Barrie & Hibbert Blog

Too dominant in the ESG space?

2009-08-26T09:11:34Z

We are sometimes asked to comment on the claim that we are just "too dominant in the ESG space". Notably, European insurance regulators raise the issue deep in the 'white text' of their July 2009 Consultation Paper #56 on internal model approval in paragraph 10.29

Whilst we could choose to interpret these questions as backhanded compliments, they do raise some interesting questions for us (as a commercial organisation) and our clients and prospective users of our models and services. Below I address some possible concerns here.

Abuse of dominant position

Whilst there are technical challenges in developing ESG models and their development requires both financial and management resource, there are no other significant barriers to entry to the modelling business. Indeed, we do have credible competitors today and I expect this to continue in future. I like to think that our position is simply the result of the quality of our products and services. Any attempt to abuse this position through pricing would be checked by competitors.

Firms gain significantly as a result – our economies of scale allow many clients to benefit from our investment in people, intellectual property and technology. As explained before, we can offer modelling solutions at a fraction of the cost (and risk) of in-house development.

A systematic risk?

There are a number of possibilities here. Firstly, there is the possibility that a large number of firms are exposed to a common error in model implementation or calibration. However, even if firms had adopted a common approach (they do not), they are now required to apply the same level of testing and validation to external models as internal models so this possibility requires the failure of all firms’ testing and validation (and the collective failure of regulators to identify the shortfall).

Secondly, a possibility exists that software fails simultaneously and completely across a large number of firms. Whilst there must be some probability of this sort of failure, our testing and release procedures are designed to stop this happening – either accidentally or maliciously – and we believe this risk is very small. Firms have access to different versions of the software which could also mitigate the impact of such an event.

The other side of risk

Given the high cost and complexity of ESG model development (together with the paraphernalia required to meet emerging regulatory demands) it must make sense for the industry to share these costs through outsourcing. When we look at the potential operational risk associated with small in-house teams, the external model provider with critical mass in analytical areas plays a vital role in minimising this far more material risk. Now - if I were a regulator - that is the operational risk that would keep me awake at night.

Naturally, we would like to hear client views on this subject so please use the facility to comment.

Is regime-switching a cure for equity fat tails?

2009-08-17T12:31:01Z

The standard textbook model for equity returns is one which assumes that the (log) equity return in excess of the risk-free rate is normally distributed with a given mean and volatility. It is well known that this model fails to capture some of the empirical properties of equity returns, in particular the observed likelihood of extremely low (and high) returns will be much higher than predicted by the model (the so-called equity ‘fat-tails’). It also does not generate the bunching through time of extreme returns observed in real-world markets.

The inability of the Normal distribution to realistically describe observed behavior has a number of results:

The model is generally unable to accurately price equity options with different strike prices at the same time; the ‘correct’ volatility to use in order to reproduce a market price will tend to vary by moneyness of the option, to compensate for the mismatch between ‘true’ and modelled equity return distributions. The model may hence not be suitable for the market-consistent valuation of certain liability profiles.
If the mean and volatility of the normal distribution is calibrated to produce realistic average returns and return volatility, the tail percentiles of the equity return distribution are unlikely to accurately reflect historical data. Understating the likelihood of extreme events may be particularly problematic for real-world modelling work where the tails of the distribution have a material impact on the result (such as 1-in-200 Value-at-Risk calculations). The model only has two parameters so we can only fit exactly two characteristics of the real-world distribution.

A regime-switching equity model is a relatively simple technique that attempts to capture fat-tailed distributions, and is one that Barrie & Hibbert have employed in the past.

The basic idea is that, rather than describing equity returns with one normal distribution, we use a combination of two (or more) normal distributions, each with a different volatility and mean. This fits with the intuitive idea that equity markets ‘flip’ between states of low volatility (which prevails most of the time) and high volatility which occurs less frequently. Most importantly, calibrations can be developed that provide a relatively accurate fit to historical equity return data.

However, there are some important drawbacks; the assumption that volatility switches between two (or more) distinct regimes rather than changes continuously is unrealistic and means that projected option prices simply switch between two different states. Introducing further states can provide more realism but introduces a significant degree of complexity (and subjectivity) into the calibration process.

It is also not possible to capture the so-called ‘gearing effect’ where poor equity returns tend to be associated with increases in equity volatility. For these reasons, a regime-switching model would not be an ideal choice for anyone performing a real-world projection of a market-consistent balance sheet, particularly if using a replicating portfolio as a proxy: this is because the model would produce unrealistic distributions for the prices of any replicating assets that are valued using equity volatilities, such as equity options.

Note also that where the model is used for projecting at different time increments, a different calibration must be provided for each set-up. These calibrations will not normally be consistent with each other.

Finally, it may be necessary to model more than one risk driver using a regime-switching process. For example, a domestic equity index and an overseas equity index may both follow a regime-switching equity model, each with a different calibration. Should these regimes switch independently of each other, at the same time or with some specified correlation?

At Barrie & Hibbert we tend to favour models where equity volatility is itself stochastic, which provide a richer description of volatility behaviour that can also be correlated to other economic variables, such as equity returns themselves. The addition of random jumps in the equity price, where the timing and size of the jump is also stochastic, allows us to generate fat-tailed equity return distributions as well as realistic distributions of option prices.

This is the approach adopted in our recently-developed Stochastic Volatility Jump Diffusion (SVJD) equity model. Whilst any model is ultimately a simplification of reality and model choice is sometimes subjective, experience has taught us that there are compelling arguments why a model such as SVJD, rather than regime-switching, should be the model of choice for anyone wishing to capture the key risk drivers associated with equity exposure.

Using inertial and behavioural traits to boost pension contribution levels

2009-07-21T07:35:34Z

In my previous blog I highlighted the positive impact that increased pensions contributions combined with the right investment strategy can have on potential levels of retirement income.

Whilst this is clearly fine in theory, the pensions industry clearly faces a challenge in getting scheme members to sign up to this sort of contribution arrangement. In particular, recent market turmoil has resulted in increased skepticism regarding the relative benefits of long-term savings versus shorter-term ‘living’ expenditure.

However, there is evidence that inertia and other behavioral traits can be used to boost employee savings rates. For example Thaler and Benartzi (2004) propose the Save More Tomorrow (SMT) scheme as one potential solution to this issue.

In the initial SMT experiment, employees of one US company who had low savings rates were invited by a financial planner to raise their savings by up to 5% of their salary. 22% of the employees decided not to sign up to this immediate increase. These employees were then offered the SMT plan, where savings would go up 3% every time they received a pay rise. 78% of these employees accepted the SMT plan and their annual savings rose from 3.5% to 13.6% after four years. Interestingly, for those employees who opted for an immediate increase, the resulting average savings rate was only 8.8% of salary.

These findings suggest that people find future commitments easier to accept than current action, and are less resistant to forgoing future pay rises than seeing a reduction in take-home pay. As providers continue to evolve their corporate pensions propositions, building in this type of feature would support increased long-term savings levels whilst improving the chances that employees achieve meaningful salary replacement rates.

Do as I Say - How do we confirm that business decisions are being made in line with ERM vision?

2009-07-01T12:08:33Z

This relates closely to some of the discussion below around senior management’s understanding of models. For ERM to be useful, senior management needs to fully buy into the ERM vision, and what it means for their business models. If it identifies economically non-viable business policies, senior management needs to be willing to listen and learn from those insights. In the global insurance sector, this is happening, but it takes time, and some firms will embrace this more quickly and fully than others.

This can be a challenging process – it is not an easy message for ERM to take to the Board when ERM believe that the firm’s most successful product is loss-making, or that the firm is inadequately capitalized. These implementation challenges quickly move from intellectual and technological to cultural and political.

Fundamentally, for business decisions to be fully aligned with ERM, there needs to be a completely holistic and consistent approach to assessing market-related risks and costs in all product lines and in all forms of market risk. Management compensation and structure then needs to be aligned to these metrics as well as defining what exposures MUST be escalated to the boardroom table.

See Nothing, Hear Nothing, Know Nothing - How much do senior management need to understand of the models being used?

2009-06-29T09:29:41Z

This can operate on a number of levels:

First, there needs to be an understanding of the questions that the model is trying to answer and their fundamental ramifications for the firm’s business model. For example, if the model is being used to calculate 1-yr VaR based on market-consistent asset and liability values, this might mean that pricing products at a sub-market-consistent-level is suddenly going to start looking untenable. So this needs to be joined up with the way the firm more broadly measures value creation, and there may be a need to recognize that previous practices of, say, managing to GAAP accounting rules may not perform well from a more economically coherent basis.

Second, banks and insurance groups are holding and writing complex products with complex market/credit risk profiles. There has to be some deep technical understanding of these complexities at a very senior level in order to effectively and responsibly manage the business.

Third, senior management don’t need to be quants. But they need to have an awareness of the fundamental limitations of the models that are being employed by the firm in internal, regulatory capital and financial reporting applications. There needs to be an appreciation of what the models leave behind; that financial modeling always has model risk; that calibration is far from an exact science. And that these limitations may create perverse incentives for their people that may conflict with what’s in the long-term interests of the firm .

The collapse of commodity prices - is anybody listening?

2009-06-23T09:06:25Z

In March 2007 the Financial Services Authority published a report entitled ‘Growth in commodity investment’ in which they state:

‘...this [commodity] boom has mainly been caused by dramatic growth in demand (particularly from the rapidly developing economies of China and India); i.e. it is underpinned by what seem to be long-lasting fundamentals. It is widely stated that institutional investors will stay for the long term’.

Writing two years later, this assertion seems somewhat void. The FSA and others with ill-conceived perceptions of the safety and diversifying properties of commodities were proven wrong. We have seen it all before.

S&P GSCI Excess returns indices (semi-log scale)

Data Source: Bloomberg derived

Managing your risks is never easy. Recent analysis at Barrie & Hibbert in the light of developments in commodity prices in 2008 concludes that there are three important lessons worth highlighting:

Neither empirical evidence nor economic theory conclusively demonstrate that a risk premium is, or should, be earned on commodity investments over the long term.
In the short term and during periods of wider market stress commodity investments may not prove to be a good hedge.
The volatility of returns on commodities can be highly unstable over the short and medium term. Economic downturns are often related to volatile commodity markets.

Yield Curve Extrapolation Webinar

2009-06-18T13:28:28Z

We are always keen to share our latest thinking with our clients. As our client base has grown over the past few years it has become less practical to meet everyone face-to-face to share our thoughts and analysis. Fortunately the internet provides a medium for us to connect with clients all over the world and so we decided to experiment with running webinars to share our thinking.

The choice for our first webinar was easy. I have recently spent some time in Asia, speaking at conferences and visiting firms and regulators. Probably the most topical discussion point was unrelated to stochastic projection - the construction and extrapolation of yield curves. I suppose this isn’t surprise, given many of the Asian markets don’t have bonds with long maturities and many insurance companies have long-term liabilities. This maturity mismatch problem is the first hurdle to applying market consistent valuations in these markets.

We reviewed our approach to this last year and developed a methodology which extends the yield curve beyond what it available in the market, in an economically intuitive and sensible way that, we believe, is in line with broad Market-Consistent principles.

We presented this approach during two webinars on the 17th and 18th of June 2009 and have now posted these sessions on our knowledge base if you want to “listen" again.

Please help us understand if this experiement was successful by posting a comment to this entry - we welcome your feedback and suggestions for future webinars.

Can personal pensions produce meaningful retirement benefits?

2009-06-03T14:03:42Z

Given that people saving into the proposed national Pension Accounts scheme may have a contribution level as low as 8%, it is worth looking at what level of retirement income this could potentially generate. Let’s consider the example of a 25 year old male, looking to retire at age 65. Our analysis suggests that if he invested 50% equities and 50% bonds, it would give a retirement income of 27% of his final salary. If joining is delayed to age 40, the replacement rate drops to 14%.

Will this level of income be attractive, and will it be enough to support his lifestyle in retirement? Probably not.

So how do we address this issue? Clearly one solution is to persuade people to save earlier, and then to increase contribution levels. Using the example of a 25 year old male, our analysis shows that if he made a fairly modest increase of 1% per annum for the first 5 years (taking his contribution up to 13%), and if he maintained this level of contribution through to age 65, he would increase his potential retirement income to 42%. This starts to look meaningful.

That’s fine in theory, but how do we get scheme members to sign up to this sort of contribution arrangement, and how do providers and trustees communicate the benefits of this sort of arrangement to members?

More of which later…

Garbage In - Who polices the information used to run the model

2009-06-02T09:07:39Z

This is partly about finding the right regulatory balance between prescription and principles. Globally, insurance regulators have so far taken different positions on this spectrum. Some risk factors may be so common and ‘standard’ that a prescriptive approach to the criteria that the model and its calibration must satisfy seems obviously beneficial. As an example, if an insurance company has S&P 500 exposure, the regulator might state that the model should assume that the 99.5th percentile 1-yr fall in the S&P must be at least 40%. This approach can ensure some consistency in capital assessment across firms, and may seem like a no-brainer to most people.

The difficulties though, arise in the detail:

Who decides what the prescriptive measure is? The regulator? The actuarial profession? The government? How do they do that and will they want to risk getting it wrong? How often should that assumption be updated?
Are all firms S&P 500 investments really the same, or are some firms taking materially more active risk than others? Are all firms exposed to the same type of behavior in the S&P 500? What if a firm has a hedging strategy that means that they are more exposed to S&P falling 4% than 40%?

As a general rule, greater regulatory prescription usually means greater opportunity to game the system, or to find a regulatory arbitrage. It is worth emphasizing that even in a principle-based system that gives firms considerable freedom to make its own modeling assumptions, strong regulatory oversight is required. The UK ICA system is perhaps a good example of this working in an insurance industry context. However, it has highlighted the need for expert modeling analysts within the regulator that can perform effective scrutiny of firms’ assumptions and modeling methodologies.

Mark-to-market for assets of banks and insureres. Market-consistent valuation for insurance liabilities.

2009-05-28T12:26:47Z

There has been considerable debate in recent weeks around the mark-to-market valuation of credit-risky assets, and in particular whether differentiation can be made between credit losses and liquidity / risk premium losses (i.e. temporary impairments in the jargon), and if these should be treated differently in accounting valuation (i.e. the effect of temporary impairments can be removed from valuations).

Discussions around the identification and utilization of liquidity premiums have been topical in the life / annuity sector for a few years from a slightly different perspective. There, it has been argued that the uniquely illiquid nature of some insurer’s liabilities (e.g. fixed annuities) reduces the market value of these liabilities. This reduction in market value is usually assessed by trying to identify the illiquidity premium that may be embedded in long-term assets such as long-dated corporate bonds that are being held to match these liabilities. Whilst arguments for this approach can be made intellectually compelling, the estimation of such an illiquidity premium is highly subjective - arguably so much so that there is a danger that it simply becomes a mechanism for firms to fudge results.

Putting topical issues of liquidity premiums and pro-cyclicality aside, there are a number of other structural challenges faced by the insurance sector in attempting to perform mark-to-market or market-consistent valuation of its liabilities. Often the financial guarantees embedded in these liabilities are very long-term and can also contain highly complex, path-dependent forms of optionality. This necessitates subjective extrapolation of observable market prices to estimate market-consistent values for such liabilities. This issue is particularly pertinent in developing markets, or in times of market stress, but it is also a ‘day-to-day’ challenge for all insurance firms that are trying to mark embedded liabilities to market (e.g. what’s the 30-yr out-the-money S&P 500 option-impled vol?). I don’t believe this undermines the rationale for the estimation, but it is another example of how judgment has to be used alongside math in the assessment of market risk exposures and costs.

The impact of Quantitative Easing on longer-term government bond yields

2009-05-19T13:06:09Z

Central banks have found it necessary to use non-conventional monetary policy methods in an attempt to stimulate the real macro economy. By purchasing assets, the central banks are now in a position where they can increase the wider supply of money directly.

Many had expected that quantitative easing policies would lead to lower yields on long-term nominal government bonds, as quantitative easing is an attempt to lower expectations of longer term real interest rates to stimulate demand and bring inflation back to target levels in the medium term. In the UK this hasn’t yet happened – in fact there has been a relatively strong increase in longer dated nominal bond yields since the beginning of the year. The increase has been particularly strong at maturities beyond 10 years.

Figure 1 - Government spot yield curves at end December 2008 and 17 April 2009

(Data Source: Bank of England)

You could argue that the increase in longer term nominal forward rates might be due to an increase in inflation expectations offsetting the fall in real interest rates; rational agents could link increased money supply with inflation. Market participants could simply believe that the impact of quantitative easing would be substantial inflation in the future as the central bank fail to hit its medium term target for inflation. There is, however, little evidence from the yield curves to support this!

To show this point, let’s have a look at the break down of the nominal curve into the real and inflation forward curves (Figure 2, below). At maturities up to 6 years there has been a fall in the real forward curve which has been offset by an increase in implied inflation. To the extent that the real yield curve reflects expectations about real interest rates and the implied inflation curve reflects expectations about future inflation, this would mean that inflation expectations are higher. We cannot, however, rule out the possibility that the changes in the curves reflect changes in real term premia and inflation risk premia (or technical convexity effects).

The breakdown of the nominal curve also reveals that the main explanation for the increase in the longer dated bond yields is a very strong increase in real bond yields while the increase in implied inflation had a larger impact on medium term maturities. We had been surprised about the historical low level of longer dated real government bond yields since 2003. This was a global phenomenon which Allan Greenspan dubbed as the ‘bond yield conundrum’. Does the higher real forward curve reflect higher expectations of future real interest rates? We need to be careful with such interpretation! It seems unlikely that markets expected future real interest rates to be negative at the end of December 2008. The low level of longer dated real forward rates probably reflected a negative term premium embedded in longer dated real and nominal government bond prices. A negative term premium could reflect preferred habit and/or demand-supply imbalances.

A candidate explanation for the increase in the longer dated government bonds, in an environment where many expected quantitative easing to push down on nominal and real government bond yields of all maturities, is a less negative real term premium. The preferred habitat effect in longer dated government bonds may be unwinding as the government is becoming increasingly indebted and the uncertainty surrounding its capability of servicing the debt is increasing.

Figure 2 - Instantaneous Government forward curves at end December 2008 and 17 April 2009

(Data Source: Bank of England)

Interpreting movements in yield curves is never an easy task. The changes in yield curves since the beginning of the year leave us with a number of interesting questions. Does the increase in short to medium term inflation derived from the government curve in the UK imply an increase in inflation expectations as the intervention by the government and central banks push down on short to medium term real interest rates? Could the increase in longer dated (real) government bond yields reflect the unwinding of preferred habitat effects, and negative term premia, as government debt levels become higher? The yield curve is affected by a number of demand and supply factors. Interpreting the changes in yield curves is not an easy task!

Back to the Future - The past is not always indicative of the future

2009-05-18T14:38:45Z

The science of predicting changes in near-term levels of risk and the persistence of those changes has developed a great deal over the last 30 years or so. This has generally focused on predicting short-term asset volatility and using these changes to make inferences about tails, rather than being directly focused on predicting the tails. This is natural as it is a more tractable mathematical problem, and it is the right place to start. But there are fundamental limitations and dangers in using the most recent 90 days’ market behaviour to make predictions about what a 1-in-1000 or 10,000 day event might look like.

In early 2008, some financial institutions were declaring in their published financial reports that they believed they were capitalized to withstand a 99.97th percentile 1-year event. They have since been bailed out by their governments. Now, maybe we did just experience a 1-in-3000 year event. But the more likely explanation (putting aside that some risk and leverage was off balance sheet and effectively ‘model-exempted’) is that there was an extrapolation from volatility to tail that was systematically sanguine and self-serving.

Many modeling approaches exist that can do a better job of measuring tail risk – but their application and calibration will necessitate judgment as much as algorithms. Whether we like it or not, financial markets are not stable scientific systems in a perpetual equilibrium that produces infinite relevant historical data. That means that risk measurement has to be about more than a formula. So the formulas need to get more complex, but even more importantly, model calibration needs to be more rigorously scrutinized rather than being assumed to be a ‘solved’ algorithmic problem.

What just happened? Some perspectives and their implications.

2009-05-03T09:34:32Z

Perspective 1 (ex-post rationalist)
In recent years, risk, capital and leverage have been globalized and securitized in unprecedented ways. This has occurred at all levels of the economy. For example, at government level, foreign ownership of US government debt is at an all-time high, as is the overall level of the debt as a percentage of GDP. At the corporate financial services level, the securitization and distribution of traditional banking loans has globalized banks’ credit exposures; at a personal level, personal debt levels in developed economies have reached record levels.

The securitization and globalization of bank’s’ mortgage books is, in theory, a positive development – it should create more diversification for banks, and improve the efficiency of capital allocations. But its rate of growth was caused by (and then created) a number of principal-agent problems and other forms of market failure. Many loan originators had little incentive to perform effective underwriting, as they were acting primarily as distributors rather than risk underwriters. It became very difficult to know who was ultimately exposed to what. Financial innovators packaged these securitizations to keep them off-balance sheet, thereby helping banks to understate the true leverage of their economic position. Rating agencies were incentivised to rate these securities highly. Banks were incentivised not to look beyond the ratings if they could get away with it. Regulators let them get away with it – and at some point this created a fundamental dislocation of risk and capital in the global banking sector.

The net result of all of this was that leverage (government, corporate and personal) reached unprecedented levels, and some of this was inherently understated and lower quality than assumed at origination. This fuelled an asset price (real estate) bubble that was unsustainable and the level of leverage made the economy inherently very vulnerable to it bursting. It burst.

The financial crisis wasn’t caused by illiquidity; it wasn’t caused by pro-cyclicality; it wasn’t caused by mathematicians and their models. It was caused because the managers of financial institutions had incentives to write puts and not value them properly on their balance sheet or to hold capital for the risks they created. It was caused because regulators and accounting standard-setters were two steps behind financial innovators who were incentivised to stay two steps ahead. Fundamentally, regulatory capital-setting got broken in an avoidable way.

Perspective 2 (sceptical empiricist)
Ex-post rationalization is a wonderful thing, but risk is an inherent part of the financial / economic system. It is inevitable that asset prices will occasionally experience significant short-term downward shocks. This is not predictable or avoidable, except when looking backwards. But managers, regulators, rating agencies and perhaps even financial market prices systematically underestimate the frequency and severity of these events. Who cares about rationalizing the cause, it will be a different cause next time. The point is that next time is sooner than we think, and there is little we can do about that. So let’s address that issue by addressing the systematic flaws in our risk measurement techniques.

In particular, the prevailing approach of using normal distributions that are calibrated to very short-term data horizons is a structurally flawed approach to capturing the extreme low-frequency events that drive the results of interest. Our models and calibration methods need to go beyond this. That isn’t hard – we have hundreds of years of economic data to learn from; we have a range of fat-tailed financial models that can capture the inherent uncertainty in financial tail risk measurement. Managers need to understand this – not the formulas, just the quite simple principle that market risk is a complex and ill-tempered phenomena, and extreme events can be much, much worse than is implied by a normal distribution with a volatility based on last month’s market behaviour. Regulators and governments need to demand that risk analysis takes a broader, more judgment-based approach that goes further than short-term algorithmic data analysis and self-servingly sanguine extrapolation.

Perspective 3 (Anti-geek)
Markets are crazy. Let’s try not to worry about it. Worrying about it makes it worse – it creates pro-cyclical price depressions. Short-term market volatility is irrational and vastly overstates changes in investors’ expectations. Reacting to it creates a self-fulfilling prophecy. Mortgages and pensions are long-term, so we can safely ignore the short-term. These geeks and their fancy models have blinded managers and regulators and stopped them from thinking about the business fundamentals. Turn the fancy models off and let’s get back to simpler sums that the people that know the business can understand.

And my perspective?
Most of us will be able to find some grains of truth in each of these perspectives. In these fascinating times, truth is in the eye of the beholder. But the point is that there are valuable regulatory and internal risk management improvements that can be found from each of these perspectives. As a self-confessed wanna-be-a-geek-but-wasn’t-smart-enough-so-they-put-me-in-marketing person, it is perhaps not surprising that the first two perspectives resonate most strongly (and feel mutually re-enforcing).

Smarter use of more sophisticated models is essential to better appreciating the quantum of the extreme tail that market risk exposures create. Using a normal distribution with a volatility based on the last 30 days to estimate 99.97th percentiles of market risk exposures is a fundamentally bad idea, and that was as obvious before Q4 2008 as it was after it. Fat-tailed distributions and calibration approaches that are aimed at robust estimation of the tail are essential improvements to the VaR methodology. This will mean greater reliance on qualitative judgment and less on Exponential GARCH models (which are great for volatility forecasting, but not so great for 99.5th percentile forecasting). So, we need to make better use of the available science. And we need to recognize that risk managers must employ some (independent) judgment in the application of this science or ruthless innovation will again expose its weakest link (remember principles not prescription?).

But that alone isn’t enough. Regulatory and accounting systems need to incentivise holistic, bottom-up risk analysis – market risk, no matter its source or its wrapper, must have nowhere to hide on a financial institution’s balance sheet or regulatory capital assessment. Managers need to see this risk analysis as a source of insights into business decision-making, not merely a tool for regulatory appeasement. Aligning performance measurement and hence business models with the true economics of the business is a must-do. The banking sector has experienced the greatest public scrutiny as this financial crisis as unfolded, but on this latter point, the global insurance sector has at least as much to do.

So is there really a case for the in-house ESG model?

2009-04-15T07:01:16Z

One of our perennial conversations with prospective clients is the buy-vs-build discussion. Solvency II will bring an increased focus on internal models and some will make the mistake that this means they need to build models in-house. Internal in this context refers to the use of models and not where they are built.

Using models internally means understanding them, their limitations and questioning their results, it doesn’t necessarily mean building them from scratch. I am often surprised when managers believe that they should build their own Economic Scenario Generator (ESG) models and software. The expertise to build and maintain ESG modelling solutions are difficult to find and often expensive. I am acutely aware of the operational risks involved in building, maintaining and regularly calibrating an ESG modelling solution - it is the core of my business, has given me a few grey hairs and occasionally kept me awake at night.

Insurance companies have lots of other risks to worry about – and many pay us to manage the operational risk of maintaining a world-class ESG modelling solution – because it’s what we are good at.

I have heard many arguments over the years and in this note consider the pros and cons of building an in-house ESG.

Risk Management Round Table with Barrie & Hibbert Asia and Ernst & Young

2009-04-08T06:37:04Z

Introduction

In February 2009, Barrie & Hibbert Asia hosted two joint breakfast seminars with Ernst & Young. The aim was to provide an informal platform for CROs, CFOs and Chief Actuaries of the top life insurance firms to discuss the critical issues related to the use of stochastic models in the areas of risk management, capital management and asset allocation in these tumultuous times. Topics clearly at the forefront of top management thinking as demonstrated by the strong attendance at the events by senior staff, at both the Hong Kong and Singapore venues. While a lot of ground was covered in these events, these notes aim to summarise some of the more interesting and Barrie & Hibbert centric elements discussed in the Hong Kong event only.

Fitting for the location, Barrie & Hibbert founding partner John Hibbert, focussed the initial discussions on the appropriateness of using methodologies developed from more developed economies to other perhaps less well developed ones – a topic particularly relevant to a number of the Asia Pacific economies. Whilst it was acknowledged that the magnitude of some of the issues faced may be different, there was a general agreement with (and certainly no vocal objection against) the view that sound methods, grounded in economic theory, should be portable across regions. In some cases minor modifications may need to be made, but in general, that there should be no need for fundamental changes in approach.

Yield Curve Extrapolation Methodology

The recent example of Barrie & Hibbert’s new yield curve extrapolation methodology sparked the initial debate. Even in developed markets there will be the need to value liability cash flows beyond the term of the longest (suitable) market instrument, whether this is 15 years as in Hong Kong or just under 50 years in the United Kingdom. A few comments supported the Barrie & Hibbert approach as it improved stability of the longer forward rates from calibration to calibration (as supported by general economic theory) as well as removing the sensitivity of the longest rates to small changes or errors in the prices of the very longest market instruments (an issue with the method currently recommended by the CFO forum), which is typically where the market is less deep and liquid. In summary, while this is clearly a controversial topic, it’s hard to disagree (and no-one did) with the underlying principle that yield curves that are (unnecessarily) volatile at extrapolated maturities can be of no benefit to any of the key users of the financial information based on such yield curves: shareholders, analysts, regulators or management.

Ultra-Long Risk-Free Rates

The discussion turned to the estimation of ultra-long risk-free rates (an input to the above methodology) and whether these in particular, but also more generally whether all long-term assumptions, should be country or region specific. With the 120 year risk-free forward rate, there was no disagreement that it would be difficult and likely spurious to try and differentiate between the equilibrium risk-free rates over such a long time horizon, and hence that a global target is probably appropriate. In other cases, there will be arguments to the contrary, and country or region specific long-term targets will be more appropriate. A challenging area and like many aspects of long-term financial modelling, not one amenable to a one size fits all solution.

Global Market Movements

The ‘Insurer’s perfect storm’ followed with the scene being set with the global market movements over 2008 being placed in to their historic context. Even going back over 100 years, the 2008 global equity market falls are some of the most severe, both in magnitude, and the number of markets impacted. This has resulted in considerable strain being placed on market-based, risk-based capital adequacy levels and market consistent measures of profitability, which in turn has prompted discussions on market consistent reporting and Solvency II.

Model Performance & Risk Management

The year has also been a severe test for financial asset models and modellers . Though we remain in the midst of the financial tsunami, we can still start to assess how the models (and their calibrations) have performed in the face of such an extreme tail event as 2008. To set the scene, the following insightful anecdote was used, which brought an uncomfortable laugh, or knowing nod from the participants:

Barrie & Hibbert have made models (and calibrations) available to clients for quite a number of years now which would allow them to capture the key characteristics that have characterised the global equity market movements over the last 12 months: excess kurtosis or ‘fat downside tails’ in equity returns; and the correlation ‘spike’ effect between global equity markets, whereby the correlation between equity markets increased substantially, severely limiting the magnitude of the famed ‘diversification benefit’. So why then are almost none of Barrie & Hibbert’s clients using these models?

The answer must lie somewhere in the observation that by using models that can capture such extreme tail events as occurred in 2008, estimated capital will likely be higher, and quite possibly significantly higher than current levels.

Prior to 2008, it is easy to see the challenges a risk manager would have faced in asking their board for more funding for a model which demonstrates that the company requires significantly more capital to support an extreme scenario which does not feature in the historical record. Even more so when you consider that the competition may not be so capital disadvantaged.

As one of the participants pointed out however – it is only ‘one data point’. Or more precisely the case actually made was that they would be interested in seeing the error bounds on the average 10 day correlation figure of 87% between global equity markets (average between US, UK, Japan, Germany and France) in 2008 that was mentioned. A point well made, and clearly motivated by the need to be able to robustly justify to management any model or calibration changes that would have such a material impact on calculated capital. I think we all agree that while the answer is almost certainly not 87% (maybe it’s somewhere between 70% or 95%), we can be pretty sure that it doesn’t remain at a constant level (say 50%) in all scenarios, as is a common modelling assumption. So perhaps the extra data point called the financial tsunami will help companies make the decision to transition to using models that can better capture these more extreme dynamics. Then, if the conversation moves on to the topic of calibration rather than model choice, another step towards better risk management will have been taken.

Calibrating Models & Extreme Tails

Inevitably there was the question of how do you meaningfully calibrate models when the point of interest in the distribution falls in the extreme tails (i.e. 99.5% or 99.98%)? With different approaches, models and calibrations giving such a wide scope for interpretation, the question was raised whether regulators and policymakers need to be more precise about what their words actually mean? One alternative put forward was to pull the confidence level down to a point where there can be more confidence in modelling the distributions (say a 70% point rather than a 99.5%) and then apply some appropriate scaling factor. Whatever the approach though, it was noted that while models by design, are only simplified versions of reality, they still can and do add significant value, and good models can give the modeller a good sense of ‘new’ plausible future scenarios.

Conclusions

The topic was rounded off with the observation that whilst the industry will need to consider the practical challenges of applying market-based techniques, it is crucial to recognise that the benefits of a market risk-based approach to costs and capital are better risk measurement transparency and better risk management incentives. A corollary to this is that any revisions should not undermine the long-term benefits of the last five years’ improvements in market risk measurement and management.

Before handing the floor over to Ernst & Young, the first session was wrapped up with a brief look at future developments before finishing with some Q&A time.