81 Chancery Lane
0207 112 8493
View map


1 Lochrin Square
92 Fountainbridge
0207 112 8493
View map


24A Baggot Street Upper
D04 V970
0207 112 8493
View map

Registered Office

7 Bell Yard,
0207 112 8493
View map

Send us a message

Contact Us

A Playful Introduction to Lloyd’s of London

Note that the original Excel workbook was ten times slower than the optimised version and it had to be removed from the chart in order to make the other bars visible.


Working with clients on a consultancy basis involves regular communication, to ensure that client needs are being met in the most efficient way. Having established that dialogue with the client, it was relatively easy to agree to move from Excel to the R (with C++ goal-seek) solution, noting that although this was not the fastest solution, the client was more familiar with R than some of the alternatives, having already implemented some R tools within the actuarial department.

The project was not always smooth: testing uncovered many aspects that required further investigation and moving the cashflow model from Excel to R was not in the original plan. However, all project requirements were met, and the client was grateful that we had solved a thorny problem.

Phil Creswell

March 2021

Shining the Spotlight on Expert Judgement

Figure 2: Insured catastrophe losses by catastrophe type, 1970-2019, in USD billion at 2019 prices. Source: adapted from Swiss Re Institute Sigma Paper No 2/2020[5].

1: 2001 – World Trade Centre. 2: 2004 – Hurricanes Ivan, Charley, Frances. 3: 2005 – Hurricanes Katrina, Rita, Wilma. 4: 2008 – Hurricanes Ike, Gustav. 5: 2011 – Thailand flood. 6: 2011 – Japan and New Zealand earthquakes. 7: 2012 – Hurricane Sandy. 8: 2017 – Hurricanes Harvey, Irma, Maria. 9: 2018 – Camp Fire, Typhoon Jebi. 10: 2019 – Typhoons Hagibis, Faxai.

The chart in Figure 2 is inflation-adjusted, however it does not attempt to adjust for population growth, economic growth, urbanisation or insurance penetration – all of which have increased over the period (e.g. global population has more than doubled since 1970). Its temporal trends should therefore be interpreted with great caution – a point that is readily acknowledged by the authors – but it is nonetheless useful in showing the relative importance of natural versus man-made Cats.

Annual man-made Cat insured losses have typically fluctuated between $5-10bn (in 2019 dollar terms) since the 1980s, with the two notable exceptions to this being the World Trade Centre (‘WTC’) terrorist attacks in 2001 (Figure 2) and the Covid-19 pandemic (not shown in Figure 2). Covid-19 insured losses are clearly still developing, but in November 2020 Lloyd’s of London indicated it expected the pandemic would cost the global insurance industry more than $107 billion and would likely be similar to the record 2017 losses of around $150 billion[6].

In contrast, annual natural Cat insured losses have grown significantly since the 1980s, driven primarily by losses from hurricanes/typhoons[7]. As a result, the proportion of total insured Cat losses accounted for by man-made disasters has fallen from around 40%-50% in the 1970s and 1980s to roughly 10%-20% this century, 2001 excluded. Note that Figure 2 is only showing insured losses rather than total economic losses[8]. Various significant events, such as the Chernobyl nuclear disaster in 1986, are therefore not captured in the dataset as no insurance was in place.

The challenges of man-made Cats

For (re)insurance companies the key overriding question is how man-made Cats can be modelled, priced, reserved for and managed. GI companies face numerous challenges when it comes to man-made Cats, some of which are summarised below:


Unlike with natural Cats, the modelling of man-made Cats is still in its infancy. Third-party vendor products provided by companies such as Risk Management Solutions (‘RMS’) and Verisk Analytics’ AIR Worldwide are well established in the field of natural Cat modelling, but equivalent modelling packages either do not exist, or are much less established, for man-made Cats[9]. There is so much uncertainty in the model parameters that it is hard for insurance companies to justify adopting third-party models, let alone attempt to validate them.


There is huge variability in the types of risks and events that fall within the umbrella of man-made Cats. The wind/flood/fire risks and dangers of natural Cats are relatively narrow in scope and complexity compared to the variation in the characteristics of cyber/terrorism/pollution events.


Whilst natural disasters tend to be much larger in scale than man-made Cats, almost all natural risks are geography-specific. Their locations are much more predictable and reasonably well-understood – earthquakes almost always occur along tectonic boundaries and floods almost always occur within floodplains. By contrast a terrorist attack is hyper-local, but can take place almost anywhere, whilst a cyber incident could be anything from a targeted attack on an individual institution to a potentially global-scale event if a particular technology is compromised.


Hurricanes and earthquakes occur every year. These events can be easily (from the 1960s onwards) monitored, studied and recorded and there is therefore an abundance of relevant historic natural Cat data. This is not the case for man-made disasters. Man-made Cats can be hard to monitor and/or quantify, the frequency of very large man-made disasters is much lower than for natural Cats and furthermore the data that does exist for them may only be of limited relevance in today’s world. Man-made Cats such as Cyber have only existed for the last couple of decades, whilst others such as the WTC are arguably one-off events that may not be repeated.

The human element

It is stating the obvious but man-made Cats are, by definition, driven by human behaviour. This means that some man-made events, such as terrorism, deliberately maximise the loss whereas a hurricane cannot choose the path of most destruction. This human element also means that there is constant evolution. New types of man-made Cats can emerge, often driven by technology, whereas the universe of natural Cat events essentially stays the same[10]. Cyber attackers will adapt to try to break through new security firewalls whereas floods do not increase in severity in response to new flood defences. This evolution further undermines the usefulness of historic data when evaluating future risks.


Both natural and man-made Cats are driven by complex phenomena. However, whilst the complexities of oceanic/atmospheric circulation and tectonic movements can be partly understood and modelled through the application of fundamental scientific principles, the underlying drivers of man-made Cats are much more fluid and varied. Relevant inputs into a hypothetical man-made Cat model might include factors such as:

Man-made Cats – key considerations

When viewed at a global level (Figure 2), insured annual losses from man-made Cats have been surprisingly regular (2001 and 2020 excepted). Zoom out far enough and it seems that the inherent complexities of man-made Cats retreat into background noise, leaving a quasi-random but nonetheless reasonably predictable stream of human-induced disasters.

Insurance companies do not have the luxury of operating at this global level. They operate on a policy/portfolio level and must therefore concern themselves with the specifics of geography, the type of risk they are providing insurance for and who they are insuring. In this regard, even natural catastrophes are extremely difficult to price and reserve for as there is a high degree of uncertainty over how frequently they will occur, where they will occur and how severe the losses they cause will be. These uncertainties are even greater for man-made Cats, which are less generic, have a higher geographic variability and – in the case of large man-made Cats such as the WTC or Covid-19 – are less frequent and (relatively) more severe.

A full evaluation of the implications of these man-made Cat challenges for GI companies – and, by extension, for the insurance industry as a whole – is beyond the scope of this article. Instead, below we highlight some of the key aspects that companies should consider when dealing with man-made Cats:

Portfolio analysis – correlation and aggregation

It is crucial that underwriters and actuaries have a good understanding of their existing portfolio of business, and of possible common vulnerabilities within the portfolio. New risks need to be considered in relation to existing portfolios before underwriting them. Reserving actuaries may need to reflect on potential correlations and challenge underwriters.

For example, are physical assets located close to each other and therefore susceptible to all be damaged by a single fire/explosion? Do Cyber policies provide cover for the same systems/technology, thereby making them potentially all vulnerable to a targeted attack? How diverse is the book of business for Product liability insurance – could a single legislative change, or the emergence of an adverse side effect, potentially trigger multiple claims? Covid-19 has resulted in lots of small losses (and some large ones), highlighting the importance of aggregation analysis across a portfolio.

Scenario analysis

Exploring a range of scenarios can identify events that may cause the greatest losses. Given the difficulty in explicitly modelling them, scenario analysis is the primary tool used by insurers to allow for man-made Cats. Insurers will reflect on what disaster scenarios might lead to insolvency, high-level assumptions can be made, and appropriate ranges can be considered. In the London market Lloyd’s maintains a set of mandatory realistic disaster scenarios (‘RDS’). This is a set of catastrophe event stress tests that are performed by individual syndicates and provide an aggregated view of risk in the market as a whole.

Counter-factual analysis

Counter-factual analysis is a variant of scenario analysis whereby thought experiments are carried out to consider how things could have worked out differently in past real events if certain conditions or parameters had been slightly different. Taking past events and modifying them is a useful tool for exploring feasible but previously unencountered scenarios.


ENIDs stands for ‘events not in data’. It is important to be aware that many possible events are not captured in historic data and that these ENIDs can be very severe – this is the motivation behind the Solvency II regulation that mandates an allowance for ENIDs in Technical Provisions. For example, prior to the WTC there was no comparable terrorist attack in the historic dataset and prior to Covid-19 a global pandemic had not happened for 100 years. On the other hand, once such rare events have occurred, companies should be mindful not to give them too much weight when assessing future risks.

Risk mitigation

Appropriate reinsurance protection is perhaps the primary form of risk mitigation against man-made Cats. Other options include writing less of a particular class of business to reduce correlation in the portfolio, changing the terms of the insurance provided or increasing the premium being charged for new policies.

Identification and monitoring of drivers

Carrying out the scenario analysis work will give the company a greater understanding of the key drivers for each of the risks, which will need to be monitored and updated. One approach to dealing with man-made Cats, given their constantly evolving nature, is to treat them as perpetual emerging risks[11]. This means that data and methodologies need to be continually updated as both risk drivers and actual losses change and develop. The large diversity of man-made Cats requires an equally diverse range of approaches if any meaningful insights are to be gleaned regarding the drivers of these events.

AI and big data

Given this need for constant monitoring, ‘big data’ and artificial intelligence (‘AI’) may form an increasingly important part of the insurer’s toolkit. These fields are already being used in some sectors to improve risk mitigation at source, thereby potentially reducing the demand for insurance. For example, in the oil and gas industry, machine learning algorithms are used to analyse streams of sensor data from drilling and production equipment to identify impending failures before they occur[12]. Insurance companies may need to follow suit with harnessing the power of AI and big data to more accurately monitor the varying risk levels of different types of man-made Cats so that they can be priced accordingly.


Catastrophes are unusually severe events that are commonly categorised as being either natural or man-made, though in reality man-made factors always influence the severity of insured losses. Man-made Cats are generally less material to the insurance industry than natural Cats, but their diversity and pace of evolution mean that they are even harder to model, price and reserve for. There is no ‘right answer’ for how to allow for man-made Cats but (re)insurers should be mindful of what steps they can take to gain a better understanding of the risks they are exposed to.

Rob Givens

January 2021

Notes / References:

[1] Examples of man-made Cats:
Cyber: WannaCry ransomware attack, 2017 (
Shipping: Beirut port explosion, 2020 (
Aircraft: Boeing 737 MAX groundings, 2019-2020 (
Pollution: Deepwater Horizon oil spill, 2010 (

[2] The presence of flood defences is, however, obviously no guarantee of safety, as witnessed in New Orleans during Hurricane Katrina in 2005 and at the Fukushima nuclear power plant in Japan in 2011.

[3] ‘Climate Change Affected Australia’s Wildfires, Scientist Confirm’, Henry Fountain, The New York Times 4/3/2020 (

[4] ‘This pandemic is an environmental issue’, Tony Juniper; Evening Standard 7/5/2020 (

[5] Swiss Re Institute Sigma Paper No 2/2020 ‘Natural catastrophes in times of economic accumulation and climate change’, April 2020 (

[6] ‘Lloyd’s of London sees global pandemic insurance losses above previous estimate’, Reuters 18/11/2020 (

[7] Note that hurricanes, typhoons and cyclones all refer to the same weather phenomenon. The different terms are simply used in different parts of the world – e.g. ‘hurricane’ is used in the USA but ‘typhoon’ is used in Japan.

[8] The Swiss Re Institute Sigma Paper does provide estimates of insured Cat losses versus uninsured Cat losses but these figures are not split between natural and man-made Cats. Total economic losses = insured losses + uninsured losses.

[9] AIR Worldwide does provide some catastrophe modelling for terrorism and cyber incidents.

[10] The equivalent of a ‘new’ natural catastrophe would be something like a large asteroid impact.

[11] ‘Man-made catastrophes: A perpetual emerging risk’, Richard Thornton, Global Reinsurance 18/8/2016 (

[12] ‘SparkCognition Adds Artificial Intelligence to Aker BP’s Operations’, SparkCognition Press Release 26/3/2019 (

September 2023 Exam Results

The September 2023 exam results have been published and APR student staff achieved an impressive 76% pass rate across all IFoA exams. In this article, Moses Vaughan highlights our success this sitting.

Bank of England System-wide Exploratory Scenario Stress Test

In this article Alan Reed reviews the recently specified 2023 Bank of England System-wide Exploratory Scenario (SWES) stress test exercise. When is it happening, whom is taking part, how does it compare to prior stress test exercises and what features of this new exercise stand out?

IFoA Data Science Certification – Course Review

In this article, John Nicholls demonstrates outstanding dedication to content production by enrolling in the IFoA Data Science certification and producing a review that covers course content, length, assignments, and delivery. Hopefully this article will help those of you on the fence to be gently encouraged to fall off it.

The Magic of Productivity

Are distractions derailing your productivity? Dean Woodward explores the magic of focus with tips and techniques to boost your efficiency.

Python in Excel – an Analyst’s Dream?

What’s the world’s most iconic duo? Batman and Robin? Fish and chips? Morecambe and Wise? In fact it’s none of these: join Jacob as he talks about using Python in Microsoft Excel.

Carbon Permits and Offsets

How can governments use the free market to take action on climate change? Cristian tells all about carbon permits and offsets.

Actuarial Mentoring Programme

This interview-style article provides an insightful overview of the Actuarial Mentoring Programme (AMP), its objectives, and the invaluable benefits it offers to both mentors and mentees. Their experiences shed light on the AMP’s effectiveness and highlights the opportunities it provides for professional growth and development.

April 2023 Exam Results

The April 2023 exam results have been published and APR student staff achieved an impressive 83% pass rate across all IFoA exams and 100% in all CAA exams taken. In this article, Moses Vaughan highlights our success this sitting.

Office vs home – what is the ideal working environment? 

What is your ideal working environment? Cristian and Hannah discuss how and why their ways of working benefits them the most.

The Last Remaining Defined Benefit Schemes: A Deeper Look

In this article Asees explores the rise and decline of DB Pension Schemes, and considers the impact of this on secure retirement for future generations.