1
Master thesis
Search for Yield in the Dutch Pension Sector
Msc. Economics
Monetary Policy & Banking
Thesis internship at De Nederlandsche Bank
Anja van Driel
Student number: 11276908
Supervisor: D.H.J. Chen
2
Abstract
This research investigates (a) the existence of search for yield in the Dutch pension sector and (b) how search for yield differs across pension funds. Based on the definitions of Rajan (2006), Yellen (2011) and ESMA (2014), I define search for yield as: “the incentive to seek for higher expected returns when interest rates are low through investing in assets with higher credit risk, interest rate risk or liquidity risk”. For the empirical tests a balanced panel of 185 pension funds over 40 quarters in the period 2007 Q1 until 2016 Q4 is used. I use a fixed effects regression model to prove the existence of search for yield in the Dutch pension sector. The dependent variable reflects holdings of different asset categories and the independent variable is the risk-free rate. The effect of the risk-free rate on asset holdings is quadratic. At low and declining rates pension funds increase risk-taking behavior, but the effect diminishes when rates become higher. Pension funds mainly increase credit risk and less significantly liquidity risk when searching for yield. Search for yield is more severe among larger funds. No significant effect of the average age of pension fund participants and a fund’s funding position on search for yield has been found.
3
Table of contents
Abstract ...2 1. Introduction ...4 2. Literature Review ...6 3. Institutional Framework ...8 4. Data ... 10 4.1 Descriptive statistics ... 11 4.1.1 Credit risk ... 144.1.2 Interest rate risk ... 17
4.1.3 Liquidity risk ... 18
5. Methodology ... 20
6. Empirical results ... 24
6.1 Regression of the baseline model ... 24
6.2 Extension of the model ... 27
6.2.1. Robustness ... 32 7. Conclusion ... 34 References ... 37 Appendix 1 ... 40 Appendix 2 ... 43 Appendix 3 ... 44 Appendix 4 ... 45 Appendix 5 ... 46 Appendix 6 ... 47
4
1. Introduction
After the global financial crisis, the ECB has been decreasing the interest rates to boost inflation and economic growth. One of the channels through which monetary policy affects the real economy, is the
risk-taking channel (Borio & Zhu, 2012). The low policy rates lower the public’s expectation of future
policy rates, which reflect in lower long-term interest rates. The low-interest-rate environment in the Euro Area may have incentivized financial institutions to take more risks. Rajan (2006) labelled this phenomenon as search for yield.
Especially for pension funds this search for yield is relevant. Low interest rates directly affect their balance sheet. Liabilities increase, because pension funds use a discount rate linked to the swap curve to discount their future cash flows. On the asset side, return on different assets may decrease, because low interest rates typically reflect lower growth (Franzen, 2010, Antolin et al., 2011). This might result in a worsening funding position for pension funds. Therefore, the current low-interest-rate
environment in Europe and the Netherlands might be an incentive for pension funds to increase risky investments, with a higher expected return. Especially for Dutch pension funds, increasing risk in order to achieve higher expected return can be very challenging. Dutch law prohibits Dutch pension funds in a funding shortage to increase their risk profile. Therefore, asset managers have limited options when they want to improve the funding position of the fund even though search for yield incentives are high.
Rajan (2006) defined search for yield as “the incentive to search for risk when interest rates are low”. Yellen (2011) gives a more specific definition: “(But) a sustained period of very low and stable yields may incent a phenomenon commonly referred to as “reaching for yield”, in which investors seek higher returns by purchasing assets with greater duration or increased credit risk.” The European Securities and Markets Authority (ESMA) complements that search for yield mainly takes place through investments in assets that “carry higher risk, longer maturities or lower liquidities” (ESMA, 2014). Combining these definitions, I define search for yield as: “the incentive to seek for higher expected returns when interest rates are low through investing in assets with higher credit risk, interest rate risk or liquidity risk”.
Search for yield is not necessarily harmful. When pension fund successfully obtain higher returns, they can (partly) close the funding gap and prevent pension cuts. However, in a worse scenario when too much risk is taken and losses on investments occur, funds might even worsen their funding position. In this thesis, I will examine the following questions:
1. Does search for yield exist in the Dutch pension sector? 2. Does search for yield differ across pension funds?
5 This research question is relevant for several reasons. First, a search for yield involves investments in riskier assets, which exposes the financial sector to greater risk and could increase systemic risk (see e.g. Antolin et al., 2011, Rocha et al., 1999, Claeys & Darvas, 2015). Second, search for yield can create financial instability, because bubbles can be created by overinvesting in risky assets (Acharya & Naqvi, 2016). Proof of the existence of search for yield in the pension sector could thus be used to implement macroprudential policies, to prevent the further build-up of systemic risks. Third, this paper contributes to the existing literature on the subject, by executing an empirical research for Dutch pension funds on search for yield, which has not been done before. This is especially relevant because the Dutch pension sector has many defined benefit (DB) pension funds. DB pension funds are directly harmed by a low-interest-rate environment, because a lower discount rate increases their liabilities. Especially when benefits are fixed, pension funds might have an incentive to search for yield.
Furthermore, the Dutch pension sector is relatively large, accounting for almost twice Dutch GDP. In 2015, the Dutch pension sector as a share of GDP was the largest among the OECD countries (OECD, 2017). Because of its size, the financial health of the Dutch pension sector is important for financial stability in the Netherlands.
I will use a dataset consisting of 185 pension funds over 40 quarters. The panel is balanced. Data for most variables was available for the period 2007 Q1 till 2016 Q4. In order to prove the existence of search for yield over this period, I will examine shifts of investments to riskier asset categories. Furthermore, I will investigate the impact of a decreasing risk-free rate on holdings in a specific asset category (fixed income rated AAA, AA, A, BBB, <BBB, or not rated, equity and illiquid assets), by estimating a fixed effect regression model. This methodology is in line with Becker & Ivashina (2012), who examine reaching for yield in the U.S. bond market by insurance companies.
In order to answer the second question, I included variables in the regression that represent pension fund characteristics. These variables are fund size, average age of fund participants and the fund’s funding position. I interact these variables with the risk-free rate to see how search for yield differs across pension funds. Furthermore, I included a variable on the term structure of interest rates (slope of the yield curve) to control for variations in asset holdings that are caused by changes in long term interest rates.
The main results are that search for yield indeed exists in the Dutch pension sector. This is shown by shifts in fixed income investments to lower rated categories and equity. In the regression model, a positive significant coefficient on the risk-free rate for AAA-rated assets and a negative significant coefficient for lower rated fixed income is estimated. The coefficient on the risk-free rate is also negative in the equity-models. The effect of the risk-free rate on movements in those asset holdings is quadratic: it is stronger when the risk-free rate is low and fades out when the rate increases. This indicates that a decreasing risk-free rate increases risky investments especially in a low-interest-rate
6 environment, i.e. search for yield exists. The proof for increasing liquidity risk in a low-interest-rate environment is less convincing. The results in the baseline model shows significant search for yield in mortgage loans and indirect real estate, but the effect becomes insignificant when the model is
extended. The results for fixed income and equity are still significant after extending the regression model and performing some robustness checks. Risk-taking behavior of pension funds is lower when the slope of the yield curve flattens. Search for yield is more severe among large pension funds. This thesis is structured as follows. Section 2 gives an overview of the existing literature on search for yield and the risk-taking channel of monetary policy. Section 3 shortly discusses the composition of the Dutch pension sector in an institutional framework. Section 4 describes the data used for the analysis. Section 5 elaborates on the methodology and includes the econometric model which is tested. In section 6 I discuss the empirical result. Finally, I will discuss the implications of these results and give some concluding remarks.
2. Literature Review
Search for yield is broadly analysed in the literature as the risk-taking channel of monetary policy. Borio and Zhu (2012) argued that when a central bank changes its policy rate, the risk perception and pricing of banks is affected. Several empirical studies prove that banks in a low-interest-rate
environment take more risk. For example, Altunbas et al. (2011) who conducted a study with data from 15 countries for the period 1999-2009. Buch et al. (2014) also did an empirical study using a factor-augmented VAR model for the U.S. for the period 1997-2008. They found that banks in a low-interest-rate environment take on more risk, though this is mostly true for small domestic banks. For non-bank financial institutions, evidence on the risk-taking channel of monetary policy exists as well. A decrease in the policy rate increases the term spread, which makes additional lending more profitable. The income from interest increases, whereas interest payments decrease. Thus, capital appreciates, which in turn increases the risk-taking capacity of these institutions. Adrian and Shin (2010) elaborate on this idea, and develop a model in which they prove that the risk-taking channel not only holds for banks, but also for non-bank financial institutions. Beck et al. (2016) support this view, when they tested the response of financial intermediaries (Monetary Financial Institutions, Insurance Companies and Pension Funds and Other Financial Intermediaries) to the ECB’s monetary policy. In addition, some literature exists on risk-taking incentives for pension funds in a low-interest-rate environment. Because of longer duration of their liabilities compared to their assets, low interest rates are a problem for a pension fund’s balance sheet (BIS, 2011). De Nicolò et al. (2010) provide a theoretical argument that financial institutions like pension funds and insurance companies are therefore incentivized to search for yield. When because of low interest rates the yield on their assets is low, they need to invest in riskier assets to match the yield they promised on their liabilities. Antolin
7 et al. (2011) theoretically complement that this is especially true for pension funds that offer defined-benefit promises, since they are subject to a long-term commitment.
Besides low interest rates, a few other incentives for risk-taking are mentioned in the literature. First, the funding status of a pension funds affects investment decisions. Two opposing hypotheses appear in the literature: the risk-shifting hypothesis and the risk-management hypothesis. Rauh (2009) argues that for DB pension funds both risk-shifting and risk-management incentives are present. On the one hand, funds in financial distress might gamble for resurrection through investing in riskier assets like equity. On the other hand, if pension funds manage to avoid bankruptcy by taking more risk, the overall poor performance could result in too less liquidity for paying out pensions, which is an incentive to manage risk. Several empirical studies on these two hypotheses are done. Rauh has tested this several times, and finds support for the risk-management hypothesis. Firms in a poor funding scenario, invest significantly more in safe bonds than firms that are financially healthy (see e.g. Rauh 2009 and Rauh 2005). An et al. (2013) also find empirical evidence for the risk-management
hypothesis in their study among U.S. DB pension funds.
Second, according to the model developed by Bodie et al. (1992), an individual will move his investments to more conservative fixed income securities when he reaches retirement age. The underlying idea is that the most powerful asset of younger people is their future earning power by expended human capital. When one becomes older, human capital depreciates. Therefore, it is
profitable for younger people to invest in riskier assets whereas older people move their investments to (more secure) fixed income bonds. In line with this reasoning, the average age of pension fund
participants can also be an incentive to increase risk-taking behavior. When the average age is lower, pension funds could increase risky investments. On the contrary, when the average age of a pension fund’s participants is higher, it is more secure to invest in less risky assets.
There exists little empirical research on search for yield by financial institutions other than banks. The study of Becker & Ivashina (2012) proves for insurance companies that reaching for yield exists in the bond market. The main result of their research is that U.S. insurance companies choose among bonds within investment grade, those with the highest returns. Another example is the study of Choi & Kronlund (2015), who find that U.S. corporate bond mutual funds engage in a search for yield. They provide evidence that the result is stronger when the level and slope of the yield curve is low and the default rate is small, using a time series regression. Chodorow-Reich (2014) investigates the effects of unconventional monetary policy on financial institutions, among which private DB pension funds in the U.S. The research reports that after the global financial crisis, pension funds searched for yield, but only for a short period. I will use elements of these researches. First, like Becker & Ivashina (2012) I will estimate a regression model with the risk-free rate as an explanatory variable and holdings of specific assets by pension funds as the dependent variable. Becker & Ivashina (2012) use the CDS and
8 Treasury spread as the independent variable and insurers’ bond holdings as the dependent variable, and prove that an increase in the CDS or Treasury spread increases insurers’ bond holdings, i.e. search for yield. I will execute a similar method by showing that a decrease in the risk-free rate leads to an increase in lower rated asset categories and a decrease in higher rated asset categories. Second, Choi & Kronlund (2015) investigate the time series of search for yield and a cross sectional analysis, in order to find out what incentives there have been for mutual funds to increase risk-taking behavior. Like Becker & Ivashina (2012) they find that search for yield is mainly persistent in investment grade bonds. Choi & Kronlund focus on variables concerning the term structure of interest rates in their time series analysis and find that search for yield is lower when the slope of the yield curve flattens. In their cross sectional analysis, they add variables on fund age, fund size and the expense ratio. They find that funds that are larger, younger and have higher expense ratio search for yield to a greater extent. I will use this cross sectional element, but use different variables that are relevant for pension funds: average age of pension fund participants, the fund’s funding position and fund size. Third, like Chodorow-Reich (2014) I will use a measure that reflects the difference between actual and benchmark returns in order to determine the extent to which pension funds searched for yield. Unlike these three studies, I will study a quadratic effect of the risk-free rate so that I can distinguish between search for yield in a low-interest-rate environment and a period in which interest rates are high.
So, existing literature has provided evidence that the risk-taking channel of monetary policy exists for banks and non-bank financial institutions. Search for yield has been proved empirically for banks and a few other financial institutions (insurance companies, mutual funds, U.S. pension funds). This research contributes to the literature in several ways. First, it provides statistical evidence on searching for yield by Dutch pension funds, which has not been empirically proved before. Second, it shows that low interest rates are indeed one of the incentives for searching for yield by pension funds, as has been analysed theoretically by Antolin et al. (2011). Third, the empirical studies mentioned above all focus on the U.S. Like in the Netherlands, low interest rates increase the liabilities of American pension funds due to a smaller discount rate. Unlike the Netherlands however, American pension funds are allowed to discount their liabilities using expected returns. So, when expected returns on their assets are higher, liabilities decrease. Therefore, U.S. pension funds have a direct incentive to search for higher expected returns in a low-interest-rate environment. (Chodorow-Reich, 2014).
3. Institutional Framework
The pension system in the Netherlands consists of three pillars. The first pillar is a general old age pension (Algemene Ouderdomswet, AOW), which is a public flat pension for all Dutch citizens reaching retirement age. The second pillar consists of employer’s collective pension plans, in which employees save for additional pensions. The savings are managed by pension funds, which can have a Defined Benefit or Defined Contribution scheme, or a combination of those two (hybrid scheme). The
9 third pillar consists of private savings, which people can arrange individually in addition to the first and second pillar of the pension system (Ponds & van Riel, 2007). In 2016 there existed over 200 pension funds in the Netherlands that were under supervision of the Dutch central bank (De
Nederlandsche Bank, DNB). Total assets of these pension funds were around 1.371 billion euros in
2016 (DNB, 2017). This accounts for almost two times GDP in the Netherlands in the same year (CBS, 2017).
The investment process of pension funds can be summarized in the following steps. First the pension fund determines its strategic investment policy for the long run. Important in this phase are the goals the pension fund sets and its policy principles, to which its investment beliefs contribute. Next an investment plan is established, which is an application of the strategic policy in the short run. In this phase investment beliefs and economic vision play a role. Third, the investment plan is implemented. Last, the implementation of the investment plan is monitored and periodically evaluated. In every step regulations play an important role and should be taken into consideration (DNB, 2015a).
Dutch pension funds are subject to regulations, which are determined in the Pension Law. Part of the Pension Law describes a Financial Assessment Framework (Financieel Toetsingskader, FTK). The most relevant regulations in the FTK are the Required Own Funds (Vereist eigen vermogen, VEV), the Minimal Required Own Funds (Minimaal vereist eigen vermogen, MVEV) and the prudent person rule. The VEV requires that every pension fund has own funding. The law determines that a pension fund should prevent having less assets than the size of its technical provisions in a one-year-period, with a certainty of 97.5% (Pensioenwet, 2006). The VEV is calculated using a standard model which includes several risk categories. Those risk categories are interest rate risk, equity / real estate risk, currency risk, commodity risk, credit risk, technical insurance risk, liquidity risk, concentration risk, operational risk and active management risk. For every category the sensitivity of the asset portfolio to risk is measured. Next, the VEV is calculated by a predetermined formula, which accounts for
diversification of risk and correlation between the different categories. When a pension fund does not have the required VEV-level, it should implement a recovery plan. This plan should make sure that the pension fund will achieve the required level of the VEV in 10 years again.
The MVEV requires the minimum amount of own funding a pension fund should hold. The MVEV should be 4-5% of technical provisions. The exact amount depends on the extent to which pension funds are exposed to investment risks.1 A pension fund that does not meet the MVEV for five sequent years, should immediately take measures in order to return to the required MVEV-level. Those
1 A fund exposed to investment risk should have a MVEV of 4% of gross technical provision multiplied by the
ratio between gross technical provisions excluding insurance transfers and the gross technical provision at the end of the previous financial year. For a fund exposed to venture capital at demise, 0.3% of the venture capital multiplied by the ratio between venture capital excluding insurance transfers and the venture capital at the end of previous financial year is added. (More detail: Besluit financieel toetsingskader pensioenfondsen, 2006).
10 measures include pension cuts or immediate depositing, until the current funding ratio is at the
required MVEV-level. (DNB, 2015a).
The prudent person rule impacts the investment strategy of pension funds. The prudent person rule covers amongst others the following points:
1. The goals and policy principles, e.g. risk appetite, should be established;
2. The strategic investment policy is in line with the goals and policy principles of the pension fund;
3. The strategic investment policy is based on solid research (DNB, 2015a).
Pension funds with a funding ratio below the MVEV are not allowed to increase their risk profile. However, an increase in the risk profile caused by market developments is allowed. The most important measures used by DNB to test whether pension funds have unfairly changed their risk profile are the VEV and a feasibility test (Haalbaarheidstoets, HBT). The VEV is used as a short term risk measure and represents the risk profile of a pension fund. Pension funds are allowed to change risk-taking behavior, as long as they maintain the desired VEV. This means that pension funds are allowed to increase risk in one VEV-category, as long as risk in another category is decreased and net VEV remains equal to the level before the change. The HBT is used as a long term measure. Every pension fund should periodically take a HBT. Based on a stochastic analysis, the HBT provides insight in the (financial) health of a pension fund and the risks that the fund encounters in its policy.
4. Data
Pension funds in the Netherlands report their financial statements to DNB through a system called
eLine. The reported data is collected by DNB. All data on pension funds was obtained from these
eLine reports. Data was available for 212 pension funds. These pension funds are categorized in different supervision classes, which range from T1 till T4. T4 is the highest class, and is therefore subject to the most intensive (risk) analysis. 5 pension funds are classified as T4, 27 pension funds as T3, and 180 as T2. There are no pension funds classified as T1. The investments of the five T4-pension funds are more than 50% of total investments by all T4-pension funds.
I eliminated 27 pension funds from the dataset, because for these funds data missed for several quarters in the selected time period. Two of the eliminated funds were categorized as T3, the
remaining 25 funds were categorized as T2. The dataset thus consists of a panel of 185 pension funds. Data was available for 40 quarters (Q1 2007 until Q4 2016), so total number of observations is 7400. The panel is balanced.
I used quarterly data on the return on fixed income and stocks in order to determine the extent to which pension funds search for yield. This data is also reported in the eLine statements. Along with actual return, pension funds report a benchmark return. This benchmark return differs per pension
11 fund, but are usually based on indices like the MSCI world index or Bloomberg Barclays indices. In order to analyse excess return, I calculated the difference between the actual and benchmark returns. For all 185 pension funds I collected quarterly data on their investments from the eLine reports. To measure credit risk I collected data on investments in fixed income and stocks. The available data is measured in euros and reflects market value excluding derivatives. Total fixed income consists of government bonds (mature and emerging markets), index-linked bonds, mortgages, credits, and short term claims and cash. These assets are categorized into rating categories AAA, AA, A, BBB, smaller than BBB, or no rating, according to the Fitch rating scale. All categories exclude derivatives. Furthermore, I used data on total investments in stocks in mature markets and stocks in emerging markets. This data is measured in euros and reflects market value excluding derivatives as well. In order to determine the weight of a specific investment category in a pension fund’s portfolio, I divided the amount of an investment category by total investments by that pension fund in the same period. Other variables that were collected through the eLine reports are average age of pension fund participants; the funding ratio; VEV; MVEV; norm weight of asset categories; and data on interest-rate risk. Quarterly data on the independent variable risk-free swap interest-rate was also provided by DNB. Because data on the average age for 2016 was not yet reported by pension funds at time of this research, the variable average age is collected over the period 2007 Q1 until 2015 Q4.
For the other independent variables I used external sources. From Eurostat I obtained quarterly data on the zero-coupon yield curve spot rate with maturity 1 year and maturity 30 years. The yield curves are for AAA-rated Euro Area central government bonds. I subtracted the 1-year yield from the 30-year yield to obtain an estimate of the slope of the yield curve. More detailed information on all the variables in the dataset is attached in appendix 1. This appendix also contains some additional summary statistics that are not presented in this section.
One limitation of this dataset is that especially for the return on assets, no detailed data was available. That is, only data on the return on total investments in fixed income and total investments in stock was available. Second, pension funds determine the benchmark return themselves, which makes the variable not entirely independent and the same for each observation. Third, there is insufficient data available to properly measure interest rate risk among Dutch pension funds. Finally, another limitation is that the sample period includes the global financial crisis, which might affect the results. During and after a crisis investment preferences change, which might affect the extent to which investors search for yield. For example, investors become more risk averse (Guiso et al., 2013).
4.1 Descriptive statistics
Search for yield was defined as “the incentive to seek for higher returns when interest rates are low through investing in assets with higher credit risk, interest rate risk or liquidity risk”. Figure 1 below
12 displays the main independent variable, the risk-free rate. This rate is the one year maturity interest rate term structure that is derived from the swap rate on the interbank market and is approximately risk-free. Dutch pension funds use this rate to discount their liabilities (DNB, 2015b). From the figure it becomes clear that the risk-free rate has been decreasing after the financial crisis in 2008. A small increase occurred in 2011, which can be traced back to the increase in the policy rate by the ECB in April 2011. However, from July 2011 onwards, the ECB kept decreasing its policy rates and the risk-free rate followed a similar path. In June 2014 the ECB policy rate first became negative (ECB, 2017). The risk-free rate followed the same pattern and dropped below zero shortly after in 2015 Q4. The question that now rises is if due to the declining risk-free rate pension funds searched for higher returns by increasing credit risk, interest rate risk and liquidity risk.
Figure 1: Interest rate term structure for pension funds
Table 1 below presents summary statistics for the actual return on fixed income and stocks and the benchmark return on fixed income and stocks. The summary statistics are taken over the period 2012 Q1 until 2016 Q4, because data on actual return and benchmark return was not available for the years before 2012. The third and sixth row in the table show excess return on fixed income and equity. The excess return is calculated as the difference between the actual and benchmark returns. The table shows that the means of excess return on fixed income and equity are positive and highly significant, so the actual returns on fixed income and equity are significantly higher than the benchmark return. This is a first sign that in the low-interest-rate environment pension funds indeed searched for higher returns. Outperformance compared to the benchmark could mean that investment managers took higher risk or actively managed the portfolio (Siegel, 2003).
13
Table 1: Summary statistics on actual and benchmark return (in %) for 2012 Q1 - 2016 Q4
Is the average excess return accompanied by higher risk taking? One way to find increasing risk taking behavior would be to check how VEV has developed. As explained before, the VEV is the
representation of a pension fund’s risk profile. The development of average VEV from 2012 Q1 until 2016 Q4 is shown in figure 2 below. Data on VEV missed for the years before 2012. On average VEV has an increasing trend. The steep increase in 2015 is due to the implementation of the new Financial Assessment Framework (nieuw Financieel Toetsingskader, nFTK). The nFTK allowed for a one-time increase of pension funds’ risk profiles. Also pension funds in a funding shortage could increase their risk profile, provided that they fulfil at least the MVEV-requirement at the moment of change (DNB, 2015a). It can be seen from the figure that pension funds indeed increased their risk profile after the implementation of the new assessment framework in 2015.2 However, because data was not available
for all years before 2012 it is difficult to see whether the average VEV has increased as a consequence of the low-interest-rate environment. Furthermore, funds could have increased risk in one category and decreased it in another in order to seek for higher returns. The VEV-level would stay the same then however, and risk-taking behavior would not be noticed. Therefore, I will analyse the three
aforementioned categories of risk: credit risk, interest rate risk and liquidity risk separately for a longer time period.
2 In fact the introduction of the nFTK causes a structural break in the time series. The effects of the structural
change are analyzed in section 6.4.1
3700 observations Mean St. dev. Median Min Max
Actual return on fixed income 1.57 3.58 1.7 -21.8 22.5
Benchmark return on fixed income 1.51 3.68 1.5 -22.2 21.1
Average excess return fixed income 0.07***
(0.02)
1.37 0 -16.0 14.6
Actual return on equity 3.17 5.35 4.1 -13.6 24.2
Benchmark return on equity 3.08 5.35 4.1 -14.8 18.7
Average excess return equity 0.09***
(0.02)
1.18 0.1 -9.9 18.2
*** presents a significance at the 1%-level. The standard error of the mean is reported in brackets
14
Figure 2: Average VEV
4.1.1 Credit risk
Did pension funds increase credit risk in the low-interest-rate environment? Rating agencies like Standard’s & Poor, Moody’s or Fitch give a higher rating to bonds with lower credit risks. So an increase in credit risk could be perceived by higher holdings in lower rated fixed income categories. Table 2 below provides summary statistics on holdings of fixed income and stocks for the entire sample. Fixed income is divided into the rating categories AAA, AA, A, BBB, <BBB, or not rated according to the Fitch rating scale. All numbers in table 2 reflect a percentage of total investments. The column ‘Total observations’ shows the amount of pension funds over 40 quarters that correctly reported the holdings in the specific asset category (which is 7400 if all 185 pension funds reported correctly over 40 quarters). For AA, A and not rated holdings there are less than 7400 observations, because of reporting errors by several pension funds. The column ‘Observations holdings > 0’ shows the amount of observations with holdings larger than zero. From this column it appears that some pension funds do not invest in lower rated fixed income or equity. Pension funds invest most in fixed income (mean = 56.91%, median = 57.71%). On average, almost 30% of the portfolio is invested in equity. 114 115 116 117 118 119 2012 q1 2013 q1 2014 q1 2015 q1 2016 q1 2017 q1 period
Average VEV
15
Table 2: Summary statistics of fixed income and equity holdings (2007 Q1 – 2016 Q4)
Figures 3 and 4 below show the evolution of average fixed income and stock holdings from 2007 Q1 until 2016 Q4. Figure 3 shows the division of fixed income holdings over the six rating categories. The rating categories are presented as a percentage of total fixed income and reflect the unweighted average of fixed income holdings in the entire pension sector from 2007 until 2016. The average investment mix of the pension sector has changed over time. From the picture it becomes clear that the most remarkable changes have occurred in the rating categories AAA, AA and BBB. In the beginning of 2007, pension funds held on average almost 60% AAA-rated fixed income, and around 40% fixed income in lower rating categories. By the end of 2016 pension funds only held about 35% AAA-rated fixed income, and about 65% lower rated fixed income. AA-rated fixed income has more than doubled from about 10% in 2007 Q2 of total fixed income to almost 25% of fixed income in 2016 Q4. Fixed income in rating category BBB has increased from less than 5% in 2007 Q2 to almost 15% in 2016 Q4. These results seem to support the search for yield hypothesis. However, it should be taken into account that during the sovereign debt crisis in Europe, many government bonds have been
downgraded from AAA to AA (e.g. France’s government bonds) (Baum et al., 2016). This could also partly explain the drop in AAA-rated holdings, and the accompanying increase in AA-rated holdings. However, after the drop in 2012 AAA-holdings stayed at a lower level than before the crisis, so pension funds have not rebalanced their portfolio. This might be an indication that pension funds have increased credit risk, which is one of the determinants of search for yield.
Figure 4 shows the evolution of average equity holdings by Dutch pension funds over the selected time period. On average stock holdings in emerging markets have increased since the global financial crisis in 2008. It is assumed broadly in the literature that equity investments are more risky than investments in other assets (like governments bunds) and thus have a higher expected return (see e.g. Gorter & Bikker, 2011, Lee et al., 1997, and Ibbotson & Sinquefield, 1989). In line with this
Total Observations 5th 95th
% of total investments Observations holdings >0 Mean St. dev. Median Min Perc. Perc. Max
Total fixed income 7400 7396 57.04 12.88 57.76 0 36.03 77.45 94.16
AAA AA A BBB <BBB Not rated 7400 7390 7392 7400 7400 7282 7288 7182 7057 6672 6131 6347 26.65 10.16 7.10 5.57 3.16 4.52 13.57 8.73 5.27 4.85 3.39 7.83 25.28 8.24 6.05 4.50 2.39 1.90 0 0 0 0 0 0 6.37 0.73 0.26 0 0 0 50.49 25.83 16.68 15.06 9.42 16.61 83.74 79.67 72.81 34.57 28.59 80.61 Total equity 7400 7141 28.73 10.95 29.42 0 9.93 45.46 73.74 Mature markets Emerging markets 7400 7400 6704 6157 21.94 3.57 11.19 2.85 23.30 3.43 0 0 0 0 38.16 8.16 70.02 65.59
16 assumption, the increasing weight of equity in pension funds’ portfolios, reflects higher exposure to credit risk. This in turn might indicate search for yield.
Comparing figures 1 and 3, it is interesting to notice that the decline in AAA-holdings (and the increase in AA and BBB-holdings) starts around 2011 Q4, which is also the period in which the risk-free rate started declining and reached historically low levels. As discussed before, following the economic reasoning of e.g. Rajan (2006), Yellen (2011) or Antolin et al. (2011), the low interest rates might indeed have been an incentive for Dutch pension funds to increase credit risk.
Figure 3: Average fixed income holdings categorized by rating
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Average fixed income holdings of the pension sector
(% of total fixed income)
17
Figure 4: Average equity holdings by Dutch pension funds
4.1.2 Interest rate risk
Another way to search for yield is by increasing interest rate risk. This is based on Yellen (2011) who suggests that search for yield involves investing in assets with a greater duration. Interest rate risk occurs when investing in fixed income bonds. There are several ways to measure interest rate risk, including a duration analysis or a full valuation approach (Fabozzi & Mann, 2005). In a full valuation approach the sensitivity of a fund’s portfolio to a change in interest rate is measured for a given interest-rate change scenario. I cannot perform a proper duration analysis, because only data on the duration of total fixed income was available for the selected time period. Duration of total fixed income is no good measure for interest rate risk, because long duration AAA-bonds can be used to hedge against interest rate risk (DNB, 2015b). Pension funds use the full valuation approach to report to DNB the extent to which they are exposed to interest rate risk. The funds revalue their assets and liabilities after a given interest-rate change scenario. The net loss of the worst-case scenario is used as input for the VEV of pension funds (DNB, 2015c). Pension funds report this exposure to interest rate risk through eLine. I linearized this variable by taking the natural logarithm. Figure 5 below depicts the average interest rate risk in the Dutch pension sector. Interest rate risk has increased over time. The Harris-Tzavalis unit root test confirms that the trend is stationary.3 The steep increase from 2014 to 2015 again indicates that pension funds used the implementation of the nFTK to increase their risk profile.
3 This test assumes that the number of panels tends to infinity whereas the number of periods is fixed. Because of
the large number of panels (185) relative to the number of periods (40), this is the right test to use. 0.0% 0.5% 1.0% 1.5% 2.0% 2.5% 3.0% 3.5% 4.0% 4.5% 5.0% 0% 5% 10% 15% 20% 25% 30%
Average equity holdings
(% of total investments)
18
Figure 5: Average interest rate risk in the Dutch pension sector
Unfortunately this measure does not indicate in which asset categories pension fund increased interest rate risk. Moreover, because this measure is an entirely different measure than the ones used for measuring credit risk and liquidity risk it is hard to compare them. Therefore, interest rate risk is excluded from the further (regression) analysis.
4.1.3 Liquidity risk
Liquidity risk arises from investing in illiquid investments that cannot easily be traded for cash or cash equivalents. The advantage of investing in illiquid assets for pension funds is that the expected return is higher than on illiquid assets, because of an additional liquidity premium that is paid out. A
disadvantage could be that a pension fund ends up having too less liquidity to pay out its pensions and consequently has to sell some of its assets at a price below market value. An increase in investments in illiquid assets could indicate that pension funds search for yield by increasing liquidity risk. An increase in investments in illiquid assets thus could indicate that pension funds search for yield by increasing liquidity risk.
I will account for real estate and mortgages as illiquid investments. Real estate and mortgages cannot easily be traded for cash at market value, because the investor is bound to these obligations for a specified period of time.4 Table 3 provides summary statistics on these investments for the entire
sample period. Dutch pension funds hold on average 6.18% real estate investments and 1.84% mortgage loans in their portfolio. Pension funds invest more in indirect real estate than direct real estate. Though indirect real estate is more liquid than direct real estate and mortgages, it is still less liquid than bonds and stocks.
4 Other illiquid investments are investments in infrastructure and structured notes. I did not include these
investments into this research, because data was only available for 2015 and 2016. 9.80 9.90 10.00 10.10 10.20 10.30 10.40 10.50
Interest rate risk
(logarithms)
19
Table 3: Summary statistics on illiquid investments
Figure 6 presents the evolution of illiquid investments over time. From figure 6 it becomes clear that on average pension funds increased their investments in mortgages (the average weight increased from about 1.4% in 2007 Q1 to 3.3% in 2016 Q4). The drop in 2015 reflects that during the implementation of the nFTK, pension funds suddenly changed their investments preferences. After 2015 mortgages gradually increase again. Also investments in indirect real estate show an increasing trend, contrary to direct real estate which shows a decreasing trend. The decreasing trend in direct real estate can be explained by the international orientation of real estate investors. Directly investing in foreign real estate requires good management, so it is easier to invest abroad indirectly through a venture capitalist (CBS, 2013). To test whether the increase in mortgage loans and indirect real estate is stationary, I conducted a Harris-Tzavalis unit root test. The test is significant at the 1%-level, so the
null-hypothesis that the panels contain unit roots is rejected, the increase in these asset holdings is indeed stationary.
Figure 7 depicts the number of pension funds investing in illiquid assets. From 2007 Q1 to 2016 Q4 the number of pension funds investing in mortgage loans has almost tripled (increase from 49 to 134). The steepest increase took place from 2011 Q4 to 2012 Q1 (+22 pension funds), which is at the same moment that the risk-free rate reached historical lows. The number of pension funds investing in real estate has also increased from 143 to 169 over the entire period. This indicates that across the entire sector more pension funds started investing in illiquid assets. Furthermore, the average weight of total illiquid investments also increased over time, so pension funds increased their liquidity risk. This is especially true considering mortgage investments and indirect real estate. This might be another indicator of search for yield.
(% of total investments) Total
Observations
Observations
holdings >0 Mean St. dev. Median Min 5th
Perc. 95th
Perc. Max
Total real estate 7400 6421 6.18 5.38 5.32 0 0 15.70 40.2
Direct real estate 7398 2072 1.62 4.23 0 0 0 9.71 39.56
Indirect real estate 7295 6058 4.66 4.43 3.79 0 0 12.96 27.10
20
Figure 6: Average weight of illiquid investments in the entire Dutch pension sector
Figure 7: Number of pension funds investing in illiquid assets
5. Methodology
I will use empirical tests to examine the following questions:
1. Does search for yield exist in the Dutch pension sector? 2. Does search for yield differ across pension funds?
The aforementioned definition of search for yield: “the incentive to seek for higher expected returns when interest rates are low through investing in assets with higher credit risk, interest rate risk or
0 2 4 6 2007 q1 2009 q3 2012 q1 2014 q3 2017 q1 period
Direct real estate Indirect real estate
Mortgage loans (% of total investments)
Illiquid Assets
125 130 135 140 145 150 155 160 165 170 175 0 20 40 60 80 100 120 140 160Number of pension funds investing in illiquid assets
21 liquidity risk”, determines the approach to the first question. The results in section 4 imply that
pension funds on average outperform their benchmark. Also, shifts to riskier asset categories occur in the period that the risk-free rate reached historically low levels. This indicates an increase in
investments in assets with higher credit risk, interest rate risk and liquidity risk. However, as described above, these results neglect other factors that influence the investment behavior of pension funds. Therefore, the next step is to estimate the effect of movements in the risk-free rate on investment shifts. So, in line with Becker & Ivashina (2012), I will estimate a regression model in which the independent variable is the risk-free rate and the dependent variable reflects holding of fixed income in a certain rating category, equity and illiquid assets. I will correct for a possible trend in the holdings by including lagged holdings, so that exogeneity of the other explanatory variables is ensured. A possible concern is that this model would just estimate the impact of the risk-free rate on holdings of assets. However, the purpose of this research is to prove that search for yield is more severe in a low-interest-rate environment. To control for this, I will add the squared risk-free rate into the regression besides the risk-free rate. This way it is possible to see what the impact of the risk-free rate on asset holdings is when the rate is low compared to when it is high. Since some pension funds might have certain investment preferences ex ante, I will include pension fund fixed effects to control for unobserved heterogeneity. Since the risk-free rate is fixed over time for all pension funds, I will exclude time fixed effects. Much of the variation in holdings caused by movements in the risk-free rate would be absorbed by including time fixed effects.
The baseline model for the regression is:
(1) 𝐻𝑖,𝑡 = 𝛽0+ 𝛽1𝐻𝑖,𝑡−1 + 𝛽2𝑅𝐹𝑅𝑡 + 𝛽3𝑅𝐹𝑅𝑡2+ 𝛼𝑖+ 𝜀𝑖,𝑡
The variables are explained in table 4 below.
Table 4: Explanation of baseline model
Variable Definition Type
𝐻𝑖,𝑡 Holdings of assets in a specific category by pension fund i in quarter
t. The categories are: Fixed-income assets rated AAA, AA, A, BBB,
<BBB or not rated, stocks in emerging and mature markets, direct real estate, indirect real estate and mortgage loans.
Dependent
𝐻𝑖,𝑡−1 Lagged holdings Control variable
𝑅𝐹𝑅𝑡 Risk-free rate in quarter t. Measured in percentages. Independent
𝛼𝑖 Pension fund fixed effects Fixed effects
𝜀𝑖,𝑡 Error term Error term
Using this model I will first test whether credit risk has increased. In this case, the dependent variable equals holdings of fixed income in the six rating categories and equity. Search for yield takes place when the coefficient on the risk-free rate in the high rated (AAA) fixed income model is positive,
22 indicating that an increase (decrease) in the risk-free rate will increase (decrease) AAA-holdings and decrease (increase) credit risk. Simultaneously, the coefficient on the squared risk-free rate should be negative. A negative coefficient indicates that the risk-free rate only increases AAA-holdings until a certain point. When it is high enough, the effect diminishes. The expected coefficients for fixed income rated lower than AAA is negative, indicating that a decline in the risk-free rate increases those riskier holdings and credit risk. The coefficient on the squared risk-free rate should be positive in this case, indicating that the effect diminishes when the risk-free rate increases. Considering equity investments, I expect the same as for lower than AAA-rated fixed income investments. To check whether a decrease in the risk-free rate increases liquidity risk taken by pension funds, I will also estimate a model in which the dependent variable reflects the weight of direct real estate, indirect real estate and mortgage loans. For this asset category I expect the same coefficients as for the lower than AAA-rated fixed income and equity categories.
With regard to the first research question about the existence of search for yield in the Dutch pension sector, I hypothesize the following:
H1: Search for yield exists in the Dutch pension sector.
I will continue my research by extending the baseline model with other variables that might affect risk-taking behavior of pension funds. First, following the method of Choi & Kronlund (2015), I will include the slope of the yield curve into the regression model (𝑌𝐶𝑡) as a control variable. I will roughly estimate the slope of the yield curve by calculating the difference between Euro Area AAA-rated government bonds with 30 year maturity and 1 year maturity. This measure reflects a liquidity premium and expectations of future interest rates, which also affect holdings of assets in a specific category. When the slope of the yield curve decreases, investment opportunities become scarcer, because of smaller term premia. Assuming rational expectations, this might incentivize pension funds to increase risk-taking (Choi & Kronlund, 2015). To capture the effect of changes in the interest rate term structure on investment decisions, I included the slope of the yield curve as a control variable. In order to answer the second question, I will include variables that reflect pension fund
characteristics. Hence I will discover how (1) a fund’s funding position, (2) the average age of pension fund participants and (3) fund size affects search for yield.
A pension fund’s funding position could influence its risk appetite. As explained before in section 2, two opposing theories about taking by firms in financial distress appear in the literature: the risk-shifting and the risk-management hypothesis. Gorter & Bikker (2011) find empirical evidence for the risk management hypothesis, when they conduct a study on risk-taking behavior by institutional investors in the Netherlands. Also Rauh (2009) and An et al. (2013) find empirical evidence in favor of the risk management hypothesis. Dutch pension law prohibits pension funds to increase their risk
23 profile when they have a funding ratio below MVEV. To control for this I will create a dummy variable which equals 1 if the funding ratio is higher than MVEV, and pension funds are thus allowed to increase their risk profile (𝐹𝑅𝑖,𝑡). I will interact this dummy variable with the risk-free rate, to find out how the funding position of pension funds affects risk-taking behavior in a low-interest-rate environment, i.e. search for yield. I expect to find support for the risk management hypothesis, in line with the aforementioned studies. The coefficient on the interaction term should be positive for high rated assets, indicating that a declining risk-free rate decreases AAA-holdings for pension funds in a stable funding position. For lower rated assets, equity and illiquid investments the coefficient should be negative on the contrary.
H2: Search for yield is more severe among pension funds with a funding ratio above MVEV than among pension funds in a funding shortage.
Second, demographic characteristics of pension funds might influence search for yield. Following the theory of Bodie et al. (1992) younger people have more incentives to engage in riskier investments (see section 2). In the end, pension funds invest for individuals which could have preferences related to their age. Therefore, I will include the average age of pension fund participants into the model and interact it again with the risk-free rate. For AAA-holdings, the coefficient on the interaction term should be positive, indicating that for a given increase (decrease) in age, the effect on AAA-holdings depends positively on the risk-free rate, i.e. AAA-holdings will increase (decrease). For fixed income rated lower than AAA, equity and illiquid assets the coefficient should be negative. This indicates that for a given increase (decrease) in age, the effect depends negatively on the risk-free rate, i.e. holdings of ‘riskier’ assets will decrease (increase).
H3: A lower average age of pension fund participants increases search for yield.
Finally, I will include pension fund size into the regression, measured by total assets. I took the natural logarithm of this variable to linearize it. Larger firms benefit more from diversification and scale economies. Large funds can easier organize risk management than smaller organizations, and hence can better evaluate the actual risk they are taking. Because of better risk management, larger pension funds can thus more efficiently engage in risky investments (Gorter & Bikker, 2011). I expect that because of these efficiency benefits, search for yield will be higher among larger pension funds. I interact the variable again with the risk-free rate. For ‘safe’ assets (AAA), the expected coefficient on this interaction term is negative. Indicating that when fund size increases (decreases), it depends negatively on the risk-free rate, and thus decreases (decreases) AAA holdings. For fixed income rated lower than AAA, equity and illiquid assets I expect the opposite effect.
24 I will still include pension fund fixed effects and exclude time fixed effects. So eventually, I will estimate the following regression model:
(2) 𝐻𝑖,𝑡 = 𝛽0+ 𝛽1𝐻𝑖,𝑡−1+ 𝛽2𝑅𝐹𝑅𝑡+ 𝛽3𝑅𝐹𝑅𝑡2+ 𝛽4𝑌𝐶𝑡+ 𝛽5𝐹𝑅𝑖,𝑡 + 𝛽6𝐴𝐺𝐸𝑖,𝑡+ 𝛽7𝑆𝐼𝑍𝐸𝑖,𝑡+
𝛽8(𝐹𝑅𝑖,𝑡∗ 𝑅𝐹𝑅𝑡) + 𝛽9(𝐴𝐺𝐸𝑖,𝑡∗ 𝑅𝐹𝑅𝑡) + 𝛽10(𝑆𝐼𝑍𝐸𝑖,𝑡∗ 𝑅𝐹𝑅𝑡) + 𝛼𝑖+ 𝜀𝑖,𝑡
The variables are explained in table 5 below. The hypotheses with regard to the signs of the coefficients in equation (2) are summarized in appendix 2.
Table 5: Explanation of the variables in the extended regression model
Variable Definition Type
𝐻𝑖,𝑡 Holdings of fixed income in a specific rating category, equity and illiquid assets by pension fund i in quarter t. The categories are: fixed income rate AAA, AA, A, BBB, <BBB or not rated, stocks in emerging and mature markets and direct real estate, indirect real estate and mortgage loans.
Dependent
𝐻𝑖,𝑡−1 Lagged holdings Control variable
𝑅𝐹𝑅𝑡 Risk-free rate in quarter t. Measured in percentages. Independent
𝑌𝐶𝑡 Slope of the par yield curve in quarter t, calculated as the difference between the yield on 30-year and 1-year Euro Area AAA-rated government bonds. Measured in percentages.
Control variable
𝐹𝑅𝑖,𝑡 Dummy variable which equals 1 if pension fund i has a funding ratio higher than MVEV in quarter t
Independent
𝐴𝐺𝐸𝑖,𝑡 The average age of the participants of pension fund i in quarter t (measured in years)
Independent
𝑆𝐼𝑍𝐸𝑖,𝑡 (The logarithm of) total investments of pension fund i in quarter t Independent
𝛼𝑖 Pension fund fixed effects Fixed effects
𝜀𝑖,𝑡 Error term Error term
6. Empirical results
6.1 Regression of the baseline model
Following the method of Becker & Ivashina (2012) I will estimate the impact of the declining risk-free rate on holdings in a specific asset category. As explained in section 5, I estimated the baseline
regression model in equation (1). In this model the dependent variable reflects fixed income holdings in a specific rating category (AAA, AA, A, BBB, <BBB, and not rated), equity or illiquid assets. Individual pension funds might ex ante have different investment preferences due to for example different managerial skills. To control for this unobserved heterogeneity as a possible source of
25 endogeneity I added pension fund fixed effect.5 I clustered standard errors by pension fund to correct for heteroskedasticity and serial correlation.6
The results of the baseline regression are displayed in table 6 below. Columns (1)-(6) represent fixed income holdings in the six rating categories. Columns (7)-(9) represent equity holdings, respectively total stock holdings, stock in emerging markets and stock in mature markets. Finally, columns (10)-(12) represent illiquid assets, respectively direct real estate, indirect real estate and mortgage loans. As expected, a decline in the risk-free rate increases AAA-holdings (see column 1). The effect fades out when the risk-free rate increases, which appears from the negative coefficient for the squared risk-free rate. The highly significant F-statistic implies that these results can be interpreted jointly. In other words: an increase (decrease) in the risk-free rate increases (decreases) AAA-holdings, until the point where the risk-free rate equals 3%. From this point the effect of the risk-free rate on AAA-holdings diminishes. Both coefficients are significant at the 1%-level, so indicating search for yield. The R-squared indicates that 68% of the within variation of AAA-holdings is explained by the risk-free rate. For AA, BBB and <BBB-rated fixed income the estimated coefficients resemble the hypothesized coefficients: negative for the risk-free rate and positive for the squared risk-free rate. This indicates that an increase (decrease) in the risk-free rate decreases (increases) these holdings, when interest rates are low. The effect diminishes when interest rates are high, which is shown by the positive coefficient on the squared risk-free rates. So, downward movements of the risk-free rate in a low-interest-rate environment are an incentive for pension funds to increase risk-taking behavior and invest more in lower rated, riskier assets with an expected higher return. In model (3), the coefficient on the risk-free rate is positive, instead of the negative hypothesized coefficient. A declining risk-free rate is
associated with a decrease in these holdings. The positive coefficient for the squared risk-free rate shows that this effect deteriorates when risk-free rates are high. The R-squaredis low compared to the other model, indicating that less of the variation in A-holdings is explained by movements in the risk-free rate. One way to explain this coefficient is that pension funds either choose to invest in the relatively safe (AA) holdings or search for the higher yield in even lower rating categories (BBB, <BBB or not rated) in a low-interest-rate environment. The negative sign on the squared risk-free rate implies that when the risk-free rate is high enough, pension funds stop moving away from this
category.7
5 The Hausman test is significant at the 1%-level, so the null hypothesis of using a random effects model is
rejected.
6 The statistic for the modified Wald-test for groupwise heteroskedasticity in the residuals of a fixed effect
regression model is significant at the 1%-level. The null-hypothesis of homoskedasticity is rejected.
7 A-holdings do not fluctuate much over time (see e.g. figure 3). When the model is extended, the positive sign
becomes statistically insignificant (see section 6.2). This supports the idea that A-rated fixed income investments are nor very safe, neither promise high yields. Pension funds either play it safe by investing in higher rated fixed income or search for yield by investing in lower rated fixed income. A-rated fixed income is just captured in the middle between those two categories.
26 In order to check whether movements in the risk-free rate affect stock investments by pension funds, I estimate the same regression model, where 𝐻𝑖,𝑡 and 𝐻𝑖,𝑡−1 reflect total equity and lagged equity-holdings respectively. The results are shown in panel B of table 6, columns (7)-(9). The coefficients for the risk-free rate is not statistically significant in any of the models. It shows the expected sign in model (9) reflecting stocks in emerging markets. Opposite signs appear in models (7) and (8)
concerning total equity investments and stocks in mature markets. However, since the results are not statistically significantly different from zero, they cannot be interpreted.
Columns (10)-(12) present the results for direct real estate, indirect real estate and mortgages. The risk-free rate coefficient is significant in all three models. For indirect real estate and mortgages loans, search for yields exists. A (one percent point) decreasing risk-free rate increases indirect real estate with 0.22 percent point and mortgage loans with 0.18 percent point. The effect diminishes when the risk-free rate increases, considering the positive coefficient on the squared risk-free rate. For direct real estate, the effect is opposite. When the risk-free rate decreases, direct real estate decreases as well. Though this differs from the hypothesized effect, it might be explained by the international orientation of real estate investors, as explained before.
From the baseline model it appears that pension funds increase credit risk due to the low interest rates. They decrease safe AAA investments, and increase <AAA rated fixed income. In the baseline model, no evidence has been found for an increase in equity investments. Pension funds also increase liquidity risk, but mainly by increasing investments in indirect real estate and mortgage loans. These results imply that Dutch pension funds increased their credit and liquidity risk as a consequence of the declining risk-free rate. Hypothesis 1 is confirmed, i.e. search for yield exists in the Dutch pension sector.
27
Table 6: Estimation of the baseline regression model
Panel A
Dependent variable: holdings of fixed income AAA (1) AA (2) A (3) BBB (4) <BBB (5) Not rated (6) Lagged holdings 0.7791*** (0.0154) 0.7405*** (0.0220) 0.6839*** (0.0300) 0.8400*** (0.0128) 0.7699*** (0.0262) 0.6684*** (0.0441) Risk-free rate (%) 2.0275*** (0.1670) -1.3836*** (0.1516) 0.4237*** (0.1168) -0.5179*** (0.0647) -0.2825*** (0.0378) -0.1141 (0.1075) Risk-free rate squared (%) -0.3375*** (0.1670) 0.2203*** (0.0249) -0.0629*** (0.0206) 0.0610*** (0.0100) 0.0389*** (0.0062) 0.0481** (0.0235) Pension fund fixed effects
Yes Yes Yes Yes Yes Yes
F-stat. Joint significance RFR and RFR2 74.90*** 41.67*** 8.16*** 60.75*** 29.47*** 3.53** N 7215 7202 7202 7215 7215 7023 R2 0.6796 0.6048 0.4891 0.8230 0.6356 0.4985 Panel B
Dependent variable: holdings of equity Dependent variable: holdings of illiquid assets Total (7) Mature markets (8) Emerging markets (9) Direct real estate (10) Indirect real estate (11) Mortgage loans (12) Lagged holdings 0.8727*** (0.0129) 0.8414*** (0.0107) 0.6473*** (0.0893) 0.8176*** (0.0297) 0.7683*** (0.0250) 0.7883*** (0.0397) Risk-free rate (%) 0.0330 (0.1399) 0.2016 (0.1481) -0.0564 (0.0599) 0.1169*** (0.0312) -0.2223*** (0.0514) -0.1801*** (0.0589) Risk-free rate squared (%) -0.0155 (0.0250) -0.0824*** (0.0294) -0.0134 (0.0086) -0.0073 (0.0058) 0.0391*** (0.0091) 0.0204** (0.0093) Pension fund fixed effects
Yes Yes Yes Yes Yes Yes
F-stat. Joint significance RFR and RFR2 2.06 18.39*** 10.29*** 19.36*** 9.53*** 7.07*** N 7215 7215 7215 7211 7106 7110 R2 0.7437 0.7192 0.4656 0.6973 0.6073 0.6549
Standard errors are clustered by pension fund and reported in brackets. ***, **, and * report 1%, 5% and 10% significance respectively.
6.2 Extension of the model
The purpose of the second part of this investigation is to determine how search for yield differs across pension funds. In order to do so, I extended the baseline regression model. I included three variables on pension fund characteristics: 𝐹𝑅𝑖,𝑡 is a dummy variable that equals 1 if a pension fund has a
funding ratio above MVEV and is thus allowed to increase its risk profile; 𝐴𝐺𝐸𝑖,𝑡 reflects the average age, of pension fund participants; and 𝑆𝐼𝑍𝐸𝑖,𝑡 is (the logarithm of) total assets. These variables are included both individually and interacted with the risk-free rate. 𝐻𝑖,𝑡 reflects holdings of fixed income
in the six rating categories, stock and illiquid assets. As explained before: 𝑌𝐶𝑡 reflects the slope of the
28 The Hausman test suggests that pension fund fixed effects should still be included. I will exclude time fixed effects, because the risk-free rate is fixed over time for all pension funds. Including time fixed effects would thus absorb much of the variation in the dependent variable that is caused by movements in the risk-free rate. The model being estimated is presented in equation (2) in section 5. The results of the regression are shown in table 7 below. The significant coefficients on the ‘slope of the yield curve’ variable show that it indeed affects risk-taking behavior. A flattening yield curve increases ‘safe’ AAA-rated fixed income investments and is accompanied by a decreasing allocation to A, BBB and <BBB-rated fixed income, equity and mortgages. This effect can be explained by the use of (long duration) AAA bonds to hedge against yield curve risk (i.e. a flattening or steepening yield curve). When long-term interest rates decrease, the increase in liabilities becomes more nonlinear. Therefore, pension funds have an additional incentive to match the duration of their assets and liabilities better by investing more in (longer duration) AAA-bonds (Domanski et al., 2017).
Now the next question is whether search for yield differs across pension funds. I will answer this question by analysing the results based on the four hypothesis that were formulated in section 5.
H1: Search for yield exists in the Dutch pension sector.
Dutch pension funds search for yield if 1) they increase credit risk due to a decreasing risk-free rate in the low-interest-rate environment and 2) they increase liquidity risk due to a decreasing risk-free rate in the low-interest-rate environment. In column (1) concerning AAA-rated fixed income, the
coefficient on 𝑅𝐹𝑅𝑡 is positive and the coefficient on 𝑅𝐹𝑅𝑡2 is negative, which is as expected. When
interest rates are low, a decrease of one percent point in the risk-free rate causes a decrease in AAA-holdings. Until the point where the risk-free rate equals 3.8 percent, then this effect diminishes, ceteris paribus.8 Both coefficients are highly significant, and also the F-statistic testing the joint significance of 𝑅𝐹𝑅𝑡 and 𝑅𝐹𝑅𝑡2 is significant at the 1%-level. Columns (2)-(6) show the results for fixed income
rated lower than AAA. The coefficients on the risk-free rate and the squared risk-free rate have the expected sign in columns (2) and (4)-(6): negative for 𝑅𝐹𝑅𝑡 and positive for 𝑅𝐹𝑅𝑡2. This means that
pension funds increase these holdings when the risk-free rate is low and declining. The sign reverts at a point where the risk-free rate is higher, meaning that holdings of these assets respond less to changes in the risk-free rate when it becomes higher. For A-rated fixed income, the coefficient shows an opposite sign on the risk-free rate, but is statistically insignificant. In columns (5) and (6), reflecting <BBB-rated and not rated fixed income, the coefficients on the risk-free rate are not significant. This is in line with the findings of Becker & Ivashina (2012) and Choi & Kronlund (2015) that search for yield mainly occurs in the investment grade bonds.
29 The equity models in columns (7)-(9) also show the expected sign on the risk-free rate and the squared risk-free rate. The coefficient for 𝑅𝐹𝑅𝑡 and 𝑅𝐹𝑅𝑡2 are jointly significant in the models for total stock
holdings and stock in mature market holdings (respectively at a 1% and a 5%-level). Pension funds increase total equity holdings when the risk-free rate is low and decreasing, until the point where the risk-free rate equals 2.7%. At this point equity holdings react less severe to a declining risk-free rate. For mature markets the turning point is at 5.9%. Finally, the results in columns (10)-(12) show that pension funds do not search for yield through investing more in illiquid assets: the coefficient is on the risk-free rate is statistically insignificant in all three models.
These results imply that pension funds indeed increase credit risk more when the risk-free rate is low and declining, compared to when the risk-free rate is high. So, also in the extended model, hypothesis 1 is confirmed, i.e. search for yield exists in the Dutch pension sector.
H2: Pension funds with a funding ratio > MVEV have higher search for yield than pension funds in a funding shortage.
When funds have a stable funding position (𝐹𝑅𝑖,𝑡 = 1), they invest less in AAA-rated fixed income
(column 1) and more in AA-rated fixed income (column 2). This can be seen from the coefficient on the variable ‘funding ratio’. Does this reflect search for yield? Looking at the interaction term (𝐹𝑅𝑖,𝑡 ∗
𝑅𝐹𝑅𝑡) in column 1, the coefficient shows a negative sign. The F-statistic testing joint significance
between the risk-free rate and this interaction term is significant at the 1%-level (p-value=0.0003). This indicates that for pension funds with a stable funding position, increase (decrease) AAA-holdings when the risk-free rate increases (decreases). Column (2) presents AA-holdings. Pension funds with a stable funding position decrease (increase) their allocations to AA-holdings, when the risk-free rate increases (decreases). For BBB-rated fixed income (column 4), the effect is opposite. Pension funds with a stable funding position will decrease BBB-holdings when the risk-free rate declines. For total equity and stock in mature markets, the funding position of a fund also positively depends on the risk-free rate. These results are opposite of the hypothesized sign on the interaction term. So the hypothesis that pension funds in a stable funding position increase search for yield, cannot be confirmed.
H3: A lower average age of pension fund participants increases search for yield.
This hypothesis remains inconclusive. The coefficients on the individual variable is only weakly significant in models (4) and (6), and the coefficient on the interaction term is only weakly significant in model (10). The result is ambiguous however. In none of the other models the result is statistically significant. The age variable does not change much over the selected time period, so the insignificance of the results might also be explained by the little variation in the variable. Any effect of average participant age on holdings is possibly absorbed by the inclusion of pension fund fixed effect.
