Rita Chretien, a fifty six year old from Canada, survived seven weeks trapped in the winter mountains of Nevada after a road trip to Las Vegas went spectacularly wrong. Sadly her husband Albert who set out on foot for rescue on the third day of the ordeal was less fortunate. He has never been seen since. The reason they got so lost was a fatal decision to follow the GPS for a shortcut to their hotel. Unfortunately their navigation system omitted seasonal conditions. The old mining track the Chretiens took became impassable and they buried their vehicle axle deep in a mud hole; a tragic story of the dangers that lurk in a world where man increasingly relies on the machine to take decisions.
Not a matter of life and death of course but what is going on in the banking sector as financial institutions bed in the Basel III regulatory framework is also testing the faith of investors in what the computer is telling them. As research from Barcap revealed this week, there is a huge variation between banks and year on year within the same bank in the estimates of risk and capital required for the same or similar categories of loans. Evidently there is a widespread belief, according to the Financial Times, that banks are massaging their models to cut their capital requirements. If true, that the banks are in a position to do this, is because the regulations permit them to use internal models rather than a common industry standard formula. Such is the level of concern it also reported that some senior officials in the Bank of England and Federal Deposit Insurance Corporation are calling for an end on the reliance on internal models.
Leaders of the insurance industry will be watching nervously as the European regulators start to tackle this issue. The implementation of Solvency II lags Basel III but closely follows its principles and structure. Given the choice, London Market specialist insurers, like the bankers, have so far opted to develop internal models to set their capital. Currently all are engaged with FSA in an internal model approval process, IMAP as it has become known. Despite the challenges of this lengthy validation exercise, insurers would be less happy, to put it mildly, if instead they were mandated to set (higher) capital levels by applying the standard formula within the Solvency II regulations, as ultimately might be the case for the banks under Basle III.
Banks and insurers will argue the case for a tailored modelling approach that incorporates a range of factors unique to their own business and experience. So the argument goes BNP Paribas, for instance, is correct in holding half the capital that Lloyds Bank assigns to its book of A-rated corporate loans because their historical default rate has been markedly lower. We know there is also a wide variation between insurers in risk profiles, degrees of volatility and underwriting performance that are all but impossible to factor into a standard one-size-fits-all formula.
In demonstrating that the internal models they are developing assess risk and allocate capital accurately, our business leaders have to convince the regulators that they will use them wisely. They may need to be super-persuasive. A survey of global insurers conducted last month by KPMG found a surprisingly high forty percent of respondents had only a limited management understanding of modelling. If such a knowledge gap exists then the consequences of not addressing it at board level could be quite damaging.
Where relative ignorance exists the temptation might be for managers to cling on to old style metrics, build strategies based more upon intuition and gut feel; to ignore or override the models if the outcomes are undesirable. Yet worse still, according to the authors of the research, is the other extreme that too much reliance could be placed by managers on model results without fully considering the boundaries within which they should be interpreted. Models may become expensive tick box exercises that can mislead managers into falsely believing that risks are adequately covered prompting actions that go against sound business principles.
By slavishly following their GPS the Chretiens discovered to their cost in Death Valley that models are not perfect; they do at times behave badly. Had the unlucky couple thought about and given weight in their decision making to the time of year, weather conditions, terrain and the type of vehicle they were driving then things may have turned out differently. Likewise, as the debate with regulators about internal models and industry standard formulas intensifies, our senior management will need to prove that they have a sufficient grasp of the models and their limitations to avoid driving their businesses like poor Rita and Albert down the corporate equivalent of a track heading to a mud hole in the desert.
One thought on “On Models Behaving Badly”
The folly of an industry-wide “addiction to prediction” is all the evidence anyone should need as to the lack of understanding of the capabilities of models!
This sorry tale reiterates the dangers of over-reliance on technology in unfamiliar or unpredictable environments…like the current economic landscape. One that simply cannot be predicted, even with on onboard supercomputer and numerous PhD’s using data that relates to an, artificially created, period of relative stability!