Can pension funds rely on value at risk models given unprecedented events like Brexit or climate change – and should they continue using them? 

 

Have models just been debunked after the UK’s initial pandemic response relied on them heavily yet was proven wrong, or are they, on the contrary, becoming ever more relied upon? Can a model really offer any prediction in the absence of permanent fixtures like the laws of physics, or are we simply ‘fitting the data’ to the model? 

 

 

Former Bank of England governor Lord Mervyn King and economist John Kay have criticised the way risk is viewed in modern finance, pointing out that the probability of a risk occurring is not distributed equally and that models are poor at capturing extreme events, and unable to capture events that are not represented in the past data they are based on. 

 

Value at risk is used to assess the investment strategies of pension funds up and down the country, to see how big their potential losses could be, whether the sponsor could afford to cover them or if the investment strategy needs adjusting. 

 

But there are some well-known issues with VaR. For example, there is no protocol about which data to use, so there is a potential for picking data from a period that gives a more benign projection. The other issue is that high impact, low probability risks – so-called black swans – might not be captured well with normal distribution probabilities. 

 

 

Independent actuary Jeff Brown also thinks the main problem in quantifying risk is the non-stationarity of the underlying probability distributions and that therefore “VaR is not fit for purpose”. 

 

The technical weaknesses have been known for several years, he says, and wonders why the actuarial profession continues to use VaR: “The $64,000 question is, why are actuaries using these models?” 

 

There could be some actuaries “who may not understand the technicalities”, he suggests, or otherwise “they do understand the technical issues but feel that they must offer a VaR tool for commercial reasons”. 

 

Another possible explanation for the continued use of VaR could be that actuaries are instructed to use a black box model “and accept without question anything that is produced”, he speculates. 

 

Historic data does not exist to model current conditions, so the underlying probability density distribution is not known, he says: “Given today's big issues – Covid, climate change, Brexit and record levels of global debt – I do not know how anyone can put their hand on their heart and say they can give a precise number to quantify investment risk.” 

 

 

Despite these issues, VaR has become a relatively standard measure of risk for UK pension funds, often applied via a simplified algorithm instead of a full stochastic projection of assets and liabilities. David Fogarty, a professional trustee at Dalriada Trustees, says that VaR is “definitely” useful. 

 

“Understanding the potential for funding level volatility is a challenge for all trustees, and asset liability modelling is complex and expensive, so if the trustee can access a single number, even if the calculation of the number itself is simplified, that is undoubtedly helpful,” he argues. 

 

But like all models, VaR must be taken with a pinch of salt, it seems. Fogarty admits that the assessment is heavily dependent on the assumptions used and therefore trustees should look more closely at the investment risk level on a periodic basis, for example during their actuarial valuation. 

 

This could help them to have “a good understanding of the full distribution of possible outcomes as well as the nuances of the VaR calculation and how it might vary with different assumptions, as well as with changes to the investment strategy”, he says. 

 

Fogarty describes VaR as just one of many characters in a story, saying that it is not one character but “the relationship and interaction between all the characters in our story that determines whether we end up with a happy ending”. 

 

Marian Elliott, pensions board and council member at the Institute and Faculty of Actuaries, said VaR is a useful metric but admitted that “there is a danger that stakeholders place too much reliance on this figure and use it for purposes for which it isn’t intended”. 

 

As a short-term indicator of volatility, it should be one in a suite of tools used to understand risk but “should not be used to consider long-term risk or, on its own, to decide between different potential courses of action”, she says. “If used on its own, it can be dangerous.” 

  

Minimising short term VaR, for example, could reduce the probability of success over the long term if a particular return is needed to achieve a funding goal within a given timeframe, she says. “It is also susceptible to model risk – the assumptions made about correlations and diversification benefit can have an impact on the VaR figure, so it is important for stakeholders to understand these assumptions before placing undue reliance on this number. As with all model outputs, VaR gives a useful indicator and helps build up a picture of the risk exposure but it is not a ‘correct’ figure and nor does it give the whole picture.” 

 

 

The difficulty with a model like VaR, it seems, is ensuring that it is understood for what it is – a mere indication rather than an absolute assurance, rather like a weather forecast. One would not rely on the forecast alone when leaving the house but check actual conditions by looking out of the window. 

 

Lord King, in a lecture given 10 years ago, noted that “much of the value of forecasts is in their being understood – in all their subtleties – by the general public”. A forecast cannot be 100% accurate, yet “to publish forecasts only when the outcome is virtually certain would be an admission of at least partial defeat” – suggesting that the trick lies in conveying the limits of a model. 

 

Marian Elliott

Jeff Brown

Hugh Nolan

Charles Cowling