Evolving risk and capital modelling

In part one of this InsuranceERM and SS&C Algorithmics roundtable, chief risk officers and risk experts discuss how the UK's Solvency II reforms, and the growing requirement to perform scenario testing on liquidity and climate change, are affecting their approach to modelling risk and capital

Q: What are the implications of the Solvency II reforms, and in particular, the impacts on risk and capital modelling?

Chris Craig: It's making the industry further aligned to where HM Treasury are, in terms of wanting to show there's better capacity to invest in UK assets. It's also an acknowledgement of some of the issues we might have had in getting approvals in the past.

In terms of modelling, when you put Solvency II reforms alongside other reforms around model risk management, resilience and stress and scenario testing, it's bringing a more holistic approach to how we ought to think about our modelling. That's going to require a bigger range of inputs and, critically, better communication skills for the people doing the modelling.

David Harrison: The immediate impact is minimal, but there is a real will from the regulator to support investments in sustainable and productive assets. The matching adjustment (MA) rules maybe aren't the panacea that they could have been to unlock that investment, but they are helpful.

As this evolves, hopefully, it will develop a more holistic view of modelling and potentially a bit more of a conversational approach with the regulator. If it does, we will have the opportunity to say 'this is what to do, this is why we think it's a good idea, this is why we think it's in everybody's best interests' – and that will drive improvements in the modelling approach.

Keith Davies: Solvency UK should create the flexibility to do things differently. But supervisors also want to ensure consistency and a level playing field, so may be reticent to grant flexibility to new incumbents. The will is there, but it will take time to filter through.

Guy Barton: To Keith's point, credit models are quite different, so that makes opening it up more difficult, because it's not obvious how it relates to someone else's model. The complexity of the different credit models out there probably makes it harder for the regulator to become more comfortable.

Patrick van Beek: The 'highly predictable' treatment in the MA gives you a bit more flexibility, but it's a maximum 10% of the MA portfolio, so it's going to limit its applicability.

William Diffey: For multinational groups – which might have internal models in several jurisdictions – there is a level of regulatory divergence, which is itself creating additional work and complexity. My personal view is that the cost / benefit of this needs further work to meet the needs of stakeholders.

Guy Barton: On the modelling side, it's probably only notching [of credit ratings] that directly induces a change to the modelling. The rest of the changes – the MA testing and the 'highly predictable' cash flow testing – are making sure you consider all the risks, and that's potentially encouraging you to become more granular in modelling those asset classes.

Patrick van Beek: You may end up with a bifurcation of your modelling estate. Ideally, you would use the same model for your risk and capital, but the MA attestation may require something more granular that you wouldn't necessarily put into your internal model.

David Harrison: It becomes a materiality question: if you making an adjustment to the fundamental spread of an asset because there are additional risks to the asset that are not captured in the credit rating, you should probably be modelling that in your SCR [solvency capital requirement]. But at a portfolio level, it's unlikely to be proportionate to do so.

Ravin Jagatiya: I agree the reforms are taking the industry in the right direction. The challenge from the internal model perspective is how to allow for all the elements in a proportionate manner. There is a balance to be struck between model complexity, operational efficiency and cost. I expect the first couple of years for firms might be challenging as new processes are embedded into the business.

David Chen: One of the reform changes is the removal of the 'cliff edge' effect in the MA [which limited the MA benefit from sub-investment grade assets]. From a risk manager's perspective, we need to be mindful of the risk of a gradual drift in credit ratings – especially in the current tight spread environment.

Guy Barton: There are good and bad things about more granularity. One good thing is in helping There are good and bad things about more granularity. One good thing is in helping understand the risk concentration in a portfolio. By going into detail, it might help you think about the correlations between asset classes, but complexity can also make your model harder to understand.

Also, most people overestimate how much data there is on credit risk. Until you have a crisis, the volume of the data is much more limited. Look at the transport sector: it would have the worst rating of any sector because of the railroad companies that went bust in the 1920s, but it's not relevant today. The skill is understanding the risk while avoiding the trap of over-using data.

Chris Craig: Building a model can be relatively easy. Understanding the limitations of the data, the model and the output – and the understanding of the people who have to use the output – is probably an area where we need to invest as an industry.

Q: How do we use the outputs of scenario testing – for example, for climate change and liquidity – in our capital models?

Patrick van Beek: It's always been an interesting dynamic between liquidity and capital. In our capital model, we might find the biting stress that gives rise to the biggest SCR involves a scenario where yields and interest rates are falling. Whereas from a liquidity perspective, it might be the opposite way around, as in we're having to face collateral calls if interest rates rise.

Trying to catch that in one model is quite difficult, so we often end up with two separate frameworks: one telling you what the SCR should be, the other saying how much liquidity you should be holding. That creates challenges because you don't have a consistent modelling architecture.

David Chen: I agree you need two models and two separate frameworks – liquidity is really different to capital. But climate is slightly different. Liquidity is something that could kill the business tomorrow, whereas in the context for annuity firms, the timeframe for climate change could be much longer.

William Diffey: In the non-life business, capital and liquidity can be very linked. If you're insuring property exposed to North Atlantic hurricanes, one of the key risks is around liquidity to settle claims. The modelling is very much event-driven, and you have to deal with the complexities of catastrophe modelling and how that links back to capital model, in a consistent framework. This includes thinking about climate change trends.

Ravin Jagatiya: In the internal model, you might not want to – or be able to – model liquidity in the same way as you would in a separate model dedicated to measuring liquidity risk. However, you could consider from a qualitative lens how liquidity events could affect solvency (or vice versa) and assess whether this impacts the calibration of the internal model.

David Chen: The way I think about it is the risk drivers could be different. Long-term liquidity risks are driven by longevity stresses and credit events. For short-term liquidity, it's about volatility in interest rates and foreign exchange rates, and the impact on collateral. We just need to be cognisant of that.

Keith Davies: There are two main types of climate risk: physical and transition. In the general insurance business, we are seeing natural catastrophes becoming more frequent and more severe because of climate change. Those physical risks to liabilities have currently meant more focus on liquidity (to meet claims) and pricing than capital. On the life side, where insurers hold long-term assets, there's more about the transition risk and perhaps felt more on capital rather than liquidity.

Chris Craig: We've been talking about liquidity and capital models, but if you think about the potential causes of failure of an enterprise, we should also mention the operational side. You might have plenty of capital and liquidity, but a material event to software suppliers or ransomware may bring different challenges. That brings us back to the holistic nature of modelling, and how we ensure our operational, capital and liquidity modelling all fit together and we're able to tell a consistent story.

Q: How do you bring the information you get from climate scenario modelling into your models?

Keith Davies: It's typically through adjustments to catastrophe model settings impacting different perils. For the different types of weather event, through the models you would predict what the risk and severity is for the next year or over longer time horizons. That obviously depends on the climate scenarios and the data set you use. In individual pricing, you can drill down a bit more into risk selection, but it's very much a developing area. Indeed, some types of events are not modelled in industry tools – freeze events, for example. There, we have developed our own model and rely even more on our own data and expert judgement.

David Harrison: Climate is an issue for life insurance too, particularly for long-dated assets. When you're looking at a private asset with a 20-year duration, it's difficult to think about what the physical and transition risk impacts would be. For example, gas pipelines are a secure and stable asset, but equally, it's easy to imagine a government saying they will transition everyone to using heat pumps, or hydrogen. That gets considered at the point of investment and is not necessarily baked into the model. But going back to the fundamental spread adjustment in the MA, that is the kind of thing we should be starting to consider.

Patrick van Beek: It's quite challenging to think how you're going to use MA attestation to address climate risk, especially when there is a lack of data on credit generally.

David Chen: And the rules say it has to be done with 'a high degree of confidence'.

Patrick van Beek: You might have some estimation of how ratings might change over a one-year time horizon, but trying to forecast that 20 years into the future and allowing for transition risk is very challenging.

Leo Armer: SS&C Algorithmics has been working with academic experts and data vendors to integrate climate within our market and credit risk framework. Our clients require [support] to evaluate the impact of physical and transition scenarios on investment assets and portfolios at the most granular level, with the ability to calculate climate VaR in a stochastic way. The rest of the financial industry is also looking at insurance, because you have a lot more familiarity with this data on the liability side than the banking world has.

Ravin Jagatiya: If we can overcome the challenges of incorporating climate risk in the capital models then it is important for the industry to do this collectively. There is no incentive to be the first mover: if you are the only insurer doing this it could adversely impact your competitive position. This is where clearer guidelines from the regulator would be helpful.

Guy Barton: There is value in trying to understand how physical and transition risks relate to each of your asset classes, and how worried you should be about that from an overall portfolio construction perspective.

Chris Craig: That brings with it the need to ensure the climate risk assessment is properly embedded in our enterprise risk frameworks. It could be quite easy to separate it – and in the early days some did see it as a separate discipline – but we need to make all our teams aware of climate risk and it can start feeding through.

Q: How do you go about choosing your climate scenarios?

Keith Davies: You have to go for something that is understood and recognised, because it is very much about judgement. You can use NGFS scenarios as a base, and as you modify and understand the scenarios, it's about the actions you put into practice rather than the actual path.

The problem is it's impossible to model with confidence. I make the comparison with the trend in UK longevity: everyone knew it was going up, and every year it was going up by more. It feels like that to me with climate change. We know weather events will become more frequent and more severe, but we don't know if it's in three, five or 15 years' time, especially as it's possible there is a tipping point and step change in climate. 

InsuranceERM will publish part two of the SS&C Algorithmics roundtable next week.