In this Guest post , Jackie Grant and Mark Taylor-Batty wonder why estimates of USS future service costs go up and down in line with the gilt yield.
Jackie and Mark have expressed thanks to Con Keating for his expertise and time in providing additional commentary, checking the analysis and editing a longer piece here.
The future cost of our pensions, according to USS monitoring, is remarkably almost entirely predicted by that month’s value of the long-dated gilt yield (see figure 3, below, for detail).
This is a remarkable and alarming finding; pensions do not vary with the vagaries of government bond markets nor do their costs of provision. Conventional gilt yields for 15 and 20 year maturities have today varied by more than 3% in price in a day (e.g from 4.227% to 4.344% yesterday). To import this volatility into the cost of pension provision and valuation is simply senseless.
USS has just published their latest valuation monitoring update. These USS monitoring reports appear every three months, and they outline how the value of the pension scheme has changed since the 2020 valuation ‘snapshot’ date of 31 March 2020. That valuation, we all remember, led to huge cuts to pension benefits.
It was argued at the time that the effects of the covid linked panic in markets rendered that valuation unreliable. The statistics now show that it was just that, unreliable, and that the cuts should not have been made.
The date of the latest data is 31 March 2023. This date is significant because this is the ‘snapshot’ date of the 2023 valuation. What this data now shows, as the USS Trustee has repeatedly indicated, is that the likely outcome of the 2023 valuation will be that the contribution costs of fully restored benefits would be around 26% or less of salary. Under the 2020 valuation now in force, the cost of the current benefits, in their reduced form since April 2022, is 25.2% (on top of which 6.2% is paid toward the ‘deficit’ defined in 2020, making a total current cost of 31.4% of salary).
In fact, market movements in the period since this was written reinforce the affordability of the reinstatement of benefits.
Analysing volatility over three years
Looking at the latest USS data, the most immediate feature that stands out is the shocking amount of volatility, which is to say how much the data has fluctuated over time. For example, in March 2020 USS recorded a theoretical deficit of £14.1bn, while three years later USS is now recording a £7.6bn surplus. A £21.7bn movement in surplus/deficit over three years, against assets of £70–80bn is an alarming level of volatility and, we are often told, presents huge challenges for Universities’ financial planning. This has led to cycles of cuts and industrial action from understandably furious pension scheme members.
Here, using USS published data and GCSE-level maths, we seek to unpick the patterns in this volatility. This is presented below by plotting the sets of USS data against the gilt yield over that period. The gilt yield is the return earned on holding gilts (government bonds) given as an annual percentage of that holding, akin to an interest rate. The gilt yield used here is what is known as the ‘single equivalent gilt yield’. This is calculated by USS as the average of the 20–30 year UK gilt yield curve, where this average is weighted by the pensions liabilities that come due at these different points in time. This is approximately the market value of the 20 year UK gilt yield.
To be clear, in terms of our ongoing negotiations over USS, we are absolutely committed to prioritising the USS accelerated 2023 timeline to fully restore benefits on 1 April 2024, for the likely cost of 26% or less, as given by the USS Trustee. To achieve this implies that we have to operate within the valuation methodology as it is currently being applied by USS, though with some important and welcome changes to the assumptions and to the implementation of the self-sufficiency definition.
We will continue on this path as long as this approach is expected to fully restore benefits for a sensible price in around 10 months time. However, we are utterly convinced that the USS valuation methodology absolutely needs to change for future valuations. It needs to be more stable and much less sensitive to the gilt yield. USS needs to make public, and be crystal clear, how it is calculating the costs in valuations and monitoring.
We have plotted the following the data sets against this USS gilt yield:
(i) Future Service Costs (FSCs). This is the amount USS asks us to pay towards our future pensions in our own contributions from salary, and our employer contributions.
(ii) Technical Provisions Liabilities (TP liabilities). This is the amount that USS calculate (on a prudent basis) is needed in order to pay all pensions that have been promised so far, as they fall due. This calculation also provides the Surplus or Deficit defined as: Assets minus Technical Provisions Liabilities.
It is notable that, when plotted against the gilt yield, the data sets all demonstrate a very high correlation.
Astonishingly, the highest predictability is the Future Service Costs, with coefficient of determination R² of 0.99 (a very high correlation) [see footnote]. What this logically implies is that the cost of our pension can be entirely predicted by the gilt yield.
This of course makes little sense for a scheme which doesn’t invest predominantly in gilts.
The FSC are costs for all staff members currently paying into pensions, and it therefore excludes deferred or retired members. This part of the scheme should be (and is indeed) invested for the long-term in a portfolio heavily weighted by return seeking assets. This cost should not be entirely determined by the gilt yield calculated on any single given day.
Below is an investigation of the volatility in the data sets outlined above.
(i) Future service costs are almost perfectly predicted by the gilt yield
We first look at the USS graph of FSC over time (figure 1) and also their graph of nominal gilt yields over time (figure 2). We notice an inverted relationship, FSCs go down when gilt yields go up.
Taking the values of FSC and gilt yield, as a pair for each date, we plot FSC on vertical axis and gilt yield on horizontal axis to get figure 3 below. This shows that all the points lie almost perfectly on a straight line with a negative slope of 3.6% FSC per percentage change in gilt yield. The coefficient of determination is R²=0.99. The gilt yield alone is almost perfectly predicting the cost of future pensions, meaning FSC costs are determined by the simplest one parameter model possible: a linear relationship.
Figure 3: Plot of USS FSC data for cost of pensions with post-April 2022 benefit cuts against USS gilt yield data. The coefficient of determination, R²=99%, is the highest correlation of any two sets of USS data. It suggests USS FSCs can be entirely predicted by the gilt yield. The dependency is also significant, ranging from FSC of 17% to 29% over a 3.5% change in the gilt yield.
Cut and Uncut Benefits
USS monitoring, as given in the Financial Monitoring Plan reports, which uses the USS Integrated Risk Management scheme, focuses on the monitoring of the scheme with post-April 2022 benefits imposed.
UCU, supported by UUK, have repeatedly requested costs for fully restored benefits, at the pre-April 2022 level, to be included in monitoring and so we now have some data for what it would now cost to fully restore benefits to pre-2022 levels.
We understand that USS do not calculate the costs of restored benefits in the same way as they do current (cut) benefits. Instead, they are estimated from the data sets of costs under post-April 2022 benefits. Figure 4 shows the data available for pre-April 2022 benefits in the same form as figure 3. The same feature is clearly present albeit with lower predictability (R² of 0.92), and the rate of change of costs with varying gilt yield is almost double, as the negative slope is 6.7 FSC% per percentage change in gilt yield.
This increase in the sensitivity of the Future Service Cost to the gilt yield as the level of projected benefits increases is entirely intuitive. Its scale is determined by the increase in the effective duration of the projected liabilities resulting from the additional liabilities, and is the product of their magnitude and timing.
Figure 4: Plot of USS FSC data for cost of pensions with pre-April 2022 benefit cuts against USS gilt yield data. The coefficient of determination is R²=92%. The dependency is also significant, ranging from FSC of 21% to 43% over 3.5% change in the gilt yield.
(ii) Technical Provisions Liabilities is highly predicted by the gilt yield
Technical Provision Liabilities are calculated from the projected cashflows (which are roughly stable over valuations and monitoring) but discounted to a single value at the valuation date using the TP discount rate (which is not stable over time). Again, we first consider the USS graph, included in their FMP reports, of TP Liabilities over time since March 2020.
Liabilities are in effect the present values of all the pensions that have already been earned, and are to be paid in the future as they fall due, though of course factors that affect their ultimate size and duration (such as career profiles, demographics, inflation and longevity) mean that some estimation is involved.
Figure 5: USS graph of TP liabilities (referred to as ‘Actual liabilities’) for post-April 2022 benefit cuts from March 2020 to March 2023.
If we compare figure 5 of TP liabilities to figure 2 of gilt yield, again we see clearly TP liabilities go down as gilt yield goes up. Again, taking pairs of points at each date and plotting TP liabilities against gilt yield we see a high dependence. This is shown in figure 6 below.
The predictability is lower (R² of 0.92) but there is still a clear trend with good predictability. The TP liabilities change by around £25bn for a 3.0% change in the gilt yield. This is the source of the volatility of the surplus and deficit. This is critically important for two reasons: the large deficit (when gilt yields were low in 2020) gave a high monitoring value of the TP liabilities against the assets, producing a huge deficit, requiring the sector to pay over half a billion in deficit payments per year rather than leave our pensions uncut. But secondly: the market value of the USS assets are, of course, by nature volatile. Subtracting two large numbers of similar size, both with a large volatility gives a much smaller number (the surplus or deficit) but with a similar level of volatility of the two original large numbers, often more so.
Figure 6: Plot of USS TP liabilities data with post-April 2022 benefit cuts against USS gilt yield data. The coefficient of determination, R²=92%. The dependency is also significant, ranging from FSC of £66bn to £95bn over a 3.5% change in the gilt yield.
In a later blog, we will consider in detail USS’s other metric of scheme funding, self-sufficiency. At risk of spoiling that, we will note here that it also shows a very similar relation to gilt yields.
— — — — — — — — — — — — — — — — — — — — — — — — — — —
- The R-squared (R²) statistic (the coefficient of determination) measures how well observed outcomes are predicted by a model, so the R² measure is an indication of how much variation the model explains, with values between 0 (for no predictability) to 1 or 100% (perfect predictability). See more here. In the case in this blog, the Future Service Costs are the observed outcomes that are predicted by a linear model with the single input of the long-dated gilt yield. A linear model plots an outcome variable (here, the FCS) against a potential explanatory variable (here, the long-dated gilt yield) to test their relationship, and (as shown in figure 3) this relationship has almost perfect predictability of R² as 0.99 or 99%.
- The numbers: You can download an excel file containing the data explored here.
When the gilts roll over and buyers desert DB will get stuffed full of new toxic gilts
Do you think that LDI 2 could happen?
“ Do you think that LDI 2 could happen?”
It’s a ‘when’, not if. The ‘collateral’ is the protection for the banks, not the scheme. Hence, the fact they are looking for more collateral not less, does that infer there is more or less risk.
The judgement the LDI pedlars are making is that by the time this all collapses in on itself, it will be far enough down the line that they will have spent the bonuses and moved their own assets elsewhere, leaving the next generation to suffer the consequences and pick up the mess.
On Wednesday May 25th I will be talking about the ongoing aspects of LDI and the gilt market turmoil at an open university event –
The Open University event is on Thursday, 25 May.
Thank you Byron I had put it in my diary on the wrong day – I have been known to do that before. It is on Thursday.
Same link – hope you booked your train tickets for Thursday!