Victoria Jenkins, Managing Director
The Middle East and North Africa (MENA) region is acknowledged to be a key growth area for (re)insurance. Insurance penetration is rapidly increasing but still has some way to go to reach comparable levels with Europe or the United States. In the period 2003-2012, most countries in the region achieved triple-digit percentage increases in premium volume, with some exceeding 600 percent growth (source: Swiss Re Sigma).
It is quite normal by historical standards for investment in catastrophe risk quantification to lag insurance growth considerably. Building robust catastrophe models can be expensive and to make the investment commercially viable, the premium volume has to reach a critical mass.
The first available models for selected countries in the region appeared in the late 1990s. There has been a gradual expansion in coverage since this time, largely focused on earthquake and wind perils. The table below shows the current position in terms of the model availability from the three main vendor modeling companies, with an “x” designating no model provided by that vendor.
Rising Catastrophe Losses
In contrast to the comparative lack of investment in models, the risk from natural catastrophe losses appears to be increasing. With rapid urbanization, population growth and economic development compounded by climate change, catastrophe losses are on the rise around the world. This is particularly observable in the MENA region, where the average number of natural disasters has tripled since the 1980s, according to a study by the World Bank.
Floods have been the most recurrent hazard in MENA. Flash floods triggered by heavy rainfall on dry, drought prone land are a specific problem and can be disastrous. The 2008 flooding over the eastern governorates of Yemen, Hadramount and Al-Mahara resulted in an economic loss of YER327,551 million (USD1.5 billion), equivalent to 6 percent of Yemen’s GDP, according to the Global Facility for Disaster Reduction and Recovery.
Earthquakes are the second most prevalent hazard in the region, with almost all cities located in an area of seismic risk. The region sits at the junction of the Arabian, African and Eurasian plates, resulting in high tectonic activity.
The risk is real and the availability of models is patchy, so how are companies typically factoring catastrophe risk into their risk management processes? Models are used for rating agency, internal control purposes and to some extent, for reinsurance structuring and pricing. There is still a wide use of exposure-based methods in practice, for example, considering a percentage of country- or city-wide total sums insured to control accumulations and to calculate the amount of catastrophe limit to purchase. These factor-based methods can produce a useful cross-check against model results but can also produce much larger probable maximum loss numbers than the models themselves. This has led to reinsurers putting pressure on the size of the event limits offered in some areas.
Data Entered Into Models
Before we consider the quality of available models for the region we need to focus on the accuracy, completeness and appropriateness of the data that can be entered into them.
For catastrophe loss modeling to be accurate, knowing the location and the value of the risk is vitally important. This is the most basic information the model needs, but even this level of detail is not always available. Country-level total sums insured can often be the only data provided and so a method to disaggregate this information to a more localized level is required. Such techniques will use metrics like population density to inform the disaggregation. Companies should aim to understand the uncertainties that surround a lack of location-based knowledge and the implications for their risk profile. This can be achieved through sensitivity testing the hazard assumptions within the models by geographic location.
Catastrophe models can process much more information than just value and location. Additional detail on the occupancy or construction type of a risk can also be influential on the level of loss. It is also vitally important to capture good information in respect of policy terms and facultative protection to make sure an accurate financial perspective is produced. Using catastrophe models without good original information can lead to misleading results and suboptimal decisions, so investing in data improvements is a wise investment and is likely to pay dividends in terms of understanding the risk, giving a competitive advantage and can reduce loadings reinsurers may impose for data uncertainty.
Testing and Validation
Testing models is not a new activity. Brokers have been active in this space since catastrophe modeling began. Recent regulatory advances across the globe have been a catalyst for a renewed focus on model validation. It is no longer acceptable in certain regulatory regimes to blindly accept a model vendor’s results without challenge. The use of a specific catastrophe model or a blend of model results requires justification. Justification requires understanding and so the level of model documentation and transparency can be influential here.
At Guy Carpenter, we believe in a very structured approach to model validation. All model tests can essentially be categorized into three areas, sensitivity testing, loss validation and scientific appraisal. Sensitivity testing measures how the model’s output changes with changes to the data inputs. This can be illuminating in terms of understanding how the model works, identifying any anomalies and also in prioritizing any data improvement exercises. Loss validation tests how well the model reproduces historic losses where data is available. Scientific appraisal considers the underlying scientific assumptions of the model and where possible compares with independent data or academic studies.
The test results can be used for many things. You can justify the choice of one model over another based on performance against the tests. Deficiencies in a model can be highlighted and adjustments made to compensate for them. Where multiple models are available, the test results can suggest suitable blending weights.
This testing process works when there are models actually available to test, but what about territories and perils where there is no available model? How do we quantify a non modeled peril? The Association of British Insurers in the United Kingdom recently published a very useful report on this topic outlining many helpful techniques.
One useful technique can involve what we call “model transportation,” using a well-accepted model from one territory and peril as a proxy for modeling the same peril in another part of the world. Taking earthquake as an example, this can involve using the available seismic data for the region in question and finding similar events in the stochastic catalogues of the vendor models and applying these, with some necessary modifications, to the new territory.
Improvements Likely as Industry Matures
As we have seen, there are ways to improve risk quantification and reduce uncertainty in catastrophe modeling for the MENA region, and further advancements are likely as the insurance industry grows. The recent developments in catastrophe modeling platforms will lead to better understanding of the risks, allowing better transparency and more control over the model levers. The existence of open source modeling platforms such as OASIS could stimulate a more comprehensive and global perspective on model building by attracting more local and specialist academic input into hazard quantification, not just for insurance purposes, but also for disaster mitigation reasons.
The existing circumstances present a healthy challenge to accurately quantify catastrophe risk in the MENA region but I think the future is bright and improvements will be forthcoming as industry techniques mature.
This article first appeared in the June 2014 edition of Middle East Insurance Review.