Published with kind permission of www.insuranceerm.com. Original article July 2017: https://www.insuranceerm.com/analysis/ma-optimisation-the-experience-at-lv.html
When artificial intelligence (AI) defeated the world champion of the board game ‘Go’ in 2016 it was seen as a surprise step change in AI technology. One of the less well known surprises was how AI achieved this – the heuristics and strategies developed by the best human players over centuries of human thinking were superseded by a few minutes of machine thinking and powerful mathematics. The machines still played by the rules; they just didn’t play the same way.
OptiMA is a software tool developed to optimise and control a Matching Adjustment (MA) portfolio using mathematical techniques rather than brute force. Its results are surprising and take just minutes to reduce a firms’ BEL by up to 1% whilst simultaneously providing control over the outcome of the PRA’s tests. What also surprised LV=, when they implemented the tool in 2017, was the manner in which OptiMA achieved this.
The matching adjustment is an addition to the risk-free rate used to value annuity business under Solvency II, estimated to be worth around £59bn1 to insurance firms in the UK. It is based primarily on the credit spreads in excess of default and downgrade costs of credit assets held to back annuity liabilities. To this extent it is similar to the illiquidity premium under Solvency I, but with some key differences in the calculation detail and portfolio management / eligibility requirements. Two of the more material changes, which introduce fresh challenges – and opportunities – for firms, are as follows:
1. The ring-fencing of assets within the MA fund and hypothecation of these assets between three components, A, B and C (of which only component A is used to calculate the MA). Tight regulations around eligibility criteria and the ability to release assets from the ringfenced fund also create additional operational considerations.
2. Firms are required to demonstrate (on a monthly basis) that their MA portfolio satisfies a number of PRA tests. Should the portfolio fail to meet the PRA tests, the firm has a period of two months to ensure the tests are met or risk losing the MA for two years. With an industry value estimated at £59bn, this represents a material operational risk.
Whilst the principle of the MA has many similarities to the Illiquidity Premium under Solvency I, the value itself is only calculated using assets hypothecated to component A. How the assets are hypothecated within the ring-fenced fund is entirely up to the firm (subject to passing the PRA tests), and can significantly affect the size of the MA.
Maximising the MA for the given ring-fenced assets presents an optimisation opportunity. But with typical investment portfolios containing hundreds of bonds and other assets, manual optimisation can be challenging and time-consuming with no guarantee of an optimal result. Furthermore, it must be done in a manner that ensures the fund continues to pass the PRA tests. To mitigate the risk of failing a test and potentially losing the benefit of MA for two years, firms may hold a significant buffer against the PRA test tolerances. This buffer, in turn, will limit the optimisation of the MA.
For further background on the portfolio management requirements attached to the Matching Adjustment and the different components of the portfolio, see https://www.insuranceerm.com/analysis/optimising-hypothecation-in-matching-adjustmentportfolios.html.
Optimising the MA immediately reduces the Solvency II Best Estimate Liabilities (BEL).
Inforce: This improves the surplus before any allowance for the Transitional Measures for Technical Provisions (TMTP) and reduces firms’ reliance on TMTP afterwards. [Note TMTP is a transitional measure between the Solvency I and Solvency II technical provisions that effectively spreads the difference between these over 16 years to ease companies into the new Solvency II regime.] This is important for two main reasons:
1. Solvency both with and without TMTP must be published. In many cases investment analysts have expressed more of an interest in the Solvency II figures without TMTP
2. As TMTP runs off this causes a strain on the Solvency II balance sheet if liabilities extend beyond the 16 year run-off period. If a firm’s solvency is dependent upon TMTP they must demonstrate annually that this run-off can be managed via a phasing-in plan.
New business; For firms who price or monitor new business profit on a Solvency II basis, increasing the level of MA either increases the level of surplus recognisable on day one or increases the competitiveness of their pricing. Both of these are likely to improve various new business metrics.
A reduced BEL, on both in-force and new business, can result in additional surplus that is immediately available for distribution as a dividend should the firm desire to do this.
As well as financial benefits, the results of each of the PRA tests can be carefully controlled and capital requirements potentially reduced through the hypothecation process.
Improved control; Given the risks attached to failing the PRA tests, firms are likely to:
1. Operate their MA portfolios with an appropriate buffer against the PRA test statistic tolerances, in order to reduce the risk of breaches from fluctuations (caused by simple evolution of the portfolio, or their manual optimisation process).
2. Hold a larger buffer of assets in the ring-fenced fund than strictly necessary; this provides more scope to amend the manual hypothecation and reduce the test scores.
Both of these behaviours create inefficiencies on balance sheets; additional buffers over test statistics limit the optimisation of the MA, and additional assets in the ring-fenced fund mean liquidity is tied up and cannot be used elsewhere.
With a mathematical optimisation, a surprising degree of control can be placed on the outcome of the test statistics simply through changes in hypothecation. This gives firms scope to further optimise capital and liquidity. Furthermore, limiting the test scores in OptiMA has surprisingly modest impacts on the level of optimisation achieved compared to a manual process.
Reduced Capital Requirements; OptiMA currently maximises the MA which minimises the BEL. The tool can also be used to optimise capital requirements. This can be done for the standard formula calculations as well as different Internal Models. The task of optimising the BEL and the capital position have some conflicting objectives, in that some assets that might be better in component A from a BEL perspective are sub-optimal from a capital perspective. To allow for both of these objectives a firm might seek to optimise its Solvency ratio rather than just BEL or just capital.
Playing by our rules, not our strategies; When carrying out the hypothecation process for the MA calculation many firms have used similar heuristics to guide which assets should be put in component A, B or C. For example:
• Cash is better allocated to components B or C and component A is better for higher yielding corporate bonds which will pass through to the MA value
• A higher yielding corporate bond is always preferable to a gilt in component A
• The higher the credit spread net of fundamental spread, the better the asset is for component A
Such heuristics are a good guide to ensuring the MA is maximised when carrying out the hypothecation by hand. However, they pay little consideration to the PRA tests which are also dependent on cash flow term. In solving for a mathematically optimal solution, OptiMA ignores all strategies that the manual process has developed with an end result that is surprisingly at odds with many of them. For example, in many cases the allocation of cash to component A improves the PRA test 1 statistic, giving more scope for higher yielding longer term bonds to also be included in component A, increasing the MA overall.
Years to spare each month? To optimise the MA it is possible to consider all combinations of existing assets (and possibly all proportions of those assets) in components A, B and C until a maximum value for the MA is achieved. This would be a “brute force” approach to MA optimisation. However, considering all combinations of assets for a portfolio of say 400 bonds, with a calculation time of just 1 second for each MA calculation, would take many years and is clearly not practical in a monthly process! This means the algorithm used to optimise the MA is crucial and must be carefully designed to ensure it optimises the MA in a practical time limit.
LV= have an MA portfolio of approximately £2.5bn, backed by a portfolio of corporate bonds, gilts and swaps. In 2016 LV= purchased OptiMA and have been using it in their monthly MA calculation process. Previously LV= were applying a manual hypothecation process using heuristics, in keeping with the rest of the industry. The shift to OptiMA has demonstrated:
• A reduction in BEL of 0.7%, reducing TMTP reliance and improving new business metrics
• Control and confidence over the PRA test statistics, reducing potential operational risk considerations and improving liquidity
• A clear process for carrying out the hypothecation which has a short run time, reducing operational overheads on a monthly basis
• Greater flexibility and understanding of the management of the MA portfolio.
• The ability to assess the impact of new assets on the MA, to see whether new credit assets are a suitable addition to the MA portfolio
Ed Rayson, Head of ALM at LV=; “The driver for us was improvfed efficiency; Solvency II has brought with it many new challenges and LV=, like other firms, are actively seeking to optimise our balance sheet under Solvency II. OptiMA has been a cost-effective way of reducing prudent margins and improving the efficiency of both our solvency and liquidity positions. It has also delivered improved control and reduced the operational risk and operational overheads attached to running our MA portfolio”.
OptiMA was developed by Sharpe Actuarial Ltd. For further information on the items in this article, or OptiMA itself, contact james@sharpeactuarial.co.uk.
