16 minute read

Changed in a heartbeat

Next Article
Net gains

Net gains

William Russell has played a crucial role in helping expats whose lives have been upended by politics and a pandemic. Inez Cooper, the specialist insurer’s MD and Co-founder, and Avin Talabani, Transformation Manager, explain their ongoing response

Life as an expat has never been so uncertain – or, potentially, so expensive. It can be difficult enough, in normal circumstances, being separated from your homeland by family arrangements, study, work or happenstance. But, as we know, the past 18 months have been anything but normal.

A global pandemic and Brexit have left tens of millions of people unsure of their status, and feeling confused and vulnerable, especially when it comes to accessing medical support.

In May, one of the biggest players in the international private medical insurance market (IPMI), UK-based William Russell, commissioned a survey among nearly 1,200 randomly-selected expats in five countries, to find out how they were holding up under the strain: the results were stark. Thirty-eight per cent said their mental health had declined during the COVID-19 crisis (12 per cent seriously), with 44 per cent regretting that they weren’t in their home country.

Commenting on the response at the time, Inez Cooper, MD and co-founder of William Russell, said expats’ experience of the pandemic had been made more acute due to isolation from family and friends, and language and culture barriers, while a lack of local knowledge may have prevented many of them from seeking the mental health support they needed.

For those whose lives straddle the UK and the EU, the situation is doubly troubling. Many have found it difficult to register their residential status in their host country, post-Brexit, which can mean they’re denied state healthcare, including COVID-19 vaccines. But neither are they able to return home to protect themselves because of travel restrictions.

It’s at times like these that an insurer that can not only deliver services fast and simply through multiple digital channels, but also offer responsive human contact, comes into its own.

“We’ve been doing phenomenally well, if you look at our Feefo score. We’re rated 4.5 out of 5, which, in the insurance space, is unique and really differentiates us,” says Avin Talabani, transformation manager for William Russell, who has been in post since the pandemic began. “We definitely want to maintain that human service, but we are also ramping up the ability for customers to interact with their policies digitally, both in terms of managing them and submitting a claim, as well as managing their healthcare more holistically.”

The insurer’s recent experience of how clients wished to access its services has mirrored that of others in financial services – online, and, increasingly, via mobile.

“That people would like a hyper-personalised service, and more self-service, has been an ongoing trend. But the pandemic effectively catapulted us three, four years into the future,” says Talabani. “Due to things like test and trace in different countries, and tracking the results of tests and submitting them online, people have got a lot closer to their mobile phones… it’s almost become an expectation that they want to be able to interact with us in that way, too – in their own space and in their own time. But then, if they do need to speak to us, we absolutely need to have somebody there who can help them resolve their issue, first time around.”

Cooper adds: “We’re real people, and we have a real responsibility to treat others as we would want to be treated ourselves. But, on the other hand, we’re a boutique provider, we’re competing with the insurance giants;

our advantage is our ability to accelerate the digital transformation programme.”

The three-year transformation roadmap the company had in place at the beginning of 2020 for delivering its health, life and income protection business, has been condensed into an 18-month delivery timeline. And one of the tools it’s using to achieve that rate of change is Microsoft Power Apps Portals.

We’re real people, and we have a real responsibility to treat others as we would want to be treated ourselves

Inez Cooper

HUMAN-ENHANCING TECH

Portals has allowed William Russell to provide customers with an interface that will give them access to, and the ability to interact with, various aspects of their quote and policy that are contained in the Microsoft Dynamics data lying beneath.

“It’s been really successful for us, in terms of our online quotation capabilities. Since we’ve released it, our conversion rate has increased by 50 per cent,” says Talabani. “And that’s just on the quotations side. Our intent is to start extending that from just a quote, to quote, apply and buy, and then the onwards management of policies.”

It’s more than a revenue-generating model, though. Allowing a digital exchange of information between insurer and the insured moves the customer relationship on in a dynamic way.

If the pandemic has taught us anything, it’s that looking after our health is a priority. And, if we’re far from home, the penalty for not doing so is often financially severe.

William Russell releases annual data on claims made under international private medical insurance. In the year 2019/2020, one claim amounted to US$390,000, while the most expensive medical evacuation was $31,125. Cancer treatments were the most costly, while the biggest bills overall were incurred by expats living in Asia. Even routine healthcare costs abroad can be crippling. In Hong Kong and the UK, the price of getting a child vaccinated is upwards of £800, and don’t even think about starting a family in Hong Kong, where maternity care for expats runs into tens of thousands of dollars.

Worldwide, treatment costs have been increasing, year on year: poor lifestyles have prompted a rise in chronic conditions, and there are more expensive drugs and high-cost treatments available. Which is why investing in technology that helps both insurers and customers work in partnership to keep people out of the healthcare system, is seen as a priority from both an economic and personal perspective.

“We want to start including services that look at the customer’s health more holistically, using self-service capabilities,” says Talabani. “The data that flows from that allows us to assess whether somebody is heading towards a medical condition that can be avoided through lifestyle changes, so that they can avoid that diagnosis further down the line.

“People are themselves starting to look at health insurance and ask ‘how can this insurer really help me manage my health?’, not just ‘how easy is it to make a claim?’. And there are additional partnerships, like telemedicine, virtual health and mental wellbeing, that we’ll be able to extend to customers, which they are now starting to expect more of as a result of the pandemic.”

There are also economic and regulatory advantages of moving towards a more data-driven model with AI: the opportunity to contain expenses for both company and client, as well as better address fraud.

Unfortunately, says Talabani, there are providers who will try to play the system.

“They’ll include lab tests and an x-ray as additional components of a claim, where they’re not medically necessary,” she says. “We’ve got the opportunity now to partner with third parties that can use the medical data that’s coming from that provider to assess whether a treatment is medically necessary or not, and we can then interject and advise if a person doesn’t need it.”

She believes the super-fast timeframe the firm has set itself for achieving these goals is doable, partly because it’s not operating on an ancient legacy system.

“Being a smaller company also makes us more agile,” she adds. “Quite often, we’ll have an idea and get the decision-makers in the same room, on the same day, to say ‘let’s go for it!’.”

It’s also a workforce that’s balanced in both gender and ethnicity.

“The different backgrounds of our people, and their varied experiences, help us understand whether an application is going to be a one-size-fits-all, or needs customising to be personal to the different geographies and cultures our expats are based in, whether that’s using different languages or different conversational tone, all while being respectful of those local

cultures; even how we can be more mindful of people who may have mental health issues, or have suffered a particular trauma,” says Talabani.

The reality is that, post-pandemic, people will likely be a lot more circumspect about basing themselves abroad – and, certainly, they’ll be looking for protection they can depend upon if they do.

“We are here to help people when they’re going through hard times, and make sure they feel supported, financially and emotionally, when they contact us; to feel that they can rely on us,” says Cooper.

Wherever you are in the world, that’s a comforting thought.

Since we’ve released the customer interface using Microsoft Power Apps Portals, our conversion rate has increased by 50 per cent

Avin Talabani

Graphical Representation and Regression Formulation of Link Ratios

Thomas Mack identified the stochastic regression model that underlies volume weighted average link ratios. Other authors, including Murphy and Venter, have developed these ideas further. A graphical representation and regression formulation of link ratios makes it clear what assumptions underpin the methods and extensions thereof.

"There is pleasure in recognizing old things from a new viewpoint."

Richard Feynman Consider the (diagonally opposite) Incurred Loss triangular data from the American Reinsurance Association.

In general, each link ratio (y/x) is the slope of the line from the number pair (x,y) to the origin.

The graph below plots the cumulatives in development year one versus the cumulatives in development year zero for accident years 1981 to 1989. The caption on the right is for the point (5,655, 11,555) corresponding to accident year 1984. The caption on the left is for the point (1,092, 9,565) corresponding to accident year 1985. The slope of the blue lines represent the corresponding link ratios – which is 2.043 for 1984 and 8.759 for 1985.

Accordingly, an average link ratio, equivalently average trend, is an average slope through the origin.

This means that the method can be formulated as a regression (Mack (1993)).

Let y(w) denote the cumulative in development period j for accident year w and x(w) the cumulative in the previous development period, j-1.

We can write,

y(w) = b * x(w) + e(w),… (1)

where b is the slope of the line (equivalently, the average link ratio), and e(w) is the difference between the actual value y(w) and the corresponding point on the average link ratio line (b * x(w)).

Acc: 1985 Value Dev 0: 1,092 Value Dev 1: 9,565 Ratio: 8.759 Acc: 1984 Value Dev 0: 5,655 Value Dev 1: 11,555

Cum.(1) vs Cum.(0)

Ratio: 2.043

When actuaries use link ratios there are two critical assumptions:

• The expected value of the next cumulative is conditional on the previous cumulative multiplied by an unknown factor.

• The selected link ratio (factor) is optimal for prediction.

The optimum value of b is found by weighted least squares estimation according to the scale of the error terms e(w). Let the variance of e(w) = v * x(w)delta

For the following values of delta (0, 1, 2):

• 0, or constant variance, the weighted least squares estimated of b is the volume squared weighted average link ratio.

• 1, the weighted least squares estimate of b is the volume weighted average link ratio – sometimes called the chain ladder ratio.

• 2, the weighted least squares estimate of b is the arithmetic average link ratio. In the graph (previous page), the red line is the best least squares line through the origin and the green line is the best least squares line that includes an intercept. The latter appears to be a better model.

Murphy (1994) extended the regression formulation to include an intercept term.

y(w) = a + b * x(w) + e(w), … (2)

where a is the intercept term, but b is no longer the average link ratio.

Given that the intercept is positive in the previous graph, the slope of the line with an intercept term is less than any average link ratio (through the origin).

We can obtain visual indications of whether a line with an intercept (Murphy (1994) method) or a line through the origin (Mack (1993) method) is better.

Most importantly, the focus should be on the incremental model, Venter(1998), even if a = 0:

y(w) – x(w) = a + (b-1)*x(w) + e(w), … (3)

where y(w) – x(w) is the incremental data point.

When you use a link ratio to project the cumulative in the next period in essence you are only projecting the next incremental as you know the current cumulative. This is the reason all the focus should be on equation (3) not (2).

But what if b in equation (3) is statistically equal to 1, (Venter(1998))?

Then the incrementals in development periods (j) are not correlated to the cumulatives in the previous development period (j-1). That is, any ratio applied to the cumulatives does not predict the incrementals!

Here is a graph (right) of the incrementals in development year 1 versus the cumulatives in development year 0.

Incr.(1) vs Cum.(0)

Corr. = -0.117, P-value = 0.764

Note that the correlation is zero (slope not statistically significant). Equivalently b – 1 = 0.

In this case, the reduced model only contains an intercept term.

y(w) – x(w) = a + e(w) … (4)

In this model, the incrementals across the accident years are random numbers from a distribution with mean a, and variance, Var(e(w)). If e(w) has a constant variance, then the ordinary least squares estimate of a is the arithmetic average of the incrementals y(w) – x(w).

It turns out, if you graph the incrementals in any development period against the cumulatives in the previous period, you will note that there are no statistically significant correlations. All the b-1 parameters are statistically zero.

The assumption that the incrementals are random, might not be true. A case in point, is development period two. This suggests that we need to include an accident year trend parameter in model (3).

The equation that includes the intercept, accident year trend and slope can be written:

y(w) – x(w) = a0 + a1 * w + (b-1)*x(w) + e(w), … (5)

where a0 is the intercept, a1 is the accident year trend parameter and b-1 is the incremental coefficient.

The family of models included in the Extended Link Ratio Family (ELRF) are represented by equation (5) between each two consecutive development years. The significance of the parameters is determined by the data.

Link ratios have no predictive power for this incurred loss development array. The optimal combination of parameters uses simply an intercept term with the exception of the regression equation between development periods 1 and 2 where an accident year trend is also statistically significant.

Mack, T. (1993). Distribution-free calculation of the standard error of chain ladder reserve estimates. ASTIN Bulletin: The Journal of the IAA, 23(2), 213-225. Murphy, D. M. (1994, March). Unbiased loss development factors. In CAS Forum (Vol. 1, p. 183). Venter, G. G. (1998). Testing the assumptions of age-to-age factors. In Proceedings of the Casualty Actuarial Society (Vol. 85, pp. 807-847).

Incr. (2) vs Year

Corr. = -0.841, P-value = 0.009

Volume weighted average link ratios do not distinguish between accident years and development years

Accidental Years Developmental Years Incremental Data Set Consider any triangle with incremental values where:

• alpha denotes the sum of the values in the red rectangle,

• beta denotes the sum of the values in the green rectangle (one development year), and

• gamma is the sum of the values in the orange rectangle (one accident year). Let p denote the incremental value projected for the accident year represented by the gamma values for the next development year.

The value alpha represents both the aggregate of the row sums in the red rectangle and the aggregate of the column sums.

The volume weighted average when you cumulate the triangle in the traditional way is (alpha + beta) / alpha. If you cumulate the triangle for each development year down the accident years, then the volume weighted average is (alpha + gamma) / alpha. Accordingly:

If you cumulate along the development years, and

If you cumulate along the accident years. QED.

We know that development years are not like accident years.

CONCLUSION: Link ratios have got nothing to do with the structure of the data.

For the incurred array we plot the incremental values versus development year. We also plot the values versus accident year. Note the different structure.

Clearly, we expect any incremental loss development array to decay to zero, but you would not expect the same pattern down the accident years.

ELRF™ 2020 ELRF™ 2020

ELRF™ 2020 is for P&C actuaries who want to take advantage of the graphical representation and regression formulation of link ratios, and extensions thereof.

All this, coupled with the power of a relational database are included in ELRF™ 2020. All the information in the database including data, models, and results, are a mouse click away. Accessing data and information through the ELRF™ 2020 application is a pleasure.

The Extended Link Ratio Family (ELRF) modeling framework provides diagnostics for testing assumptions.

Residual plots versus development period, accident period and calendar period are also used to assess model specification error. Any patterns in the residual plots show features of the data that the method is not describing.

The Y versus X and Y - X versus X plots (left) provide diagnostic testing of the intercept and ratio minus one. Formal tests are provided in the regression tables.

Here there is no relationship between the incremental Incurred in development period 3 with the cumulative Incurred in development period 2. Link ratios do not have predictive power.

ELRF™ 2020 Standard:

• Over 144 link ratio methods including Bornhuetter-Ferguson and Expected Loss Ratio Methods • Link ratio methods formulated as regression estimators • Extensions including intercept (Murphy) and constant accident year trends for each development year • Diagnostic tools • Bootstrap distributions by accident year, calendar year and total

ELRF™ 2020 Professional:

• COM API • Extended report templates • Server database (Oracle & SQL Server)

ELRF™ 2020 affords benefits at warp speed unlike any other reserving product.

ELRF™ Best’s Schedule P:

Included with a Best’s Financial Suite - P/C, US subscription! • Offline access to Schedule P data from AM Best and derived financial metrics; • All analytical tools included in ELRF™ 2020! Contact AM Best at sales@ambest.com to learn more.

This article is from: