Maximum Information, the London-based start-up specialising in catastrophe modelling analytics, has launched a new tool aimed at automating cat model evaluation and understanding how climate change could affect modelled losses.
The Magniphi app – launched at RAA’s Cat Risk Management Conference in Orlando, Florida last month – is designed to assist insurers and other model users with the production of model evaluation documentation, the ability to stress-test different views of hazard and understand whether the model tallies with historical losses.
The app will also allow users to understand how various views of climate risk affect modelled losses, which will enable firms to better grasp potential climate impacts and respond to regulatory climate stress tests.
Regulators and investors are increasingly demanding that insurers demonstrate an informed view of their risk. While larger re/insurers will have in-house model evaluation teams, most of the market relies on model vendors or brokers for model evaluation, as independent assessment is time-consuming and expensive. Maximum Information said the Magniphi app will facilitate more firms performing their own evaluations.
Chief executive Tom Philp said the app originated in a desire to provide cat modellers and exposure managers with the tools to rapidly stress different views of risk and “get under the hood” of the model. “That very quickly iterated towards formalising model evaluation questions that could be tested with this type of platform,” he explained to InsuranceERM.
The interactive app is populated with historical data on loss and hazard, which will allow rapid comparisons with vendor event sets and loss tables.
“By connecting directly into vendors’ stochastic event sets and loss tables derived from industry exposure databases, Magniphi automates the heavy lifting associated with model evaluation,” Philp said.
The app produces a cohesive template for model evaluation, but Philp emphasised “it’s also an interactive web app, so people can see, for example, how different views of hazards impact the overall frequency-intensity distribution, as well as local impacts. If you have more exposure in Florida, say, compared to the rest of the US, users can drill down and examine the data there.”
At launch, the app will cover four peril regions: US hurricane; US severe convective storm; US earthquake; and US wildfire.
“A key reason for choosing all US major perils was that we knew we could rely on data from PCS [Verisk’s loss data service] to give an independent view of losses for the loss validation piece. In other parts of the world, it can be more difficult to get this information,” Philp said.
Magniphi operates on a software-as-a-service basis and Maximum Information is currently inviting users to become “early adopters” and participate in development, before it moves to normal pricing on 30 June. Philp said the company was already implementing user suggestions: for example, the ability to generate documentation to support a regulatory request for model version change impacts.
Philp said the app had generated strong interest at the RAA event and “we seem to have meetings every day now with potential clients”.
More philosophically, Philp sees Magniphi as part of a change in approach to model evaluation that he’d like to see adopted across the industry.
“While George Box’s famous quote that ‘All models are wrong, but some are useful’ is undoubtedly important to keep in mind, I think it has the potential to be quite dangerous. The use of the word ‘wrong’ implies the model can be the system that it's trying to predict, when in reality a model is just a model,” he said.
“I like to reframe the quote and say: ‘All models are right because they are models, not the actual system.’
“If you come to model evaluation with this in mind, I believe it changes the model evaluation process. It’s no longer just about trying to find out how much the model is wrong by; it’s about discovering what one model offers that another model doesn’t. And I think that's what Magniphi and the exhibits it produces are very neat for doing.”
Philp spoke to InsuranceERM last summer about how insurers should reconsider the use of catastrophe models.