TPR LDI Report
A Continuing Commentary
Part Three
-
- Con Keating
-
- Iain Cacher
This is the third of our blogs commenting on TPR’s Report to Work and Pensions Committee on LDI. The previous blogs can be found here:
This blog is concerned principally with TPR’s modelling of the distribution of scheme outcomes.
The data TPR collects is described as:
“46. From the annual DB scheme return, we receive the following data that we use for our
scheme modelling purposes:
o Market value of assets (based on the date of last effective valuation).*
o Technical provisions (based on the date of the last effective valuation).*
o Buyout liabilities (based on the date of the last effective valuation).*
o S.179 (PPF) liabilities (based on the date of the last effective s.179 valuation).
o Summary of the financial assumptions used to calculate the technical provisions for those schemes in surplus on the date of the last effective valuation.
o Summary of mortality assumptions used to calculate the technical provisions on the date of last effective valuation.*
o Summary of the asset allocation (based on scheme’s most recently audited accounts).
o Summary of Asset Backed Contributions (ABCs).
o Liability and asset hedging information using PV01 and IE01 (voluntary).
* For those schemes in deficit on a technical provisions basis at the date of last effective valuation this is the same information as provided through the recovery plan submission.”
While the limitations of the data are outlined. Simply put, the key issue is that the lags in data collection mean that they do not cover the period of interest, 2022-23, in any meaningful way.
“52. Furthermore, it is important to note that:
(i) The data is historical. Schemes have up to 15 months to complete a valuation and need to only undertake a valuation every 3 years. As such, the valuation data held is generally between 1 year and 4 years out of date. This is shown in the following table:
(ii) Similarly, the asset breakdown is based on the last audited accounts over the scheme
return year. Schemes have up to 7 months from the year end to complete their audited accounts. As such, the asset breakdown is generally around 2 years out of date. This is shown in the following table:
[Emphasis added]
(iii) The data is a high-level summary of the results of the valuation i.e. we do not have individual member data that schemes have access to.
(iv) We only have a high-level summary of scheme benefit structure i.e. we do not have access to the Trust deed and rules that determine the actual scheme benefits.
(v) Although we are provided with URLs for Statement of Investment Principles for schemes where there are over 100 members, which may include some of the WPC requested information, on a universe level we do not collect details of the investment funds that schemes’ assets are invested with, nor currently do we have details of schemes’ collateral waterfalls i.e. which assets are intended to be used for LDI or collateral purposes.
[Emphasis added]
Given the significance of pooled LDI funds, this is a material shortcoming. The ONS reports pooled LDI funds held by schemes as having declined in value from £231 billion to £163 billion, to which must be added the £51 billion[1] of recapitalisations contributed by schemes in response to ‘collateral calls’ in the course of 2022. This is a total loss of 48.5% to schemes totalling £119 billion,. This is the largest proportional loss of any asset category.
The Report (section 98, Table 3) shows a range of gilt index returns for the period.
“(vi) In the Scheme Return we ask for data breakdown at a high level for asset classes. A review of the asset information was conducted in conjunction with the PPF in 2021 and following a review of that consultation, the DB scheme return was updated in 2023.
(vii) A summary of the asset allocation recorded in the 2023 scheme return is shown in Appendix 4, alongside the indices we use when adjusting asset values to different dates.
(viii) We estimate the level of leverage that schemes have based on the breakdown of the scheme assets and/or the PV01 data and estimate how this may have changed over time.”
[Emphasis added]
There are 54 mentions of leverage in the Report, but no estimation shown, anywhere, of its magnitude. The ONS survey reports explicit leverage through repo as shown in Table 1 below:
Table 1: Leverage using repo
The overall loss of value between Q4 2021 and Q1 2023, shown in the final column, is £71 billion but actually £87 billion from its high in Q2 2022. Similarly, the degree of change in leverage is only 0.8% overall but a decline of 5.9% from its high. It should also be noted that the majority of the decline occurred after September 2022 and continued into 2023. Bank of England monetary statistics also show further declines in repo outstanding in the second quarter of 2023.
Much of the leverage exposure of schemes was undertaken through the use of interest rate swap derivatives. We show in Table 2, the gross swap exposure, and the change (loss) in this.
Table 2: Gross and net swap exposure (ONS)
These swap exposures are materially larger than the explicit leverage of repo. We would note here that the maximum change within the period was £41 billion. The implicit overall leverage (excluding that within pooled LDI funds) of schemes is shown as Table 3.
Table 3: Overall effective leverage
We see that overall leverage was maximal in the third quarter of 2022 and has continued to decline in the first quarter of 2023. These leverage estimates are consistent with the average leverage seen in our data collections.
The quarterly and aggregate losses for 2022 due to explicit leverage are shown below as Table 4. These losses have been calculated on the basis that the losses experienced were the average of all assets. Elsewhere in these blogs, we have shown the loss due to repo as £51 billion, this was calculated on the basis that the repo proceeds were invested in growth assets, which experienced smaller losses than assets overall.
Table 4: Losses due to leverage
This has implications for the relative performance of net assets relative to liabilities. This may be illustrated by the following.
Suppose we have liabilities of £100 and use £100 pounds of assets to hedge these perfectly, then at end September 2022, there would be £36.2 exposure to market movements (Q3 2022 from Table 3 above). With markets falling, net assets would be expected to decline by more than liabilities. The precise impact in funding ratios is ambiguous depending upon the initial level of funding – losses can be larger than liability declines, but the funding ratio still improves.
We would also caution against the use of PV01 metrics in such analysis when market movements are orders of magnitude larger than that. This is due to convexity effects and so the losses will tend to be far larger than estimated by those methods for movements of the observed order.
(ix) Likewise, we do not hold data on which assets trustees choose to sell to meet any cashflow needs or collateral calls, or which assets they choose to invest in if they have funds to invest.”
In other words, none of TPR’s scheme data covers the year in question, which means that the entire quantitative content of this report is the result of their modelling. To rephrase George Box “All models are wrong, but some are (not) useful.” The confidence expressed by TPR in their modelling seems to us entirely misplaced as the quality of the data is simply not good enough for any of the modelling to generate the types of insight one would hope to see in such a crucial report.
“53. We are content that the data used in the modelling is appropriate to provide results regarding the potential universe impacts in aggregate over 2022 and updated funding positions into 2023. However, as the data has not been verified on a scheme-by-scheme level, the model should not be used to draw conclusions for individual schemes.”
[Emphasis added]
The report commences its narrative on individual schemes with Section 18:
“18. There were however some schemes for which our modelling shows funding deteriorated
over 2022 as can be seen in the following chart:
- The vast majority (87%) of schemes are modelled to have experienced improved funding levels over 2022, with only 13% showing a funding level deterioration on a technical provisions basis. Of those that showed a funding level deterioration, nearly two thirds were either still in surplus at the end of 2022 or had a fall in the size of their deficit over 2022. Therefore, only 5% of the DB scheme universe had both a deterioration in their funding level and either an increase in their existing deficit or a movement from surplus to deficit over 2022.
In early July, we published an empirical analysis of scheme funding at December 2022 – to quote from that, with some emphasis added:
We have been collecting the reported funding ratios of DB schemes since December 2022. Through this analysis we are finding some fairly large discrepancies between reported funding ratios and the widely broadcast narrative of highly significant improvements of those ratios across the sector. One basic figure from our analysis is that the range of funding ratios spans 50% to 161%.
In this note, we touch upon some of the notable points of the collected sample of 350 schemes on which we based that published analysis. Of course, at slightly less than 7% of the universe of private sector funded DB schemes, this sample is not large enough, nor sufficiently assured of being representative, to prove or disprove anything, but it is sufficiently large to raise questions and concerns.
First, there is no scheme in this sample which reported a positive return on assets in 2022. The best three reported results are losses of 3.8%, 4.6%, and 5.1%; the worst performances show losses in excess of 40%. This means that any improvements in funding ratios must have been derived solely from declines in the present value of scheme liabilities.
Second, 32% of our sample saw their funding ratios deteriorate over the year. The median deterioration was 4.1%. There is a pronounced difference in the symmetry of the distribution of improvements/deterioration. The median improvement was 11.6%. The median funding level of our overall sample was 95.4% in December 2021, and this improves to just 102.1% by the end of December 2022.
It is also evident from the sample that schemes which were in deficit in 2021 were far more prone to experience deteriorations in the 2022 funding ratios than schemes that were in surplus, which exhibited a tendency to improve further.
There is a pronounced difference between the 32% which saw their funding ratios decline in this sample and the 13% of PPF modelling. Figure 1 below shows, the funding ratio distributions of our sample at December 31st 2021 and December 31st 2022. The observation that there were both gains and losses is evident from this Figure. For example, the increase in 2022 of schemes funded between 60% and 70% can only be explained by more schemes experiencing declines in scheme funding than from improvements in funding, as the number of schemes in this range in 2022 is greater than the total number of schemes combined in the 50-60% and less than 50% range in 2021.
Figure 1: Distribution of Funding Ratios, 2021 and 2022
On liabilities that empirical analysis also contained:
The PPF reports liabilities having declined by 38.8% over the year to December 2022. For comparison, a 15-year duration discount function would have declined by 34% and a 20-year by 42.7%. Our sample shows very considerable heterogeneity with liability declines from as little as 7% and 8% to as much as 41% and 42%. However, the vast majority of our sample are clustered between 30% and 35%. Of course, changes to inflation and mortality assumptions will affect liability estimates, but these observed differences still appear to be very large by comparison with plausible estimates of those potential effects taken account of.
The Report offers some further insight in TPR’s modelling in the section:
Results Technical Provisions.
As we have commented on much of this previously, we show only the analysis provided there.
“113. As shown in the chart, the biggest contributing factor to the movement in surplus was the change in gilt yields reducing the value of liabilities. This improvement to funding was partly offset by the large overall negative asset returns over the year (due primarily through the fall in bond and gilt prices).”
Our principal difference with TPR is in the magnitude of the returns on assets, this would reduce the increase in surplus shown, £147 billion (£205 – £57), to a deficit of £18 billion, using the difference in asset values reported by ONS. The Report offers a further breakdown of changes in deficits and surpluses but as this is based upon the higher TPR asset estimate, we will not comment further.
It would be most interesting to see the equivalent of this impact analysis for the changes in Buy-Out and Low Dependency bases, as that might offer some insight into otherwise inexplicable changes in prices.
We have been asked why we have greater confidence in the ONS asset value statistics that those of TPR (and PPF) – we address this question in Box 1.
| Box 1
We have collected some 1800 scheme and sponsor reports of assets (ca 900 schemes) as at December 2021 and March 2022, and December 2022 and March 2023. The ONS survey sampling is described in the table below:
We do not know the ONS response rates but believe it to be in the 78% – 88% range. We decided to conduct an analysis based on the data we possessed. The first step of which was to divide our sample according to the ONS classification. We then fitted distributions to these subsets. The next step was to conduct a 10,000-iteration simulation, drawing samples from those distributions in the proportions used by ONS for the 606 largest schemes. The median results are shown below in the blue cells labelled Survey.
These are respectively 70% and 75% of the ONS reported values. They are 71% and 65% of the PPF reported values, and 72% and 66% for TPR. The next step was to deduct these values from reported values; these are the results shown in the final three tan coloured columns. This is the implicit contribution to the published data from the 4,625 very small schemes. Note that the ONS reported values for these small schemes show a decline of £210 billion, 40.7%, while both the PPF and TPR figures show increases in the values of assets held by these 4,625 smaller schemes. Our sample of smaller schemes shows an average decline in assets of 36% The increases in small scheme asset values (for PPF and TPR) are simply not credible as we have not observed an increase in asset value for any of the schemes in our collected sample. |
The Report contains a section concerned with Buy-out funding levels:
This 99% funding level is inconsistent with the earlier figure of 41% of schemes being in buy-out surplus. However, our first concern is with the asset values, which we will simply restate using ONS asset values. as Table 5:
Table 5: Buy-out Funding with ONS asset values
The improvement in funding ratio using ONS data (10.1%) is less than half that of TPR’s modelling (21%). This 89.5% funding implies that some 19% of schemes are funded to buy-out level, not the 40% claimed elsewhere in the Report.
The Report covers Low-Dependency portfolios.
If we again recast TPR’s Table 6, using ONS data, Table 6 below, the low-dependency funding ratio improves from 89.4% to 97.6% in December 2022. This is an improvement in funding ratio of 8.2% not the 20% claimed by TPR. With the funding ratio at this level, it is clear that the earlier claim that 65% of schemes are funded to low dependency levels cannot be correct.
Table 6: Low Dependency funding using ONS data
We are also concerned with the declines in liabilities shown in the Report for both Buy-Out and Low Dependency, both of which have declined far more than TPs. We show the decline of buy-out and low-dependency liability values relative to technical provisions estimates in Table 7 below.
Table 7: Buy-out and Low Dependency Liabilities relative to Technical Provisions estimates
TPR’s models show the relative cost of buy-out and low-dependency liabilities declining by 42% and 47% respectively. The reported valuation of buy-in policies in our sample show declines of 24% – 28%. The appendices to TPR’s Report offer no insight into the models which generate these results, and we are at a loss to explain them.
If we look to the PPF estimates, Table 8 below, we do not see a similar decline in relative cost to the TPR’s technical provisions level and the PPF liabilities contain an actuarial revision in May 2023. Over the 2021-2023 period, the relative cost of the PPF liabilities declined by 13.8%, while buy-out cost declined by 50.4% and low dependency by 64.8%.
Table 8: Relative Liability Performance
Table 7 also shows the differences between buy-out, low dependency and PPF and TPR’s technical provisions estimate in monetary terms (£ Billions). These differences are material amounts.
There is a significant omission from the Report, an explicit counterfactual group of schemes which did not engage in LDI. Local Authority schemes are just such a counterfactual. They are not regulated by TPR and did not pursue LDI investment strategies. Table 9 below shows Local Authority schemes assets and their quarterly returns.
Table 9: Local Authority assets and returns
Local Authority schemes experienced a loss of 8.4% of assets in 2022, while TPR’s estimated loss is 23.8% and ONS 31.7%. One interpretation here would be that in 2022, LDI in TPR’s modelled form has an excess cost of 15.4% of the initial endowment and the ONS full effect of LDI, rebalancing and all, 24.3% of that endowment.
Final thoughts
It is also worth noting that there is a pronounced difference in the returns in the fourth quarter of 2022 – with TPR reporting a positive return of 4.4% and ONS a negative return of 3.5%. This difference which amounts to £102 billion is the most significant single period contributor to the differences seen between ONS and TPR statistics. It is, of course, the period when most crisis actions, such as rebalancing, not captured by TPR’s models took place. It is also interesting that in all cases, the majority of asset declines occurred before September’s crisis.
[1] Author’s calculations and data collection.
Pingback: Final thoughts on the impact of the “LDI episode”. | AgeWage: Making your money work as hard as you do