Investment in real estate and the effect on coverage ratios of Dutch pension funds

(1)

1

Investment in Real Estate and the Effect on

Coverage Ratios of Dutch Pension Funds.

Jeroen B. Meester

Student number: 5877059

Address: Lange Acker 3, 1965TK Heemskerk Email: Info@JBMeester.nl

Universiteit van Amsterdam

Faculty: Faculty of Economics and Business

Master-track: Double specialization: Finance and real estate Finance Address: Roetersstraat 11, 1018WB Amsterdam

Supervisor: dhr. dr. M.I. Dröes

(2)

2

Hierbij verklaar ik, Jeroen Meester, dat ik deze scriptie zelf geschreven heb en dat ik de volledige verantwoordelijkheid op me neem voor de inhoud ervan.

Ik bevestig dat de tekst en het werk dat in deze scriptie gepresenteerd wordt origineel is en dat ik geen gebruik heb gemaakt van andere bronnen dan die welke in de tekst en in de referenties worden genoemd. De Faculteit Economie en Bedrijfskunde is alleen verantwoordelijk voor de begeleiding tot het inleveren van de scriptie, niet voor de inhoud.

(3)

3

Abstract

In this thesis, the effects of real estate investments on the coverage ratios of Dutch pension funds are examined. The aim of the research is twofold. First, the role of the financial crisis is examined to confirm that real estate became less of a shock absorber for the portfolio of pension funds since the financial crisis. Secondly, it is hypothesized that funds investing largely in more volatile indirect real estate suffered a deeper crisis than funds investing in less-volatile direct real estate. The main part of this research focusses on the financial crisis period starting in 2008, because it pressurized coverage ratios of pension funds. Resulting in undercoverage, payment cuts had to be implemented. Particularly in the Netherlands, where pension funds’ investments cover a total of twice the GDP in value, financial distress can have significant impact. Due to low or negative correlation of real estate with other asset categories, real estate can form a shock absorber against financial market downturns to prevent pension funds form the problem of undercoverage. However, real estate as an asset class experienced downturns in the crisis as well. This research makes use of portfolio data on the 31 largest pension funds in the Netherlands. The composition of the real estate portfolio is compared with the coverage ratio over the years 2006-2013. It is found that on the one hand, during the crisis period, having real estate in the portfolio positively influenced the coverage ratio of the fund. But on the other hand – and contrary to the hypothesis - if the funds had

direct real estate in the portfolio, the effects on the coverage ratio were negative. Hence, it is

found that the indirect real estate plays an important role in stabilizing the portfolio. Funds which invested merely in direct real estate have suffered a deeper decline in coverage ratios. Likely, this is caused by either a lack of diversification in direct real estate, or a lack of foreign market knowledge if the funds do diversify internationally.

(4)

4

1. Introduction

In the final quarter of 2013, the Dutch pension funds’ total balance reached the one trillion euro border for the first time in history.1 Considering the Dutch economy, which totals €557 billion in terms of GDP in 2012, the Netherlands has the largest pension funds compared to GDP of all OECD countries.2 This makes Dutch pension funds unique in their role in the economy, as the high investments implicate high vulnerability towards the pension system if coverage ratio’s – which are the amount of assets relative to pension liabilities of the fund - get pressurized. Among the investments of pension funds is €91.3 billion in real estate, which is around ten percent of the total investments Dutch pension funds made in 20132. Real estate can be invested in in different types. Direct real estate is the type of real estate that is most tangible, wherein the pension fund actually owns the property. Indirect real estate can exist of two sub-types: listed and non-listed real estate. Listed real estate is participation in an

exchange-listed fund so there is vulnerability to cyclicality and stock market fluctuations. Indirect real estate is real estate held by a fund, in which can be invested. In indirect non-listed real estate, there is less vulnerability to daily fluctuations since this type of investment is in a private fund, not listed on a stock exchange. Therefore the type of real estate which is invested in can have significant impact on the performance of the fund. In order to prevent financial distress as a result of low solvability, the Dutch regulator DNB has set the coverage ratio to a minimum required rate of 105%.

Inevitably, the recent financial crisis has influenced Dutch pension funds’ investment performance negatively, and caused coverage to fall below this 105% rate. In May 2012, for example, the ABP pension fund, the largest Dutch pension fund, announced that its coverage ratio had declined to 94%. It was not the only pension fund to announce a below-100% coverage ratio. In fact, in the final quarter of 2011, the total average coverage ratio of all Dutch pension funds fell to 98%. Therefore, the funds couldn’t cover their liabilities with assets if they would have to at that moment. As a consequence, funds were forced to stop indexing pensions, increase premiums and even cut pay-outs in order to regain solvability. Direct real estate performed positively at the start of the crisis in 2008, but when looking at the 2006-2012 time period, overall, the return on global real estate went from 13% in 2006 to 3% in 2012, according to an IPD summary of returns on the global real estate market. It

1_Link: http://www.statistics.dnb.nl/financieele-instellingen/pensioenfondsen/macro-economische-statistiek-pensioenfondsen/index.jsp (accessed on January 5th_{, 2015)}

(6)

6

should be noted here that there is a difference in return of the different types of real estate: where direct real estate returns moved in the range +10% to -10% return on a yearly basis, indirect real estate stocks moved from +40% to -40% in the same 2006-2012 time period3_.

This implies that direct real estate suffered less shocks and therefore it seems that a quick conclusion can be drawn that funds that invested largely in direct real estate suffered a smaller dip in coverage ratio.

Despite the benefits, real estate is not the only asset Dutch pension funds invest in. Every asset category added to the portfolio diversifies this portfolio, lowers the standard deviation and thus increases stability towards its return. In fact, real estate is known for its significant differences with other categories regarding liquidity, transaction costs and its high initial investment. In his article, Van Gool (2010) confirms that correlation of real estate with other asset categories is low to zero. In other words, the presence of - direct - real estate should make a safety net against downfalls in return of the portfolio according to the theory. This raises the question whether if pension funds had chosen stable direct real estate instead of indirect real estate, they would have had the problem of undercoverage in the first place. This study aims to answer this question by proposing the following research question:

To what extent does investing in real estate influence the coverage ratios of Dutch pension funds?

In addition, the following sub-questions are examined:

- Did pension funds that invested mainly in real estate experience different coverage ratios as a result of the financial crisis?

- Did funds that have invested merely in the more volatile indirect real estate suffer a deeper crisis compared to the funds that mainly invested in direct real estate? This analysis is based on data from the 31 largest Dutch pension funds. These funds are chosen because of their total Dutch pension market share of over 80%, even though the Netherlands had over 600 pension funds in 2013. Of every fund, the investment portfolio data is extracted from its annual report over the years 2006-2013 in order to gain insight in the preface and aftermath of the financial crisis. This data includes total euro amount invested, total euro amount invested in real estate, sub-categorized in direct, indirect listed and indirect non listed real estate. Also, the annual year-end coverage ratio of the fund is distilled from the

(7)

7

annual report in order to be able to highlight the effect of the real estate investment dynamics on the coverage ratio.

A novel aspect of this study is that although many studies revealed the beneficial

characteristics of real estate in an institutional investor’s portfolio, like van Gool (2010), research has only very scarcely focussed on the role of the crisis and the different types of real estate. For example, Chun (2004) argues a higher allocation towards real estate, without concluding on which type of real estate. Moreover, the effect of real estate investments on the riskiness of the portfolio likely has changed after the financial crisis because of an urgent need for liquidity in the portfolio felt by most institutional investors (Allen et al, 2009), which makes pre-crisis studies possibly wrongly conclude optimal allocation towards (direct) real estate in the portfolio. This thesis studies the effect of the crisis on real estate investments and can provide insight in the types of real estate a pension fund should ideally invest in, showing the lessons that can be learned from the most recent crisis.

The results of this thesis show that those pension funds that invested highly in real estate before the crisis actually had a higher coverage ratio during the crisis. However, the benefit is entirely mitigated for those pension funds that invested substantially in direct real estate. The remainder of this research is structured as follows: Chapter 2 discusses the existing literature on real estate investments by institutions, along with the implications of investment decisions of Dutch pension funds. Chapter 3 states the hypothetical expectations of the research question based upon the literature. A section dedicated to the data and descriptive statistics is presented to uncover the portfolio dynamics of Dutch pension funds in Chapter 4. Accordingly, in Chapter 5, the methodology is elaborated on using the aforementioned panel data in a differences-in-differences regression model to capture the effect of the crisis period on the performance of pension funds regarding their real estate investments. In Chapter 6, the results of the proposed model are presented, after which the model is checked for robustness. Chapter 8 contains the main findings and gives suggestions for further research.

(8)

8

2. Literature review

2.1 The Dutch Pension System

In order to get a grip on the relevance of the real estate portfolio dynamics within Dutch pension funds, it is crucial to gain insight in the Dutch pension system and its importance for the Dutch macro economy. The Dutch pension system is a three-pillar system, in which the first pillar is the state pension. The state pension (or AOW in Dutch) is funded by the working population between 15 and 67 years of age. Each year, an employer builds up two percent of the state pension right, which amounts to 50% of the minimum wage. The focus in this thesis, however, lies in the second pillar. This pillar houses the collective pension scheme, funded by pension funds. These funds manage the contributions of the accompanying employers (capital funding) and are separated from the companies. Therefore, the financial position of the

pension fund is not directly related to the performance of the company. The aforementioned capital funding implies that the final pay-out to the pensioner consists of the contributions of the employers, plus any capital gain resulting from investments. As with any capital investor, the aim is to gain more return with respect to the obligations the investor has. That is, the coverage ratio should ideally remain above the 100% level. Finally, there is the third pillar, which is the privately held, non-compulsory insurances an individual can obtain to secure a certain periodic pay-out when a specific age is reached. This type of pension is mostly demanded from people whose sector of industry does not provide a collective pension

arrangement. This third pillar is known for its freedom of choice in premium and pay-out-age. Also, this pillar is fiscally encouraged by the government since these premiums are income-tax-deductible (OECD, 2013).

The performance of Dutch pension funds is determined by different factors. First, investment decisions the fund managers make significantly influence the performance of the fund and directly determine the pay-out and eventually the pension fund’s coverage ratio. As the article by Huang and Mahieu (2012) states, these decisions are based upon asset

management-specialized managers. Their role to the fund managers in determining the strategic asset allocation is providing them with advice. This allocation is summarized in a benchmark portfolio which the fund aims to outperform. The out- or underperformance is measured in the so-called z-score of the fund. The authors find that for Dutch pension funds, there is a

structural underperformance in terms of this z-score based upon the asset management investment decisions. This creates a situation in which there is demand for a more persistent annual investment plan, as the critique on the z-score as a performance measure grows.

(9)

9

Second, and inherent to the previous annual investment plan, a pension fund’s coverage ratio is determined by shocks on a macroeconomic level. Around the year 2009, the latter became reality as the global financial crisis took its toll. Dutch pension funds had to cut pay-outs and increase nominal pension premiums due to a severe undercoverage with ratios around the 90-to-95% level (Bonenkamp, 2009). Interestingly, this severe impact in the Netherlands is unique, as Bloom and Freeman (1992) only find little effect of the institutionalism on coverage ratio of pension funds. The authors conclude that falling coverage ratios affected mostly less well educated men and the downfall had only little effect on other income groups. This is striking with the Dutch situation, as the capital coverage system is more levelled towards groups.

Third, the prognosis for future returns a pension fund makes is bounded by rules. These governmentally set rules significantly influence the performance of pension funds, since they are committed to the FTK (Financieel Toetsingskader) in which the limitations are set. For instance, the interest at which a pension fund forecasts its returns for the coming year – the actuarial interest – is given by the Dutch Central Bank. This actuarial interest is derived from the capital markets interest rate, and pension funds have the obligation to use this actuarial rate. Consequential towards the financial crisis, the actuarial interest has declined analogous to the capital markets interest rate. For a pension fund, this implicates a lower growth potential on paper, making the coverage ratio decline, and eventually this can lead to cut of pay-outs and increased premiums while the underlying result of the fund is not affected by the actuarial interest rate. A more detailed description of the Dutch regulatory interest rates is given in section 3.7.

The consensus of the above is that both internal and external factors play a role in determining the coverage ratio, or performance of pension funds. Besides the political influence fund managers have on easing restrictions such as riskiness or actuarial interest, the most

significant instrument towards positive return on investments is assumed to be the influence amendable internal factors. Therefore, in the model presented later on, the external effects of actuarial interest are assumed to be incorporated in the final model (time dummies), and any deeper investigation on this topic is left for further research. The only externality focussed on is the financial crisis period.

(10)

10

2.2 Real Estate efficient portfolio dynamics.

As aforementioned, pension funds typically set a norm portfolio in order to determine the weighted amount of each asset category. Extensive literature has been published providing answers to the question on how to quantify the optimal allocation to real estate for

institutional investors. For example,Chun (2004) concludes that there are reasons for

investing in real estate, and moreover, that real estate investment in general helps to eliminate specific risk in the portfolio.In addition to this, Chun (2000) provides an explanation for the under- or overinvestment. Compared to the mean and variance of the real return, real estate is not highly correlated with other pension-plan liabilities, the author concludes. The consensus derived from these articles is that due to the low to negative correlation of real estate with other asset categories, a lower standard deviation of the portfolio is achieved, which helps pension funds stabilize their return on investments while reducing the risk of a sudden decline in coverage ratio (Bikker, 2012). Key to this insight is that an optimal mix of assets is

generated, as any added asset diversifies the portfolio. This optimal mix is widely argued upon by many authors. For instance, it was already in 1952 that Markowitz introduced an algorithm which identifies an all efficient portfolio. Also, as Friedmann (1971) concludes that general investors should hold real estate portfolios rather than stock portfolios given their higher return, the eagerness towards optimal allocation is expected to be even stronger. However, changing regulation like the introduction of Basel III for banks and a negative market outlook in times of crisis make institutional investors like pension funds change their exposure to real estate on a regular basis (DNB statistics4). This particularly holds for indirect participations in real estate funds which can be sold in a matter of seconds. Not only in indirect, but also in direct real estate, diversification can occur on an international scale. Eicholtz (2000) concludes that international diversification works better for real estate than it does for stocks and bonds. However, segmentation of real estate in direct and indirect makes defining the optimal amount a complex task, which can be derived from the literature, as Chun (2004) and other authors conclude that the widely used mean-variance model to

estimate the efficient level of real estate investments does not work well enough compared to a surplus-risk-return model in an asset-liability framework. The authors conclude that

optimally, an allocation towards real estate of 6% to 10% works best in an asset-liability framework. Another advantage, according to the article, is that real estate is benefitted from when it is most needed. Due to the low correlation with other asset categories, real estate

4_Link: http://www.statistics.dnb.nl/en/financial-institutions/pension-funds/supervisory-data-on-pension-funds/index.jsp accessed on December 17th₂₀₁₄

(11)

11

performs when the consumption is low. Also, especially for pension funds, the conclusion that real estate returns are equally predictable as stock returns can help pension funds to improve on their long-term horizons, and possibly prevent pension funds from ending up with a below-100% coverage ratio.

Changing exposure towards real estate on a regular basis requires the asset category to be liquid. However, real estate is seen as the most illiquid asset class in a portfolio (IVBN, 2010). This is why pension funds usually invest more in indirect real estate rather than in direct real estate (DNB statistics). And as the recent period of turmoil has shown, a pension fund might get ‘undercovered’ where the value of the investments have declined to a level lower than the level of long term pay-out obligations the fund has. This caused pension funds to cut back pension payments to pensioners.

As a similar liquidity risk existed for banks, where the difference is that banks became less able to provide mortgages to consumers, the Bank of International Settlements added to the existing regulation to banks that their investments need to be more liquid in the Basel III banking supervision regulation rule which was agreed upon in 2011. Since the increased liquidity pressure, pension funds have not been able to increase their exposure towards direct real estate (see figure 6, data from CBS4_{). However, as mentioned earlier, theory suggests a}

higher real estate allocation, due to low volatility and low correlation with other asset classes. Ennis (1991), for instance, concludes that the allocation to real estate in pension funds’

portfolios should be around 10 to 15 percent in order to derive optimal risk weighted return benefits. These authors however focus on the United States, which is a market under different regulations compared to the Dutch market (Ezra, 2011). On the one hand this seems to make the comparison rather difficult, but the characteristics of real estate remain comparable. On the other hand, financial assets are allocated among an institutional investor’s portfolio using modern portfolio theory (MPT), which applies the risk-return trade-off to measure portfolio weight. Real estate is an asset class to incorporate in an institutional investor’s portfolio, but MPT cannot be used here for three reasons:

First, real estate is traded infrequently and usually in private markets. This makes the returns on real estate not normally distributed over time, and therefore basic OLS regression analysis will not be useful. Second, real estate is illiquid. Among other non-variance factors, illiquidity brings along a problem when wanting to be included in portfolio theory: the inability to capture risks in the variance in a mean-variance analysis. Last, real estate brings along higher

(12)

12

transaction costs. While this is also the cause for the infrequent trading, it also has an effect on the willingness to incorporate real estate in the portfolio of pension funds, as there can be noticed a shift towards liquid assets in the portfolio.

Also, as Cheng et al (2013) mention, there are several studies which show that real estate allocation should be around 15 to 40 percent. They - among others - refer to Hartzell et al. (1986), Fogler (1984), Firstenberg et al. (1988), and Hudson- Wilson et al. (2005), by directly applying modern portfolio theory. This raises the question whether the model that is used is consistent.

The characteristics mentioned above are the reason why, according to Brown (2000), real estate cannot be quantified in a portfolio using a simple risk-return model, or the Modern Portfolio Theory (MPT). Instead, a more sophisticated model needs to be used coping with the specific characteristics of real estate mentioned above. Cheng et al (2013) find the solution to this drawback of MPT in an adapted model that takes into account the three special

characteristics of real estate of heterogeneity, illiquidity and transaction costs. Their conclusion is that their model is more in line with the real percentages of allocation, but it must be seen as a first step in the right direction of optimal allocation of real estate, and some mathematical inconveniences are surpassed by making simplifying assumptions.

Despite the problematic determination whether real estate should account for a larger or smaller share in the portfolio of institutional investors, the fact that this research focusses on pension funds is also an addition to the existing literature. Pension funds operate in an institutional setting, meaning the investment management is professional and it faces less protective regulation and should protect themselves against downturns. This institutional setting may cause pension funds to hedge their portfolios by investing in derivative financial instruments which is also a real estate investment.

2.3 Real estate liquidity problems

According to the literature mentioned above, it cannot be made clear what the ‘justified’ allocation of real estate actually is. Chun et al. (2000) noticed this as well and stated its

dependency on the framework that is used, and also the holding period of the real estate. What the authors - among other authors, like Brounen et al. (2010) - use is the asset-liability

framework, in which the holding period is accompanied by the value development of the asset. Considering pension funds, they have liabilities that are unique to any kind of institutional investor. Mainly given the participants pay-out liabilities when they reach

(13)

13

retirement age, a pension fund needs to be able to fulfil this liability. Therefore, the fund needs to estimate their return on investments needed for stable operation. This liability introduces a problem that a 2012 Ortec Finance report stated5_{, and has become increasingly}

important since the recent (2008) crisis. This report learns that pension funds are under increasing pressure by the authorities to be able to quickly liquidate assets in order to fulfil liability problems if necessary. This was already the case before 2008, but the liquidity

pressure has increased significant since the crisis. Because real estate is – as stated earlier – an illiquid asset, pension funds are less able to fulfil this need of quickly liquidating this type of asset. The result is that pension funds tend to lower the percentage of real estate in their portfolio on purpose. Moreover, the argument that real estate – among certain securities as well – is yielding significant value to the pension fund in the long term, has been countered by the focus on the shorter term because of lowered coverage ratios followed after 2008 (DNB). This raises the question whether the crisis period made the pension funds react to the liquidity problem in an exaggerated way by losing focus on the longer term and therefore decreasing allocation to real estate. Theory does not provide evidence that the liquidity-based-policy of fund managers actually caused the below-100% coverage ratios of pension funds. After all, there are externalities, which are mentioned earlier, that had their effect on the performance of pension funds all over the world. Changes in the actuarial interest rate, legal limits to

investment decision and macroeconomic shocks are examples of these factors. But future research could prove that had the funds chosen to maintain or increase their real estate investments, whether they would have the problem of a shortage in coverage ratio.

2.4 REITs

As real estate as an asset class is usually illiquid as mentioned above, this obviously holds for direct real non-listed real estate more than for publicly traded, indirect real estate. Chun (1991) also notices this as the author uses a Real Estate Investment Trust (REIT) index in order to analyse the benefits of the asset-liability framework. The importance of allocation towards direct versus indirect real estate can, for this research, not be ignored. REITs form a significant subcategory in total real estate investments, and they form a stable addition to the institution’s portfolio (Kallberg, 2000). As the optimal allocation is on average around 20% real estate, Ennis (1991) concludes that using continuous auction market trading instead of

5_{Ortec Finance report “Trend vastgoed bij pensioenfondsen” August 2012, provided by Ortec} Finance

(14)

14

appraisal based returns, and using expected returns derived form a simple equilibrium model instead of using historical returns, the optimal level of investment should be 10-15%. Fisher et al. (2005) examined the performance of real estate portfolios and found that usual portfolio analysis does not capture the relationship between marketable securities and real assets in a portfolio.

2.5 Economic shocks

What is also noticeable, is that allocation of assets within pension funds continually changes, but this change becomes larger when a major financial shock occurs (An et al., 2013). They also say that labour unionization, or sponsor incentives as maximizing tax benefits influence asset allocation. But most importantly, it is stated that when underfunding is being recovered from, which was the case after the financial crisis, an aggressive investment strategy is

usually implemented. This is mainly due to the higher ‘risk budget’ when the coverage ratio is high.

Pension funds can thus bare the downside risk, while sustaining large upward potential. It needs to be noted however that this U.S. literature can differ for the Dutch market. When looking at Dutch funds, for instance Ahold Pension Fund, their investment policy is

analogous to the theory of ‘risk-budget’ mentioned above. A high coverage means a higher risk budget which enables the possibility to increase the allocation towards higher-risk asset classes. However, when looking at the Shell Pension Fund for example, the complete opposite can be found. The Shell fund decreases the level of risk in the portfolio if coverage ratio is high (IPE, 2014). The managers of this fund did so in order to act in line with their initial investment policy. This example shows that besides of the economic situation, fund managers have to cope with other factors as investment policy and participants’ votes as well.

Considering the market of research for this thesis, the Dutch market is chosen not only for convenience purposes, but also because of pension funds’ size in the Dutch economy. As mentioned earlier, the Dutch pension funds are in relative terms as large as the total economy. This is nowhere else to be seen in the world, and makes investment decisions more impactful towards the pensioners.

2.6 Real estate diversification: direct versus indirect real estate

In terms of the low volatility characteristics of real estate and the changing allocation towards real estate at financial shocks, the suggestion that real estate works as a shock absorber in the portfolio is confirmed by van Gool (2010) in an article, showing that correlation of real estate with other asset classes is low to zero. It is not surprising that indirect, publicly traded real

(15)

15

estate does show some correlation with other Dutch stocks (0,6%), but correlation between direct real estate and Dutch stocks is virtually zero. Striking with the Shell fund policy, however, is the finding of this author that real estate is a ‘capital increasing’ asset, with a high return/risk ratio. This would imply that when coverage ratios are low, real estate share should be increased in order to regain capital levels and assure stable return.

But not only does real estate act as a shock absorber in the portfolio of a pension fund. By looking at real estate on a firm level, the collateral that real estate provides gives firms a relatively cheap opportunity to leverage debt. Because of its tangibility, real estate succeeds in giving firms more debt capacity (Giambona, 2009). In fact, the effect of real estate on the leverage of firms is twice as large as other tangible assets leverage capabilities. Therefore, it can be concluded from the literature above that real estate as an asset class has unique characteristics that need to be taken into account when incorporating in any investor’s portfolio, pension funds have unique obligations that need to be taken into account when comparing to other institutional investors, and the crisis period of 2008 onwards has

influenced both the real estate asset class and pension funds investment behaviour. Also, the literature shows that pension funds do not always invest in line with theory, due to policy characteristics. This makes investment decisions sometimes irrational, hard to predict, and should be taken into account when further researching pension fund portfolios. It is

nonetheless worth investigating the effects of the real estate investments on the coverage ratios of pension funds in times of a global financial crisis.

2.7 Summary of hypotheses

As mentioned in the introduction, the aim of this research is to answer whether investing in real estate influences the coverage ratio of Dutch pension funds, by specifically focussing on the effect of the financial crisis and the type of real estate invested in.

That the real estate sector acts as a volatility-decreasing diversifier within the portfolio, is already confirmed by numerous studies, as has been mentioned in the literature review. However, whether this effect has weakened since the recent period of turmoil cannot be made clear from this literature, and therefore demands further research. Investigating what the effect of real estate is on the coverage ratio of Dutch pension funds puts forward the hypotheses that:

- Since the 2008 financial crisis and the changes in economic environment that followed, pension funds that did invest largely in real estate experienced different coverage ratios than funds who were less exposed to real estate.

(16)

16

- Dutch pension funds with a larger share of more volatile real estate have suffered a larger decrease in coverage ratio than funds with a low share in volatile real estate.

(17)

17

3. Data and Descriptive Statistics

3.1 Data

In order to quantify the relationship between real estate investments and the coverage ratios of Dutch pension funds, data is investigated on these two topics. Regarding the pension fund coverage, data on the yearly nominal coverage ratio is obtained, predominantly using the annual reports of the 30 largest Dutch pension funds. The nominal coverage is chosen in order to cover direct changes in the solvability of pension funds. A yearly rate is chosen due to standardized reporting use. Regarding the real estate data, records on several levels of real estate investments are used in order to gain insight in the effect of different types of real estate on pension funds’ coverage ratios. Real estate investment balance at year-end is used in order to correspond with the accompanying coverage ratio. Levels of real estate investments needed include direct, indirect total, indirect listed and indirect non-listed.

However, this data is only partly available from pension fund annual reports. Combined with data received from a large pension fund investment advisor in the Netherlands, the dataset was completed. Data of 30 Dutch pension funds obtained with portfolios of over

€250.000.000,- worth of real estate investments. Together, these funds have invested a total amount of €83.3 billion in real estate. Compared to the total investments in real estate of all institutional investors in the Netherlands, which amounts to €92.8 billion6_{, the 30 pension}

funds cover over 89% of the total institutional investor’s market as far as real estate capital at risk is concerned. These variables are obtained from the 30 funds for the years 2006-2013, leading to a total number of observations of 240 in a balanced panel. A more detailed description of the variables is given below:

Fundid i Unique number per fund for identification purposes. Namei Name of the pension fund.

Yeart Data origins 31 December of this year.

Coveragei,t Nominal coverage ratio of the fund matched for accompanying year. Resharei,t Share of real estate compared to the total portfolio size of the fund. Rei,t Total real estate investments of the fund in €mln.

Totinvi,t Total size of the portfolio in €mln.

Indirecti,t Percentage of indirect real estate investments compared to the total real

estate investments.

IndirListi,t Percentage of indirect listed real estate investments compared to the

total real estate investments.

6_{IVBN paper: M Mosselman (2013) “samenstelling en rendement van Nederlandse institutionele}

(18)

18

IndirUnlisti,t Percentage of indirect unlisted real estate investments compared to the

total real estate investments.

GDPt Dutch Gross Domestic Product for year t is used to correct for

macroeconomic circumstances in the main model.

TBILLt Dutch 10 year treasury bill rate generated from datastream used to

correct for actuarial interest pension funds need to calculate their funding ratios upon. As the actual actuarial interest is based on duration, the T-bill rate is assumed to be a proxy for the actuarial interest

These variables enable testing the main hypothesis by regressing the composition of the real estate portfolio to the coverage ratio of a fund. This is done for the years prior to - and after crisis. The result obtained is further compared to other specific characteristics of the fund’s portfolio. Prior to this analysis, Dutch Central Bureau of Statistics’ data is used in order to summarize the total portfolio composition of all Dutch pension funds aggregated for the use and presentation of descriptive statistics. This is quarterly data ranging from 2005 until 2013, containing both the percentage wise as the euro specified allocation of the total investment portfolio, divided into eight categories. This chapter is structured as follows: in section 3.2, the cumulative portfolio of all Dutch pension funds aggregated is described. The section is divided into two categories. Section 3.2.1 describes the Dutch pension fund portfolio dynamics in a way that creates a view on the relative size of the real estate part of the

portfolio. Section 3.2.2 Describes the Euro amount of the real estate investments and had the aim to give a view on the year-on-year development of the investments of Dutch pension funds in real estate. Section 3.3 describes the dataset used for this research. It comprises the 31 pension funds of which data is gathered in order to arrive at the results of this research.

3.2. Descriptive Statistics

In this section, the descriptive statistics are given firstly for the total of all Dutch pension funds regarding their real estate investments. This provides an insight in the point at which the Dutch pension funds are standing, and provides justification for the dataset used afterwards. After this, the section is narrowed down towards the 30 pension funds used to investigate the research question. This is more detailed data on not only the coverage ratio, but also on the different levels of real estate a pension fund invests in.

3.2.1 General Descriptive Statistics about Dutch Pension Funds, relative.

In this part, a brief insight is given into where the pension funds are standing at the end of 2013. At that time, Dutch pension fund portfolios existed of €968 billion of total investments, where at the beginning of 2005, this was €565 billion. The total portfolio size (prior to the

(19)

19

crisis) in 2007 reached €731 billion, after which it declined to €606 billion in the first quarter of 2009. It was already in the first quarter of 2010 that the pension funds aggregated portfolio totalled the height of 2007 again, so a rather quick recovery was made. It cannot be said, however, that the crisis was over at this point in time. Since the contributions to the funds are not taken into account here. So the size of the total portfolio has been changing significantly, and continues changing.

The composition, moreover, changed as well. The Dutch Central Bank keeps track of this data, and publishes it on its website, under the link Assets invested for risk of the fund (T8.9). To gain insight in the dynamics of portfolio allocation, a graphical overview of the total portfolio composition for all Dutch pension funds is given in figure 1, 2 and 3 for the fourth quarter of the years 2007, 2009 and 2012 respectively. These years were chosen because the relative allocation of assets moves within these boundaries. Percentages are rounded at a 3-decimal level, given the relatively limited relative bandwidth of movement.

 Figure 1: Summary portfolio of all Dutch pension funds in the final quarter of 2007 (in %)

Source: Supervisory databank of DNB Statistics7

As can be seen, the total share of real estate is around 11% at the beginning of 2007. As the theory supports, the share of direct real estate is lower than the share of indirect real estate. In order to compare this portfolio to the after-crisis portfolio, where the higher liquidity pressure, lowered interest rate and (overall) lowered coverage ratios have taken shape, two points in time have been chosen to show the medium term effect of the crisis measurements on the

7_{Figure made from data collected on}

https://www.statistics.dnb.nl/financieele-instellingen/pensioenfondsen/toezichtgegevens-pensioenfondsen/index.jsp, accessed on 16 May 2015

3,308% _7,959% 38,212% 2,292% 2,134% 2,095% 0,395% 43,607%

2007 Q1, total portfolio equals €666.7 bln

Direct Real Estate Indrect Real Estate Stocks

Private Equity Hedge Funds Commodities Other

(20)

20

pension funds’ portfolios. These periods are the fourth quarter of 2009 and the fourth quarter of 2012. See Figure 2 and Figure 3.

In terms of the real estate sector, these figures show that direct real estate has declined to almost half of the value of 2007, and indirect real recovered to near pre-crisis relative levels. Also note the growth of the total portfolio size in combination with the halved allocation towards direct real estate. Given this growth of the total portfolio and the increased volatility of the real estate sector, it is expected that coverage ratio pressure made pension funds react to the crisis in an exaggerated way by reducing the amount of real estate in the portfolio so much that the benefits of investing in real estate are being surpassed.

2,618% 7,496% 33,031% 2,849% 3,104% 1,529% 0,274% 49,098%

2009 Q4, total portfolio equals €655.2 bln

Fixed Income Securities

1,667% 7,932% 31,053% 4,807% 3,068% 0,108% -1,626% 52,990%

2012 Q4, total portfolio equals €862.1 bln

(21)

21

3.2.2 General Descriptive Statistics about Dutch Pension Funds, absolute

The crisis caused three effects to take shape in the real estate portfolio, according to van Gool (2010): a capital loss which declined the share of total real estate, less appetite for investing in direct real estate and more appetite for more liquid indirect real estate. These effects are represented in the aforementioned figures, but the effect in euros is more distinguishable in the following table:

 Table 1: Aggregate value of the real estate investment portfolio of Dutch pension funds in bln. euro compared to the total investment portfolio of Dutch pension funds.

2006 2007 2008 2009 2010 2011 2012 Direct Real Estate €19.8 €19.2 €17 €15.2 €9.9 €9.7 €9.1 Indirect Real Estate €54 €52.6 €45.2 €49.2 €64.8 €68.6 €77.2 Total Real Estate €73.8 €71.8 €62.2 €64.5 €74.9 €78.2 €86.3 Total Portfolio €709,96 €720,02 €622,77 €697,43 762,21 €817,34 €748,66 Source: DNB Statistics8

In Table 1, one of the main characteristics of the different types of real estate is put forward, namely the lagged effect of the performance of real estate in the portfolio. On the one hand, the table shows a rather late decline in value of direct real estate which does not show any sign of recovery. On the other hand, a rather quick recovery of indirect real estate

immediately following the dip of the crisis in 2008 can be noticed. Whether this has to do solely with capital loss, cannot be said regarding the table since it excludes the added and subtracted number of properties to and from the portfolio giving insight to conclude whether the cause of the movement in the portfolio is purely based on capital loss. Given the non-recovering value of direct real estate, it seems likely that the funds did not do much effort to bring the value of direct real estate back to the pre-crisis level. The bad economic outlook affecting the vacancy in the market and reluctantly behaviour of people mowing homes in that period can explain why the interest was limited.

8_{Table created form data published on}

(22)

22

3.3 Descriptive statistics main dataset

The data used for this research comprises 240 observations, namely 30 funds over an 8 year time span. The descriptive statistics are given in Table 2:

 Table 2: Descriptive statistics for the dataset used in this research

Variable: Mean: Std. Deviation: Minimum: Maximum:

Direct 0.278 0.365 0 1 Indirect 0.721 0.367 0 1 Indirlist 0.179 0.196 0 0.576 Indirunlist 0.532 0.318 0 1 Coverage 115.502 17.080 16.4 188.6 Reshare 0.110 0.065 0.006 0.462 Re 2109.558 5003.692 52 33647 Totinv 19467.43 46164.46 4061.2 322894 GDP 624194.1 22477.61 573444 642929 Tbill 3.063 0.885 1.56 4.34

Over the complete sample period, and for all 30 pension funds in the data set, it holds that they have invested on average €2.1 billion in real estate. The differences in investments in real estate are significant: the fund with the smallest amount of real estate investments has €52 million, where the largest amount of real estate in a single pension funds amounts to €33.6 billion. On average, the funds invested 27.8% of their real estate investments in direct real estate, and 72.1% in indirect real estate. There are funds with zero direct, and zero indirect real estate, and in contrast, there are also funds who only invested in direct real estate, where there are also funds who only invested in indirect real estate. It also holds that, compared to the aggregate portfolio, the data of these 30 funds is reflecting the truth by means of the 10% investment in real estate. Where the total investments average €19.5 billion, the real estate amounts to roughly 10% of this number. What is also interesting, is that there is one fund that had a year in which it had a total amount of euros invested that was more than half of the average Dutch GDP in the sample period. This shows the relative importance of asset

allocation of Dutch pension funds. The treasury bill-rate of Dutch 10-year government bonds, used as a control variable in the model of this research, collapsed by around 64% over the sample period. As described in the theory, the performance of pension funds is influenced, among other factors, by the interest rate. It needs to be noted that the interest rates Dutch

(23)

23

pension funds have to use differ from the T-Bill variable. In fact, for every duration of a bond there is a different interest rate applicable. These duration differ from 1 to 60 years. For the sake of simplicity, however, the T-Bill variable is assumed to be representative for the total actuarial interest rate that needs to be corrected for in this research.

For GDP it is chosen to derive data from the Central Bureau of Statistics, (CBS) and the T-Bill rate data originates from the Dutch Central Bank (DNB) database. Both are publicly accessible.

3.4 Descriptive statistics main dataset: a year-on-year approach

Not only does the data show a dip in the real estate portfolio, but also in the development in coverage ratio for the largest Dutch pension funds over the period 2006-2013 (Figure 4). Because it takes time to invest in (direct) real estate, a change in portfolio can’t be made quickly. Data shows that over a period of seven years, pension funds have not been able to recover towards increasing coverage ratios, and multiple funds show a liability-covered asset wealth share of less than 100% for more than three years.

 Figure 4: Average year-end coverage ratio of the 30 largest Dutch pension funds from 2006-2013 (in %).

However, the bars from figure four also show a moderate upward movement towards 2013. In line with macro-economic recovery, recovery of the pension funds’ capital at risk is

increasing analogously. The pace of decline of the coverage ratio over the data period is around -3.99%-point per annum.

0 50 1 0 0 1 5 0 m e a n o f c o v e ra g e 2006 2007 2008 2009 2010 2011 2012 2013

(24)

24

Where the dip in coverage ratio was followed by a recovery from 2012 onwards, the

investments in real estate suffer a continuous decline, shown in Figure 5. During the crisis, the share of direct real estate declined faster than indirect real estate. There are two possible reasons why this is the case. First, the institutional investors could have anticipated the declining real estate market, and considering the time it takes to perform a transaction on the direct real estate market, they started divesting this direct real estate. On the other hand, the decline in share of direct real estate could be caused by capital loss solely. The latter seems unlikely due to the ROZ IPD report mentioned in the introduction. This report states a +10% to -10% movement in the global direct real estate market from 2006 to 2012. Exact data on the number of properties in the portfolio is lacking, but the bar figures from Table 5 support the theory that fund managers appreciate the liquidity aspects of indirect real estate and likely lowered their exposure to illiquid direct real estate, even though the volatility, and therefore riskiness, of indirect real estate is higher.

 Figure 5: On the left, the average share of direct real estate, and on the right the average share of indirect real estate in the portfolio of Dutch pension funds compared to the total investments.

3.5. Co-Movement

What is also interesting is the co-movement in the data. According to Figure 6, it seems that portfolios of all observed funds roughly follow the same pattern, despite the fact that

individual funds show significant differences within their portfolios. A reason behind this correlation can be the external management of the funds’ portfolios. Investment management firms who actively manage the pension funds’ capital follow roughly use the same investment model and funds probably have roughly the same risk appetite because they have to follow the same regulation. Moreover, each fund’s mandate is to perform just above target coverage set by the government. 0 .0 2 .0 4 .0 6 m e a n o f d ir s h a re 2006 2007 2008 2009 2010 2011 2012 2013 0 .0 2 .0 4 .0 6 .0 8 .1 m e a n o f in d ir s h a re 2006 2007 2008 2009 2010 2011 2012 2013

(25)

25

Figure 6, derived from DNB (Dutch Central Bank) statistics website7_{shows the total coverage}

development for all the Dutch pension funds, combined with the levels of investment in direct and indirect real estate. This shows the impact of regulation and recommendation on pension funds where funds have been encouraged since 2008 to restore their buffer for losses in the near future.

 Figure 6: Dutch pension funds have not been able to recover from undercoverage substantially until 2013. Moreover, the real total assets of the funds grew steadily towards in excess of one trillion euro (DNB Statistics7_{). On the left axis, the euro amount invested in real estate is}

displayed (in billions), corresponding with the blue and orange dots. On the right axis, the coverage ratio is displayed, corresponding with the grey dots.

Interestingly, the figure shows that coverage ratio dos not move up together with the strengthened buffer of the ‘average’ pension fund. In fact, it seems that, mainly in the years prior to the crisis (<2009), when the real estate share in the portfolio roughly maintained their level, the coverage only moves down. And only a ‘stable’ coverage ratio is obtained in post-2009 period as the pension funds’ assets reach record-high levels of investments. A reason for the rather low correlation between coverage ratio and (indirect) real estate could be in the actuarial interest rate. Because this rate is used for pension funds to calculate their coverage ratio and the rate has its origin in the capital markets interest rate, it influence the direct real estate only slightly. However, indirect real estate comprises a fraction which is listed on exchanges and is therefore more vulnerable to changes in the capital markets interest rate. Clearly, Figure 5 shows a relationship between indirect real estate, but given the slope of the grey and orange dotted lines, one might have expected this relationship to be stronger.

0% 50% 100% 150% 200% 0 20.000 40.000 60.000 80.000 100.000 2007Q 1 2007Q 2 2007Q 3 2007Q 4 2008Q 1 2008Q 2 2008Q 3 2008Q 4 2009Q 1 2009Q 2 2009Q 3 20 09 Q 4 20 10 Q 1 2010Q 2 2010Q 3 2010Q 4 2011Q 1 2011Q 2 2011Q 3 2011Q 4 2012Q 1 2012Q 2 2012Q 3 2012Q 4 2013Q 1 2013Q 2 2013Q 3 2013Q 4 2014Q 1 2014Q 2 2014Q 3 2014Q 4

Accumulated bln. euro-amount of real estate invested in by

Dutch Pension funds (Quarterly, 2007-2014)

(26)

26

3.6 Treatment and control group descriptive statistics

A difference-in-differences model is used in order to conclude on the purpose of this research, comprising a treatment group and a control group. Although the actual model is elaborated on in the methodology section, this section comprises the descriptive statistics of both the

treatment and control group. In short, the treatment group consists of those funds which have had more than the average amount of real estate in the portfolio prior to the crisis, that is, 10% of real estate in 2006.

 Table 3: Summary statistics of the treatment group, followed by the control group.

Treatment group Mean Std. deviation Minimum Maximum

Number of funds 19 Direct 0.275 0.369 0 1 Indirect 0.722 0.373 0 1 Indirlist 0.184 0.205 0 0.576 Indirunlist 0.514 0.319 0 1 Coverage 114.65 17.415 84.7 171.3 Reshare 0.129 0.073 0.023 0.462 Re 2913.042 6149.684 85.2 33647 Totinv 24977.15 57237.31 1061.2 322894 Control group Number of funds 11 Direct 0.281 0.360 0 0.930 Indirect 0.719 0.360 0.069 1 Indirlist 0.170 0.182 0 0.574 Indirunlist 0.527 0.319 0 0.986 Coverage 116.91 18.52 88.0 188.6 Reshare 0.078 0.0284 0.006 0.153 Re 730.853 494.109 52 2253 Totinv 10013.258 6204.848 2496.7 32269

From Table 3, it can be derived that the treatment group - which is the group of pension funds which had 10% or more invested in real estate in the portfolio in 2006 – comprises 19 funds that had divided their real estate investments in 27.5% of direct real estate, and 72.2% of indirect real estate. These percentages are relative to total real estate portfolio. These ‘high real estate funds’ had an average coverage ratio of 114.65% within the sample period of 2006-2013. Another important descriptive is the difference in absolute size of the real estate

portfolio between the treatment group and control group. Despite the fact that the groups are split at the point of 10% real estate allocation, the treatment group held on average 12.9% real estate with an average of €2.91 billion where the total investment portfolio was €24.98 billion.

(27)

27

The control group, on the other hand, held an average of 7.8% in real estate with an average amount of €730 million in real estate, and a total investment portfolio of €10 billion.

Interestingly, the funds with a low percentage of real estate are also the funds with the lowest euro amount of investments. Funds that hold more than 10% real estate in their portfolio are on average 2.5 times larger regarding their total amount of capital invested. The reason for this is that, as mentioned in section 3.3, the largest Dutch pension fund consists of half the Dutch average GDP over the sample period. Furthermore the control group consists of funds which had less than 10% real estate in the portfolio in 2006, had an average of 28.1% direct real estate, and an average of 71.9% indirect real estate. So the funds that had less real estate in 2006, had relatively more direct and less indirect real estate. The coverage ratio of the control group is on average a slightly higher 116.91%.

3.7 Interest risk of pension funds: Term Structure and Actuarial interest rates

One of the main factors influencing the coverage ratios of Dutch pension funds is the interest at which the future cash flow and liability expectations are discounted, Part of the data set for this research contains a proxy for this interest rate, and this rate is being controlled for in the model.

To arrive at the interest rate Dutch pension funds discount their future expected earnings in order to calculate whether they can still maintain pay-outs to their sponsors, the Dutch Central Bank used a fixed rate of 4% until 2006 (van Ewisk, 2005). From 2006 onwards, a new system of actuarial interest rates was implemented, so that the coverage ratio of pension funds was not only influenced by the asset return performance, but also the liabilities were valued based upon the interest rate defined from by term structure to obtain a ‘fair value’ valuation method for both assets and liabilities. However, this new approach impacted the coverage of pension funds negatively in times of low interest rates. Therefore, in 2012, the Dutch Central Bank implemented a discounting method which is related to the Ultimate Forward Rate (UFR). This rate comprises a mix of forward rates based on duration of between 1 and 60 years duration, and a fixed forward rate. The danger in this way of regulating is in the fact that although this rate is based on the market, a stabilizer in implemented with the UFR, making the coverage ratio’s rise, even is a declining market. As part of a newly adopted regulation regime, the Financial Assessment Framework (FTK), pension funds were obliged to use this UFR.

(28)

28

Because of the recent financial crisis and its effects, pension funds faces the problem of undercoverage. The Dutch government saw the persisting undercoverage of pension funds as a reason to implement a new version if the FTK, the FTKII. The Dutch Pension Fund

Federation set out on which subjects this plan is different from the earlier plan. In this new Financial Assessment Framework, changed regulation is implemented regarding the pension fund’s ambition, the absorption of financial shocks, the absorption of rising life expectancy, a feasibility test, cost-covering premiums and the discount curve. The FTK2 framework also increases stability of the coverage ratio of pension funds, even in a stable stock market (CPB, 2013). However, as the new FTK2 is to be implemented in 2015, and therefore lies outside the scope of this research, the effect is noted but will not take part in the model.

(29)

29

4 Methodology

This section describes the methodology and accompanying model with its extensions used to investigate the research question and hypotheses. It is investigated what the effect was of the real estate investments of Dutch pension funds on their coverage ratios, thereby looking at the effects of the crisis and the different types of real estate (see introduction).

4 .1 Main Methodology.

In order to investigate these effects, an analysis is made of pension fund portfolio-data on their share of real estate investments. This data suits a differences-in-differences model best, thereby estimating the average treatment effect of having real estate prior to the crisis and the effects of it. The real estate data is therefore broken down in parts in order to measure the levels of direct and indirect real estate, both listed and non-listed, researching the relationship between the coverage ratio and the real estate portfolio share and its composition: both before and after the crisis.

A note needs to be made here in advance: besides the impact of real estate on coverage ratios, there will also be an impact of coverage ratios on the real estate investments of pension funds (reverse causality problem). After all, funds who face the problem of undercoverage are likely to adjust the portfolio towards recovery of the solvability of the fund. This problem of

endogeneity (meaning that part of the explanatory power of the model comes from within the model itself) could be accounted for by taking instruments to assess the unexplained part of the effect on coverage ratios that does not influence the investments of the fund and hereby proving the direction of the causality for this problem. However, there are not many

instruments one can use for this scenario, except for a change in policy that is equal for all funds and affects their coverage but not the composition of the portfolio itself. What needs to be used is an instrument that that gives a view of pre-crisis level real estate impact on

coverage ratios during the crisis. Since finding the right instrument to do an IV regression is outside the scope of this research, an IV approach is left to further research. It is assumed for this research that the reverse causality problem is accounted for in the lagged effect of real estate on the portfolio.

In order to arrive at a conclusion on the effects of the crisis and type of real estate investment on the coverage ratios of pension funds, there first needs to be made a distinction of which funds invested in real estate largely prior to the crisis, in order to create a treatment group. This needs to be done because then the question whether those funds that invested heavily in

(30)

30

real estate suffered less (or more) from the crisis than funds who did not invest heavily in real estate can be answered. To this extent, it is chosen that funds who have invested more in real estate than the average amount of 10% - as can be derived from the general descriptive section 4.2 – are heavily investing in real estate. Funds which are below this level are assumed to be investing in real estate only slightly. A treatment group indicator variable is created accordingly which has the value one for funds who are above the average level of real estate investments in 2006, and zero for the funds which are below the average of 10%. The ten percent level is chosen because real estate represents on average 10% of the total portfolio of a Dutch pension fund, which can be derived from figures 1, 2 and 3.

After distinguishing these groups, in order to conclude on the effect of the crisis, a crisis indicator is added to the dataset which defines the before or after crisis period of the sample period. It is assumed in this research that the crisis stating point is in 2008, since end-of-year data is used. This indicator has the value one for the years 2009 until 2013, and zero for the period 2006-2008.

These dummy variables are accordingly interacted to the real estate characteristics of the fund to obtain the final treatment effect needed for this research. From this point onwards, a

response function can be defined which tells in more detail the path of recovery the pension funds travelled until 2013.

Two control variables are added to the model, namely the GDP and Tbill variables. As mentioned in the literature, there are external factors that influence the performance of pension funds. Considering the macro economy, an extension of the model is made by controlling for the development in GDP. Also, considering the actuarial interest rates, the Dutch T-bill rate is controlled for, which is assumed to be the risk free basis for the actuarial interest rate.

Also, these data will contain a significant time trend in the real estate value within the portfolio. Making use of yearly dummies (τt), this trend is eliminated. In combination with

fund fixed effects (αi), the first hypothesis can be tested empirically.

As the control variables GDP and Tbill do not vary between funds, but only over time, they will be omitted from the regression if the regression already captures fund dummies and time dummies. These capture both time trend effects and fund effects within the model. Therefore, it is assumed for this research that both the GDP as the interest rate have the same influence on both the treatment group and the control group. That is, regressing them in the model will

(31)

31

cause multicolinearity problems, so the variables GDP and T-Bill have to be taken into account when not controlling for any other time trends.

Based on the outcome, another dummy is introduced and interacted with the treatment effect mentioned earlier. This dummy has the value one for funds that have invested in direct real estate, and zero otherwise. This gives insight in the second hypothesis of whether funds with direct real estate have suffered from the crisis less than funds that have no direct real estate in the portfolio.

Prior to addressing these issues, the main model, which is a classical difference in differences model, will look as follows:

𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑖,𝑡= 𝛽0+ 𝛽1𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑔𝑟𝑜𝑢𝑝𝑖+ 𝛽2𝐶𝑟𝑖𝑠𝑖𝑠𝑡+ 𝛽3𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑒𝑓𝑓𝑒𝑐𝑡𝑖,𝑡+ 𝜀𝑖,𝑡 (1)

Where Coveragei,t = Coverage ratio for fund i in year t

treatmentgroupi = Dummy for high real estate exposure

Crisist = Dummy for determining crisis period

Treatmenteffecti,t = Combined treatment effect for the treatment group in the crisis.

Where β0 is a constant, β1 comprises the treatment group indicator, so the group of funds that

had 10% or more real estate in 2006. The coefficient measures what the effect is of having a relatively large amount of real estate in the portfolio on the coverage ratio. β2 determines the

crisis period, which is set to 2009 until 2013 and this coefficient reflects the effect of being in a crisis period on the coverage ratio of a pension fund. β3 is the coefficient of interest, because

it comprises the treatment effect of the difference-in-differences estimation. This coefficient estimates whether the funds that did invest largely in real estate prior to the crisis, actually had a different coverage ratio during the crisis in comparison to the funds that were underinvested in real estate (the control group)

The effect of this regression, however, is based on binominal information in terms of

indicating the crisis period with a simple ‘yes’ or ‘no’. Therefore time dummies τt are added,

which will replace the crisis dummy for more detail in the regression and elimination of the time trend. The time trend will eliminate any unmeasured effect on the dependant variable

(32)

32

𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑖,𝑡= 𝛽0+ 𝛽1𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑔𝑟𝑜𝑢𝑝𝑖+ 𝜏𝑡+ 𝛽3𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑒𝑓𝑓𝑒𝑐𝑡𝑖,𝑡+ 𝛽4𝐺𝐷𝑃𝑡+ 𝛽5𝑇𝑏𝑖𝑙𝑙𝑡+ 𝜀𝑖,𝑡 (2) Added: 𝜏𝑡 = The year-to-year time trend dummies in order to filter out a general time

trend form he data.

GDPt = Control variable to control for the effect of the Dutch gross domestic product

on the coverage ratio.

Tbillt = Control variable to control for the effect of the interest rate on the coverage

ratio.

Now, the 𝛽1 still comprises the same interpretation as in the previous specification, only

corrected for GDP and interest rate. Control variable β4 shows the effect of the economic

situation in the Netherlands on the coverage ratio of Dutch pension funds, as this is, according to the theory a significant impactor on the performance of Dutch pension funds. The second control variable 𝛽5shows the effect of the interest rate on the coverage ratio of Dutch pension

funds.

Additionally, in order to control for any remaining unobserved time-constant omitted

variables, fund fixed effects are added. It is assumed from this point onwards, that the control variables GDP and Tbill are incorporated in the fund fixed effects, as the control variables have the same effect on every fund (because it is a proxy, in reality the rate is different for each fund). If both fund fixed effects and the two control variables would have been presented in the same regression, the control variables would have been omitted from the regression. This results in specification 3:

𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒_𝑖,𝑡 = 𝛼_𝑖 + 𝜏_𝑡+ 𝛽₃𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑒𝑓𝑓𝑒𝑐𝑡_𝑖,𝑡+ 𝜀_𝑖,𝑡 (3) In terms of the fixed effects, on the one hand the fund fixed effects can be interpreted as any unmeasured effect giving the fund a certain coverage ratio that does not vary over time. Examples of these fixed effects are size of the fund, any fixed allocation scheme, or a risk appetite that is higher for one fund than for the other (and fixed over time). To this extent, fund dummies are created in order to control for these unmeasured effects. On the other hand, time fixed effect are effects equal for every fund, but vary over time. An example here is the development in actuarial interest, which directly affects the way the coverage ratio is

measured, is fixed for every fund and varies over time. To this extent, table 4 shows the year dummies that control for these linear time trends.

Investment in real estate and the effect on coverage ratios of Dutch pension funds