Test 1 de-risking is dead, long live de-risking!
What’s wrong with USS’s approach to the valuation
[UPDATE 29 Oct: Here I provide further documentation, via reference to statements and a graph in a May 2018 video by their Chief Risk Officer, that USS has made a large and demonstrable mistake regarding Test 1.]
1. The death of the Test 1 case for de-risking
1.1. Certificate of evidence of death
USS claims that Test 1 — which measures and sets limits to (1) “the difference between the scheme’s technical provisions [liabilities] and the self-sufficiency approach” (Sept 2017, p. 41)— serves as a measure of reliance on the employer covenant, which is defined as (2) “the difference between the assets held by the scheme to fund the promised benefits, and those required by a low risk investment portfolio (a funding approach known as self-sufficiency)” (Sept 2017, p. 40; see also pp. 26, 34 and 58).
USS maintains that Test 1 measures reliance because it assumes that the technical provisions liabilities in 20 years’ time will be equal to the assets held by the scheme at that time. (See Sept 2017, p. 41, in the light of Nov 2016, pp. 5–6 and 10–11.)
Sam Marsh has blown this italicised assumption out of the water with his finding, which USS confirmed last Friday, that column 1 (assets) of row 1 in Figure 1 below is much larger than column 2 (technical provisions liabilities) of row 1:
As I show, moreover, in Section 2.4 of this linked blog post plus the blogs to which I link in that section, the underlying rationale for Test 1 requires a measure of reliance in terms of the difference between columns 1 (assets) and 3. This is because what’s relevant is whether the scheme has sufficient assets in year 20 to purchase a self-sufficiency portfolio by year 40 via supplementation of its assets with a 7% rise in employer contributions from year 20 to year 40. (See Feb 2017, p. 8.)
USS’s in my words “large and demonstrable mistake” was to measure reliance on the covenant in terms of the gap between columns 2 (liabilities) and 3 above rather than 1 (assets) and 3.
Question: Would the scheme pass Test 1 if we carried on accruing benefits at their existing level via contributions at their current 26% of salary, while remaining invested in the current reference portfolio, which is weighted towards growth assets?
USS’s answer: No. This is because the gap between columns 2 and 3 of row 1 exceeds the current £10bn maximum permitted reliance on the employer covenant. We must, therefore, de-risk the assets, and move down to row 2 or row 3 of Figure 1.
The right answer: Yes. This is because, on a proper understanding of USS’s underlying rationale for Test 1, what matters is the gap between columns 1 and 3 of row 1. That gap is within the current permissible £10bn limit.
In arriving at the wrong answer, USS made a huge and consequential mistake, arising from a misunderstanding of a fundamental aspect of their valuation.
1.2. USS’s response to the above critique
In his response to my blog post entitled “USS’s valuation rests on a large and demonstrable mistake”, USS’s Chief Risk Officer Guy Coughlan writes: “The analysis by Dr Marsh and the blog by Dr Otsuka are both based on the erroneous premise that setting contributions which are adequate on average over 20 years is sufficient to fund the scheme.”
Here Coughlan blurs the distinction between (i) the assumptions that are appropriate for determining the projected level of assets in year 20 for the purposes of measuring the reliance gap and (ii) the future service contributions it is ultimately appropriate to charge in each given year.
It’s one thing to determine what the level of the scheme’s assets will be in 20 years’ time under current contribution rates, benefit, and portfolio, in order to determine the size of the gap to self-sufficiency then. That calculation is made in order to determine whether the scheme would be within affordable distance of self-sufficiency by then if we carried on as at present. If it would be within affordable distance, then there is no Test 1 call for de-risking the portfolio.
We then determine the costs per year of future service, given retention of the current (non-de-risked) portfolio, followed by a further discussion of what sort of smoothing of contributions would be warranted if these costs vary from year to year.
Coughlan fails to distinguish these two separate steps, the different considerations they involve, and the need to carry out the first step before the second one.
Moreover, in continuing to maintain that USS has established that contributions would need to go up by 11.4% in order to maintain current benefits, Coughlan begs the question in favour of his contested approach of measuring reliance as the gap between columns 2 and 3 of row 1. This is because the 11.4% figure is derived from the Figure 1, row 3 November de-risking of the scheme’s assets which presupposes his contested approach to Test 1.
On the correct application of Test 1, which measures the gap between columns 1 and 3, there is no call to de-risk this existing portfolio. The current and future costs of accruing future service needs to be recalculated on the basis of the (prudently adjusted) technical provisions discount rate for the existing portfolio, not the de-risked portfolio. On such a recalculation, the cost of future service in 2018 will not give rise to a 11.4% increase in the current level of contributions in order to continue to provide benefits at their current level. That figure will be lower. Perhaps it will nevertheless be a positive number, in which case we will need to carry out the step two discussion I mention above.
This completes my demonstration that USS’s Test 1 based case for higher contributions rests on a “large and demonstrable mistake”.
2. Long live de-risking!
USS implicitly concedes that their Test 1 based case for higher contributions is flawed. This is implicit both in their acknowledgement that “Dr Marsh’s analysis is not wrong in isolation” and in their failure to attempt to defend their confused appeal to the gap between columns 2 and 3 of row 1 in making their Test 1 based case for de-risking. It it telling that the words ‘Test 1’ appear nowhere either in USS’s 16 October statement or in Coughlan’s more technical note to which their statement links.
USS’s response to Marsh and me largely consists of appealing to different arguments, based on considerations other than Test 1, in favour of raising the level of contributions to preserve existing benefits. This, however, exposes USS’s approach to the valuation to the following critique, by Professor Tom Pike of Imperial College:
But that’s moving the goal posts! If I understand correctly, Test 1 is based on a specific risk of failing to meet the liabilities, so if the expectation value changes, that risk is reduced. If their argument is that lower risk is always better, then Test 1 is just a cover story for their underlying motivation. Test 1 according to Popper would be classified as pseudoscience, whatever its outcome, the conclusion is the same — de-risk. (private correspondence)
USS’s response to Marsh and me can therefore be summarised as follows: “Test 1 de-risking is dead, long live de-risking!”
