Insurers are demanding ever greater realism when modelling risk and this has led to a step change in demand for computational power. Alun Marriott, managing partner at RPC Consulting, discusses how technology can help with the changing needs of the industry
What are the key modelling issues in the market today?
The big issue facing all insurers is how to ensure their analytic models actually add value to the business and do so in a timely fashion. The latter has often meant trading off complexity with computational effort, yet firms seek to model ever more extreme and complex risks – where, for example, simulation density in the tail is important. Generally, the more simulations you run, the more stable but slower the model (assuming that such models can be parameterised sensibly).
But complexity brings challenges – not least that a complex model may be harder to follow. As technology vendors, we have a duty to make modelling more intelligible. Users need to be able to develop, test, audit and deploy models in a collaborative framework and we’ve built Tyche to specifically address these needs.
We rely heavily on an intuitive visualisation of the analytics to make the process more comprehensible, coupled with a new actuarially friendly language called T# that picks the best bits of a range of different languages to make it easier for actuaries to support the demands of the business. Though T# is easy, fast to learn and specifically tuned for insurance actuaries, we also support more traditional languages too.
You say the need for speed is greater than ever. How does the industry address this?
With regards to speed, these days it is not just one actuary with one model. You are dealing with multiple users and many models. Some models can be enormous. In the past you could do the processing on a laptop. Now some larger models use CPU and memory than spans multiple servers.
Tyche has been built to address a wide range of hardware options, from the single-user-with-a-laptop through to the multi-user, multi-server multi-Terabyte solutions. Our business is high speed analytics.
The focus is on getting the business analysis to the client quickly and, as such, we have developed proprietary and innovative technologies that speed up low-level analytics whether on single laptops or across multiple machines working together seamlessly and invisibly to the user.
Tasks do not now need to be independent to scale across multiple machines – we call this Hive computing. The power of this Hive capability supports larger models with far higher accuracy than ever before, and in runtimes tens if not hundreds of times quicker.
Have insurer needs changed?
Though it may appear Insurer needs have changed little, there is an ever greater demand for improvements in model stability and runtime. The industry is demanding more and technology is answering the call.
As one example, Tyche’s power allows composites to model both life and non-life capital on a single integrated platform. Tyche’s power has also opened up the ability for high accuracy nested stochastic modelling – something previously thought a pipe dream.
Certainly the benefits for composites are huge, but the wider goal of full enterprise-level risk management where we further embed pension risk and other non-core risks into the general insurance, life or composite model, is the ultimate goal.
How important is it for insurers to test their models?
Model testing is absolutely key yet something often placed low or late on the priority list. As models change, model tests are often the last to be updated, if at all.
Tyche embeds testing into the heart of the process. Test and model scripts sit alongside each other and the deployment life cycle is managed concurrently. Code is worked on within the same develop, test, release environment, allowing for documentation to be automated and the life cycle from development to production usage tightly controlled.
Are spreadsheets still valuable for actuaries?
We think spreadsheets are great but the problem is governance and audit. Spreadsheets are good for storing summarised static data and presenting it. But we believe analytics should be maintained in a far more resilient environment.
It can be very hard to govern change where the model resides in a spreadsheet. Saying this, Tyche embraces and includes spreadsheet functionality, but we use it mainly for data in, data out and would discourage its use for calculations – not least as spreadsheets are very slow.