Strategy Report - The Norwegian Government Pension Fund - Global

Norges Bank Investment Strategy Research Report Strategy Report - The Norwegian Government Pension Fund - Global Staff Memo, No. 2007/1 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Norges Bank Investment Strategy (2007) : Strategy Report - The Norwegian Government Pension Fund - Global, Staff Memo, No. 2007/1, ISBN 978-82-7553-382-9, Norges Bank, Oslo, https://hdl.handle.net/11250/2507975 This Version is available at: https://hdl.handle.net/10419/210180 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/deed.no NO 2007/1 Oslo 29 November 2006 Staff Memo Norges Bank Investment Strategy Strategy Report - The Norwegian Government Pension Fund - Global Publications from Norges Bank can be ordered by e-mail: [email protected] or from:Norges Bank, Subscription service, P.O.Box. 1179 Sentrum N-0107 Oslo, Norway. Tel. +47 22 31 63 83, Fax. +47 22 41 31 05 Publications in the series Staff Memo are available as pdf-files on the bank’s web site: www.norges-bank.no, under "Publications". Staff Memos present reports on key issues written by staff members of Norges Bank, the central bank of Norway - and are intended to encourage comments from colleagues and other interested parties. Views and conclusions expressed in Staff Memos can not be taken to represent the views of Norges Bank. © 2005 Norges Bank The text may be quoted or referred to, provided that due acknowledgement is given to source. Publikasjoner fra Norges Bank kan bestilles over e-post: [email protected] eller ved henvendelse til: Norges Bank, Abonnementsservice Postboks 1179 Sentrum 0107 Oslo Telefon 22 31 63 83, Telefaks 22 41 31 05 Utgivelser i serien Staff Memo er tilgjengelige som pdf-filer på www.norges-bank.no, under «Publikasjoner». Staff Memo inneholder utredninger som inngår i bankens arbeid med sentrale problemstillinger. Hensikten er å motta kommentarer fra kolleger og andre interesserte. Synspunkter og konklusjoner i arbeidene representerer ikke nødvendigvis Norges Banks synspunkter. © 2005 Norges Bank Det kan siteres fra eller henvises til dette arbeid, gitt at forfatter og Norges Bank oppgis som kilde. ISSN 1504-2596 (online only) ISBN 978-82-7553-382-9 (online only) Norges Bank Investment Strategy 29 November 2006 Strategy Report - The Norwegian Government Pension Fund – Global Contents Summary 1. The basics 2 1.1. The objective function 1.2. Construction of the benchmark portfolio 1.3. Potential changes to the benchmark portfolio 1.4. Simulation modelling 2. Portfolio analysis of existing asset classes 8 2.1. The equity portion 2.2. The regional weighting of the fixed income benchmark 2.3. The regional weighting of the equity benchmark 3. New market segments for equity or fixed income? 15 3.1. Small cap equity markets 3.2. High yield fixed income markets 4. New asset classes 22 4.1. Real estate and infrastructure 4.2. Private equity 4.3. The investment mandate for alternative asset classes 5. The recommendations 27 1 Summary Our report on “Long term market outlook” discussed the future market conditions both in the bond and equity markets, where the Pension Fund is currently invested, and in alternative asset classes, where the Fund may invest in the future. Here we summarise the findings and employ a simulation modelling framework to determine how changes in the investment strategy can be expected to affect the return distribution for the entire portfolio. We are looking both at the expected return and at different measures of risk exposure. The basic assumptions used in the modelling exercise are presented in section 1 below. We provide numerical details on the macroeconomic scenarios introduced in the Market Report, and specify the covariance matrices that will be assumed in each scenario. In section 2 we consider the equity/bond ratio in the existing benchmark portfolio, and the regional weightings within the equity and bond benchmarks. This analysis essentially confirms the conclusions from our 2005 Strategy Report. In section 3 we consider two market segments that are not currently part of the equity or bond benchmark. The small cap segment in the equity markets is the largest segment of listed markets outside the benchmark. We believe that the diversification benefits outweighs the operational burdens and recommend that small cap should now be included in the equity benchmark. The high yield segment of the bond markets is much smaller and heavily biased towards USD issues. But the expected return looks attractive, even taking the risk into account. We thus believe that the Fund should be invested in this segment. Replicating the Lehman High Yield index may on the other hand be a bad idea, because many of the papers in the index are difficult or impossible to buy, and because pricing may not always reflect risk in a reasonable way. We therefore do not recommend including the high yield segment in the bond benchmark. In section 4 we consider new asset classes. Real estate is the largest asset class where the Fund is currently not invested. We believe that real estate investments will provide attractive diversification benefits in the portfolio and recommend that it be included in the Fund’s investment strategy. The size of the market limits the exposure the Fund can realistically obtain, however. Infrastructure investments have very similar characteristics to real estate, but the market is so far substantially smaller. We are proposing a common allocation to these two asset classes, with a strategic target of 10 percent of the total Fund. This is in line with other large investment funds, but we realise that this target will not be reached in the first few years. The Fund shall need to set up a new investment organisation and gradually build its real estate and infrastructure portfolios. Private equity is another asset class commonly included in the strategic allocations of large funds. We believe that returns on private equity are highly correlated with returns in the listed equity markets, and that the diversification benefits are limited. Furthermore, we do not find any evidence that an average investor can expect higher returns in the private equity markets. Private equity consequently looks only moderately attractive in the modelling exercise. We still recommend a strategic target of 5 percent of the Fund invested in this asset class. The reason is that returns vary substantially between managers (“leading partners”) and that the differences tend to be persistent. A large investor such as the Pension Fund has a better probability of identifying and getting access to above average quality managers, and will then have a fair possibility of earning an excess return. 2 1. The basics 1.1. The objective function The Government Pension Fund Act states that the Fund “shall support central government saving to finance the National Insurance Scheme’s expenditure on pensions and long term considerations in the application of petroleum revenues”. The proposition to Parliament1 underscores the need for large financial reserves to meet the future government expenditures as the age structure of the population changes in the next few decades. The share of the population outside the working age cohorts, and thus dependent on pensions or other government programmes, will gradually increase for most of the next fifty years. These statements do not amount to an objective function. But they indicate that one objective is to have a maximum amount of savings over a very long time horizon. The amount saved at any time will to a large extent depend on the volume of petroleum extraction and the level of petroleum prices on the one hand, and on government spending on the other hand. But the return on investments will become relatively more important as the size of the Fund increases. The statements quoted above indicate that the Fund is meant for government spending. Since the government budgets are in Norwegian currency, the objective function should at first glance also be in units of Norwegian currency. There has, however, been generally accepted that the returns on the investments are best measured in a basket of foreign currencies, to reflect the Fund’s future purchasing power on the world market. The rationale behind that view is that the Fund basically represents a national wealth. On a national level the net effect of spending from the Fund will be an increase in imports of about the same magnitude; thus the imports are implicitly financed by the Fund investments. More intuitively we could say that the purpose of the investments is to maximise the share of (relevant) global supply that can be bought by the Fund in the future. We shall retain this choice of a basket as the base currency of the Fund, in line with the assumptions set out in our 2005 Strategy Report. The composition of this foreign currency basket was updated in chapter 2 of the Market Report. It is meant to reflect the expected composition of future imports to Norway. Taking both the current composition and expected changes into account, we have chosen a basket that consists of approximately 60 per cent of European currencies, and approximately 20 per cent each of American and Asian currencies. The Norwegian Parliament has in 2001 approved a spending rule whereby on average four per cent of the Fund may be spent every year. This is meant to correspond to the expected real return on Fund investments, which implies that the expected lifetime of the Fund will be (nearly) infinitely long. The liability side of the Fund may thus be represented by an infinite vector of cash flows. The investment strategy of the Fund can not be static; it should be updated as new information arrives. Furthermore, even with an infinite projected lifetime for the Fund, finite horizons are relevant for the political viability of the Fund construction. Extended periods of negative returns may endanger the entire savings project. In this Strategy Report we focus on the probability distribution for accumulated return over a 15 year period. The analytical objective is to maximise the expected accumulated return as measured in the currency basket, within 1 O.t.prp. nr. 2 (2005-2006) 3 acceptable limits for risk exposure. Year-on-year variations in returns will not be considered important. Benchmark selection should not overlap with the active management of the Fund. Return variations over short horizons should be the responsibility of the investment management organisation. We believe that our horizon of 15 years is sufficiently long that the relevant information set for selecting the benchmark will be different from the information set that is relevant for active position taking. Our Market Report has been focused on the long term trends in the capital markets. In this Strategy Report the objective for the investments is thus to maximise the expected accumulated return over the 15 year horizon as measured in the base currency (basket), with an acceptable level of risk. There is very little guidance in official documents as to how the portfolio risk should be measured. The statement that the Fund shall be buying a share of the global supply of goods might intuitively lead us to a risk minimising strategy which consists of buying a share in the proceeds from global production capital, weighted to reflect the likely future import pattern of Norway. Those proceeds come as remuneration to stockholders and creditors, and the Fund can get a share of the proceeds by holding equity and bonds. But there would be significant elements of capital remuneration where the Fund could not buy its share, such as bank debts and equity in companies that are privately held or held by the government sector. This intuitive approach does not provide a sufficient basis for selecting an investment strategy. The conventional analytical risk measure is the standard deviation of the return. This is a symmetrical risk measure, and the risk minimising portfolio is the one that minimises the return variation as measured by the standard deviation. But the downside risk may arguably be more relevant than the upside, given the spending rule and the aim to make the Fund permanent. We shall therefore also be looking at shortfall risk measures, such as the probability of negative accumulated real returns over the 15 year period. The risk minimising portfolio will then be the one that minimises either the probability of a shortfall or the probability weighted by the conditional expected value of the shortfall. 1.2. Construction of the benchmark portfolio The building blocks of the benchmark portfolio consist of a large number of assets, in principle down to the individual securities. For practical analytical reasons the individual securities must be lumped together in sets, which we have chosen to delimit by asset class and region. For transparency reasons we employ the composition of widely used indices to represent these assets. The benchmark asset allocation should then in principle be determined by simultaneous optimisation of the risk and return trade-off over the outcome space spanned by this asset vector. In practice the analysis has been carried out in steps. In the current approach to benchmark construction the allocation between asset classes is determined first, and then the regional distributions within each class are determined in a second step. The asset allocation within each region thus becomes a function of the regional distributions within each asset class. 4 There is no direct correspondence between the location where an equity instrument is listed or the currency in which a corporate bond is issued, and the location of the underlying production capital. A large share of global production takes place in multinational companies that are typically listed on one of the major stock exchanges. Corporate borrowings are often made in one of the major currencies rather than in the home currency of the company. Furthermore imports from one country are not necessarily priced in the home currency of that country. It is difficult to evaluate the importance of these arguments. Most stock markets include the listings of large companies with worldwide production activities, along with local companies that have mainly domestic activities. Our presumption will be that there is a significant positive correlation between where the company is listed and the location of its production activities, and that its location therefore continues to be an important determinant of the currency exposure implicit in holding its stock. The main deviations from these correspondences are probably the listing of multinational companies at the US and UK exchanges. Listings in these markets will be a weaker indication of true currency exposure than listings in other markets. It would therefore make sense to overweight these stock markets relative to the countries’ weightings in the import basket. In the current benchmark that overweight is massive, with 31 and 17 per cent of the equity benchmark in these markets, as compared to 19 and 6 per cent weightings in our projected import basket (section 2 of the Market Report). On the fixed income side the benchmark is defined in terms of currencies, not in terms of where issuers are domiciled. A relevant question both there and on the equity side concerns the correspondence between the composition of the import basket on the one hand and the currencies in which the imports are effectively priced on the other. Again, the presumption is that there is a high degree of correlation. The main deviations are probably in the commodities markets where prices are normally quoted in one of the major currencies, but these markets are of relatively minor importance in Norway’s import basket. We thus proceed to analyse the benchmark in terms of currency exposure as indicated by the domicile of listed companies on the equity side and by the currency denomination on the fixed income side. 1.3. Potential changes to the benchmark portfolio The analysis in sections 2 - 4 below will not be based on explicit optimisation procedures. In stead we shall take the current benchmark portfolio as our point of departure. We shall be investigating whether potential changes to that benchmark will improve on the return and risk properties of the portfolio. We concentrate the analysis on those changes that we ex ante believe will be most beneficial. With regard to the existing asset classes of listed equities and investment grade fixed income, we shall be looking at changes in the regional weightings, as well as changes in the overall asset class weightings. Norges Bank has in a letter of 10 February 2006 recommended that the equity portion should be increased to 50 or 60 per cent of the total portfolio. That recommendation is still being considered by the Ministry of Finance. 5 In an earlier letter of 22 August 2005 Norges Bank recommended that the regional weightings of the equity benchmark should be changed by increasing the portion invested in Asia/Oceania relative to the investments in the Americas, with a corresponding reduction in the portion invested in America/Africa2. Norges Bank also recommended that the portion of the fixed income benchmark invested in Asia/Oceania should be reduced below 10 per cent, with a corresponding increase in the portion invested in Europe. An underlying recommendation was that the total exposure to Asia/Oceania should be left approximately as before. The recommendations on regional weightings were approved in the Revised National Budget 2006, where Asia/Oceania weighting in the equity benchmark was raised to 15 per cent, whereas the Asia/Oceania weighting in the fixed income benchmark was reduced to 5 per cent. With the total equity portion at 40 per cent of the portfolio, this meant a reduction in the total Asian exposure from 10.6 to 9.0 per cent. On the other hand, if the equity portion were to increase to 50 or 60 per cent, the total Asian exposure would be raised to 10.0 or 11.0 per cent. In the ensuing portfolio analysis in section 2 we shall be considering changes in the equity portion and in the regional weightings within equities and fixed income. We shall be looking at five regions rather than the three regions currently used in the benchmark. This does not imply that the number of regions specified in the benchmark shall be increased, but we shall get an evaluation of what difference that would have made to return distributions. In section 3 we ask whether the small cap equity markets or the high yield bond market should be in the benchmark portfolio. Both market segments are permitted within the investment mandate and the question becomes whether it will be advantageous to buy an average market exposure in addition to the selective exposure we get through active bets in these market segments. The alternative asset classes are considered in section 4. We shall see how private equity, real estate and infrastructure investments will fit into the existing equity and fixed income benchmark. We shall estimate the consequences for the probability distribution of portfolio return, and employ that as a criterion for introducing one or more of these asset classes into the Fund portfolio. 1.4. Simulation modelling Part of the analysis below will be done within tailor made simulation models for financial market returns. The first version of the model will be employed for the analysis of equity and bond markets in section 2. It consists of simple inter-correlated stochastic price processes for five equity and five fixed income assets. The main deviation from random walk is some modest mean reverting in the equity price processes. The model simulates the market developments over a 15 year period, and produces a probability distribution for returns on each asset and on the portfolio. We shall focus on the distribution for accumulated portfolio returns over the 15 year period, which corresponds to the time horizon we have chosen for our analysis, confer section 1.1 above. 2 South Africa is the only African country in the benchmark. For practical reasons it is treated as part of the American region. 12 are in Europe approximately 10 percent in UK and 50 percent in Europe ex UK, and in Asia/Oceania approximately 4 percent in Japan and 1 percent in Asia/Oceania ex Japan. The simulations are done within the framework of our base scenario with stable economic growth and inflation. The risk scenarios are only used to check the robustness of any recommendations. We set out by keeping the Europe weighting constant at 60 percent. We then change the weightings of America on the one hand and the two regions of Asia/Oceania on the other, while keeping the relative weighting within Asia/Oceania constant. Similarly, in the next steps we shall be keeping the America or Asia/Oceania weightings constant, while changing the two other weightings. The simulation results from these alternative allocations are compared to the current benchmark allocation in table 2.6 below. Fixed income regional weightings (UK, Europe, America, Japan, Asia/Oceania Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 10-50-35-4-1 3.54 % 1.61 % 3.74 % 6.24 % 1.23 % Europe constant 10-50-40-0-0 3.59 % 1.64 % 3.79 % 6.34 % 1.32 % 10-50-30-8-2 3.50 % 1.59 % 3.69 % 6.15 % 1.25 % America constant 11-54-35-0-0 3.56 % 1.61 % 3.75 % 6.25 % 1.27 % 9-46-35-8-2 3.54 % 1.61 % 3.74 % 6.23 % 1.17 % Asia/Oceania constant 11-54-30-4-1 3.51 % 1.59 % 3.70 % 6.16 % 1.23 % 9-46-40-4-1 3.58 % 1.63 % 3.78 % 6.32 % 1.25 % Table 2.6: Alternative regional weightings of the fixed income benchmark portfolio in the base scenario. Probability distributions of real return based on 6000 simulations in each case. None of the alternatives implies a probability distribution that is significantly different from the one that follows from the current regional allocation. The most attractive alternative involves the down weighting of Asian bonds, which is also what we proposed last year. A number of the other alternatives are actually less attractive than the current allocation, and none of them clearly preferable to it. We also look at changing the internal weightings in Europe and Asia/Oceania away from the market weightings in the current benchmark. As alternatives to 10/50 between UK and Europe ex UK we look 0/60 (no UK bonds) and 20/40 (one third UK bonds). As alternatives to 4/1 between Japan and Asia/Oceania we look at 0/5 (no Japanese bonds) and 5/0 (only Japanese bonds). The simulation results from these alternative allocations are compared to the current benchmark allocation in table 2.7 below. The most favourable effect on the probability distribution is now clearly obtained by eliminating the Japan weighting in Asia. This will increase the expected returns, slightly reduce the standard deviation and reduce the downside risk as measured by the probability for negative accumulated real return over the entire 15 year period. We notice again that this is 13 the kind of recommendation that we made in our 2005 analysis. The main reason for the result in the present analysis is naturally the lower expected bond return that we have assumed for Japan. This could easily be counteracted by an appreciation of the Japanese currency, which is the reason that we last year combined this recommendation with an increase in the equity portion in Japan and the rest of Asia/Oceania. In conclusion, we can see no clear benefit of changing any of the fixed income regional weightings, or of introducing separate weights for UK and Europe ex UK or for Japan and Asia/Oceania ex Japan. Fixed income regional weightings (UK, Europe, America, Japan, Asia/Oceania Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 10-50-35-4-1 3.54 % 1.61 % 3.74 % 6.24 % 1.23 % Changes within Europe 0-60-35-4-1 3.50 % 1.58 % 3.69 % 6.14 % 1.30 % 20-40-35-4-1 3.59 % 1.65 % 3.80 % 6.39 % 1.28 % Changes within Asia 10-50-35-5-0 3.54 % 1.61 % 3.73 % 6.24 % 1.27 % 10-50-35-0-5 3.60 % 1.61 % 3.79 % 6.22 % 1.17 % Table 2.7: Changed weightings of the fixed income benchmark portfolio within Europe or within Asia in the base scenario. 2.3. The regional weighting of the equity benchmark The equity benchmark has a regional weighting of 50 percent in Europe, 35 percent in America and 15 percent in Asia/Oceania. In this section we shall search in the neighbourhood of that allocation to see whether other weightings can be expected to improve on the properties of the probability distribution for accumulated return over the 15 year evaluation period. The simulation model will have a more detailed specification, with two regions in Europe and two regions in Asia/Oceania. The current market value weightings are for Europe approximately 18 percent in UK and 32 percent in Europe ex UK, and for Asia/Oceania approximately 9 percent in Japan and 6 percent in Asia/Oceania ex Japan. As for the fixed income weightings, the simulations are done within the framework of our main scenario with stable economic growth and inflation. The risk scenarios will only be used to check the robustness of any recommendations. We set out by keeping the Europe weighting constant at 50 percent. We change the weightings of America on the one hand and the two regions of Asia/Oceania on the other, while keeping the relative weighting within Asia/Oceania constant. Similarly we shall in the next steps be keeping the America or Asia/Oceania weightings constant and changing the two other weightings. The simulation results from these alternative allocations are compared to the current benchmark allocation in table 2.8. 14 In parallel with our findings for the fixed income benchmark, none of the alternative weightings implies a probability distribution that is significantly different from the one that follows from the current regional allocation. Some of them are clearly less attractive than the current allocation. The most favourable effect on the probability distribution is now obtained by increasing the weighting in Asia. This will not increase the expected returns, but it will reduce the standard deviation and the downside risk as measured by the probability for negative accumulated real return over the entire 15 year period. We notice that this is the kind of change we also recommended in our 2005 strategy report. The main reason behind this result in the present analysis is the increase in the Asian weighting will bring it closer to the 20 percent weight of Asian currencies in our currency basket, confer section 1.1 above. A lower currency weighting for Asia would have eliminated the benefits of a higher Asia portion in the equity portfolio. Equity regional weightings (UK, Europe, America, Japan, Asia/Oceania Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 18-32-35-9-6 3.54 % 1.61 % 3.74 % 6.24 % 1.23 % Europe constant 18-32-40-6-4 3.54 % 1.63 % 3.74 % 6.31 % 1.33 % 18-32-30-12-8 3.55 % 1.59 % 3.74 % 6.17 % 1.08 % America constant 20-35-35-6-4 3.54 % 1.62 % 3.74 % 6.29 % 1.37 % 16-29-35-12-8 3.55 % 1.60 % 3.75 % 6.21 % 1.12 % Asia/Oceania constant 20-35-30-9-6 3.55 % 1.60 % 3.74 % 6.21 % 1.18 % 16-29-40-9-6 3.55 % 1.62 % 3.75 % 6.28 % 1.32 % Table 2.8: Alternative regional weightings of the equity benchmark portfolio in the base scenario. Probability distributions of real return based on 6000 simulations in each case. Equity regional weightings (UK, Europe, America, Japan, Asia/Oceania Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 18-32-35-9-6 3.54 % 1.61 % 3.74 % 6.24 % 1.23 % Changes within Europe 10-40-35-9-6 3.49 % 1.60 % 3.68 % 6.21 % 1.32 % 25-25-35-9-6 3.60 % 1.62 % 3.80 % 6.29 % 1.22 % Changes within Asia 18-32-35-15-0 3.51 % 1.61 % 3.70 % 6.23 % 1.20 % 18-32-35-0-15 3.60 % 1.64 % 3.80 % 6.35 % 1.28 % Table 2.9: Changed weightings of the equity benchmark portfolio within Europe or within Asia in the base scenario. 15 We also look at changing the internal weightings in Europe and Asia/Oceania away from the market weightings in the current benchmark. As alternatives to 18/32 between UK and Europe ex UK we look 10/40 (less in the UK) and 25/25 (more in the UK). As alternatives to 9/6 between Japan and Asia/Oceania we look at 15/0 (only Japanese stocks) and 0/15 (no Japanese stocks). The simulation results from these alternative allocations are compared to the current benchmark allocation in table 2.9. Favourable effects on the probability distribution are now obtained by eliminating the Japan weighting in Asia, and by increasing the UK weighting in Europe. Both of these changes will increase the expected returns, keep the standard deviation the same and reduce the downside risk as measured by the probability for negative accumulated real return over the entire 15 year period. The main reason behind these results in the present analysis is naturally the lower expected equity return that we have assumed for Japan, and the higher return we have assumed for the UK. In Asia the low correlation assumed between Japanese and other Asian equities is also important. We hesitate to make a recommendation based solely on these assumptions. In conclusion, we can see no clear benefit of changing any of the equity regional weightings, or of introducing separate weights for UK and Europe ex UK or for Japan and Asia/Oceania ex Japan. 3. New market segments for equity or fixed income? In the Market Report we looked at the question of adding more market segments to the equity and fixed income benchmarks. The small cap and the high yield segments of the equity and fixed income markets, respectively, were discussed in detail. In this Strategy Report we summarize the main findings and present some supplementary results from model simulations. 3.1. Small cap equity markets The small cap segment is the largest of the candidates to be included in the benchmark, with a market value of more than ten per cent of the mid and large cap segments currently included in the equity benchmark. The basic principle for portfolio construction is that a large fund, in particular when aiming to be a pure financial investor, should diversify its investments as broadly as possible. This is relevant for the small cap segment even if our Market Report only indicated modest diversification benefits for the Pension Fund in terms of reduced volatility. That was mainly due to the limited size of even this market segment, and will be true for any new segment we may consider. Table 3.1 shows the size of the small cap segment relative to the large and mid cap segments already included in the Pension Fund benchmark. In the countries that are currently part of the benchmark there are in total 4500 small cap stocks in the FTSE Global Index. Their average market value is USD 838 millions, which is far smaller than the average size of large and mid cap companies. There are considerable differences between the developed markets in America and Europe on the one hand and Asia/Oceania and the emerging markets on the other hand. In the latter regions the average size of a small cap company is only USD 300-400 millions. In North America and Europe the corresponding number is USD 1000-1100 millions. The 16 average small cap companies in these two regions are comparable in size to the average large and mid cap company in New Zealand, which is the smallest developed market in the current benchmark. Large/mid cap Small cap Region / Country Market value (mill USD) Number of stocks Mean market value per stock Market value (mill USD) Number of stocks Mean market value per stock America / Africa Brazil 263 746 66 3 996 13 738 30 458 Canada 829 972 62 13 386 229 345 178 1288 Mexico 175 203 31 5 651 6 137 14 438 US 13 023 887 707 18 421 1 952 418 1730 1129 South Africa 229 288 82 2 796 9 589 37 259 Sum all developed 13 853 859 769 18 015 2 181 763 1908 1143 Sum all emerging 668 237 179 3 733 29 464 81 364 Sum 14 522 096 948 15 318 2 211 227 1989 1112 Europe Austria 47 357 8 5 919 25 472 18 1415 Belgium 124 017 16 7 751 23 157 32 724 Denmark 84 774 12 7 064 30 997 26 1192 Finland 147 772 11 13 433 41 525 36 1153 France 1 315 239 68 19 341 86 346 88 981 Germany 867 924 49 17 712 76 787 79 972 Greece 77 710 12 6 475 19 713 43 458 Ireland 91 783 8 11 472 23 764 17 1398 Italy 534 116 44 12 139 75 357 91 828 Netherlands 440 091 20 22 004 59 983 46 1304 Portugal 46 220 8 5 777 4 395 8 549 Spain 512 475 33 15 529 44 775 32 1399 Sweden 289 349 30 9 644 58 854 54 1090 Switzerland 823 814 32 25 744 85 616 86 996 UK 2 843 991 133 21 383 374 877 315 1190 Sum 8 246 632 484 17 038 1 031 618 971 1062 Asia / Oceania Australia 675 770 117 5 775 66 994 138 485 Hong Kong 354 786 107 3 315 31 294 109 287 Japan 2 786 834 484 5 757 314 527 854 368 Korea 419 965 99 4 242 56 259 142 396 New Zealand 17 904 15 1 193 2 487 13 191 Singapore 103 235 46 2 244 18 445 56 329 Taiwan 299 331 138 2 169 56 446 249 226 Sum all developed 3 938 529 769 5 121 433 747 1 170 370 Sum all emerging 719 296 237 3 035 112 705 391 288 Sum 4 657 825 1 006 4 630 546 452 1 561 350 Global sum 27 426 553 2 438 11 249 3 789 297 4 521 838 Table 3.1: Stocks and market values in the large, mid and small cap segments of the FTSE Global Equity Index Series per August 2006. 17 The small cap segment has outperformed the large and mid cap segments for extended periods of time, last time during the past 5-6 years. But there has also been extended periods when the small cap segment has underperformed, last time in the 1990’s. This record has led analysts to consider small cap characteristics as a priced factor in the equity market. The ex post premium relative to large and mid cap has varied considerably, but on average it has been slightly positive over the years and in the markets for which data are available. We thus believe that there are small but positive benefits both in terms of diversification and expected return. It should be noted that the pricing of the small cap segment may not be attractive at present (August 2006). But the evidence of excessive pricing is too weak to make it relevant for the question of whether the segment should be included in the benchmark. There are, however, a couple of other important counter arguments to be considered. First, the transaction costs are higher for smaller companies. Implementation costs for establishing the new portfolio will be higher than for the large and mid cap segment. The exact costs will depend on market conditions and on the speed of implementation. The estimates of market impact in table 3.2 are based on the StockFactsPRO model. Buying small cap Selling large and mid cap Implementation period Commissions Taxes & charges Impact cost Total cost Commissions Taxes & charges Impact cost Total cost 1 month 5.85 11.97 110.42 128.23 5.92 1.21 17.31 24.43 3 months 5.85 11.97 49.92 67.74 5.92 1.21 6.75 13.87 10 months 5.85 11.97 31.50 49.32 5.92 1.21 6.03 13.16 Table 3.2: Initial implementation cost estimates (millions USD). If the entire small cap portfolio is bought within one month and paid for by selling large and mid cap stocks, the estimated total implementation costs are USD 153 millions. That number can be substantially reduced by stretching the implementation period. With a ten month implementation period the estimated total costs are USD 62 millions. Even that cost could be somewhat reduced by using inflows to the fund for buying the small cap stock rather than selling large and mid cap stocks. Review FTSE Global Small Cap Index FTSE All-World Index FTSE Global All Cap March 2006 3.24 0.37 0.98 December 2005 10.67 0.19 1.81 September 2005 6.76 0.82 1.46 June 2005 2.24 0.55 1.07 March 2005 5.52 0.66 1.49 December 2004 23.89 5.01 3.34 September 2004 31.28 1.67 22.43 June 2004 2.12 0.69 1.16 March 2004 4.00 0.65 1.25 December 2003 1.59 0.17 0.73 September 2003 6.00 13.61 1.60 Table 3.3: Portfolio turnover in the FTSE global equity indices. There will also be higher maintenance costs due to more frequent exits from and entries into the small cap benchmark index. The FTSE global small cap index is reviewed quarterly. The turnover figures since inception in September 2003 are reported in the first column of table 3.3. There are large variations over time. The last four quarterly reviews for which data are 18 available (June 2005 to March 2006) have in total required transactions (buys plus sells) equivalent to 23 percent of total small cap market value. However, this number does not allow for the fact that exits to the mid or large cap segments will not require transactions for the Pension Fund. Inclusion of the small cap segment means going from a benchmark defined by the FTSE All- World (i.e. large and mid cap) index to a benchmark defined by the FTSE All Cap index. It is therefore more relevant to compare the turnover of these two indices, which are reported in the two last columns of table 3.3. Looking again at the last four quarters, the required indexing transactions increase from 1.93 percent to 5.32 percent of market value. Taking the annual average from all quarterly reviews reported in table 3.3 gives higher numbers for both indexes, but does not significantly affect the difference between them. Replication of the index also requires a large number of transactions in between the quarterly reviews, mainly because of IPOs and reinvestments of dividends. With the current All-World (i.e. large and mid cap) equity benchmark, these transactions are fully 60 percent of the total replication transactions, making for a total transaction requirement of 4.7 percent of market value. If we apply that same ratio to the small cap segment, the total transaction volume requirement for replicating the All Cap index is 13 percent. A more reasonable assumption may be that transactions due to IPOs and reinvestment of dividends represent the same percentage of total market value in all market segments. With that more conservative assumption the transaction volume required to replicate the All Cap index will be 8.1 per cent of market value as compared to the 4.7 per cent for the All-World Index. The trading costs can also be expected to be somewhat higher in the small cap than in the large and mid cap segments, but the difference does not on average appear to be very large. Employing the StockFactsPRO model the average trading cost associated with changes in the large and mid cap index is estimated to 46 basis points, whereas trading costs associated with changes in the small cap index is 59 basis points. Together with the estimates of transaction volumes this implies that replication costs for the AllWorld index is approximately 3 basis points, whereas the costs for the AllCap index is between 5 and 8 basis point, depending on what assumption we use for the volume of transactions between the quarterly reviews. The main source of increased transaction costs is the higher turnover of the index. All investors in the small cap market will to some extent be exposed to higher turnover requirements, and the higher transaction costs will thus at least to some extent be reflected in a higher required gross return. In a fully efficient market this would correspond to a higher equilibrium return. The second question concerns the limit of five percent maximum ownership in any company imposed by the Ministry of Finance. That is a rule which other large funds do not need to observe. Given market cap weighting between segments, the average ownership of the Pension Fund will not be higher in the small cap segment than in the existing benchmark companies. NBIM presently replicates the equity benchmark by essentially buying all companies included. In the small cap segment there may be good reasons for choosing a different indexing strategy, where only stocks from a representative sample of companies are bought. If the NBIM chooses a sampling strategy the average ownership share would increase above the average in the rest of the equity portfolio, and could limit the room for active management in the small cap segment. 19 This problem primarily concerns the investments in Europe, where the average ownership in stocks held by the Pension Fund is highest, at present approximately 0.7 per cent. That average would be somewhat diluted if the small cap segment was included, but it will nevertheless increase as the size of the Fund increases in the years ahead. Assume for illustration purposes that the indexing in the small cap segment is done by buying a representative sample comprising one third of the companies in the FTSE index. Average ownership share in these companies would then be close to two per cent and increasing. The room for active management would be limited to three per cent of the stocks in each of these companies, as compared to more than four per cent in the case of full replication. This would of course be an impediment to active management. There is a trade-off between the degree of sampling in indexing and the room left for active management. More exact replication implies higher indexing costs and larger room for active management. More use of sampling techniques will reduce the indexing costs, but increase the tracking error and reduce the room for active management. This trade-off should be the responsibility of the operational manager (NBIM). A third question concerns the corporate governance activities towards the small cap companies. The numbers of stocks held by the Pension Fund will more than double or perhaps even triple from today, depending on the indexing strategy chosen. That will not pose a technical problem for NBIM, but it will require more resources to handle the proxy voting, and thus a larger organisation. The costs of the corporate governance activities are still small compared to other management costs. Assume for instance that the addition of small cap companies would require a doubling of the five man-years now employed for corporate governance activities. The initial cost of this would be only in the order of one basis point of the small cap portfolio, and should not in itself constitute an important argument against including the small cap segment in the benchmark for the Pension Fund. But it naturally adds to the operational burden of the management organisation. As a supplement to these arguments, we have done a model analysis to illustrate how the small cap segment would fit into the benchmark portfolio. For illustrative purposes we assume the equity portion to be 40 %, of which one tenth or 4 % is in the small cap segment. The key assumptions made for small cap equities in the simulation model are listed in table 3.4. For the modelling exercise we assume the same expected return as in the large and mid cap segments, to check whether other factors still make small cap attractive. We assume a significantly higher volatility than in the large and mid cap segments, and the correlations with other equity market are assumed to be 0.7-0.85. There is also a significant and positive correlation with real estate and infrastructure returns. In line with our assumptions for equity in general we assume no correlation with investment grade bond returns. Expected excess return over existing equity benchmark 0.0 % Volatility of small cap returns (memo: large/mid cap 15 %) 18 % Correlation between small cap and investment grade bonds 0.00 high yield bonds 0.70 large and mid cap equities 0.70 private equity 0.85 real estate 0.60 infrastructure 0.60 Table 3.4: Key assumptions for small cap equities in the simulation model. 20 The results are reported in table 3.5, where the model parameters have been calibrated to give approximately the same properties to the current benchmark as in the five region model employed in section 2. We notice that the expected return with small cap included increases by three basis points even if no higher expected return has been assumed, whereas the risk is reduced, whether measured by the standard deviation or the probability of negative accumulated real returns. The increased expected return comes from introducing a higher volatility asset, whereas the reduced total risk exposure is due to the diversification effect obtained by introducing the small cap segment. Small cap equities Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 3.54 % 1.61 % 3.74 % 6.24 % 1.23 % Small cap included 3.57 % 1.61 % 3.77 % 6.23 % 1.05 % Table 3.5: Small cap equities in the benchmark portfolio in the base scenario: Probability distributions of real return based on 6000 simulations in each case. In conclusion we recommend that that the small cap segment should be included in the benchmark portfolio for the Pension Fund. As per August 2006 this means that we recommend introducing approximately 4500 additional companies to the 2500 currently in the benchmark. With these 7000 companies, the equity benchmark will represent 96 percent of the equity markets included in the FTSE index, as compared to 85 percent today. The remaining four percent are in emerging markets that are currently not included in the Fund’s equity benchmark. An alternative to including the entire small cap segment could be to include the developed markets in North America and Europe and exclude the Asia/Pacific and emerging markets small cap segments. The rationale for doing that would be that the average size of the small cap companies in these latter regions is much smaller than in Europe and America, and that the bulk of small cap market value is in Europe and North America. The diversification gain will thus not be much reduced, but the operational burdens could be alleviated. 3.2. High yield fixed income markets The market for corporate bonds below investment grade had in early 2006 a market value corresponding to approximately four per cent of the current Lehman Global Aggregate benchmark index for the Pension Fund. The market share has been growing rapidly during the last two decades. The market segment is most developed in the US, which constitutes more than 60 per cent of the Lehman Global High Yield Index. The high yield segment can over time be expected to earn a positive premium over investment grade returns. The expected premium can reasonably be estimated at 1-2 percentage points, when the expected losses from defaults have been deducted. It should be noted that the data basis for the estimate is limited to the last 20 years’ experience in the US. Based on the same limited data set, there seems to be diversification benefits within the fixed income portfolio, with a relatively low correlation between high yield and investment grade returns. The 21 correlation between high yield and equity returns are, however, quite high, and this reduces the benefits for the total portfolio. Transaction costs are high, with average bid-ask spreads significantly higher than in the investment grade segment. Furthermore, in many cases the issues included in the high yield indexes may be difficult or impossible to buy at all. The Lehman Global High Yield Index is thus not investable in a strict sense. We have still looked at how the high yield segment would fit into the total benchmark portfolio. The key assumptions made for high yield bonds in the simulation model are listed in table 3.6. We have assumed a 1.4 percentage point higher expected return than in the investment grade segment, and a higher volatility. The correlation with investment grade bonds is assumed to be 0.6, and we also assume high correlation with the equity markets, in particular the small cap and private equity segments where high yield financing is important. There is also a positive correlation assumed with real estate and infrastructure returns. Expected excess return over existing fixed income benchmark 1.4 % Volatility of high yield returns (memo: investment grade 4.5 %) 6 % Correlation between high yield and investment grade bonds 0.60 large and mid cap equities 0.60 small cap equities 0.70 private equity 0.65 real estate 0.55 infrastructure 0.55 Table 3.6: Key assumptions for high yield bonds in the simulation model. The high yield segment will be about four percent of the total fixed income benchmark, and for the simulation exercise we assume that a fixed income portion of 60 percent becomes 57.5 percent investment grade bonds and 2.5 percent high yield bonds. The results are reported in table 3.7. We notice that the expected return with high yield included increases in line with the higher expected return assumed, and that the standard deviation of returns also increases, whereas the probability of negative accumulated real returns may be reduced. The marginal return-risk trade-off is 0.29, which is relatively attractive. High yield bonds Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 3.54 % 1.61 % 3.74 % 6.24 % 1.23 % High yield included 3.58 % 1.65 % 3.79 % 6.41 % 1.20 % Table 3.7: High yield bonds in the benchmark portfolio in the base scenario: Probability distributions of real return based on 6000 simulations in each case. Large funds will typically have a significant allocation to the high yield segment. The fixed income investment programme of the largest US pension fund CalPERS sets a ceiling on high yield investments of 10 per cent of the fixed income portfolio. Similarly, the largest European pension fund ABP has approximately five per cent of its fixed income portfolio invested in the high yield segment. These investment programmes are mandates for active management. We are not aware of funds that employ passive indexing strategies to the high yield segment. 28 The investment universe in terms of countries should be the same for the alternative portfolios as for the equity and fixed income investments, since this has been determined by criteria for political and economic stability. As is the case today, the restriction could apply to the location of the main underlying investments when indirect instruments are employed. Any further limitations could aim at reducing the risk for unacceptable low returns from the alternative investments. That is probably best done by requiring the portfolio to be well diversified across relevant return factors. It will be important not to define very strict limits, which could under certain circumstances force the manager to make very specific investments. Limits are a potential threat to the optimal choice of investments, and should be kept relatively wide. Wide limits could be combined with an ex post evaluation of the actual diversification in the portfolio held by the manager. That naturally implies some additional reporting requirements. 5. The recommendations We have recommended three major changes in the long term investment strategy of the Pension Fund. Above we have discussed each change in isolation, and we now need to check that the combination of changes will also be beneficial. The new asset allocation recommended is exemplified in table 5.1, where the first column assumes an equity-bond ratio of 40/60 as in the current benchmark. To illustrate the properties of the proposed investments strategies we have assumed that the 10 per cent allocation to real estate and infrastructure are taken in a first step, and that the rest of the Fund is allocated in a 40/60 ratio between equity and fixed income instruments. Private equity is then considered part of the equity allocation. In the second column of table 5.1 we have alternatively assumed that the allocations to the alternative asset classes are combined with a 60 percent allocation to equity. This latter allocation is closest to the asset allocation we typically find at the largest pension funds. Asset class Alternatives combined with 40% equity Alternatives combined with 60% equity Large and mid cap stocks 28 % 44 % Small cap stocks 3 % 5 % Private equity 5 % 5 % Sum of equity investments 36 % 54 % Investment grade bonds 54 % 36 % Real estate and infrastructure 10 % 10 % Sum of stable return assets 64 % 46 % Table 5.1: The recommended strategic asset allocations, conditional on the equity portion. The results from the simulation model are reported in table 5.2. Notice that the expected return on small cap and private equity investments have been assumed equal to large and mid cap equity, and that the expected return on real estate has been set to the average of expected equity and investment grade bond returns. Even with these very conservative assumptions, the volatility effect is sufficiently strong to increase the estimated expected return by 21 basis points in the case with 40 percent equity, and by 18 basis points in the case with 60 percent equity. The standard deviation of annualised returns increases by 2 basis points in the first 29 case and decreases by 6 basis points in the second case. The probability of negative accumulated real return is substantially reduced in both cases. Current and recommended asset allocation Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 3.54 % 1.61 % 3.74 % 6.24 % 1.23 % Alternatives with 40% equity 3.75 % 1.63 % 3.95 % 6.32 % 0.68 % 60 % equity, no alternative assets 3.95 % 2.22 % 4.32 % 8.61 % 3.45 % Alternatives with 60% equity 4.13 % 2.16 % 4.48 % 8.37 % 2.38 % Table 5.2: Proposed changes in the benchmark portfolio in the base scenario: Probability distributions of real return based on 6000 simulations in each case. Looking at the probability distribution for the one year return, the expected value is estimated to increase by 21 basis points and the volatility by 8 basis points in the case with 40 percent equity. The return-risk trade-off is thus very attractive. With 60 percent listed equity, the addition of alternatives is even more attractive since the 16 basis points increase in expected return comes with a reduction of 24 basis points in volatility. The above simulations are conditional on our base scenario for the world economy the next 15 years. We have also tried to evaluate the benefits of alternative asset classes in our two risk scenarios. We have, however, very little basis for making assumptions about the expected returns and the covariance matrix under these scenarios. The available data simply do not contain sufficient information. The estimates given below should therefore be treated as very uncertain. With these reservations, the key assumptions for the deflation scenario are listed in table 5.3. The expected returns of stocks and bonds are lower than in the base scenario, in line with the assumptions made in section 2, and the returns of alternative assets are also generally assumed to be lower. Notice in particular the real estate returns are assumed to fall in tandem with the equity return. The correlations are not listed in the table, but are assumed to be the same as in the base case. Expected equity premium over bonds - 0.55 % Private equity expected excess return over listed equity 0.00 % Real estate expected excess return over bonds -0.55 % Volatility of investment grade bonds 4.0 % large and mid cap equities 16.0 % small cap equities 20.0 % private equity 25.0 % real estate 7.0 % infrastructure 5.0 % Table 5.3: Key assumptions for alternative investments in the simulation model for the deflation scenario. 30 For the simulations reported in table 5.4 the model parameters have been calibrated to provide approximately the same output for the current benchmark portfolio as in the disaggregated version in section 2. We depart from that benchmark and add the alternative assets, employing the parameters in table 5.3. The inclusion of alternative assets is less attractive than in the base scenario, mainly because real estate and infrastructure expected returns are assumed to be low. But an increase in expected annual return of 7 basis points and an increase in volatility of 12 basis points still represent an acceptable risk return trade-off. Starting from a 60 percent equity share, the inclusion looks like an even better deal for the Fund. Current and recommended asset allocation Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 3.09 % 1.66 % 3.29 % 6.43 % 2.83 % Alternatives with 40% equity 3.15 % 1.69 % 3.36 % 6.55 % 2.83 % 60 % equity, no alternative assets 2.99 % 2.30 % 3.39 % 8.92 % 9.35 % Alternatives with 60% equity 3.08 % 2.25 % 3.46 % 8.71 % 8.32 % Table 5.4: Proposed changes in the benchmark portfolio in the deflation scenario: Probability distributions of real return based on 6000 simulations in each case. A similar analysis for the stagflation scenario is reported in tables 5.5 and 5.6. In this scenario both volatilities and stock-bond correlations are assumed to be higher than in the other scenarios. The alternative asset classes still continue to be attractive. The most important assumption behind that conclusion in the stagflation scenario is that the excess return assumed for real estate and infrastructure investments is well above the expected returns for stocks and bonds. The low volatility of real estate and infrastructure returns also contributes. These important assumptions are based on the fact that the Market Report did not identify differences in real estate returns between high and low inflation environments. But it must be kept in mind that the assumptions are somewhat speculative, and that the simulations reported in tables 5.3 to 5.6 must be interpreted with a high degree of caution. Expected equity premium over bonds - 0.90 % Private equity expected excess return over listed equity 1.00 % Real estate expected excess return over bonds 1.20 % Volatility of investment grade bonds 8.0 % large and mid cap equities 18.0 % small cap equities 22.0 % private equity 27.0 % real estate 7.0 % infrastructure 5.0 % Table 5.5: Key assumptions for alternative investments in the simulation model in the stagflation scenario. 31 Current and recommended asset allocation Annualised real return (geometric average) Standard deviation of annualised return Mean real return per year (arithmetic average) Standard deviation of return per year Probability of negative accumulated real return Current benchmark 1.10 % 2.57 % 1.59 % 9.95 % 33.92 % Alternatives + 40% listed equity 1.40 % 2.53 % 1.88 % 9.80 % 29.42 % 60 % equity, no alternative assets 0.90 % 3.00 % 1.57 % 11.60 % 39.00 % Alternatives + 60% listed equity 1.23 % 2.81 % 1.82 % 10.87 % 33.41 % Table 5.6: Proposed changes in the benchmark portfolio in the stagflation scenario: Probability distributions of real return based on 6000 simulations in each case. Strategy Report - The Norwegian Government Pension Fund - Global Staff Memo 2007/1