22 Jan 2021 HMT Review of Solvency II: My submission on reform of the MA

I have submitted some responses to the HM Treasury Review of Solvency II: Call for Evidence in the areas of Risk Margin and Matching Adjustment. My Risk Margin submission is simply a restatement of the arguments I presented in my October 2019 article and which were subsequently published in the British Actuarial Journal (Volume 25, 2020). My Matching Adjustment submission is set out below.

3.14 What changes, if any, should be made to the eligibility of assets for the matching adjustment?

In the approach outlined below in my response to 3.15, there would be no need for any eligibility test for assets backing Matching Adjustment liabilities. These assets would simply be subject to the same regulatory framework that applies to non-Matching Adjustment assets.

3.15 What changes, if any, should be made to the calculation of the matching adjustment?

The technical framework of Solvency II is based on the cost of transferring insurance liabilities to a third party of good financial standing. The solvency measure is based on the principle that firms should have assets of sufficient market value to fund the cost of this transfer, even in highly improbable and stressed conditions. For a liquid (i.e. non-Matching Adjustment) liability with a fixed set of cashflows, this cost of transfer is assessed as the discounted present value of those cashflows, where the discount rate is based on the published Solvency II risk-free yield curve. The assets that are used to derive this risk-free yield curve are generally very liquid. This discounted present value represents an estimate of the cost of manufacturing or replicating the liability cashflows; this provides a natural estimate of the cost of transferring the liabilities to a third-party.

How does the Matching Adjustment (MA) logically fit into this ‘cost of transfer’ framework? The MA recognises that some insurance liabilities may be illiquid and this characteristic may reduce the cost of transferring the liability to a third-party. It may reduce the cost of transfer because it widens the set of assets that may be used to replicate the liability cashflows. Specifically, the illiquidity of the liability permits illiquid risk-free assets to be used in the manufacturing or replication of the liability’s fixed cashflows. So, whilst the cost of transfer for the liquid liability is estimated as the cost of the liquid risk-free assets that replicate the liability cashflows, the cost of transfer for the illiquid liability may be estimated as the cost of the illiquid risk-free assets that replicate the liability cashflows (if, indeed, those illiquid risk-free assets cost less than their liquid equivalents).

From this perspective, the technical provisions for liabilities that are illiquid should be assessed using the illiquid risk-free rate as the liability discount rate (and this rate may or may not be higher than the liquid risk-free rate, depending on whether an illiquidity premium in risk-free asset yields can be observed at the time of valuation).

Estimating the illiquid risk-free rate

The illiquid risk-free yield curve may be difficult to observe as illiquid asset prices are, by a natural consequence of their definition, generally difficult to observe. A comparison of government debt instruments of varying degrees of liquidity (for example, strips and gilts) may provide some insight into the market reward for bearing illiquidity in risk-free assets. However, the degree of illiquidity found in the less liquid of these government debt instruments is unlikely to correspond to the degree of illiquidity found in illiquid insurance liabilities such as annuities. Such a price comparison may therefore under-state the relevant level of illiquidity premium available in risk-free assets with the degree of illiquidity found in insurance liabilities. It is difficult to observe the market prices of assets with this degree of illiquidity, as, by definition, such assets do not frequently trade.

In some cases, transaction prices for highly illiquid assets can be observed at their time of origination (these prices may only be ‘semi-observable’ in the sense that they are not regularly published but may be disclosed by market participants; this is similar to some other market price data used in Solvency II liability valuations such as long-dated OTC equity option prices).

These illiquid assets, however, will generally not be risk-free. It may therefore be useful to estimate the relevant illiquid risk-free rate by extrapolating the observed prices or interest rates of low credit risk assets where such assets can be found with a relevant degree of illiquidity. For example, the observed pattern of origination equity release mortgage rates relative to their Loan-to-Value (LTV) may permit a reasonable estimation of the 0% LTV mortgage rate that corresponds to the illiquid risk-free rate (before allowance for the costs associated with the acquisition and administration of the mortgage)[1].

The origination terms of other illiquid loans and mortgages may provide similar data for this purpose. A key point to note in such analyses is that the logic of adjusting replication costs to allow for illiquidity implies it is the illiquidity premium generated by illiquid risk-free assets that should be the measure of interest. Estimates of the illiquidity premia offered by risky illiquid assets are not relevant to the cost of replicating the fixed cashflows of an illiquid liability (except insofar as they are used in the estimation of the illiquidity premia available from investing in illiquid risk-free assets).

A valuation adjustment for illiquidity, not matching

It may be noted that the above account of the rationale for the Matching Adjustment is based on the potential valuation impact of liability illiquidity and makes no mention of cashflow matching.

Discussion of the MA often refers to the idea that insurance firms that have matched asset and liability cashflows are not sensitive to short-term asset price volatility (for example, Section 3.4 of this Call for Evidence). This is an argument for using a measure of risk that excludes a component of short-term asset volatility in the 1-year VaR capital assessment (i.e. the component that does not impact on the expected cashflows produced by the matching strategy). The use of such a measure would have a natural implication for the assessment of capital requirements.

But it is unclear why this perspective on asset risk should have any effect on the valuation of liabilities, if the liability valuation is to be based on the cost of transfer to a third party. It is perfectly true that variations in, say, credit risk premia, do not impact on the expected cashflows that will be produced by an in-force buy-and-hold investment strategy. But the near-term ability to fund the cost of transfer of the liabilities to a third-party does depend on short-term asset price volatility (and is affected by increases in risk premia that the asset portfolio value is exposed to).

If a prudential system that is not focused on the ability to fund the cost of near-term transfer of liabilities to a third party is deemed preferable to Solvency II’s 1-year VaR framework, then the logical implication of this would be to fully depart from the 1-year VaR framework and use an explicitly longer-term risk horizon in the capital calculation. This surely must be preferable to the very complicated system that arises from starting with a 1-yr VaR capital measure and then attempting to ‘switch off’ a component of short-term volatility because the 1-year horizon is judged to be the wrong horizon for assessing insurance solvency capital requirements.

But if the use of a 1-year VaR framework is motivated by an intention to ensure that the near-term ability to fund the market cost of transfer of the liabilities is robust, then the variability of asset risk premia is a relevant risk factor in capital assessment. And, in any case, the degree of cashflow matching generated by the investment strategy is not relevant to the valuation of fixed liabilities when valuation refers to the cost of transfer of the liabilities.

The risk profile of the insurance firm’s asset strategy has no bearing on the current cost of transferring the liabilities to a third-party. The current capital requirement is naturally impacted by the risks in the asset strategy, and the (‘pre-MA’) solvency capital requirement framework can naturally capture the risk reduction benefits that are generated by a cashflow matching strategy. In short, there is nothing special about cashflow matching that motivates an alteration to the 1-year VaR solvency framework.

There is, however, a rationale for liability illiquidity impacting on liability valuation (if risk-free illiquidity premia can be observed in or estimated from market prices at the time of valuation). Seen from this perspective, the Matching Adjustment might be more accurately named the Illiquidity Adjustment.

Under this approach, the solvency capital requirements of assets backing MA-eligible illiquid liabilities would only be reduced by the MA to the extent that variation in the illiquidity premium that applies to the MA discount rate is a relevant risk factor in the asset solvency capital requirement. If a firm chose to back illiquid MA-qualifying liabilities with liquid assets, the firm’s net asset-liability position would be exposed to a fall in the (risk-free) illiquidity premium, and the capital requirement should recognise that exposure.

A radically simplified Matching Adjustment (the Illiquidity Adjustment)

The above argument that the MA discount rate should simply be the illiquid risk-free rate has some significant ramifications for the implementation of the Matching Adjustment:

1.      The MA discount rate would be an estimate of the risk-free rate that is attainable from risk-free assets that have the same degree of illiquidity as the MA liabilities. This would be derived from market prices observed at the valuation date.

2.      The MA discount rate would therefore not be a function of the asset strategy that the firm chooses to pursue for the assets backing the MA liabilities. The MA discount rate would not vary with the asset allocation or credit risk of a firm’s MA asset portfolio. In this setting, much of the complicated apparatus associated with the existing MA system is rendered redundant. In particular:

a.      There would be no need to estimate an MA fundamental spread for any asset (each asset’s fundamental spread would effectively be defined as the difference between its gross redemption yield and the illiquid risk-free yield of the same duration);

b.      There would be no need for additional regulatory diagnostic tests such as the Effective Value Test that are currently applied to some MA-eligible asset classes (equity release mortgages and their securitisations), as the MA benefit would no longer be a function of the asset strategy;

c.      There would be no need for firms to demonstrate that their asset strategies deliver a particular level of liability cashflow matching in order to qualify for the MA.

3.      All firms would therefore use the same MA discount rate yield curve. It would perhaps be most natural for this yield curve to be estimated and prescribed on a regular basis by the PRA or a body that the PRA delegates this technical task to (this would be similar to the way EIOPA currently specifies Solvency II risk-free yield curves and MA fundamental spreads).

4.      There would be no eligibility requirement for assets backing MA liabilities. MA assets would simply be subject to the same regulatory treatment as non-MA assets.

The above approach would deliver a radical simplification of the current (very complicated) Matching Adjustment regime. It would also remove the features that currently have the potential to result in inadequate (net) prudential capital requirements for MA-eligible liabilities.

3.16 What changes, if any, should be made to the matching adjustment approval process?

With the MA approach outlined in my response to 3.15 above, the MA approval process would only consider whether the liabilities meet the specified criteria to be regarded as sufficiently illiquid for the purposes of the MA. There would be no need for specific MA approval of any aspect of the strategy for the assets backing the MA liabilities (including asset eligibility and adequacy of cashflow matching). (For the avoidance of doubt, these assets would be subject to the existing regulatory requirements that apply to non-MA assets with respect to solvency capital, the prudent person principle and asset reporting).

3.19 What are the costs and benefits of any changes proposed in response to the above questions? How should any risks to the safety and soundness of insurance and / or to policyholder protection be mitigated?

Firms’ MA regulatory compliance and MA asset ‘optimisation’ activities must currently represent a very material cost that is ultimately likely to be borne, at least in part, by policyholders. The reforms proposed above would radically simplify the MA regulatory regime and would reduce the costs associated with it commensurately. I believe these proposed reforms are also very unlikely to reduce policyholder protection relative to the current MA system.

[1] Please see my December 2020 article for some further discussion of this approach.