Not drowning but waving?

Transcription

1 Not drowning but waving? Quantitative easing and UK pension schemes FOR INSTITUTIONAL INVESTORS ONLY NOT FOR USE BY OR DISTRIBTION TO RETAIL INVESTORS

2 Table of contents 3 Introduction 5 What is QE? Modelling the impact of QE 7 The results of the counterfactual analysis 10 The end game for unconventional monetary policy 11 Conclusion 12 Appendix 8 The impact of QE on pension schemes 2 Not drowning but waving? Quantitative easing and UK pension schemes

3 Introduction Over the last few years, central banks have made increasing use of unconventional monetary policy to stimulate economic growth. While growth has been the motivation for this policy, pension schemes have also been seriously affected. Pension liabilities are driven by Gilt yields: if yields are pushed down, liabilities and, potentially, contributions are forced up. On the face of it, the impact of quantitative easing (QE) has been immense. As Exhibit 1 shows, ten-year Gilt yields were around 150 basis points lower than their natural levels during both phases of QE. J.P. Morgan asset management 3

4 Exhibit 1: The impact of QE on the ten-year Gilt yield Difference between predictred and smoothed 10 year Gilt yields (basis points) QE1 Inflation report (telegraphed QE) QE2 Jan 2008 Apr 2008 Jul 2008 Oct 2008 Jan 2009 Apr 2009 Jul 2009 Oct 2009 Jan 2010 Apr 2010 Jul 2010 Oct 2010 Jan 2011 Apr 2011 Jul 2011 Oct 2011 Jan 2012 Apr 2012 Jul 2012 Oct 2012 BoE purchases Uncertainty zone 10-year basis Announcements Date Source: DataStream, Bank of England; J.P. Morgan Asset Management analysis. The Uncertainty Zone represents the area in which we cannot attribute the effect to QE, due to the inherent imperfection of the model; specifically, it is the model error during the fitting period, as represented by the Root Mean Square Error. The nature and duration of pension schemes mean that their reaction to changes in bond yields is not necessarily straightforward. The impact of QE on both nominal and real rates must be assessed and understood for the full term of the yield curves. For this paper, we have built a model to allow this degree of analysis, and the results are interesting. At the height of the first round of QE (QE1), we estimate that the impact on bond yields both through the increases in liabilities and the offsetting rise in bonds pushed pension scheme deficits as much as GBP 150 billion above the levels that would have been seen without QE. Since this peak value occurred on 30 September 2009, contributions assessed as at this date could have been similarly elevated. The impact of the second round of QE (QE2) on pension schemes was more measured, and the impact had effectively disappeared by the end of An important aspect of this effect is that it was felt largely through the impact of QE on real yields despite the fact that the Bank of England bought only conventional Gilts. The impact of QE on pension schemes is not just an issue of historical interest. Transitory effects on rates can have long lasting effects on contributions. In many cases, 2013 pension contributions are driven by 2012 bond yields. There may even be further bond repurchases as recently as February 2013, one third of the Bank of England s Monetary Policy Committee voted to expand the programme of QE 1, so the potential impact on pension schemes is as relevant as ever. Even in the absence of additional QE, the Bank of England still buys more Gilts from time to time: when Gilts reach their redemption date the proceeds are used to buy more Gilts. It is unlikely that any one-off gains due to reduced perceptions of tail risk would be seen with future instances of QE; however, the fall in yields due to excess demand would likely occur with any future episodes of QE, or other Gilt purchases. Understanding the impact of QE can also help us to appreciate the potential impact that an unwinding of QE might have. It is unlikely that QE would be unwound in such a way as to increase the fear of tail risk, so risk assets would not be significantly affected. However, bond yields could be driven higher. While this would increase the cost of borrowing for the government, it would also reduce pension scheme liabilities. It has also been argued that QE has had an impact on risk assets. However, this effect is thought to have had a different profile. Exhibit 1 suggests that QE has had an impact when bonds were being bought, but not otherwise. In other words, there appears to be a flow effect, caused by the excess demand for Gilts. However, the impact on risk assets is thought to have been seen only at the start of QE. This is consistent with the idea that as well as a demand and supply effect on assets, there was also a reduction in the fear of tail risk that provided a one-off boost to risk assets. Having said this, we have found little evidence for such an impact. Of course, QE is not the only economic event to have happened in the last few years. The problems in the eurozone, together with the broader liquidity crisis, have also had an impact. In particular, the rise in government bond spreads for countries that were previously seen as safe significantly reduced the pool of high-quality assets. This has meant that the demand for such assets including Gilts could well have driven their prices up. It is difficult to control for these events in a counterfactual 1 Bank of England (February, 2013), Minutes of the Monetary Policy Committee Meeting, 6 and 7 February Not drowning but waving? Quantitative easing and UK pension schemes

5 setting. Variables such as peripheral European credit spreads grew substantially at various times in these crises. However, in the training period before the crisis, these variables were low and stable, so they do not appear as significant factors in counterfactual models. This means that if rates were lower than expected in the period that QE was in place, there is a risk that QE is not the only factor to have caused this effect. In this regard, it is important to see what happened between the end of QE1 and the start of QE2. If rates appear to be out of line at this time, then it is reasonable to assume that QE is not the only driver; however, if rates appear to normalise, it can be argued that QE is a dominant factor in distorting rates. Exhibit 1 appears to suggest that the latter is a better explanation. Nor is the UK the only country to have used unconventional monetary policy and there is every reason to expect that the policy of other central banks will have had an impact on UK rates. However, Exhibit 1 again suggests that it is UK QE that has been the main driver of the difference in rates. Because the issues around QE are so fundamental, we start this paper by looking in more detail at the relevance of QE to pension schemes, before exploring the nature of QE and assessing its impact. What is QE? 2 QE is a form of unconventional monetary policy. It is a technique used by the Bank of England and other central banks to inject liquidity into the economy, and ultimately to reduce the cost of capital for companies. This essentially means pushing yields downwards and asset values upwards. As well as being beneficial to companies, the support for asset values benefits investors. In broad terms, QE in the UK is implemented by the Bank of England increasing the money supply by creating electronic money, and using the proceeds to buy securities from financial institutions through its Asset Purchase Facility (APF). The securities bought have been almost overwhelmingly government bonds: at the end of 2012, the Bank of England held GBP 375 billion in Gilts, whilst it holds only GBP 25 million in corporate bonds and no commercial paper at all. Even 2 More detail on QE and how it relates to monetary and fiscal policy is given in Appendix 1 3 The 1 ¼% Gilt maturing in 2018 was not purchased due to its limited liquidity when private sector debt was being traded more actively, the levels of corporate bonds and commercial paper held reached only GBP 1.5 billion and GBP 2.1 billion respectively. The Bank of England has purchased Gilts of virtually all terms from three years upwards. It has tried to ensure that it holds no more than 70% of any issue, in order to maintain liquidity in the market, and while it has purchased a proportion of all but one issue 3, it has concentrated on the medium-term (seven to fifteen year) market: it holds around 40% of the securities here, as compared with 35% of the short- and longterm Gilt markets. 4 The Gilts bought were exclusively conventional, as opposed to index-linked. This is important for UK pension schemes, because the inflation protection built into benefits means that it is the impact on the real yield of index-linked Gilts that is more important. This is something we investigate later. Modelling the impact of QE There are a number of ways in which the impact of QE can be assessed, and a number of approaches have been tried. Some have simply commented on the movement in rates, noting that Gilt yields fell by around 100 basis points when QE was first announced, while investment grade corporate bond yields had seen a 70 basis point fall in yield. A range of studies 6, 7, 8, 9, 10, 11 put the impact of the first round of QE on longterm Gilt yields at between 30 and 125 basis points, noting anecdotally a similar fall in corporate bond yields, a modest weakening of sterling and a strong rise in equities. 4 Full details of the Bank of England s actions are available on the bank s website, 5 D. Miles (October 2011), Monetary Policy and Financial Dislocation, Speech on 10 October M.A.S. Joyce, A. Lasaosa, I. Stevens and M. Tong (September 2011), The Financial Market Impact of Quantitative Easing in the United Kindgom, International Journal of Central Banking 7 J.H.E Christensen and G.D Rudebusch (May 2012), The Response of Interest Rates to U.S. and U.K Quantitative Easing, Federal Reserve Bank of San Francisco Working Paper A. Meier, Panacea, Curse or Nonevent? Unconventional Monetary Policy in the United Kingdom, International Monetary Fund Working Paper 09/163, August G. Kapetanios, H. Mumtaz, I. Stevens and K.Theodoridis (January 2012), Assessing the Economy-Wide Effects of Quantitative Easing, Bank of England Working Paper F. Breedon, J.S. Chadha and A. Water (2012), The financial market impact of UK quantitative easing, BIS Paper No J. Bridges and R. Thomas (January 2012), The impact of QE on the UK economy some supportive monetarist arithmetic, Bank of England Working Paper 442 J.P. Morgan asset management 5

6 In terms of equities, it is noted 12 that although UK equities rose by around 50% between March 2009 and the end of May 2010 it is hard to identify precisely what the UK-specific effect may have been, as more than half of earnings of FTSE All Share earnings originate from overseas and other countries have also engaged in unprecedented monetary stimulus. In our analysis, we concentrate on the impact of QE on bond yields. To do this, we use a counterfactual model. The principle here is to build a financial model that describes the movement of bond yields before the introduction of QE (the training period ), and use this model to show how yields would have behaved if QE had not been introduced (the projection period ). The training period we use for Gilts is February 1993 to December 2008, and the projection period is January 2009 to December The main yields that we model are conventional Gilt yields and index-linked Gilt real yields. We do not try to describe the movements in the yields directly, but for both conventional and index-linked Gilts we look at the movement of the yield curve as a whole. We do this by breaking the movement down into changes in the level of curve, in the slope and in the shape 13. Once we have modelled these features of the curve in the training period, we project them forward and recombine them to give the counterfactual yields. Constructing a counterfactual analysis makes an explicit assumption that the relationships exhibited in the training period should be maintained in the projection period. This is no small assumption; one need only look to the history of sovereign debt yields in the eurozone to observe that fundamental factors, at least on a cross-sectional basis, can be disconnected from bond yields. 15 This analysis is therefore conducted on the basis that absent the effects of QE, markets would value UK government debt in line with the relationships seen between 1993 and 2009, and further that QE has not significantly affected these macroeconomic variables. For equities (public and private), property and hedge funds, we find that the results of our counterfactual model are not particularly stable. In particular, the calculated impact seems to differ significantly depending on the period used to fit the counterfactual model. The end result is that we find no clear evidence using a counterfactual approach that the values of risk assets were increased due to QE; however, this position makes no judgement on the relative valuation of assets, fair or otherwise, compared to their historical macroeconomic relationships. The variables that we use to model the Gilt curves are: 14 UK policy rate the Bank of England Base Rate Sterling effective rate the relative strength of sterling against a basket of currencies BD manufacturing orders survey a survey of German industrial activity BD Ifo survey a survey of the business climate for German industrial companies Federal Reserve funds rate a measure of short-term US interest rates Liquidity risk proxy the spread between Libor and the Bank of England Base Rate VIX the level of market volatility implied by S&P Index option prices UK index of production an index that reflects levels of industrial production in the UK US non-farm payrolls a survey of US employment UK unemployment rate UK inflation shock proxy the difference between surveyed RPI inflation expectations reported by HMRC and realised RPI inflation UK inflation expectations RPI inflation expectations reported by HMRC 12 M. Joyce, A. Lasaosa, I. Stevens and M. Tong (August 2010), The Financial Market Impact of Quantitative Easing in the United Kindgom, Bank of England Working Paper Full details of this approach are given in Appendix 2 14 Details of the fitting approach are given in Appendix 3, together with the significance of the factors 15 De Grauwe, P., and Ji, Y., (January 2012), Mispricing of Sovereign Risk and Multiple Equilibria in the Eurozone, CEPS Working Document no Not drowning but waving? Quantitative easing and UK pension schemes

7 Exhibit 2: The impact of QE on the ten-year Gilt yield and basis Redemption yield and Basis (basis points) Inflation report (telegraphed QE) QE1 QE Date BoE purchases 10-year Gilt (smoothed) 10-year Gilt (counterfactual) Announcements 10-year basis Source: DataStream, Bank of England; J.P. Morgan Asset Management analysis the counterfactual alternative. At the twenty-year term, QE1 again had a slightly lower impact, 137 basis points, but QE2 only reduced yields by 103 basis points. Importantly, while five-year yields still appeared to be depressed by 123 basis points at the end of 2012, the impact on twenty-year yields was only 45 basis points. This is within one standard deviation of the errors in the fitting period in other words, a difference of 45 basis points could be thought of as noise. Table 1 summarises the difference between the smoothed observed and the counterfactual yields at their peaks in QE1 and QE2 and in December 2012 at the five-, ten- and twentyyear terms. table 1: Difference between counterfactual and smoothed observed Gilt yields Country 5 Year 10 Year 20 Year Maximum basis during QE Maximum basis during QE Basis in December Source: DataStream, Bank of England; J.P. Morgan Asset Management analysis The results of the counterfactual analysis The broad picture for conventional Gilts is shown at the start of this paper, in Exhibit 1, and in Exhibit 2 this information is shown again, in the context of the full training period. The apparent impact is clear here: Gilt yields at the ten-year term started falling behind the rising counterfactual yield as soon as QE was telegraphed in the Bank of England s inflation report. Yields remained at these relatively depressed levels for much of QE1 with the difference peaking at 162 basis points. Importantly, however, as soon as QE1 ceased, ten-year yields rose to be broadly in line with the yields predicted by the counterfactual model, and stayed there until around the time that QE2 was implemented. Here, the difference between actual and predicted yields was as high as 199 basis points at one point. The difference has since reduced following the end of QE, but at the end of 2012 yields still appeared to be 109 basis points lower than they should have been according to the counterfactual model. Because we model the full Gilt curve, it is also possible to look at the impact of QE at different terms. While the broad picture is the same, the level of the impact is different. At the fiveyear term, the impact of QE1 seems to have been fractionally less, with the difference between counterfactual and observed yields being 145 basis points. However, QE2 appears to have depressed five-year yields to well over 200 basis points below The difference in the impacts at different terms is important for pension schemes, since their liabilities are typically longer term. However, as noted earlier, pension schemes are typically more interested in real yields, since these are what drive the valuation of inflation-linked liabilities. The Bank of England did not buy any index-linked Gilts. However, did QE nonetheless have an impact on real yields? The answer, shown in Exhibit 3, is clearly yes. One distinctive feature of this chart is the spike in the real yield in November 2008, a feature not reflected in the counterfactual model. This reflects short-lived fears of severe deflation in the wake of the collapse of Lehman Brothers. The feature is equally clear for five-year real yields, but absent from the picture of 20-year yields. The more pertinent feature of this chart is that despite the Bank of England buying only conventional Gilts, QE appears to have affected index-linked Gilts in exactly the same way. During both QE1 and QE2, real yields were depressed by up to 155 basis points and at the end of 2012, this apparent mispricing largely persisted, falling only to 144 basis points. At the five-year term, QE2 appears to have had a marginally greater effect than QE1, with the impact being 224 basis points in the later periods compared with an earlier figure of 195 basis points, while at the twenty-year term QE1 had an J.P. Morgan asset management 7

8 impact of 120 basis points and QE2 had an impact of only 72 basis points. For both of these terms, as with the ten-year picture, the impact of QE2 persisted to the end of Table 2 summarises the difference between the smoothed observed and the counterfactual real yields at their peaks in QE1 and QE2 and in December 2012 at the five-, ten- and twenty-year terms. Exhibit 3: The impact of QE on the ten-year index-linked Gilt yield and basis 500 Redemption yield and basis (basis points) Inflation report (telegraphed QE) Date BoE purchases 10-year Gilt (smoothed) 10-year Gilt (counterfactual) Announcements 10-year basis QE Source: DataStream, Bank of England; J.P. Morgan Asset Management analysis Table 2: Difference between counterfactual and smoothed observed index-linked Gilt yields QE 2 Country 5 year 10 year 20 year Maximum basis during QE Maximum basis during QE Basis in December Source: DataStream, Bank of England; J.P. Morgan Asset Management analysis However, rather than trying to infer the impact on pension schemes by simply looking at the yields, it is possible to calculate the effect more directly. The impact of QE on pension schemes From the analysis above, it is clear that QE has had an impact on Gilt yields, both fixed and real. It is therefore worth extending the counterfactual analysis to consider the impact on the assets and the liabilities of pension schemes. 16 Unfortunately it is not straightforward to track the progress of funding valuations since the start of QE, as such valuations are generally carried out on a triennial basis, and on differing dates over these three-year periods. A more frequent source of valuation information is available from the Pension Protection Fund (PPF), a compensation scheme for members of under-funded schemes whose sponsoring employers are insolvent. The PPF gives monthly updates on the aggregate level of solvency that is, ratio of pension scheme assets to pension scheme liabilities for all schemes it covers. The solvency level quoted is calculated on the PPF s own basis and for PPF benefits. It is difficult to infer an exact relationship between the PPF and funding levels of solvency, but we estimate that funding solvency is around 94.2% of the PPF measure 17, 18. If the assets are assumed to be the same for both the PPF and funding valuations, then a lower funding level implies a higher level of liabilities. A funding level of 94.2% of the PPF measure therefore implies that the funding liabilities are 106.2% of the PPF liabilities. This approximation allows us to track the estimated funding liabilities on a monthly basis from the start of 2008 to the end of By creating a proxy for the PPF liabilities and seeing how this proxy changes when a counterfactual yield curve is used for valuation rather than the smoothed actual yield curve, we can then estimate how differently the funding liabilities would have behaved over the last five years 19. The indirect impact of QE on real yields is important for a number of reasons. First, it shows that pension scheme liabilities could well have been elevated by QE, since real yields were depressed. It also shows that longer-term yields were less affected, mitigating the impact on pension schemes though the effect continued after QE2 had ended. The starting point for creating the proxy liabilities is to extend both our smoothed and counterfactual yield curves. As 16 The construction of the counterfactual asset and liability values is discussed in detail in Appendix 6 17 The Pensions Regulator (June 2012), Scheme funding: Recovery plans of UK defined benefit and hybrid schemes 18 More information on the relationship between PPF and funding solvency is given in Appendix 4 19 A full description of the methodology used to calculate the proxy is given in Appendix 5 8 Not drowning but waving? Quantitative easing and UK pension schemes

9 discussed in Appendix 2, yield curves are fitted out only to 24 years. For terms beyond this, we assume that the one-year rate 24 years forward (in other words, the yield implied for the period between 24 and 25 years) is the forward rate that applies in all future years. While this gives 50-year rates that differ from the quoted yields on 50-year conventional and index-linked Gilts, the magnitude and direction of the movement over the period 2008 to 2012 is consistent with the movement in our model. The change in the smoothed and counterfactual yields from the middle of QE2 to the end of 2012 from five to 50 years is shown in Exhibit 4 for nominal yields and Exhibit 5 for real yields. The blue lines here show the yields as at June 2012, while the orange lines show the yields at December 2012; the solid lines (to the left of the vertical line) show the counterfactual yields, while the dashed lines (to the left of the vertical lines) show the smoothed actual yields; and the continuations of these curves to right of the vertical line show extrapolations of the yields assuming a fixed forward rate. Exhibit 4: Change in nominal yields from June 2012 to December Yield (basis points) Maturity (years) Actual (Jun 2012) Counterfactual (Jun 2012) Constant forward rate Actual (Dec 2012) Counterfactual (Dec 2012) Source: DataStream, Bank of England; J.P. Morgan Asset Management analysis Exhibit 5: Change in index-linked yields from June 2012 to December Yield (basis points) Constant forward rate Maturity (years) Actual (Jun 2012) Counterfactual (Jun 2012) Actual (Dec 2012) Counterfactual (Dec 2012) Source: DataStream, Bank of England; J.P. Morgan Asset Management analysis It is interesting to note that for both fixed and index-linked yields, the counterfactual yield is actually below the smoothed yield for longer maturities in December This again suggests that the QE hangover for pension schemes might be limited. Tracking the various yield curves over time, we are able to see how our estimated funding liabilities would have moved both allowing for QE and under the counterfactual scenario. However, QE would not have changed only the liabilities; the assets would also have been affected. At this point, a decision needs to be taken as to which asset values would also be altered, and how. The principle guiding the decisions is that for an asset to be changed, it would need to be likely for additional QE to have the same impact as previous instances, or for QE unwinding to have the reverse effect. The value of conventional and index-linked Gilts would clearly change with Gilt yields, so these are adjusted. For corporate bonds, an assumption is made that the spread would have remained the same in the counterfactual case as in reality. For other asset classes, it is assumed that performance would have been the same in both cases. This is clearly a simplification the reduction in perceived tail risk probably had a positive impact on equity values. However, we were not able to find clear evidence for an effect associated directly with QE. More importantly, it is not clear that a repeat of QE would have the same effect, or that unwinding QE would cause equity values J.P. Morgan asset management 9

10 to fall. Indeed, it would require perverse signalling from the Bank of England, of both a fragile economy and an unwillingness to provide monetary support, for this to be the case. It is also interesting to see how pension schemes would have fared between the end of QE1 and the start of QE2. It appears that in this period, pension schemes were actually better off than they would otherwise have been. Exhibit 6: Estimated actual and counterfactual funding solvency 140 Funding solvency (%) Inflation report (telegraphed QE) QE1 QE2 Dec 2007 Apr 2008 Aug 2008 Dec 2008 Apr 2009 Aug 2009 Dec 2009 Apr 2010 Aug 2010 Dec 2010 Apr 2011 Aug 2011 Dec 2011 Apr 2012 Aug 2012 Dec 2012 BoE Purchases Actual Date Counterfactual Announcements Source: DataStream, Bank of England, the Pensions Regulator, the Pension Protection Fund; J.P. Morgan Asset Management analysis Combining all of this information together gives an interesting result for funding solvency, shown in Exhibit 6. This shows that even allowing for the impact on bond values, QE1 had a significant impact on funding solvency, reducing solvency by up to 16 percentage points. In monetary terms, this is equivalent to an increase of around GBP 150 billion in the aggregate deficit of UK pension schemes. Importantly, this point of maximum divergence occurred on 30 September 2009, so any pension schemes carrying out valuations with this effective date would have seen their funding solvency depressed by a substantial degree and there is a risk that this impact could have flowed through to give higher contribution rates. Even if equities had increased by 20%, as suggested by Bridges and Thomas (2012), the deficits at 30 September 2009 would not have been cleared: at this time, only half of portfolios were invested in equities and similar assets. The end game for unconventional monetary policy Even though additional QE seems more likely than an unwinding of existing positions, it is important to consider how monetary policy may, eventually, be normalised, as this gives an indication of how rates will be affected. The infrequency of QE as a central bank policy makes it difficult to identify the most likely exit policy. We have one practical example of an exit from unconventional monetary policy, following the Bank of Japan s actions in the early 2000s. QE as undertaken by the Bank of Japan targeted the current account balances of corporate banks held at the central bank. This was achieved partially through conventional central banks tools for the management of the money supply in particular, the provision of short-term liquidity and partially through the purchase of securities 20. Unwinding the policy involved reverting short-term liquidity facilities to normal levels, and making sales of securities. The exit of this phase of Japanese QE was abrupt, as measured by the current account balances of commercial banks, with a fall from approximately JPY 30 trillion to less than JPY 10 trillion between March and July Over the same period, shortterm liquidity facilities operated by the Bank of Japan reduced by JPY 20 trillion 22. Sales of government bonds purchased during the operation were far more gradual, to the extent that they were not complete by the recommencement of QE in Japan to combat the latest financial crisis. However, the Japanese experience is not necessarily the best guide to the likely course of action for the Bank of England. In contrast, it might be more appropriate to look at comments that have been made in relation to possible strategies for other markets. In contrast, QE2 had a much more moderate effect, though funding levels still appear to be depressed. However, by the end of 2012 funding solvency for both the actual and counterfactual models is broadly similar. 20 H. Yamaoka and M. Syed (May 2010), Managing the Exit: Lessons from Japan s Reversal of Unconventional Monetary Policy, IMF Working Paper 21 A.S. Blinder (November/December 2010), Quantitative Easing: Entrance and Exit Strategies, Federal Reserve Bank of St. Louis Review 22 Yamaoka and Syed (2010) 10 Not drowning but waving? Quantitative easing and UK pension schemes

11 Ben Bernanke, chairman of the US Federal Reserve System, has on a number of occasions provided insight into a potential exit strategy, as summarised in Blinder (2010). Broadly, the strategy can be seen as a gradual reduction in the central bank balance sheet. A similar approach was detailed for the UK in a speech by David Miles, a Monetary Policy Committee member, in September These statements reflect an important difference between the implementation of QE in the US and UK, and in Japan; the Bank of Japan made greater use of short term credit facilities, relative to purchases of government bonds and other securities, whereas asset purchases make up the bulk of QE in the US and UK. The expectation in the US, and by extension in the UK, should be for gradual asset disposal through a combination of passive redemption (the bonds reaching maturity) and open market operations. It is anticipated that only passive redemption will be used in the US until the recovery is well underway, and that even this will precede any rise in the policy rate 24. In the UK, it is worth noting that the process of passive redemption began in March this year, as the nearestdated Gilts purchased by the Bank of England matured. However, the results of the counterfactual yield curve model shown in the previous section would suggest that the speed of exit may well be immaterial in influencing the movement of the yield curve, absent a careless disposal. While the yield curve clearly reacts to the buying pressure of QE, re-normalisation to the prevailing macroeconomic conditions has been rapid. The Pensions Regulator has offered some flexibility in deficit recovery periods 25, albeit for a relatively small number of schemes. It is also worth noting that funding rules do not change the liability cash flows that must be met. However, too stringent a funding requirement could well result in excessive contributions being paid and, more importantly, in companies being forced into distress. This issue is broader than that caused by the impact of QE, but it is related. The overall position appears to be that QE has artificially inflated liability values for the purpose of funding, and this has accelerated or is likely to accelerate the payment of deficit recovery contributions in many cases. It also suggests that the effect on solvency was temporary. However, once a contribution schedule has been agreed it might stay in place for three years. It is also possible that there will be further QE in the future. It is important to look to the future. It is clear that QE has depressed Gilt yields. However, this means that unwinding QE could inflate Gilt yields. This does, of course, depend on the way in which QE is eventually unwound, though it is likely that care would be taken to make any unwinding as controlled as possible. QE has had a major impact on pension schemes, and it is important to keep pension schemes in mind when QE is on the agenda in future. Conclusion The analysis above shows that QE has clearly had an effect on conventional and index-linked Gilt yields. Not only that, but it appears that pension scheme solvency that is, the ratio of assets to liabilities was also depressed by a massive amount, with the inflation in bond values doing little to offset the increase in liabilities. It could be that some of this deficit was offset by an increase in the values of risk assets; however, this is far from certain. Funding solvency is not easily observed for most pension schemes. Companies tend to disclose only accounting liabilities, which are affected by corporate bond rather than Gilt yields. As such, it might be thought that funding solvency is less relevant. However, it is the funding valuation that determines the contributions the hard cash that an employer must pay into a scheme. 23 D. Miles (September 2012), Winding and unwinding extraordinary monetary policy, RBS Scottish Economic Society Annual Lecture 24 B.P. Sack (March 2010), Preparing for a Smooth (Eventual) Exit, National Association for Business Economics Policy Conference, Arlington, Virginia 25 The Pensions Regulator (April 2012), Pension scheme funding in the current environment, Statement from the Pensions Regulator J.P. Morgan asset management 11

12 Appendix 1 QE and economic policy QE can be considered in the broader range of policy tools available to influence growth. Two of the key levers used to control economic growth are fiscal and monetary policy. In broad terms, fiscal policy involves altering levels of taxation and of government spending. As such it can be used to stimulate or to rein in economic growth, as well as being used as a tool for redistribution. However, fiscal policy can be quite slow to have an impact on the economy, while monetary policy is seen as faster-acting. Monetary policy is generally enacted by or through a central bank such as the Bank of England. It involves altering factors such as the volume of money in the economy as well as the rate of interest payable on deposits (and charges for lending). Raising interest rates is used to slow growth and curb inflation, while lower rates are used to encourage growth and stimulate inflation. However, if growth is still low as central bank base rates approach zero, it becomes increasingly difficult to weaken monetary policy by reducing rates further. One solution is quantitative easing. As discussed above, QE is a form of unconventional monetary policy used to inject liquidity into the economy. It is unconventional because the money supply is increased by creating electronic money and using the proceeds to buy securities from financial institutions. The programme of QE in the UK is particularly amenable to analysis because it has been so straightforward: the Bank of England has bought conventional Gilts and almost nothing else, and the purchases have been spread evenly along the maturity range. This is in contrast with, for example, the US, where a significant proportion of the purchases have been of mortgage-backed securities, and simultaneous sales and repurchases at different parts of the curve have been used to change the shape of the yield curve (Operation Twist). QE has been carried out in the UK in an attempt to ensure that the Bank of England meets its inflation target of 2% per annum. Falling growth led to fears that inflation could fall below its target level. With interest rates already low, QE was seen as a way of avoiding deflation. Since the introduction of QE, inflation has actually remained consistently above the target. However, the longer-term inflation outlook resulted in QE being used until autumn QE is intended to work by lowering the cost of funding for companies and borrowing for individuals. There are a number of ways in which this might happen. On a fundamental level, the fact that the bank is prepared to take such steps to support growth can itself increase spending and inflation, as well as supporting the price of risk assets. Such an impact could be reinforced by the fact that a policy of QE signals that rates are likely to remain low for some time. However, the main effects of QE are through the assets bought by the Bank of England. One would expect that the Bank s purchase of Gilts, in particular, would push their price up, while continuing to hold them thus restricting supply would similarly inflate values. The Bank of England s intention is that such market movements would encourage investors to sell Gilts and instead to buy substitute assets, such as corporate bonds. This should increase the price of these assets and thus reduce the cost of borrowing to companies. It might also be the case that the Bank of England s actions reduce the liquidity premium on other assets by increasing the frequency with which they are traded, again lowering funding costs. Lower longer-term interest rates could also feed through to individuals. This could discourage saving and encourage spending. Spending could also be funded by increased borrowing if lower rates are passed on to consumers. There is also the possibility that increasing Gilt prices would encourage individuals not only to sell their Gilts but to deposit the proceeds in bank accounts. This could, in theory, encourage banks to lend more, although it is more likely that additional deposits would be used to continue deleveraging as the Basel III banking rules are phased in. Appendix 2 Breaking down the yield curve Following a paper by Breedon et al (2012), we construct a number of macrofinancial yield curves for each of the nominal and real market. These yield curves are described using the following functional form, based on Svensson (1994) 26 ; this resembles closely the specification used by Diebold and Li (2006) 27, but with a fourth β parameter: where y(τ) is the yield at time τ. Broadly, β 1 may be interpreted as the level of the curve (as it affects all maturities equally), β 2 as the steepness of the front of the curve (as its value declines from 1 to 0 as τ increases) while β 3 and β 4 influence the curvature. 12 Not drowning but waving? Quantitative easing and UK pension schemes

13 For the nominal and real government curves, we make use of Bank of England yield curve data 28. The Bank of England provides yields for each term from one to 24 years, and these are the terms for which we fit the model. Factorising the curve It is necessary, therefore, to produce an estimate of the four β parameters for each curve. Following Breedon et al (2012), the λ parameters are assumed fixed. By fixing the λ parameters, the equation is linear, and, as such, we fit it using ordinary least squares (OLS) regression. Regressing historical yield curves produces an estimate of the historical β factors. While the Kalman filter approach used by Breedon et al (2012) was also considered, it was not thought to provide significantly superior results. Counterfactual yield-curve In order to determine any impact of QE, the β yield curve factors are related to macroeconomic variables by way of the seemingly unrelated regression technique; this allows for any correlation between the error terms to be explicitly considered. The training period for the nominal and real yield curve regressions is January 1993 to December The macroeconomic variables and their associated weights are detailed in Appendix 3. Making counterfactual comparisons In the charts and discussion above, we make use of a smoothed historical yield curve. This smoothed curve is constructed using the historical β parameters and fixed λ assumptions. All comparisons are made between this smoothed curve and the counterfactual yield curve. This further step is designed to reduce any noise that would be introduced by comparing a yield curve factorised using the Svensson (1994) functional form (and the fixed λ assumptions) to the actual yield curve constructed using a different, or differently specified, methodology. Exhibit A3-1: Summary of macroeconomic variables and unit root testing Variable Evidence of unit root? (10%) UK policy rate GBP effective rate BD manufacturing orders survey BD IFO survey Federal Reserve funds rate Liquidity risk proxy VIX UK index of production US Non-farm payrolls UK unemployment rate UK inflation shock proxy UK inflation expectations Source: J.P. Morgan Asset Management analysis, for illustration only; all variables sourced from DataStream except inflation expectations and the inflation shock proxy, which are based on HMRC data. This feature in the underlying data has the ability to allow spurious correlations to translate into high levels of statistical significance. A common technique to address unit roots, by modelling the changes in variables between time periods, was considered. However, this seemed unjustifiable in the context of the counterfactual experiment, as it would lead to the compounding of errors into the long forecasting horizon (by construction it was not possible to have a model of changes that updated to the last observed factor value, as the yield curve is anticipated to have been influenced by QE). All models used fixed parameters for the values of λ 1 and λ 2 of and 0.03, respectively. Appendix 3 Counterfactual yield curve specification The macroeconomic variables considered are drawn albeit not exclusively from those referenced in Breedon (2012). An element of concern is the presence of unit roots in the source data, as well as in the historical beta factors derived as described in Appendix L. Svensson (1994), Estimating and interpreting forward interest rates: Sweden , IMF Working Paper, No 114, as referenced in Breedon et al (2012) 27 F.X. Diebold and C. Li. (2006), Forecasting the term structure of government bond yields, Journal of Econometrics, 130(2), Source: DataStream J.P. Morgan asset management 13

14 Nominal Gilt model Exhibit A3-2: Nominal Gilt regression results β1 β2 β3 β4 Intercept *** *** *** *** UK policy rate 0.60 *** 0.47 *** *** *** GBP effective rate *** 0.13 *** 0.13 *** 0.10 *** BD manufacturing orders 0.01 *** (-1) *** 0.03 * BD IFO *** 0.05 *** (-1) 0.14 *** 0.15 *** Federal Reserve funds rate 0.26 *** Liquidity risk proxy 0.70 *** 1.45 *** VIX *** UK index of production *** 0.09 *** (-1) 0.68 *** Non-farm payrolls *** 0.17 *** UK unemployment rate 0.12 ** *** 1.89 *** Inflation shock proxy 0.29 *** *** ** Inflation expectations 0.51 *** *** *** R^ Source: J.P. Morgan Asset Management analysis, for illustration only; significance: 1% ***; 5% **; 10% *; (-n) indicates lag of n months. Real Gilt model Exhibit A3-3: Real Gilt regression results β1 β2 β3 β4 Intercept 4.59 *** *** *** *** UK policy rate 0.19 *** *** 1.47 *** GBP effective rate *** 0.10 *** BD manufacturing orders *** 0.08 *** *** 0.10 *** BD IFO 0.09 *** Federal Reserve funds rate 0.88 *** Liquidity risk proxy VIX ** 0.14 *** *** UK index of production 0.02 ** 0.50 *** *** Non-farm payrolls ** 1.25 *** *** *** UK unemployment rate 1.47 *** *** 0.43 *** Inflation shock proxy 0.20 *** *** 0.21 * Inflation expectations 0.52 *** *** 3.16 *** *** R^ Source: J.P. Morgan Asset Management analysis, for illustration only; significance: 1% ***; 5% **; 10% * Note: the models constructed using contemporaneous rather than lagged data give marginally worse fits as measured by the R-squared, but suggest a difference in the impact of QE of no more than 13 basis points. 14 Not drowning but waving? Quantitative easing and UK pension schemes

15 Appendix 4 PPF, buyout and funding solvency Funding valuations are carried out every three years to determine the level of contributions that a sponsor must contribute to a pension scheme. The results of these valuations are essentially approved by the Pensions Regulator (TPR). The Pension Protection Fund (PPF) is a compensation fund funded by a levy on virtually all UK pension schemes. The size of the levy for each scheme is determined by a number of factors, but the starting point is the level of solvency calculated using PPF benefits and the PPF valuation basis. The PPF benefits in other words, the benefits that the PPF would pay if a scheme entered the PPF are essentially the benefits payable from the pension scheme subject to a cap. For retired members over normal retirement age, the full pension is payable, with any pension earned after 5 April 1997 increasing in line with prices subject to a maximum of 2.5% per annum; for all other members, benefits are capped at 90% of the pension due from the scheme with a cap of around 34,000 per annum; increases in payment are as above. A third type of valuation carried out is an estimate of the buyout cost. This gives the assets as a proportion of the value required to secure the benefits with an insurance company. Exhibit A4-1. Comparison of PPF, funding and buyout solvency Funding Level (assets divided by liabilities) 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Group 1 Group 3 Group 2 September 2005 to September 2006 September 2006 to September 2007 September 2007 to September 2008 Valuation Period September 2008 to September 2009 PPF Funding Buyout September 2009 to September 2010 Because funding valuations are carried out on a triennial basis, it is difficult to assess the current level of funding solvency for all schemes. TPR publishes an annual study 29 which, among other things, compares these three measures of solvency for each of the three potential cohorts from 2005 to The results are shown in Exhibit A4-1. This shows considerable variation between the groups. However, the average ratio of funding to PPF solvency 94.2% seems reasonable. Appendix 5 Creating proxy PPF liabilities Before building a counterfactual model of pension scheme solvency, it is necessary to create a series of liability cash flows to act as a liability proxy. These cash flows are discounted using the smoothed actual and the counterfactual yield curves to show the different ways in which the liabilities would have moved under the two different scenarios. The broad approach used is to assume a stationary population with unchanging rates of mortality over time. This is a population where the impact of mortality is to leave the proportion of individuals at each age unchanged from year to year. The mortality rates are based on the mortality of members of self-administered pension schemes in the UK 30. The mortality rates of all members, both male and female, are used. A Makeham model 31 of the form below was fitted to the initial mortality rate, q x, for ages x ranging from 26 to 90: q x = Α + Βc x where Α, Β and c are constants. The data above age 90 appears unreliable, so log mortality rates are interpolated between the fitted value of q 90 and a value of q 110 set equal to one. The stationary population is then assumed closed, and rolled forward into the future until all members have died. Members are assumed to retire at age 65 with identical pensions, a proportion of which are fixed in nominal terms and the remainder of which are fixed in real terms. Source: The Pensions Regulator 29 The Pensions Regulator (June 2012), Scheme funding: Recovery plans of UK defined benefit and hybrid schemes 30 Continuous Mortality Investigation Bureau (May 2012), CMI Working Paper See Life Contingencies, A. Neill (1977) for details of this model J.P. Morgan asset management 15

16 The stream of cash flows is assumed to be identical whenever the pension is valued. The only factor that changes apart from the discount rate is the proportion of fixed and real liabilities. This is determined as the proportion in each year that gives the closest PV10 of the liabilities to the value quoted by the PPF for change in nominal interest rates and implied inflation 32. These figures are set using the 31 March figures in each year, this being the effective date of the PPF valuations. The split is assumed to apply for the whole of each calendar year. If the proxy PPF liabilities calculated using the smoothed Gilt yield curves are first rebased, then increased in line with the implied interest rate in effect in each period plus seventy basis points (to allow for the fact that new liabilities are being accrued), then it is clear that the proxies move in a similar manner to the true values, as shown in Exhibit A5-1. This also shows the fitted liability values calculated using the counterfactual yield curve. Exhibit A5-1: Proxy and actual PPF liabilities Liability Value ( bn) Dec 2007 Apr 2008 Aug 2008 Dec 2008 Apr 2009 Aug 2009 Dec 2009 Apr 2010 Aug 2010 Dec 2010 Apr 2011 Aug 2011 Dec 2011 Apr 2012 Aug 2012 Observed PPF Liabilities Date Constructed PPF Liabilities - Actual Constructed PPF Liabilities Counterfactual Source: DataStream, Bank of England, the Pensions Regulator, the Pension Protection Fund; J.P. Morgan Asset Management analysis Appendix 6 Creating counterfactual asset and liability values Having created a liability proxy as described in Appendix 5, the next stage is to calculate the percentage change in the liabilities using smoothed rates and using the counterfactual rates. The difference between these rates is taken meaning that any rebasing or artificial inflation added in Exhibit A5-1 is nullified and the result added to the change in the actual PPF liabilities. The starting PPF liability figure is then rolled forward using these counterfactual changes. This gives the counterfactual value of PPF liabilities. For the assets, a similar approach is used. The asset allocations in each year 33 are used to calculate the change in value of a proxy portfolio of PPF liabilities. These allocations are assumed constant in each calendar year. The indices used and the asset allocations are shown in Table A6-1. The return is calculated as the weighted total return on all of the indices except for the bonds, which are treated differently. For the conventional and index-linked Gilts, the index duration is obtained in each period. This is then used to calculate the weighted average of par five-year and twenty five-year Gilts that have the same approximate duration in each period for both conventional and index-linked Gilts. The return for each bond is calculated as the yield received plus the change in capital value given by the change in yield each month. In each month, it is assumed that each bond is sold and a new fullterm bond is bought. The difference in return between smoothed actual and counterfactual returns is then calculated. This is used to adjust the returns implied by the change in the value of PPF assets in each period, to give a series of counterfactual PPF asset values. The counterfactual assets are then divided by the counterfactual liabilities to give counterfactual levels of PPF solvency. This result is then multiplied by 94.2% to give estimated counterfactual levels of funding solvency. 32 Pension Protection fund ( ), The Purple Book 33 These are available from the Pension Protection Fund s Purple Book, Not drowning but waving? Quantitative easing and UK pension schemes