The Chairman (Mr P. Fulcher, F.I.A.): This is a slightly unusual sessional meeting because we have two separate presentations on the topic of the liquidity premium. This is a very important issue in life insurance, not just for insurers but in terms of its wider implications for society. To give some examples, it has implications for the ability of people to achieve good incomes in retirement through the annuity market and the ability for insurers to be long-term investors. The whole Solvency II process was, of course, held up for several years by disagreement about this topic.
A number of theoretical techniques have been developed over the years to quantify the liquidity premium. Mr van Loon and Professor Cairns will be presenting a new one that they have developed, which has some very promising applications.
At the end of the day the politicians became involved in the development of Solvency II, and the theoretical methods were not adopted. Mr Dowthwaite and Mr Gore, on behalf of the Profession’s Liquidity Premium Working Party, will give you an overview of the history of Solvency II and how it may progress going forward.
To introduce our first two speakers: Andrew Cairns is the Professor of Financial Mathematics at Heriot-Watt University. Paul van Loon is in the second year of a PhD, supervised by Professor Cairns, on the topic of the liquidity premium. The paper on which their presentation is based won an award at the International Congress of Actuaries in Washington last month. So, it is a very high quality paper. I will now ask them to make their presentation.
Prof A. J. G. Cairns, F.F.A.: Thank you very much for the introduction. I will be doing the presentation part and Mr van Loon will contribute to the discussion. I should also like to acknowledge the sponsors of the project on which Mr van Loon has been working. First of all, we have Partnership Insurance. One of the supervisors of Mr van Loon’s PhD. project is Alex Veys, who is in the audience. We are very grateful to them for having sponsored this particular project, and providing the practical insights into the work necessary to make sure that we academics did not stray off too far in the wrong direction.
Mr van Loon is one of the first cohorts of PhD students who have gone into the Actuarial Research Centre, a research centre set up by the Institute and Faculty of Actuaries a couple of years ago to build up a centre of excellence in research training and research within the actuarial profession.
To outline the contents of the talk, I will first of all describe the basic problem. I will give a stylised account of what we are trying to do in terms of quantifying the liquidity premium. That will be through modelling, initially, the bid-ask spread on corporate bonds. Then, secondly, looking at the credit spread, which is another reported statistic on corporate bonds. Thirdly, we will discuss the numerical results we are finding for the liquidity premium.
We are trying to decompose the credit spread into various components. We have, at one end of the spectrum, the default or expected default and re-rating losses. Then the top slice that we have been looking at in this project is the illiquidity premium. I will tend to use illiquidity and liquidity interchangeably for the same quantity. In between, you also have a slightly uncertain quantity. People do not quite know what to call it. Sometimes it is the risk premium that is partly in respect of the default risk itself. There are other elements of risk that are also being contemplated.
As the focus of the work, we are considering an investor, be it a life insurance company or a pension fund, which is holding a corporate bond to maturity. That particular assumption allows us to make particular inferences from the data at which we are looking.
Figure 1 is very stylised. It is building up a picture of what we are trying to estimate. At the bottom you have a yield curve for risk-free, perfectly liquid, bonds. Think of that as being the gilts curve. The next step is where we move on to look at the corporate bonds market. The green line represents the yield that you would obtain if you were rewarded only for the expected default losses for a bond that is perfectly liquid. Moving up from that, we add in a risk premium for default risk for bonds that are still liquid. The next step is illustrated by the solid and dashed lines. This is where we introduce illiquidity. In particular, the illiquidity that is presented here is the result of the difference between the bid and the ask prices. The particular feature here is that the more illiquid a bond, the higher the yield. Even if everything else is equal in terms of default risk, greater illiquidity will push down prices and push up the credit spread and the yields on those bonds.
Equivalent bond yields
What we are trying to do in terms of measuring the illiquidity premium is to measure the gap between the yield on an illiquid bond versus a perfectly liquid corporate bond. The challenge is that there is no such thing as a perfectly liquid corporate bond. It is, however, that quantity, the third line up from the bottom of figure 1, which we are trying to estimate from the data that we have.
We have been very fortunate in having access to the market corporate bond data set for investment grade bonds. It is about 10 years’ worth of data. On any particular date we have about 1,000 bonds in the data set with various bits of information. We have some contractual data, which does not change from day to day, and then we also have some data that does change from day to day such as prices, yields, credit rating and credit spread. We are using most of those types of information in order to try to estimate the price if the bonds were perfectly liquid.
There are three stages to the modelling work that we have been doing. The first is to look at the bid-ask spreads. Modelling the bid-ask spread (see Figure 2) is a function of various inputs such as duration, etc. There is a residual term when you do this. You do not achieve a perfect fit in terms of the prediction using these covariates. The residual is what we call the relative bid-ask spread. Secondly, we then use the residual as one of the inputs into modelling the credit spread itself, which is a function of that plus a number of other variables. The third stage is to ask this question: what would this bond trade at if it were perfectly liquid as opposed to being illiquid and having our bid-ask spread? The difference is what we define as being the illiquidity premium.
Figure 2 shows the first stage. It is a regression equation. We are modelling the bid-ask spread. On the left-hand side you can see we are modelling the log of the bid-ask spread. That is a function of various inputs. There are things like the duration of the bond, the notional and various indicator functions such as whether the bond has been issued by a financial or a non-financial, is it a senior bond, etc.? There is a constant and β terms, which are the regression coefficients, and these change from day to day. Finally, there is a residual term which incorporates anything that is left over in terms of the difference between the actual outcome and what we predict on any date we are estimating these various factors.
Regression equation for the bid-ask spread.
Figure 3 gives a flavour of the results. There are many more graphs that we could go through but time is limited. On the left-hand side we have a plot of the β coefficient for A-rated bonds. We are looking at the impact of duration on the bid-ask spread. You can see the red and the black lines, which are the financials and the non-financials. They both have their different regression coefficients. You can see that they are initially very similar. But then round about the end of 2007, with the Northern Rock incident and on moving into the Lehman problems, you can see that a big gap opens up between the financials and the non-financials. Of course, it was a natural reaction to the crisis that the bid-ask spreads widen for financials compared with non-financials. This is just one example. There is quite a lot of variation over time in the regression coefficient but there is a story to tell.
Regression coefficients over time for log duration and log notional amount.
There is reasonable robustness in terms of the movement in the day-to-day values that you find. They do not jump about massively. Also, the variation in these coefficients does move around in a way that reacts to the unfolding events of the credit crunch and its aftermath.
On the right-hand side we have the progression coefficient for notional. It is a similar sort of picture. It is basically indicating that under normal market conditions, the larger issues were more liquid. Those are the negative numbers in Figure 3. But then, at the peak of the credit crunch, the biggest issues were less liquid than the smaller issues. Again, you can build up your own story of why that might have been. In those kinds of market stresses it is, perhaps, a bit more difficult to sell large quantities.
Figure 4 shows histograms of the residuals. They are centred around 1. This indicates that the prediction is pretty accurate. A value of 2 would mean that the bid-ask spread was about double the size that we would have predicted.
Histograms of RBAS (the exponential of the residual in Figure 2) on two dates.
Figure 5 shows the regression equation for the credit spread. Again we use a similar regression-based approach. This time the inputs are slightly different and we also include the residual bid-ask spread that we have just derived in the previous part. The residuals at this stage are of less interest. Again, just one or two examples in order to give you a flavour of what the results were showing. First of all, in Figure 6, we are looking at the regression coefficient for duration. If there is a positive value for that coefficient, that is indicating that you have a rising credit spread curve for the A-rated bond. During the credit crunch the coefficient went negative, which indicated a falling credit spread curve. That was really indicating that there was the most immediate credit risk or default risk in the shorter term compared with the long term. So, if things were going to go wrong, they were going to go wrong sooner rather than later, which is an interesting observation. On the right-hand side we have the γ coefficient for the relative bid-ask spread. This is the key thing in terms of estimating the liquidity premium. In this case, the coefficient is telling us that if it is positive, that is a good thing, because that is what we would logically expect from our estimations. However, in the run up to the credit crunch, this coefficient was very close to 0, and occasionally slightly negative. That was really indicating that there was no premium for illiquidity. Very soon afterwards, everybody suddenly realised that this was a dangerous situation in which to be. Of course, there were many other things going on at the same time. Clearly, that was a point where the liquidity premium, as we will see, was suddenly going to start jumping up.
Regression equation for the credit spread.
That brings us onto Figure 6b, where we are again dealing with A-rated bonds. It shows the non-financial indicator, and is indicating that there is no difference between non-financial and financial in the run up to the credit crunch. Similarly, there was no difference between senior bonds and non-senior bonds in terms of their pricing and credit spreads in that period. Of course, soon as the credit crunch came along, there was a flight not to liquidity but to bonds that were more secure. Even if there was a default for senior bonds, you would receive more than for non-senior bonds. Similarly for the financials, the negative coefficients here were indicating that the financials were less secure than the non-financials.
(a) Gamma regression coefficients for log duration and the RBAS; (b) gamma regression coefficients for the non-financial and seniority indicators.
Figure 7 shows our estimates of the liquidity premium. Again, this is just for A-rated bonds, to give you a flavour of the results. Initially, in the run up to the credit crunch, we can see very low liquidity premiums. Of course, they then steeply rise. They do not immediately jump up at the time of Northern Rock, but they do peak shortly after the Lehman bankruptcy. They gradually start to drop back at the same time as credit spreads themselves were falling back. The left-hand figure shows the liquidity premium in basis points. The right-hand figure shows the liquidity premium in terms of the percentage of the credit spread. What you can see in terms of the percentage is that it does vary quite a lot. Near to 0 before the credit crunch and then more typically between 30% and 50% as time went on after the credit crunch.
Median liquidity premium for A rated bonds in absolute terms (basis points) and as a percentage of the credit spread.
Summarising, the method that we have developed using this very rich data set is a method that applies to all quoted bonds. It is done on an individual basis. We have an estimate for each individual bond rather than some broad brush percentage, which applies to all bonds regardless of whether they are illiquid or liquid and with different durations, different issuers and so on. It is a relatively objective method. It does not really require any subjective inputs compared to some of the other alternative approaches. The only assumption is which of the regressors do you put into the model? The parameter estimates that we have are reasonably robust over time, and we also have an intuitive story to tell.
I have talked already about the outputs that we obtain in terms of our estimates of the liquidity premium. Even if an average bond has a liquidity premium as a percentage of the credit spread of about 40%, there are other bonds that are more liquid than the median, so they would typically have a lower percentage of the credit spread. A typical figure would be about 32%. If you go to the most illiquid bond, you are up to rather higher levels of percentages for the percentage of the credit spread, perhaps up to about 55%. That gives you an idea of the impact of the bid-ask spread and the assessment for individual bonds.
Figure 8 depicts what happens next. I have talked about a situation where we are looking at an investor who is holding to maturity. We have the liquidity premium at the top. Expected default losses at the bottom. Some sort of risk premium or other factors in the middle. If you have a slightly different investment mandate, maybe you intend to hold for only 2 or 3 years, or perhaps you have investment mandate where you are required to sell if the bond has downgraded below BBB, you are going to be somewhere in the middle. You will lose some of that liquidity premium. You might have to sell and therefore lose out to some extent. Work on this aspect is still in progress. We know what the picture is, but we do not know what the numbers are within that picture.
Discussion: holding time
The Chairman: Does anyone have any immediate questions?
Mr M. H. D. Kemp, F.I.A.: The regression model you are using is an additive model. To what extent would the answers have changed if you had used a different sort of model, for example, one that had multiplicative components? Did you try to fit any different sorts of regressions to see whether the choice of model type might be driving the conclusions you draw from the analysis?
Prof Cairns: We have done a little bit of additional analysis perhaps more in the direction of trying out different inputs. Our equation uses logs so it can be regarded as a multiplicative scheme. Of course, there are numerous other ways that you could do it. What we can say is we have reasonably satisfactory results within the restricted range of tests that we have done.
Mr D. I. W. Reynolds, F.I.A.: Over what period do you carry out the regression? Surely there is some serial correlation between the data. You clearly could not use 10 years and then 10 years, 300 days, etc.
Prof Cairns: There is a very simple answer to that. Each trading day is taken in isolation. You do see a little bit of volatility from one day to the next but they are pretty consistent, and then you observe bigger variations over time. The next step would be to start to try to do some time-series modelling in terms of predicting where the variables are going to go in the future. This would be particularly important if you are interested in asset liability modelling. We have deliberately taken the approach of taking each day in isolation, partly as a robustness check, because if we found that particular coefficients were jumping around then that would very much invalidate the use of one particular input as a meaningful statistic.
The Chairman: I will defer further questions to the end. I think it is worth saying that we did not give the speakers a huge amount of time – but they are presenting an extended version of their work at the Risk and Investment Congress next month. There is also quite a detailed paper outlining their work on the web.
We will therefore hand over to our next two presenters. Carl Dowthwaite and Bob Gore from the Liquidity Premium Working Party. Mr Dowthwaite is Group Commercial Actuary at Legal and General and has spent much of the last 3 years working to understand, and lobbying very hard on, the proposed Solvency II rules. Mr Gore is a principal adviser at KPMG, although also spent 7 years at Legal and General, working on the financial reporting side, including financial reporting of annuities. They are both experts on this topic.
Mr R. N. Gore, F.I.A.: In terms of the agenda, this evening I will go through some brief background of the financial crisis and its impact on liquidity premiums, discuss some of the possible learning points and cover the interaction with liabilities. Mr Dowthwaite will then take us through Solvency II and the liquidity premium in that respect, its evolution, implications and challenges.
In terms of background, several studies in corporate bonds spreads have identified deviations from the theoretical price of bonds caused by the influence of illiquidity in the market. There are a variety of estimation methods, some of which we covered in the appendix to the presentation. We have seen a new method detailed in the previous presentation. To put it simply, the liquidity premium is part of the additional spread over the reference yield curve, which determines the market price of an asset. Its use has been a long held and lately a key part of UK insurers’ and annuity writers’ business models. Even with the budget announcement about annuities, liquidity premiums will still continue to be important for the insurance industry. Its existence helps insurers in their role as providers of longer term funding in the marketplace together with generating consumer value in provision of annuities. The liquidity premium has been around for some time but has appeared to be relatively small in stable markets. It was not really until the onset of the financial crisis that major and serious questions were being asked by the industry. With the increase in yield that we saw through the crisis, we have to ask whether the additional return that is being offered to us is for increased default expectations and increasing uncertainty around the yield, or is it because of an increase in the liquidity premium. Bear in mind this was coming at a time for the insurance industry when Market Consistent Embedded Value was effectively being launched in the UK. It caused somewhat of a controversy about the ability of firms to recognise liquidity premium values in market-consistent measures. At the time it was critical for firms to ascribe a value to the liquidity premium to support the valuations.
So why consider liquidity premiums now? Well, we have moved through the financial crisis and, in theory, we should know a little bit more than we did at the start. It would be useful to review the industry experience and practice to date. There are still ongoing aspects to the liquidity premium with regard to the implementation under Solvency II. There is still a lot for firms to consider. This is a good point for the working party to consider issues and report back to the profession and obtain your views on what is an important subject.
Figure 9 covers bond spreads generated by the Merrill Lynch Index, and the graphs underlying that are covering the expected defaults under the Bank of England model and the uncertainty allowance in spreads. We can see that against this there is quite a variation from the late 1990s through to the end of last year. While the cause of the crisis has been much debated around combinations of credit in liquidity issues in the marketplace, dissection of those causes is not so important as to look at the outcomes and the behaviours. Certainly, there was market dislocation. This was not reflected in severe default experience as perhaps some of the high-level spreads that you see on the chart might have indicated. Bear in mind this chart is set against a backdrop of falling yields over the last 20 years for the industry, as we can see from the allowance we expected and the uncertainty allowance and defaults, there is a clear gap between the spread of the corporate bond and those allowances. It indicates that there are other factors at play in valuing corporate bonds, including illiquidity premium. From the graph we can see several spread increases as credit events agitated the market. At the start, expected defaults and uncertainty in defaults accounted for, roughly, about 75% of the corporate bond spread. This dropped to slightly <50% at the end of 2013. The major variations we see are from the 1998 long-term capital management crisis in August of that year, the Lehman crisis and the euro debt crisis, and are marked on the graph.
Corporate bond spread index UC00 and credit default allowances (January 1997 to December 2013) (bps). Note: The allowance for expected defaults and allowance for unexpected defaults are shown cumulatively. These results should not be taken to represent the view of the Bank or FPC of the size of any illiquidity premia in sterling-denominated corporate bonds. Source: BofA Merrill Lynch Global Research, used with permission, Bank of England.
If we think about expected and unexpected liquidity as we do about expected and unexpected allowances for credit defaults, we should expect the liquidity premium to behave very differently in good times and bad times. This is what we see. Movements at times have been seemingly independent of both credit risk and standard liquidity measures. It highlights the dangers associated with taking market prices when markets are highly illiquid and imperfect. Again, it is worth observing that over this period actual default experience has not corresponded to those extreme spreads that I have mentioned.
An important question is whether the liquidity premium has been a broader indicator of uncertainty or fear in markets leading to the undermining of confidence in apparently sound institutions? Another issue is what of pro-cyclicality in the graph? Normally, we think of this in relation to the economic cycle, but equally this could be viewed as a cycle of a crisis. Was this liquidity a downward spiral? Or, as several commentators note, a de-leveraging downward spiral due to previous excesses in the market unwinding? Certainly, these have been unusual and uncertain times and therefore interesting times. But what can we learn?
First of all, models of risk during the financial crisis perhaps proved a little optimistic both for investors and credit rating agencies. Increasing globalisation, together with the increase of volatility of financial markets called into question the exactness of credit ratings, and the credit rating agencies authority. Perhaps revisions were not timely, causing loss of confidence. Perhaps, too, reconsideration of intra-asset class diversification benefits at times of stress was needed. Certainly, when macro risk shocks hit, and a particular asset class is hit, the diversification benefits are reduced.
Intra-asset class correlations increase in the crisis as credit assets were increasingly viewed by some risk-averse holders as the same. Were these changes in stress adequately modelled? Or did risk models on which much of modern finance depended in the run up to this crisis underestimate the uncertainty of outcomes? If they did, why? Was this the over-reliance on value-at-risk models, which dominated, or the Gaussian Copula, which has received some criticism? Or rather, perhaps, it was due to mono culture and the lack of questioning assumptions or inadequate stress testing?
Certainly, this was an extremely unusual situation and scenario to play out in the markets and could not have been foreseen – or could it? The Financial Crisis The Financial Inquiry Commission in the US outlined that there were warning signs. As its chair said, the tragedy is that they were ignored and discounted.
So what can we take away about the points on liquidity premium? I think, first of all, we need to consider what is the correct price in illiquid or imperfect markets? The lack of certainty can mean asset liquidity can change through time and through markets. Methods taking credit for liquidity premium need to take account of such uncertainty.
Secondly, studies of negative CDS bond basis during the crisis indicate a variety of explanations. Falling bond prices, perhaps due to de-leveraging, less perceived worth of CDS due to counterparty risk, coupled with limits on arbitrage capital and finance risk is one argument. Using a liquidity premium with undue reliance on assumed financial relationships or mechanisms should be avoided and should be stress tested.
We have also seen capital from the banking side, and capital adequacy ratios, losing credibility to a degree under the old standards. New standards, covering, among other things, limits on asset quality, countercyclical buffers and liquidity standards ensuring capital was supported by the re-emergence of stress testing, has sought to improve the protection offered. This has implications for firms in considering forward-looking assessments for solvency in demonstrating solvency under Solvency II.
The crisis for some shows the limits of markets, and for others it shows gaps in the structure of regulation from rating agencies, corporate governance, or government regulation and policy making. Is this a fair assessment of what commentators describe as a perfect storm? It does explain tensions in ascribing a value to the liquidity premium in firms’ balance sheets from a regulatory point of view.
Pro-cyclicality, the amplifying of market fluctuations, as described above, has serious implications for the insurance business, and there is a strong and clear need to make liquidity premium measures an aid to stability, and allowing adequate provision over the economic cycle or indeed that of any crisis.
I should like now to take a couple of moments to think about liquidity premiums as they apply to liabilities. We can now consider the motivation for using liquidity premium in the insurance industry. Illiquid liabilities allow investments in illiquid assets, which may be expected to achieve high yields, and so it is reasonable for the fair value of the liabilities to be reduced to reflect this benefit. This may be applied by increasing the liability discount rate. In thinking about this, there should be some concepts, which might reasonably be expected to apply when determining a fair value of the liquidity premium in valuing liabilities. Predictability of the timing of the liability cash flows should increase the liquidity premium benefit. The value of the liabilities ought not to change due to changing the assets backing the liabilities, as such assets are technically independent of the liabilities. Typical assets used in the market for backing such liabilities rather than the actual assets backing them may be a more suitable basis for determining the liquidity premium. The liquidity premium derived from assets, it is worth noting, will not be directly equivalent to a liquidity premium in the liabilities. For example, corporate bond assets do not typically contain demographic risks, which are present in annuity liabilities.
A mark to market model method, using best estimate probabilities, will not typically arrive at a true market price, which will be influenced by other factors such as capacity in the market for risk. Related to liquidity in liabilities is the situation where an insurer has only a limited ability to transfer out its liabilities. For example, where reinsurance availability is limited, which being disadvantageous to an insurer, should increase the size of its liabilities and limit its liquidity premium benefit. Perhaps, similarly higher levels of free assets reduce the company’s risk of being forced to sell assets in the market earlier than intended and so could potentially increase the level of liquidity premium benefit in their liabilities. It is also worth noting predictability of the size of liability cash flows may have less influence in liquidity premium benefit in certain instances. For example, if claims are higher than expected, extra new assets are bought rather than existing assets being sold earlier than intended.
What I have outlined are general issues that we should bear in mind when considering the liquidity premium under Solvency II. Mr Dowthwaite will now pick up that part of the presentation.
Mr C. E. Dowthwaite, F.I.A.: We are now going to move on to consideration of the liquidity premium in the context of Solvency II. I am going to argue that the final agreed approach to the liquidity premium in Solvency II and the route to arriving at that approach provide important lessons for future debates on the issue.
The evolution of the treatment of illiquidity premium in Solvency II went through at least four distinct phases. This is in part a reflection of the importance attached to this issue and the range of views on this issue which, towards the end of the process, was one of a small number of issues that delayed agreement on the final Solvency II rules. If we look at Figure 10, the original directive, on the extreme left, had no explicit allowance for liquidity premiums. Quantity Impact Study 5 (QIS5), about 18 months later, did include a liquidity premium and that was set at around 50% of a portfolio-independent spread, with the liabilities bucketed, with some buckets including 100% of the premium and others 50%, 25% or 0%. This approach is probably closest to the graphs Mr Gore used earlier, in terms of the actual liquidity premium derived from that approach.
Evolution of liquidity premium in Solvency II
But the debate evolved further and this again changed 2 years later when we saw the emergence of the matching adjustment and the countercyclical premium. Those in turn ultimately evolved into the final rules for the matching adjustment and the volatility adjustment, which we saw in Omnibus II at the beginning of last year.
The general trend in this evolution, as you can see from the descriptions, was that the allowance for liquidity in liability valuations generally increased as the debate on the issue continued.
I think it is useful at this point just to consider the matching adjustment and the volatility adjustment in their final form in a little bit more detail. What are the key features of the matching and volatility adjustments? The key points with the matching adjustment are that it is both very restrictive in the assets and liabilities that it can be applied to, and that it gives higher values in terms of the liquidity premium than other methods of assessment. For the matching adjustment, liabilities need to be fixed or inflation linked. Assets, in turn, need to have fixed cash flows and to match those liabilities. The matching adjustment is calculated as the spread on a firm’s own portfolio, less an allowance for the expected cost of defaults on that portfolio. In contrast, the volatility adjustment is much more widely applicable. It is based on a reference portfolio of assets and the calibration is set to be 65% of the spread less an allowance for defaults. In both cases we see significant regulatory safeguards, including disclosure of their impact on a firm’s balance sheet, and for the matching adjustment prior regulatory approval is required.
So, why did we end up with the matching adjustment in the final Solvency II rules? I think that we can identify four distinct factors that were influences in moving from the earlier forms of liquidity premiums in the Solvency II rules to matching adjustment in its final form. Two clear influences, as Mr Gore mentioned earlier, were the impact on consumers and investment markets. There was particular concern around creating disincentives to invest in corporate credit. This was a particular concern at a time when encouraging investment to support economic growth in Europe was seen as an important issue. There was also clear concern about pro-cyclicality, particularly if widening spreads meant the corporate bond investors could become forced sellers. Running alongside this is the simple fact that for a matched annuity portfolio, that portfolio is exposed to defaults rather than spread movements. In these circumstances, a measure that focuses on defaults does seem to have the merit of considering the actual risk to which the business is exposed.
Moving on to the implications of the rules, so far as capital and solvency are concerned, although there is still a degree of uncertainty as to how the rules are implemented, we are likely to see a lot more consistency of approach relative to the range of interpretations of the current Pillar I rules. Some firms may see a limited change in how best estimate liabilities are calculated as the new rules could be relatively close to their current approach.
It is worth bearing in mind, though, that we are seeing some significant changes in the balance sheet as a whole. For example, we are seeing for the first time the introduction of the risk margin within the calculation of technical provisions. That will, however, be mitigated by the transitional provisions included in the final Solvency II rules.
It is worth comparing the allowance for liquidity premium in the various stages of evolution of the Solvency II rules. Figure 11 summarises the explicit or implicit allowance for defaults comparing the measures at each stage of the evolution of Solvency II. Here I have used as an example, a single A-rated corporate bond with results based on the parameters in the European Insurance and Occupational Pensions Authority (EIOPA) long-term guarantees impact assessment. It should be noted, though, that the values in Figure 11 are based on 2011 values and are higher than we are currently seeing in investment markets where spreads are materially lower. The narrowing of the spreads will have compressed the results here. As you can see from Figure 11, the approach taken can make a significant impact and different approaches can give quite different results. The allowance for defaults per annum is materially lower under matching adjustment than, for example, under the QIS5 approach.
Matching adjustment calibration annuity backed by 10-year A-rated corp bond
It is worth bearing in mind, though, that firms will, of course, hold capital in addition to this allowance for defaults within the best estimate liabilities, and so the total allowance for defaults on the balance sheet will be significantly higher than implied by these values Figure 11.
What are the implications of this for investment strategy? The first key factor is likely to be whether assets qualify for the matching adjustment, with assets with variable cash flows, potentially at risk. The ability, however, to group assets together to generate fixed cash flows could significantly mitigate this requirement. The eventual calibration of the matching adjustment itself will be set by EIOPA, and again that points to having a material impact particularly on the relative attractiveness of different asset classes and of different corporate bonds. Likewise, for standard formula firms, the SCR parameters, set in Level II, will have a material influence on the relative attractiveness of those assets. There are some restrictions on trading within a matching adjustment portfolio. Firms are not restricted in respect of buy and hold strategy as long as the portfolio matching is maintained.
So, what are the challenges emerging from the final approach that was adopted in Solvency II? For the matching adjustment, the key issue is likely to be whether the calibration, being based on historic defaults, is sufficiently risk sensitive. The derivation of expected defaults will clearly react only relatively slowly as market conditions change. If we see worsening market conditions, we expect to see a relatively slow change in the matching adjustment allowance for defaults. It is worth reminding ourselves, however, that this calculation is only the calculation of technical provisions and so insurers are also required to hold capital in respect of the 1 in 200 stress on top of this allowance.
For the volatility adjustment, the key issue is likely to be whether the measures adequately reflect the risks, given that firms using the volatility adjuster can be significantly exposed, for example, to spread movements, where they have surrender options in their portfolio. Again, there is, however, a significant mitigant to this situation. Firms will be expected to hold capital in respect of that surrender risk, so that will diminish any concerns about the risks generated by this approach being reflected in the balance sheet.
Where do we go from here? After the enormous upheaval that Solvency II represents, in terms of its implementation, it would be nice to be able to pause for breath. However, although the fundamentals of Solvency II are now clearly set, there remains a lot to do before the approach can be finalised in every level of detail. We are not expecting to have final Level II rules, for example, until early next year. We can also expect a lot of debate and scrutiny both in the run up to and the aftermath of the implementation of Solvency II. Just as Solvency II is entering its final phase, we are potentially starting the whole process over again with International Capital Standards (ICS). We can only hope with ICS that some of the lessons learnt in the development of the evolution of Solvency II are taken into consideration as ICS evolves.
To sum up, what are the conclusions? We do not have a universally accepted approach to calculation and application of liquidity premium. Mr Gore has already mentioned some of the key lessons that we need to learn from the financial crisis. Running alongside the various theoretical derivations, we see the impact of real-world influences, including influences such as the impact on investment markets.
Where there is no need for firms to realise bonds, Solvency II decisively rejected reflecting short-term market spread movements in the calculation of technical provisions. The financial crisis surely had an impact on this decision.
We are now about to embark on a further exercise to develop international standards. With the lessons of Solvency II still fresh, and with a better understanding of the broader issues surrounding the use of liquidity premiums, we can hope that the debate on international standards is better informed than the difficult gestation of these measures in Solvency II. We look forward to a stimulating debate to these issues. I should finish by thanking my colleagues on the working party for their contributions so far. As our work remains work-in-progress, we very much continue to welcome ideas and suggestions for further work.
The Chairman: We will now open the floor for discussion.
MrC. D. Lewis, F.I.A.: How comfortable are we at calling the matching adjustment a liquidity premium when there is going to be a large amount of credit risk premium left in the residual spread and indeed large portfolios of very liquid assets would have generated large volumes of this matching adjustment during the crisis?
Mr Dowthwaite: It is an interesting question. I do not think any of the official Solvency II rules call them a liquidity premium. I think we know they are there as a sort of enhanced liquidity premium. Certainly, the liquidity premium is part of the matching adjustment. I am not sure how important the labelling is. It is there to fulfil a role.
Mr A. D. Smith: I should like especially to thank Mr van Loon and Professor Cairns for the clarity of their presentation and the way in which they explained how the liquidity premium arose from the data they analysed and, in particular, the way in which their analysis of liquidity premium was related directly back to specific measures of illiquidity on the underlying bond in terms of bid-offer spread rather than just being some balancing item when you have tried to explain other things. In my view, that makes it a really worthwhile and groundbreaking piece of research and is very helpful conceptually by linking it so explicitly to the illiquidity of the assets.
On the liabilities side, I do not think we have the same clarity. I think we use phrases like illiquid assets. If I look at the measures of illiquidity in the paper, they clearly do not apply to the liabilities at all. In fact, the definition in relation to liabilities seems to be about cash flow predictability. If we apply that back to the asset side, they are looking at bonds where the cash flows are almost completely predictable.
I do worry that we are using the term illiquid assets and illiquid liabilities with a completely different meaning of illiquid. A cynic might suggest that this was a rhetorical sleight of hand which, if nobody noticed, could then be used to justify taking the illiquidity premium on the asset side and adding it to the discount rate for the liabilities. I am sure that would not be the intention. But there is certainly a danger of that because of the unnecessarily confusing terminology. So I should like to invite the working party, perhaps in their response this evening or in their continuing work, to be much more explicit about why they think that there is a connection between the fundamental work that the Heriot–Watt team have done and discount rates for liabilities. It seems to me that part of the story still has to be substantially told.
The Chairman: I will ask Professor Cairns and Mr van Loon whether they want to comment because that was a thought I had as well. You talked very much about the asset side, and your liquidity premium is a function of the assets. Mr Gore and Mr Dowthwaite started from the principle that the liquidity premium should have nothing to do with the assets. It would be interesting to discuss that point and have a contribution from both sides.
Prof Cairns: I am certainly very sympathetic to Mr Smith’s comments about the distinctly different nature of the assets and the liabilities. There is a question mark in our minds as to how you actually use the results of the analysis of the corporate bond data. The principal way, we feel, is in terms of thinking about it in terms of asset liability modelling rather than valuation. On the valuation side there really is no obviously correct answer to that question. It is not a perfect world and there is no perfect replicating portfolio. Everything is uncertain. Thinking in terms of asset liability modelling is part of the work onto which Mr van Loon is now moving, looking at portfolios and thinking about the individual constituents of those portfolios and then, as a separate exercise, considering how that links to the liabilities side of things with potentially some uncertainty in the cash flows.
The Chairman: Mr Dowthwaite, did you want to comment?
Mr Dowthwaite: The only thing I would add, which may be a little controversial, is that the matching adjustment, at least, whatever one thinks of it in terms of its calibration, does reflect an explicit link between the assets and the liabilities that a firm is holding. There are interesting questions where you are expecting to earn less on your asset portfolio than allowed for in your discount rate, whether that is a reasonable approach, and that potentially is a possibility in some portfolios, but the matching adjustment does specifically recognise that link between assets and liabilities.
Mr J. A. Jenkins, F.I.A.: I have a question for Mr Dowthwaite or Mr Gore. If I understood you correctly, EIOPA calibrates this matching adjustment. But I am not sure how frequently they will do that. Companies obviously do monthly reporting. How often does the white smoke come out of the EIOPA chimney to say this is the matching adjustment? Is it monthly or is it quarterly? If it is less frequent than monthly, will companies be able to have a go at it themselves in order to do their reporting or will we have discontinuities every time that there is a recalibration? If they do a recalibration, can you then offset that against the 1-in-200 capital requirement so you do not get a discontinuity in the overall amount of money being held?
Mr Dowthwaite: I do not think we yet know. They are still working on it. We have had one set of EIOPA calibrations so far. That was the one for the Long-Term Guarantee Assessment. I am not sure that is changing materially, for example, for the stress test.
One thing I would say, though, is that, given it is based on historic defaults, I am not expecting to see very significant movements from month to month, for example. Maybe it will be less of an issue than it might seem on the surface.
The Chairman: It is probably worth clarifying what EIOPA publishes is a fundamental spread, which is the part you deduct from the market yield to find the discount rate. EIOPA referred to 30-year averages. Technically, a 30-year average ought to move pretty slowly. The volatility adjustment is something more interesting in that regard. That will be more volatile because they are telling you what the spread part is. So actually your question becomes more an issue when considering volatility adjustment.
Mr R. J. Houlston, F.I.A.: This is a question for Mr van Loon and Professor Cairns. You appear to have used quotations from the market, which have not necessarily any consistency with the actual volumes of business or the directions of transactions. Have you thought about making adjustments for these factors? Spreads can be extremely wide if somebody wants to sell and not buy. It would strike me that if the markets are illiquid then the spreads would be wide and you may not be obtaining the right answers, certainly at times of stress.
Prof Cairns: Certainly, that is a very good question. We do not really have an answer to that. We have to work with the data that are made available to us. You are quite right that if you have a lot to sell all of a sudden, then clearly you are going to do worse than the so-called market quote that has been provided to our data provider, Markit. Obviously, that sort of thing definitely needs to be taken into account. What we would need, in order to be able to extend our analysis, would be to gain access to a good quality and extensive database, which actually gives you the transaction data rather than the quotations.
In terms of the data that we have used, we clearly recognise that it is a shortcoming of the analysis that it uses quotations rather than actual trades. But, on the other hand, we do take some comfort from the fact that we do find relatively consistent results over time, and results that have intuitive explanations. You might argue over the size or some of the numbers, but at least we feel that the overall approach is reasonably robust.
It is not a model that is going to tell you the effect of liquidity if you have a very large amount to transact. It is more related to what would be applicable for a typical trade in the market. Even then my expertise is not enough to be able to say whether those numbers represent typical transactions or something else. Perhaps other people in the audience, who are closer to the coal face, might be better able to comment on what the numbers that are provided to the market actually mean in terms of the relationship to genuine trades.
Mr M. G. J. White, F.I.A.: I have a number of points to make, not all very well linked together. I am wondering to what extent this illiquidity premium discussion is just a lobbying matter. Is it simply a case of the industry wanting the rules to be changed as market prices swing between high and low? Are we making the argument that a divergence in prices between government bonds and corporate bonds, when it happens, is to be ignored? In other words, ignoring market values? When market values are high against lower risk assets of similar term, will companies ask to use the lower numbers? In the past they have not.
The term the “perfect storm” has been used. This has connotations of something unlikely and unusual. With the incredibly low interest rates we see today, can we be confident that a really good crash does not lie ahead? I think it may, but whether it will be dramatic or long drawn out I have not a clue.
I would like to finish with a much more general point. Something which seemed absolutely fundamental to actuarial work years ago, when I was studying for the exams, was to start with the liabilities first and then the suitability in term and type of the assets available to invest in to meet those liabilities. The essence of the actuarial challenge was to steer the institution concerned through an uncertain future. To what extent was the institution safe, we would ask, against fluctuating market prices and in fluctuating liquidity within the market?
Mr Smith also talked about the liabilities. I think the liabilities that are the most important to society are essentially real in nature and quite long term. I do not think that nominally fixed investments can ever be suitable for these. I work in the non-life area these days and the big question that fascinates me at the moment is what to do about Periodic Payment Orders. These are similar in nature to fully index-linked pensions to 15-year olds. The courts are handing them out in cases where people need looking after for the rest of their life, say after a motor accident. This is a matter which needs much thinking through and I know that it will be on the agenda at the forthcoming GIRO Conference.
The one important question is what is the appropriate asset to hold for real liabilities with a term of up to 90 years or more? These liabilities will ultimately be a huge proportion of the balance sheet of some non-life insurers and will totally dominate them, if not swamp and kill them, if and when they go into run-off. Another question is how much of these real assets do you have to hold if you want to have any confidence of remaining solvent over the decades ahead? The question of how you account for them is not quite the same thing. Accounting does not change reality, it just changes behaviour.
The Chairman: Does the working party have any particular points that they want to pick up?
Mr Gore: In relation to the perfect storm, and what insurers want to take credit for in valuing liabilities and what is allowed for or seen in asset prices, I tried to draw out that there is clearly a natural tension between what a regulator wants to see and what underpins annuity writers’ or insurers’ business plans. I think that there needs to be a balance for a functional industry. If the capital charges are too excessive, people do not want to write the business and people will not want to buy those products. That is my view on the solvency aspect and its use.
Mr Dowthwaite: I would just make the point that for a matched annuity portfolio fundamentally, the business is not exposed to spread movements. It is exposed to default movements. We saw some very extreme movements in spreads during the financial crisis, some of which may have influenced by liquidity movements, some of which may have been influenced by investors fluctuating views of risk. The fundamental question raised in making allowance on an insurance balance sheet is whether it is right to look at the risk those insurers are running or at something that is implied by a spread movement?
The Chairman: It is weird, because you are exposed to spread risk if the regulator tells you that you are. In other words, under the old version of Solvency II, you would have been exposed to spread risk because you would have become a forced seller as spreads widened. But if the regulator says you are not exposed to spread risk, you are not. It is a circular question.
Mr D. Brooks, F.I.A.: A question for Mr van Loon and Professor Cairns. Did you do any comparison between the results of your model and those from some of the other models that are used by industry currently to measure the liquidity premium, for example, the Merton model? Do you have any views on the strengths and the weaknesses of those models?
Prof Cairns: The short answer is that we have not done any direct comparisons. There are obviously a number of different methods. We have had some discussions over what alternative might be used. In its favour, our approach has the benefit of objectivity. Some other methods have some subjective inputs. The structural Merton type of approach, for example, which I think is mentioned as one of the approaches for measuring liquidity premium, requires subjective estimates of the expected returns on equities. There are other approaches such as looking at CDSs where there is a limited number of bonds where there are CDSs. CDSs themselves were touted as being very much more liquid. Then the credit crunch came along and they were not as liquid as they used to be. You then really are measuring relative liquidity premiums rather than absolute values.
Overall, although we have not done the particular calculations, all of the methods have their pros and cons. I would not say that there is any method that is philosophically obviously better than any other.
The Chairman: I think Mr Smith earlier drew out a very important advantage of your method, which is that the answer is not a residual. A lot of the other methods suffer from the problem that the answer is the residual, which is a polite word for the error term, which is a strange thing to have as the answer.
Mr P. O. J. Kelliher, F.I.A.: Just a few observations on the whole liquidity premium debate. I think the first thing any kind of regulatory framework should have as its principal is the Hippocratic Oath, first do no harm. Basically, this is why the liquidity premium is quite important. Without the liquidity premium I think regulatory rules can have a very destabilising effect in a financial crisis such as the one we saw around 2007–2009. There was a paper on systemic risk (Besar et al., Reference Besar, Booth, Chan, Milne and Pickles2011), which made this point about not creating these pro-cyclical effects. If we examine what happened during the financial crisis, effectively the pricing mechanism broke down. The idea of having prices that were rational went out of the window. You had prime Residential Mortgage Backed Securities trading at 50 cents in the dollar. To put that in context, that price would indicate a 70% default rate and a 70% loss given that default rate, which obviously has no bearing to reality. One of the problems was that you had fire sales of assets, which was also the reason why you had breakdowns in diversification as well. People were trying to move out of, let us say, Asset-Backed Securities. They found that they could not sell them. Then they had to dump all the better quality stuff, plain vanilla corporate bonds and the like. That is one of the reasons why you had that loss of diversification benefit, not necessarily to do with any underlying credit risk, it is just that everybody needed to sell.
If you did not have the liquidity premium, we would have these huge falls in values of assets without any corresponding fall in the value of liabilities. This would have had led to further distressed sales, a bit like life insurers ditching equities in the early noughties. The liquidity premium leaned against where the market was going. That is why I believe it is quite important.
Regarding the liabilities point of view, I take Mr Smith’s point. We are considering the liquidity of assets and the liquidity of liabilities, which are two very different things. We should also remember how illiquid insurer liabilities are relative to bank liabilities. That is one of the key issues. Whatever optionality you might have in life insurance liabilities, it pales into insignificance when you look at the short-term deposit book of banks.
A final question for Professor Cairns and Mr van Loon: Since the crisis, I have read that market maker inventories have fallen by about 75%–80% in the US. I suspect a similar fall has happened here. With Barclays laying people off and pulling back from the Fixed Income Clearing Corporation, I think we are going to see further reductions in investment bank inventories. Are you seeing any of that in bid-offer spread data?
Mr P. van Loon: I think in terms of the bid-ask spread, we do see that they have not returned to pre-2007 levels. They shot up really quickly round mid-2007 and the ending of the crisis is not easy to pinpoint. The spread has been declining for years and years, but it has not returned to normal levels, at least in the data that I can see. I do not know to what extent that answers your question. It seems that the recovery is slow but still on its way in that sense.
The Chairman: That leads to a point someone made earlier. I work for a bank. Banks’ capacity to take down large blocks of credit is definitely massively diminished. Pre-credit crunch, if you asked me to sell £1 billion of credit in an afternoon, my colleagues would have done that. Now £100 million would be a large trade. The capacity for the really big trades has diminished quite a lot. That may have some interesting implications for the model and that may come out in your regression analysis.
Mr J. N. Hayes, F.I.A.: Insurers are increasingly investing in private markets, assets or infrastructure, and real estate to back annuity-type liabilities. To what extent does the matching adjustment, volatility adjustment, apply to those assets? Can you speak a bit about the calibration of the metrics for those assets?
Mr Dowthwaite: As long as the asset fulfils the matching asset criteria, particularly around cash flows, then an asset should qualify for inclusion in a matching asset portfolio. We are still in slightly unknown territory in terms of how the calibration is going to work for those assets. EIOPA will be providing the calibrations, whether that will be applied to those sorts of assets is not yet clear.
The Chairman: One practical constraint is a lot of back books, in terms of things that banks have on their balance sheets that might more naturally sit on insurers’ balance sheets, are not going to satisfy that test. As part of the budget and the £25 billion that a number of firms agreed to invest in infrastructure at the Chancellor’s request, before he destroyed the annuity market, there was talk of the life industry telling people who wanted to borrow money the sort of format in which they needed to borrow the money in order that the borrowing would be matching adjustment compatible. So, I think the answer is yes, but only in the right format.
I think we will now end the discussion. I thank you, the audience, and may we also thank all the authors.
