You are here

Uncertainty and Coronavirus

Andrew Smith is an assistant Professor at University College Dublin, who specialises in the application of advanced mathematical and statistical methods to solve problems in the financial services industry. In his blog, Andrew discusses principles that can be used to manage uncertainty in the context of the COVID-19 crisis.

Andrew SmithAt the beginning of March, the Society of Actuaries in Ireland hosted a well-attended event considering current topics across areas of actuarial endeavour. We considered the burgeoning aviation finance market in Dublin, as well as the need for actuarial involvement in understanding the climate crisis, noting the tension between the two.

Only two months later, Coronavirus has transformed our world. The pandemic has thrown into sharp focus the uncertainty associated with the use of models. The UK has taken a comparatively relaxed approach compared to Ireland and other European neighbours, while on the other hand the UK lockdown appears severe relative to US and Brazil. All governments profess to heed scientific advice, but with future outcomes uncertain, so the advice provided is varied and sometimes conflicting.

Few actuaries are directly involved in planning hospital capacity or prescribing social distancing, yet the coronavirus affects nearly everything actuaries try to predict, including:

  • Mortality, from COVID-19 itself but also secondary impacts such as competing heath service demands and the beneficial impact of improved hygiene on the spread of other diseases.
     
  • General Insurance claims are likely to be lower for motor business, because of fewer vehicles on the road, while burglars have fewer empty houses to raid. On the other hand, many businesses are seeking to make interruption claims.
     
  • A lack of economic output and increase in indebtedness is likely to generate a recession, with many firms losing money and some failing. To what extent do stock market falls already reflect this?
     

In my particular academic role, I am trying to understand the operational impact of future lockdown rules on delivery of teaching and student assessment over the next year.

How then can we respond to this proliferation of risk and uncertainty?

Last year, an IFoA working party reported on Managing Uncertainty – Principles for Improved Decision Making. View the paper

The working party paper sets out some key principles we can apply to modelling the impacts of epidemics. Five of those are particularly relevant as follows below.

1. Deconstruct the problem

Pandemic modelling shares an important property with financial models – the model can feed back into behaviours which change what we are trying to model. Sometimes that feedback happens in a good way – a model scenario shows a lack protective equipment, so the government builds up a stockpile in preparation for a pandemic. Or a model shows dire consequences of close personal contact, which encourages people to practice social distancing. As in the story of Jonah, a modeller can be discouraged when their dire forecasts fail to materialise. That should be recognised as a good outcome, however. There is sufficient protective equipment for everybody. Disease transmission slows. The people of Nineveh are saved. And so on.

So we need to be clear not just about the mechanics of a model, but also what it is trying to achieve, what the real underlying questions are, and what success looks like.

2. Don’t be fooled

The current pandemic has more than its fair share of traps for the unwary. As in other areas of modelling, there are many ways to misinterpret data, ranging from failure to ask the right questions, through self-censorship to deliberate attempts to mislead.

The UK Government has now changed the basis on which COVID-19 deaths are published. Daily figures now include estimates of deaths at home or in care homes, as well as those in hospitals, bringing the UK into line with some neighbours. It would have been easy previously for a modeller to misunderstand the figures which excluded significant deaths in care homes.

Self-censorship and selection biases are harder to detect. When does a client select an advisor who will tell them what they want to hear? Can we really make a clear distinction between objective, science-based facts and political decisions, or is political interference in science inevitable? What incentives face those on whose data or analysis we rely?

When the dust settles, many modellers are likely to discover that the data on which they based their forecasts does not represent what they thought or intended. Let us take steps to avoid this.

3. Models can be helpful and dangerous

In addition to best estimate forecasts, actuaries are updating their stress test assumptions in the wake of the pandemic.

How should firms re-calibrate 1-in-200 stress test? Do we argue that falls to date have eaten into the stress test so further falls will be modest? Or should we strengthen stress tests given the uncertain outlook?

Firms may seek forbearance, while acknowledging the risks are more severe than we thought last month. The argument will be that now is not the time to fortify capital requirements (it never is, of course). We need to guard against tweaks to models being used to disguise changes to a firm or a pension scheme’s risk appetite.

4. Stability and resilience

In our enthusiasm to put numbers and likelihoods around outcomes, we should also be asking ourselves: what happens if the model is wrong? What control mechanisms are in place to cope with the outcomes our model says are unlikely? If we are wrong, at what point do we admit it?

At the time of writing, the UK has embarked on a number of processes whose exit strategies are unclear, to say the least.

More broadly, will we ever unwind quantitative easing, and if so, what happens? How will the government repay its huge COVID-19-related borrowing requirements? Behind COVID-19, Brexit looms, with conflicting views on whether deadlines will again be extended.

Most of us have little control over these external factors. Yet, we have to plan for multiple outcomes. Our plans need to be resilient enough to cope if the external environment does not perform as we hoped.

5. Bring people with you

The final aspect of managing uncertainty is bringing people along with you on the modelling journey. That means explaining the science, to the extent this is understood, but also being clear about trade-offs being made and the basis of decisions taken.

I am in the fortunate position of being able to do much of my work from home. I have a roof over my head and enough food. I do not have to compromise my personal safety to earn an income. However, many others are less fortunate and COVID-19 risks aggravating society’s existing inequalities. People with low-paid, insecure work, living in crowded accommodation now face losing what little they have.

For these reasons, we need to be especially aware of the impact of our models and recommendations on the most vulnerable in society. COVID-19 has challenged fundamental assumptions about the role of governments and markets, as well as the obligations that companies have towards their employees and society. I hope that we will look back on 2020 as a year when actuaries did the right thing for all in our communities.