest out of 20,000 simulated outcomes"
She noted that the European Insurance and Occupational Pensions Authority did not expect internal model details to appear in the SFCR’s summary section, which is the section meant to be understandable to policyholders and beneficiaries. Readers of the chapter most relevant to internal models, namely section E detailing capital management, are more likely to be analysts and investors with greater financial understanding.
But Drummond continued: “That said, it’s clear that describing a stochastic model to a non-actuarial audience will be challenging. The key is to stick to the key points and not get too bogged down in the detail. Instead of describing copulas, explain how it is important to model the possibility of many things going wrong at the same time, and how this might change under different circumstances.
“And instead of talking about the ‘99.5th VaR’, describe the process of ordering losses by size and picking the hundredth highest out of 20,000 simulated outcomes. The challenges aren’t insurmountable if this aspect of the reporting is led by someone who is skilled in communicating complex issues in an easy to understand way.”
Length does not correspond to size
On analysing text on internal models in group-level filings, Insurance Risk Data (the data service companion to InsuranceERM) found a wide array of methods to describe bespoke models. Some still wrote treatises. One dispensed with words almost entirely, using illustrations instead.
Among seven major internal model groups examined, Scor wrote most (3,527 words) about its IM, while Pension Insurance Corporation (PIC) was briefest, at 938 words. Axa Group struck the middle ground (1,637 words) but noted more was in its last annual report.
The size of an insurer, by gross written premium, did not determine the length of internal model discussion, however. Aspen Insurance UK (2016 GWP: £392m) wrote 1,000 words – more than PIC, whose GWP was £2.6bn – including details of its internal model governance, its independent reviewing before the regulator granted approval in December 2015, and details of continual, internal scrutiny.
"Scottish Widows honed in on the ability with its internal model to fit joint-loss functions to paired risks"
Internal model discussions were structured around governance and validation, differences to the standard formula, how widely insurers use internal models, and the data for them.
All insurers explained they decided to build internal models – not taken lightly given labour and costs involved - because the standard formula was manifestly inadequate.
PIC said the standard formula “does not represent the risks to the business [so] does not incentivise good risk management, with actions being taken to optimise a position under a formula rather than aligned to the risks.”
Munich Re added solvency capital requirements generated by standard formula are “largely inappropriate for most [reinsurers’] business segments and regions. The standard formula was designed for a mid-sized European primary insurer, not for a global reinsurer [like Munich Re].”
Scottish Widows honed in on the ability with its internal model to fit joint-loss functions to paired risks “to reflect the interaction between risks [occurring] simultaneously” - disallowed in the standard formula approach – and the possibility of diversification between matching adjustment (MA) and non-MA funds. It also criticised the standard formula’s underlying assumption all risks are Gaussian. It then showed, diagrammatically, how the internal model calculates its SCR.
Vienna Insurance Group also opted to illustrate, almost without words, how its partial internal model differs from the standard formula.
As insurers prepare for their second round of SFCR reporting, due in May-June, it remains to be see how they will alter – if at all – their descriptions of internal models.
Most initial attempts at producing an SFCR have resulted in a document that includes plenty of detail, but possibly more detail than is needed. However, just like the evolution of the own risk and solvency assessment (Orsa) document, the next iterations of SFCRs are likely to be more concise and take lessons from others that have used diagrams instead of words.
- All SFCR reports, including commentary on internal models by their users, and associated quantitative reporting templates (QRTs) are available from InsuranceERM as part of a new insurance risk data service. This service combines European insurers' financial and regulatory filings, including the new Solvency II disclosures, into a single, comprehensive and user-friendly database ideal for market/peer analysis, research and benchmarking. To find out more please email email@example.com