Flawed reliance on VAR a systemic risk for insurers

Solvency II is a great achievement but it is not without its weaknesses, says risk management writer and consultant René Doff. Chief among them is the overreliance on a single approach to measuring risk

Now that most of Western Europe is preparing for the summer holidays, it is a good moment to reflect on what we have achieved in the insurance industry.

First, let's be honest. We are rightfully grateful to move on from the Solvency I framework, so Solvency II is a big step in the right direction. Also, there are only a few doubts about the objectives of Solvency II. As such it is good that the 2016 deadline is approaching and the industry is emerging from the stalemate in the decision-making process.

To quote Mark Twain: "Continuous improvement is better than delayed perfection."

However – as always – there are some who disagree, and some who complain. Here I want to make some observations about the achievements of Solvency II so as to be ready for Solvency III, which is undoubtedly awaiting us somewhere in the future.

Rather than picking fault with the calibrations of the sub-modules of the solvency capital requirement (SCR), the ultimate forward rate or the matching adjustment, I want to comment on the central risk measure in Pillar 1: the value-at-risk.

Continuous improvement is better than delayed perfection

VAR has enabled the risk management profession to make great progress, giving us a means to express different risks in one central risk measure. This made different risks comparable, which in turn facilitated strategic decision-making. VAR also allowed consistency in modelling efforts, so Solvency II is now able to allow internal modelling based on a company's own data – be it for market risk or underwriting risks.

However, VAR also brought a serious systemic risk into the framework. All companies use VAR, and base their VAR calculations on similar data. This means all companies will make similar decisions, because their VAR models direct them in the same way. Secondly, the use of VAR encourages people to think in the same paradigm, and include the same risk types in their models. This means that all companies observe the same risks, and also ignore the same risks. There is too little variety in perspectives.

The use of VAR makes three main assumptions, but these assumptions are seldom discussed. Although many readers will recognise the list below, I doubt any of these assumptions will find their place in the documentation of risk models for Solvency II, or have been discussed in the educational sessions with boards in the preparatory phase for the directive. The assumptions are:

1. The law of large numbers: we need sufficient data before we can estimate a VAR model with sufficient statistical reliability.

2. Ceteris paribus: the environment stays the same, so that events that have occurred in the past are a good predictor for the future.

3. Mutually independent events: if events influence each other, or even trigger each other, then statistical models will underestimate the risks involved.

The question is whether all these three assumptions always hold at the same time. For instance, the financial crisis has shown that stock prices and interest rates can influence each other. This means independent stock price modelling and independent interest rate modelling and then aggregating VAR outcomes (using correlations) is wrong. There was (and still is) a causal connection between interest rates and equity prices due to monetary policy by central banks.

The assumption of ceteris paribus also does not hold. Markets have changed so that old data is not predictive for the future anymore. Stock markets have become much more internationally connected than, say, 20 years ago. This makes a 20-year loss data set no longer relevant. And, theoretically, we would need at least a 200-year history to make a reliable VAR 99.5% calculation. Likewise, climate change has influenced the occurrence and the severity of extreme weather events. This makes claim triangles outdated, and catastrophe models even more difficult to make.

The relevance of these observations is obvious. While housing prices in the US were already going down in 2007, risk models showed that mortgage-backed-securities (MBS) prices would still go up – and indeed they did for a while. At that time, VAR was extremely low, because the risk was considered low. However, fundamental analysis would have shown that risks were actually very high. A fundamental analysis would have shown the relationship between interest rates, mortgage rates, housing prices and MBS. This causal connection went unnoticed by many risk managers, even despite articles in the mainstream media at the time.

Adopted flaws

Solvency II has adopted this flaw by introducing VAR as a main risk measure. Although stress testing in Pillar 2 under the own risk and solvency assessment (Orsa) might seem to decrease the reliance on VAR alone, that is not really so. Stress testing is based on the same models as the SCR. It provides a sensitivity analysis on the VAR models, rather than a fundamentally new insight.

How serious are the consequences of this flaw? In practice, the SCR is not seen as the real calibration of capital, because most supervisors expect insurers to operate well above the SCR level. This means more capital is kept as a buffer. However, the systemic risk still exists because all insurers are biased towards making the same kind of strategic decisions.

The Orsa is the vehicle for additional risk analysis on top of the bare minimum SCR calculations. The question is whether fundamental risk analysis (as we have called it above) will find a place in firms' Orsa reports. Despite the efforts of supervisors, Orsa scenarios have been relatively modest in showing the real (unknown) risks. Partially this is because insurers are understandably reluctant to share information with supervisors who might then apply capital add-ons to the SCR.

This position opens the case for a Solvency III. A next step in the development of risk management would be for risk professionals to develop fundamental risk-analysis skills. This means looking beyond the data from risk models into the connections of real life trends in markets and economies. This implies that risk managers have time to reflect in addition to their tight deadlines for model approvals and risk reports.

When I look at stock-picking advice, I always see two components: the technical analysis (that is, the statistics) and the fundamental analysis (the outlook for the particular company in its market). It is the latter that is missing in current risk analysis. Some risk analysis has more meat on its bones, but real fundamental analysis (root-cause analysis), based on economic trends, I have not seen applied widely.

Risk managers as well as supervisors would do well to review the strategic scenario-planning practice of Royal Dutch Shell, which predicted the oil crisis in the late 1960s – years in advance of it actually happening.

This technique relies on trend analysis and logical reasoning, not only of economic variables but also of political and demographic trends. Based on these trends, the scenario-makers derive three to five multi-year scenarios and, for each scenario, sketch out the inherent risks and opportunities for the company.

The important step is that for each scenario an action plan should be developed. The trick is not to choose between potential futures but to prepare for each of them at the same time. Even though the emphasis is on risk mitigation rather than risk measurement, there is a clear role for such risk analysis. I would encourage curious risk managers and supervisors to study and master this skill in order for the entire industry to become more stable and resilient.

René Doff's Risk Management for Insurers is published on June 29 by Risk Books. To find out more, visit the Risk Books website.