Adams speech to the Economist Insurance Summit - lessons from financial crisis

Some particularly useful context setting from Julian Adams last week for anyone in the Internal Model game, with this speech to the Economist Insurance Summit around what lessons could be learned by insurance supervisors from the financial crisis.

While he amusingly interchanges between "financial crisis" and "banking crisis" to emphasise that it wasn't our fault, and drops in the now obligatory reference to the importance of insurers as long-term investors, echoing the Commission's pleas from late last year, the majority of the speech focuses on why models go wrong (not the name of a ropey catwalk reality tv show...)

Insight on where the FSA thought firms were going awry in the Solvency II modelling preparations was delivered to the industry in the middle of last year, but I found this speech helpful in the context of proportionality i.e. what elements of economic capital modelling are worth spending extra time on theorising, documenting, debating and minuting for IMAP candidates. I saw the following comments as highlights;

Reasons for internal models in the banking industry being exposed;

• "...rested on assumptions which turned out not to hold when bad times came"
• "...review period" selected when parameterising
• "...insufficient rigour and independence from the front end of the business" when parameterising
• "...management attention too often focused on those parameters considered too conservative at the expense of those that were insufficiently prudent"
• "...destabilising feedback loops" where underestimation of risk (due to data selection) plus use of the model leads to a vicious cycle of unacknowledged over-accumulation of risk
• "...flawed technical assumptions" in tail-end probability estimation where data is drawn from "normal" times

Lessons for Solvency II

• "Data [should be] sufficiently robust"
• Assumptions should be "appropriately conservative"
• "[Supervisors] can be helped...by the much greater use of imaginative tests of resilience to deeply stressed scenarios"
• "...paucity of relevant historical data for the calibration of tail dependencies between risks"
• That "...the limitations [of capturing tail dependencies] are recognised, and conservatism built in to the calibrations"
• That "...correlations in the tail are likely to be assymetric in nature" for insurers
• That "...the adoption of quantitative techniques...will not change the nature of the risk itself"
• That supervisors "...must not blindly accept the outputs of these models"

Appreciating some of this is hardly new news, any increased documentation and rigour in the areas highlighted will no doubt be well received down at the Wharf.