• No results found

The influence of negative interest rates on the investment behaviour of pension funds and insurance companies

N/A
N/A
Protected

Academic year: 2021

Share "The influence of negative interest rates on the investment behaviour of pension funds and insurance companies"

Copied!
36
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The influence of negative interest rates on the investment

behaviour of pension funds and insurance companies

Abstract:

This paper studies the effect of negative interest rates on the investment behaviour of European pension funds and insurance companies, measured by the percentage of fixed income securities in their aggregated portfolios. The results present no evidence that negative

interest rates significantly lower the percentage on fixed income securities in the aggregated portfolios of either European pension funds or insurance companies. Furthermore, I find that the interest rate itself significantly lowers lower the percentage of fixed income securities in the aggregated portfolios of European pension funds, while for insurance companies there is

no significant effect.

Master thesis MSc Finance University of Groningen

Submitted on: 12th of June

Submitted by: Laura Reinink, S3540022 Supervisor: prof. dr. K.F. Roszbach

(2)

1

Contents

1. Introduction ... 2 2. Literature review ... 4 3. Hypotheses... 7 4. Data ... 8 4.1 Pension funds ... 9 4.2 Insurance companies ... 10 5. Methodology ... 12 5.1 Model ... 12 5.2 Control variables ... 13 5.3 Robustness tests... 14

5.3.1 Change in fixed income ... 15

5.3.2 Euro countries ... 15 5.3.3 Outliers ... 15 6. Empirical analysis... 16 6.1 Pension funds ... 16 6.2 Insurance companies ... 18 6.3 Robustness ... 20 6.3.1 Pension funds ... 20 6.3.2 Insurance companies ... 22 7. Conclusions ... 24 References ... 26

Appendix A: Interest rates ... 29

Appendix B: Descriptive statistics insurance companies per enterprise ... 30

Appendix C: Regression robustness - Euro countries ... 32

(3)

2

1. Introduction

This paper studies the effect of negative interest rates on the investment behaviour of institutional investors. An institutional investor is an entity that collects money from consumers, which is collectively invested by the company with the purpose of providing some future benefit. The two largest types of institutional investors are pension funds and insurance companies. Duijm and Bisschop (2017) show that Dutch pension funds have assets under management valued at around 185% of the Dutch GDP, for insurers this is 71%.

Pension funds are institutions that collect contributions from employers and employees with the aim to provide a cashflow sometime in the future, namely when the employee has reached his or her retirement age. The purpose of this cashflow is to smoothen the consumption of its members so that retirees can maintain the same living standards when they do not receive an income from working anymore.

Insurance companies are companies that collect a premium from their policyholders. In return the policyholder receives a compensation under certain circumstances. For example, a life insurer might pay a one-time benefit to the relatives of the policyholder in the event of death, dependent on the agreed policy. An insurance on real-estate might provide a benefit to repair the building in the case of storm damage. The received premiums are invested in different assets to obtain a positive revenue when the return on the investments is higher than the pay out to the policyholders.

A study of Bouvatier and Rigot (2013) shows that pension funds in the United States and in Canada allocated on average around 30% of their portfolio in fixed income securities during the years 2000 through 2008. Investments in fixed income securities are usually seen in the literature as less risky than investments in stocks. Holders of fixed income securities are for example legally better protected than holders of common equity. Examples of fixed income securities are government bonds and debt instruments issued by organizations. This result is supported by Anson (2004), who shows that the fund value of the California Public Employees' Retirement System (CalPERS) has been slowly increasing between 1992 and 2003. The percentage of the portfolio allocated to fixed income has remained relatively constant, at slightly more than 25%.

Polacek (2018) finds that the portfolio of life insurance companies in 2008 consisted of 63% in fixed income securities, another 10% was invested in whole-loan mortgages, these have comparable characteristics as fixed income securities. Furthermore, Bendrich and Bergström (2015) state that the portfolio of European insurers in 2011 consists of 64% in bonds, of which were 28% in government bonds and 36% in corporate bonds. Finally, Berends, McMenamin, Plestis and Rosen (2013) find that 38.9% of the portfolios of life insurers in 2012 were corporate and foreign bonds.

(4)

3 In recent years, the ECB has reduced its interest rates year after year. In 2012 the deposit rate reached a level of 0% interest and in the following years the interest rate even turned negative. This monetary policy may have had a different effect on the profitability of equity investments than on fixed income securities, therefore the relative attractiveness of fixed income securities might have changed, causing changes in the preferences of institutional investors. Therefore, it is plausible that institutional investors are moving towards other forms of investing, in order to maintain a certain level of profitability.

Previous research that studies the effect of low interest rates on the investment behaviour and risk incentives of pension funds, has mostly focussed on the United States. This paper studies the effect of negative interest rates within Europe. In comparison to pension funds, insurance companies have been studied significantly less. To the best of my knowledge, there has been no previous research about the effect of negative interest rates on the investment behaviour of either insurance companies or pension funds. In this paper both matters will be closely studied using the following research question:

What is the effect of negative interest rates on the aggregated investment behaviour of European pension funds and insurance companies?

This paper aims at improving our insights in the risk incentives of institutional investors in relation to negative interest rates. This paper studies how negative interest rates influence the asset allocation of pension funds and insurance companies, the results of this paper should create a better understanding in the investment behaviour of pension funds.

(5)

4

2. Literature review

Asset allocation

The decision on how much of each asset will be in the investment portfolio of a fund is called the asset allocation. This can be divided in two categories; strategic asset allocation and tactical asset allocation.

Strategic asset allocation determines the target allocation of institutional investors for the major asset classes, representing the investment policy of the company. The purpose for a pension fund or insurance company is to ensure that the company will have enough assets to pay off their obligations and the allocation should reflect the level of risk aversion of the company. Anson (2004) states that the higher the allocation to fixed income securities the more risk averse the company is. Generally, the target is provided in the form of a percentage or a range which the asset allocation of the fund should lie within. Major asset classes are public and private equity, fixed income securities, real estate and cash (Anson, 2004).

Tactical asset allocation is used to take advantage of inefficiencies within financial markets, with the purpose of adding value to the fund. Within the range set by the strategic asset allocation the investor can attempt to add value. The tactical asset allocation determines how much of the portfolio is allocated to active accounts and how much is allocated to a passive benchmark for each of the asset classes. A main difference between strategic asset allocation and tactical asset allocation is the time frame, the latter is reassessed more often (Anson, 2004).

Previous research

Gorter and Bikker (2013) find that pension funds allocate a higher percentage of their portfolio towards equity investments compared to insurers. They conclude that pension funds take on more investment risk than insurance companies. They state that a reason for this is that insurance companies can easily lose policyholder when their solvency ratio gets too low, while for pension funds a lot of the participants cannot close their contracts. It can therefore be expected that pension funds might react differently to negative interest rates than insurance companies. Based on these findings it is meaningful to make a distinction between the effect on pension funds and the effect on insurance companies, since their risk-incentives might differ.

When studying the effect of negative interest rates on these institutional investors I will make a distinction between the different types of pension funds and insurers. As mentioned in the introduction, in this paper I will distinguish between four types of insurance companies, namely life enterprises, non-life enterprises, composite enterprises and reinsurance enterprises. Composite enterprises are insurers that do both life insurances and non-life insurances. A reinsurance company provides insurances to insurers.

(6)

5 their liabilities. This strategy protects the company against changes in the interest rate in a scenario of defined real benefits (Polacek, 2018). As stated above, the liabilities of life insurers have a longer duration compared to non-life insurers, it can therefore be concluded based on the asset-liability matching strategy that life insurers invest in assets with a higher duration than non-life insurers.

Niedring (2015) finds that during years of low interest life European insurance companies started to invest in more risky asset classes, mainly bank bonds, and less in government bonds in the short term. This practice is also known as a ‘gamble for redemption’ or a ‘search for yield’. Since the interest rate is found to be below the equilibrium rate, Niedring concludes that on the long run the interest rate would likely increase. As expected, it was found that in the long run the demand for government bonds increased. This suggests that the interest rate has an impact on the asset allocation of life insurers, since there seems to be a positive correlation between the demand of government bonds and interest rates. Antolin et al. (2011) also describe this behaviour as a ‘gamble for redemption’.

With regards to pension funds, Antolin et al. (2011) state that there are two types of pension plans, namely defined-benefit plans and defined-contribution plans. A defined-benefit plan has liabilities valued at the present value of the future promised outgoing cashflows. The cashflows are often discounted at the current interest rate, by a predicted path for future discount rates or by an average of the historical interest rates. The lower this rate, the higher the liabilities of the pension plan are valued.

Furthermore, Antolin et al. (2011) find that the value of the assets of the pension fund might also be affected by low interest rates. When the discount rate falls, the future liabilities of a defined-benefit plan rise, unless asset returns rise proportionally, or more. However, economic conditions with low interest rates are often characterized by low growth. Lower growth would result in lower profits and lower returns on assets. Hence, when the interest rate is used as discount rate the return on the portfolio investment might also be lower. According to Antolin et al. (2011) low interest rates will also affect the value of the assets of a defined-contribution plan. If the contribution of the employers and employees is kept equal, in the long run lower investment returns will result in lower benefits. To keep the benefit on a constant level, either employers and employees must contribute more to the pension plan, or the pension fund must make riskier investments in order to increase the expected return from the investment.

Antolin et al. (2011) find that low interest rates should change the investment behaviour of pension funds. This theoretical implication is substantiated by Boubaker, Gounopoulos, Nguyen and Paltalidis (2018), who find that the investment behaviour of pension funds has shifted more and more towards equity investments. They find that a decline of 5% in the 10-year US Treasury yield decreases the investment in government bonds with 18%, while the investment in equity assets increases by 17%. This effect is found to be the strongest during periods of unconventional monetary policy.

(7)

6 that both life and non-life insurers decrease their allocation to equity and increase their allocation in fixed income securities when the return of equity is lower than the return of fixed income securities. This is in line with the described theory. Duijm and Bisschop (2017), however, find that in practice this strategy is not always applied by pension funds. Pension fund are namely shown to act countercyclical when the return of equity is higher than the return of fixed income securities. This means that, contrary to the theoretical expectations, they increase their allocation in fixed income securities when they over perform.

Levy, Levy and Edry (2003) conduct research on the US market during 1926 and 2000 testing the optimal level of savings when the after-tax real interest rate on US T-bills and the average return on long term government bonds is negative. They show that the optimal level of savings might be positive even when the interest rate is negative. Furthermore, they find that when T-bills and long-term government bonds have negative real interest rates, while other assets have positive real returns, there is still a positive demand for T-bills and long-term government bonds. They suggest that the reason for this positive demand is the low correlation of T-bills and long-term government bonds with stock market returns, this yields diversification benefits. This paper thus suggests that institutional investors will still allocate a certain part of their portfolio in fixed income securities when the interest rate is negative, due to diversification reasons.

Lian, Ma & Wang (2018) show that individuals allocate a significantly higher percentage of their portfolio towards risky assets when interest rates are lower, while the excess returns of the risky asset remains the same. Célérier & Vallée (2017) find that the risk incentives of individuals influence the investment behaviour of institutions. Institutions might allocate more of their portfolio towards risky assets to meet the demand of individuals. Combining these results, it can be concluded that institutional investors are likely to increase their risk-taking behaviour in times of low interest rates. However, there are also papers that prove otherwise. One example is the study of Bams, Schotman & Tyagi (2016a), who find that pension funds have increased the percentage of their portfolio allocated to fixed income securities, while decreasing the percentage of equity securities, in a low interest environment.

(8)

7 Summarizing the above, I conclude that there is a substantial amount of literature that argues that investment in fixed income assets will increase when interest rates are low, however, there are also papers suggesting that these investments will not rise. The effect of pension funds and insurance companies must be tested separately.

3. Hypotheses

Based on the literature I expect that, when interest rates are negative, pension funds and insurance companies will allocate a smaller share of their portfolio in fixed income. The hypotheses of this paper are therefore formulated as follows:

Hypothesis 1: Pension funds

H0: Negative interest rates do not have a significant impact on the percentage of fixed income

securities in the aggregated portfolios of European pension funds.

H1: Negative interest rates significantly lower the percentage of fixed income securities in the

aggregated portfolios of European pension funds.

Hypothesis 2: Insurance companies

H0: Negative interest rates do not have a significant impact on the percentage of fixed income

securities in the aggregated portfolios of European insurance companies.

H1: Negative interest rates significantly lower the percentage of fixed income securities in the

(9)

8

4. Data

To examine whether negative interest rates affect the investment behaviour I will compare two different sample periods; one period during which the interest rates were positive and one period in which the interest rates were either zero or negative. I choose the periods in line with the official interest rate set by the central bank of the respective country. For most countries this is the European Central Bank. One of their official interest rates is the deposit facility rate for which banks can make overnight deposits at the European Central Bank (ECB, 2018). The choice to use this rate is arbitrarily. The historical ECB interest rates can be found in appendix A, figure 1 below shows the course of the rate graphically.

In the course of 2012, the deposit facility rate reached a level of 0%. The sample period for the negative/zero interest rates for Euro countries will therefore be 2012 up to and including 2016, as I use the end of the year interest rate. The sample period for the positive interest rates for Euro countries will be 2004 up to and including 2011 because there is no data available before 2004. For non-Euro countries the years will differ, the exact interest rates can be found in appendix A. For Liechtenstein the interest rate in the years 2004-2007 are unavailable. This does not cause any problems, as I have no observations for Liechtenstein in those years for pension funds. For insurance companies this results in a smaller dataset, but since I work with yearly observations this effect is minimal.

I have collected data on the asset allocation of pension funds and insurance companies in Europe for both periods. The asset allocation is extracted from the website of the European Insurance and Occupational Pensions Authority (EIOPA). The EIOPA is part of the European System of Financial Supervision network and gives independent advice to the European Commission, the European Parliament as well as to the Council of the European Union. Data is obtained from fillings by national authorities and can be deemed reliable.

Figure 1: Historical end of the year ECB deposit facility rates.

-1,00 -0,50 0,50 1,00 1,50 2,00 2,50 3,00 3,50 4,00 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 Int er es t ra te ( in %) Date

(10)

9 A limitation of this dataset is that it only contains the aggregated data per country and no firm-specific data. Most previous studies use data from CEM Benchmarking. However, as funds provide their data to them under strict conditions of confidentiality, often with non-disclosure agreements governing its use, it could not be made available to me. For the sake of time, it was not possible to manually collect data from the annual reports of pension funds and insurance companies throughout Europe. This dataset is therefore the best available.

4.1 Pension funds

Table 1: Descriptive statistics for pension funds.

Item Average Median Minimum Maximum St. dev.

Average coverage ratio (in percentages) 107,90% 104,65% 74,97% 162,00% 13,29%

Debt and other fixed income securities

(in million €) 45.860 3.260 0,01 853.203 135.999

Equity and other variable-yield

securities (in million €) 37.623 1.306 0,02 700.171 121.773

Total assets (in million €) 107.423 12.181 0,15 1.788.666 304.743

Return on assets (in percentages) 4,99% 5,06% -19,32% 27,36% 5,75%

Number of members (in thousands) 2.452 425 0,09 26.480 4.918

Number of active members (in

thousands) 1.312 255 0,09 12.287 2.219

Number of retired members (in

thousands) 487 65 0,00 4.611 1.118

The numbers display the end of the year aggregated data from 25 European countries within 2004 and 2016. The data is generated from the EIOPA website. The average coverage ratio measures the ability of a pension fund to meet its financial obligations. Active members are all the people that pay a contribution to the fund. The total amount of members includes the active members, the deferred members and the retired members. A deferred member is a person that stopped paying contribution to the pension fund, but is not yet receiving the pension benefits. Each country is equally weighted in the calculation of the average value.

Table 1 above shows the average, median, minimum and maximum value of some of the important items from the pension fund dataset, plus the standard deviation. One thing that stands out is the high standard deviation. On average 43% of the assets of a pension fund are fixed income securities, for the median value this is only 26%. Looking at the difference between the average and median value and at the minimum and maximum value it shows that the data contains outliers in terms of total assets.

(11)

10 Another thing that stands out from the data is that in the period of zero/negative interest rates the amount of total assets is higher compared to the overall value, this suggest pension funds invest more in general. I will look into the investment behaviour in greater detail in section 6.

Table 2: Descriptive statistics for pension funds.

Item Average 2004-2011 Median 2004-2011 Average 2012-2016 Median 2012-2016

Average coverage ratio (in percentages) 107,98% 104,31% 107,80% 104,93%

Debt and other fixed income securities

(in million €) 33.724 2.543 59.388 4.126

Equity and other variable-yield

securities (in million €) 35.095 1.258 40.517 1.368

Total assets (in million €) 91.465 12.599 124.647 6.747

Return on assets (in percentages) 3,95% 4,76% 6,03% 5,30%

Number of members (in thousands) 2.191 407 2.742 454

Number of active members (in

thousands) 1.180 267 1.443 215

Number of retired members (in

thousands) 473 58 502 71

The numbers display the end of the year aggregated data from 25 European countries for two time periods. 2004 until 2011 displays the period with positive ECB interest rates, 2012 until 2016 displays the period in which the end of the year ECB interest rate was either zero or negative. The data is generated from the EIOPA website. The average coverage ratio measures the ability of a pension fund to meet its financial obligations. Active members are all the people that pay a contribution to the fund. The total amount of members includes the active members, the deferred members and the retired members. A deferred member is a person that stopped paying contribution to the pension fund, but is not yet receiving the pension benefits. Each country is equally weighted in the calculation of the average value.

4.2 Insurance companies

(12)

11

Table 3: Descriptive statistics for insurance companies

Item Average Median Minimum Maximum St. dev.

Shares and other variable-yield

securities (in million €) 10.719 133 0,004 506.362 37.836

Debt and other fixed income

securities (in million €) 31.269 823 0,030 908.498 90.896

Total investment assets (in

million €) 70.877 1.465 0,004 2.186.699 221.977

Total assets (in million €) 75.691 1.881 0,250 2.197.097 230.781 The numbers display the end of the year aggregated data from 31 European countries within 2005 and 2015. The data is generated from the EIOPA website. Each country is equally weighted in the calculation of the average value. The difference between the total value of investment assets and the total value of assets is the value of reinsurers’ share of technical provisions, subscribed unpaid capital, intangible assets, debtors, prepayments, accrued income and other assets.

Table 4 shows the descriptive statistics of the period within the dataset where the interest rates where positive, and the descriptive statistics of the period within the dataset where the interest rates where either zero or negative. During the timeframe of positive interest rates on average 41% of the assets of an insurance company are fixed income securities, for the median value this is 43%. During the period of zero/negative interest rates on average 43% of the assets of an insurance company are fixed income securities, for the median value this is 55%. The average asset allocation does not seem to be correlated with the interest rate, since there are no major changes in the relative portfolios of insurance companies. However, since these statistics do not control for any other variable, we cannot base any conclusions on these descriptive statistics.

Table 4: Descriptive statistics for insurance companies.

Item Average 2005-2011 Median 2005-2011 Average 2012-2015 Median 2012-2015

Shares and other variable-yield

securities (in million €) 10.367 120 11.632 167

Debt and other fixed income

securities (in million €) 29.143 819 36.771 962

Total investment assets (in

million €) 68.075 1.574 77.663 1.310

Total assets (in million €) 71.918 1.895 84.912 1.753

(13)

12

5. Methodology

5.1 Model

I will estimate the following equations using the collected data to study the effect of interest rates on the percentage of fixed income securities in the portfolios of either pension funds or insurance companies. This will be tested using the following general model:

𝐹𝐼𝑖,𝑡 = 𝛼 + 𝛽1∗ 𝐼𝑁𝑇𝑖,𝑡 + 𝛽2∗ 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 ∗ 𝐼𝑁𝑇𝑖,𝑡 + 𝛽3∗ 𝑋𝑖,𝑡+ 𝛽4∗ 𝐹𝐼𝑖,𝑡−1+ 𝛿𝑖 + 𝜀𝑖,𝑡.

(1)

In which 𝐹𝐼𝑖,𝑡 is the percentage of fixed income securities as a share of total assets, that the

pension funds or insurance companies of a certain European country in a certain year hold. 𝐼𝑁𝑇𝑖,𝑡 is the interest rate of the central bank of a certain European country in a certain year, for most countries this is the deposit facility rate of the European Central Bank. Appendix A presents an overview of all used interest rates.

𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 is a dummy variable formed to make a distinction between the period with positive interest rates and the period with negative interest rates. It has a value of 1 if the interest rate is either negative or zero, and a value of 0 when the interest rate is positive, this means that the effect of the interest rate on the investment behaviour of either pension funds or insurance companies is allowed to be different when the interest rate is negative. The choice to include zero interest rates in the negative sample is arbitrarily. However, it does not influence the results as multiplying the dummy with the interest rate yields zero in both scenarios.

Different funds might have a different investment behaviour. Andonov et al. (2017) account for this in their model by using fund fixed effects, which has a significant effect on the percentage of a pension fund’s portfolio that is allocated towards risky assets. A similar approach is implemented in the research of Heider, Saidi and Schepens (2019), Duijm and Bisschop (2017) and CGFS (2018). Since my dataset consists of the aggregated data per country, I will use country fixed effects (𝛿𝑖) instead of fund fixed effects. Testing on my dataset,

country fixed effects are significant. A fixed effects model is thus preferred over a pooled OLS model. This also shows by comparing both models according to the Akaike Information Criteria, the fixed effects model has the lowest AIC score. Based on the Hausman test a fixed effects model is preferred over a random effects model, as the test shows that the main model contains endogeneity. A limitation of the fixed effects model is that it cannot control for factors that vary over time. However, this is easily solved by introducing time fixed effects to the model.

According to Duijm and Bisschop (2017) a lagged value of the dependent variable (𝐹𝐼𝑖,𝑡−1)

(14)

13 I will run several versions of equation (1) to document the relation between the interest rate and the percentage of fixed income securities in the portfolios of institutional investors. First, I start with the assumption that both 𝛽2 and 𝛽3 are equal to zero, this enables me to check the unconditional relationship between 𝐼𝑁𝐹𝑖,𝑡 and 𝐹𝐼𝑖,𝑡.

Next, I introduce time varying effects. I will relax the assumption that 𝛽2 is equal zero by adding the interaction term 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 ∗ 𝐼𝑁𝑇𝑖,𝑡 to the model. This allows the effect of the interest rate on the percentage of fixed income securities in the portfolio of an institutional investor to be different when the interest rate is negative. This enables me to test the hypothesis: Negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European pension funds/insurance companies.

Third, I will add time fixed effects to equation (1), this will capture the influence of time trends in the dataset. As mentioned before, this will overcome the limitation of the fixed effects model. Finally, I add a set of country specific control variables to equation (1), to account for other factors that might influence the percentage of fixed income securities in the portfolio of an institutional investor. The set of control variables are described in detail in section 5.2. As a part of the control variables I account for the effect of the crisis by introducing crisis dummies. I will discuss this further at the end of the next section.

5.2 Control variables

As described in section 2, Andonov et al. (2017) have developed a model to predict the percentage of a pension fund’s portfolio that is allocated towards risky assets. In their model they use, besides a constant, a variable to account for the percentage of retired members, time fixed effects and fund fixed effects. Bodie, Light and Morck (1987) also account for the effect of the percentage of retired members on the percentage of fixed income securities in the portfolios of pension funds, in contrast to Andonov et al. (2017) they find no significant effect. The percentage of retired members will be included in this regression for pension funds. Bodie et al. (1987) also use the size of the fund as a variable in their regression, they show that this significantly lowers the percentage of fixed income securities in the portfolio of a pension fund. This means that the larger the fund, the lower their allocation towards fixed income securities. Andonov et al. (2017) find similar results. In this regression the total amount of members per country in a specific year will be used as a proxy for the funds size in the case of pension funds. As this data is not available for insurance companies, I will use total assets as a proxy for size when testing for the effect on insurance companies.1

Committee on the Global Financial System (CGFS) (2018) studies whether the level of inflation influences the returns on assets, which in its turn can influence the asset allocation of a pension fund. However, they find no evidence that the effect was statistically different from zero. Andonov et al. (2017) include an inflation protection policy in their regression, this is

1 Using total assets as a proxy for size instead of the total amount of members per country in a specific year in

(15)

14 shown to have a significant effect on the percentage of a pension fund’s portfolio that is allocated towards risky assets. I, therefore, include the level of inflation in a certain European country in a certain year.

CGFS also shows that the real GDP growth rate positively influences the returns on assets. Since the return on assets can influence the asset allocation of a pension fund, I include the real GDP growth rate to capture this effect in a certain year and for a certain country.

The coverage ratio of a pension fund shows whether the pension fund is over- or undercapitalized. Bodie et al. (1987) argue that based on a corporate financial perspective overcapitalized pension funds should be mainly, or even entirely, invested in fixed income securities. This perspective states that overcapitalized pension funds should act as a tax shelter, since they can earn a pre-tax return on their investments and pass it on to their shareholder. This makes it beneficial to invest in highly taxed securities, like fixed income. Pension funds that are undercapitalized, on the other hand, should therefore not invest in fixed income securities. Their empirical analysis confirms this notion. The coverage ratio is also considered in the study of Boubaker et al. (2018). This suggests that the coverage ratio of a pension fund has an impact on their asset allocation and it should therefore be included in the econometric model of this paper in order to avoid a bias in the results.

Andonov et al. (2017) make a distinction between pension funds with defined contribution plans and pension funds with defined benefit plans. They find that this property has a significant effect on the percentage of fixed income holdings in the portfolios of pension funds. In this study I cannot make a distinction between a defined contribution plan or a defined benefit plan, since there is only aggregated data available and some of the countries have both defined contribution plans and defined benefit plans. However, in the case of insurance companies I can make a distinction in terms of the type of company, namely life enterprises, non-life enterprises, composite enterprises and reinsurance enterprises. I will do this by incorporating this in the fixed effects.

Both Duijm and Bisschop (2017) and CGFS (2018) include a dummy variable for the financial crisis. They both include a general dummy for all observations. Papaioannou, Park, Pihlman and Hoorn (2013) show that countries reacted differently to the financial crisis and changed their asset allocation in opposite directions. Spain, USA, Finland and Portugal lowered the amount of equity in their portfolios in 2008, while Poland, Turkey, Norway and Italy increased their amount of equity holdings. Since this research focusses on pension funds in multiple European countries, I will add a dummy variable for each country to capture the country specific effects of the crisis.

5.3 Robustness tests

(16)

15

5.3.1 Change in fixed income

As mentioned before, Bams et al. (2016b) show that institutional investors are slow in incorporating changes in their portfolios. This could cause the results to be insignificant, while in fact institutional investors might anticipate their investment behaviour on the interest rate. Another problem with the data is that the sample size is too small, as a result of that the interest rate and a few of the control variables are unfortunately non-stationary. Both problems can be solved by taking the first differences of equation (1).2 A lag of the dependent variable is no longer needed based on the Akaike Information Criteria. The control variables will be transformed to first differences.

𝐹𝐼𝑖,𝑡− 𝐹𝐼𝑖,𝑡−1 = 𝛼 + 𝛽1∗ (𝐼𝑁𝑇𝑖,𝑡− 𝐼𝑁𝑇𝑖,𝑡−1) + 𝛽2∗ 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 ∗ (𝐼𝑁𝑇𝑖,𝑡 − 𝐼𝑁𝑇𝑖,𝑡−1) + 𝛽3∗ 𝑋𝑖,𝑡+ 𝛿𝑖+ 𝜀𝑖,𝑡.

(2)

This equation will be tested using the same methodology as described at the end of section 5.1. If the results of these regressions are similar to main tests, this increases the robustness of the results.

5.3.2 Euro countries

The dataset consists mainly of observations from Euro countries. However, part of the observations come from European non-Euro countries. It is possible that the effect of the interest rate on the percentage of fixed income securities in the portfolios of European institutional investors is different in Euro countries compared to non-Euro countries. By excluding all observations from non-Euro countries and repeating the above methodology based on equation (1), it is possible to conclude whether there is a difference in the effect for both subsamples. If the results of these regressions are comparable to the full sample tests, it confirms those results.

5.3.3 Outliers

Section 4 shows that the dataset consists of observations with total assets valued far above or below the median value. Those observations are called outliers and could possibly influence the results. Excluding the observations with the lowest 10% of total assets and the observations with the highest 10% of total assets allows me to check whether the results are influenced by those outliers. If the results of these regressions are similar to main tests, this increases the robustness of the results.

2 The results of the Westerlund panel-data cointegration test suggest that there is no cointegration between the

(17)

16

6. Empirical analysis

This section displays the results of this study. First, I will discuss the results for the pension funds. Thereafter I will present the results for the insurance companies. Finally, I will show some robustness tests.

6.1 Pension funds

In this section I will discuss the results of the regressions for pension funds, following the methodology described in section 5. I will interpret the results one by one, starting with the interpretation of regression (1), as this the most basic regression. Then in each regression I add a control factor to the model and show what happens to the results.

Table 5: Fixed effects regression for pension funds.

(1) (2) (3) (4) (5) VARIABLES Official interest rate Official interest rate Time fixed effects Control variables Control variables + Crisis dummy FIt-1 0.658*** 0.658*** 0.653*** 0.643*** 0.627*** (0.133) (0.133) (0.122) (0.145) (0.151) INTi,t -0.006 -0.007 0.008 0.003 0.007 (0.006) (0.006) (0.014) (0.015) (0.016)

NEGATIVE * INTi,t -0.014 0.032 0.045 0.050

(0.041) (0.062) (0.085) (0.081)

Country fixed effects Yes Yes Yes Yes Yes

Time fixed effects No No Yes Yes Yes

Control variables No No No Yes Yes

Crisis dummy No No No No Yes

Constant 0.154** 0.155** 0.166*** 0.144** 0.126*

(0.058) (0.056) (0.052) (0.068) (0.069)

Observations 233 233 233 173 173

Adjusted R2 0.444 0.442 0.448 0.458 0.43

Number of countries 25 25 25 19 19

The table displays the coefficients of the variables listed on the left side of the table. The robust standard errors are displayed in parentheses. * means that the coefficient is significant at a 10% level, ** means that the coefficient is significant at a 5% level and *** means that the coefficient is significant at a 1% level. FIt-1 stands for the percentage of fixed income securities in the aggregated portfolios of European pension funds in the previous year. INTi,t stands for the official interest rate of a certain country. NEGATIVE * INTi,t stands for the interaction term of the dummy NEGATIVE and the interest rate. The dummy NEGATIVE equals 1 when the interest rate is either negative or zero. The control variables consist of the percentage of retired members, the number of total members per country, the inflation rate per country, the GDP growth rate per country and the average coverage ratio per country. A crisis dummy that is equal to 1 when the year is 2008 is included for each country in regression (5). The dataset consists of 25 European countries with data over the period 2004-2016.

(18)

17 income securities in the portfolios of pension funds in the current year and is significant at a confidence level of 99%. Similar to the study of Andonov et al. (2017), the effect of the interest rate is not significantly different from zero at any of the usual confidence levels. Based on this regression there is not enough evidence to infer that the interest rate effects the percentage of fixed income securities in the portfolios of pension funds.

The second regression introduces time varying effects by adding the interaction term 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 ∗ 𝐼𝑁𝑇𝑖,𝑡 to the model. This allows the effect of the interest rate on the percentage of fixed income securities in the portfolio of a pension fund to be different when the interest rate is negative. The percentage of fixed income securities in the portfolios of pension funds in the year before still has the highest coefficient and is significant at a confidence level of 99%. There is not enough evidence to infer that the official interest rate has any effect on the percentage of fixed income securities in the portfolios of pension funds. The interaction term

𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 ∗ 𝐼𝑁𝑇𝑖,𝑡 is highly insignificant at a 90% confidence level. This means that I do

not find any evidence that negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European pension funds.

Regression (3) adds time fixed effects to the model, this will capture the influence of time trends in the dataset. This slightly increases the adjusted R2. However, both the coefficient of the interest rate and the coefficient of the interaction term 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 ∗ 𝐼𝑁𝑇𝑖,𝑡 are not significant at a 10% significance level. This means that adding time fixed effects to the model does not change the outcome of this study, there is still no evidence that negative interest rates have an effect on the percentage of fixed income securities in the aggregated portfolios of European pension funds.

Next, I add country specific control variables to the regression, as described in the methodology section. Unfortunately, this lowers the number of countries and thus the number of observations in the dataset, as the percentage of retired members is not available for some of the countries. Furthermore, the inflation rate is unavailable for Liechtenstein. However, it is still valuable to report this regression, as it shows whether the results hold when I account for other factors.3 As in the earlier described regressions it can be said with 99% confidence that the percentage of fixed income securities in the portfolios of pension funds in the year before seems to have the biggest influence on the percentage of fixed income securities in the portfolios of pension funds in the current year. The effect of the official central bank interest rate on the percentage of fixed income securities in the aggregated portfolios of pension funds yields no significant result at any common confidence level.

Finally, adjusting for the possible effect of the crisis by adding country specific crisis dummies to the model confirms the earlier described results. A reason that the negative interest rates seem to have no significant impact on the percentage of fixed income securities in the portfolios of pension funds could be that pension funds are unable to change their entire portfolio at any point in time, due to liquidity constraints. Another reason pension funds might need some time to make changes to their portfolio, according to Bams et al. (2016b), is to reduce the potential

3 Excluding the percentage of retired members as a control variable, to avoid losing observations, does not

(19)

18 impact to the market. They find that pension funds are, as expected, slow in adjusting their actual investment portfolio to their strategic asset allocation. A way to get around this problem would be to use the change in the percentage of fixed income securities in the portfolios of pension funds. The results of this test will be showed in section 6.3.

6.2 Insurance companies

This section presents the results of the regressions for insurance companies, following the methodology discussed in section 5. I will interpret the results one by one, starting with the interpretation of regression (1), as this the most basic regression. Then in each regression I add a control factor to the model and show what happens to the results.

Table 6: Fixed effects regression for insurance companies.

(1) (2) (3) (4) (5) VARIABLES Official interest rate Official interest rate Time fixed effects Control variables Control variables + Crisis dummy FIt-1 0.608*** 0.608*** 0.624*** 0.627*** 0.648*** (0.057) (0.057) (0.059) (0.062) (0.068) INTi,t -0.005** -0.005** -0.001 -0.002 -0.001 (0.002) (0.002) (0.004) (0.005) (0.005)

NEGATIVE * INTi,t 0.001 0.021 0.010 0.007

(0.021) (0.051) (0.060) (0.064)

Country and enterprise fixed effects

Yes Yes Yes Yes Yes

Time fixed effects No No Yes Yes Yes

Control variables No No No Yes Yes

Crisis dummy No No No No Yes

Constant 0.190*** 0.190*** 0.195*** 0.160*** 0.146***

(0.027) (0.027) (0.033) (0.038) (0.041)

Observations 845 845 839 815 815

Number of entities 94 94 94 91 91

Adjusted R2 0.402 0.401 0.409 0.409 0.433

The table displays the coefficients of the variables listed on the left side of the table. The robust standard errors are displayed in parentheses. * means that the coefficient is significant at a 10% level, ** means that the coefficient is significant at a 5% level and *** means that the coefficient is significant at a 1% level. FIt-1 stands for the percentage of fixed income securities in the aggregated portfolios of European insurance companies in the previous year. INTi,t stands for the official interest rate of a certain country. NEGATIVE * INTi,t stands for the interaction term of the dummy NEGATIVE and the interest rate. The dummy NEGATIVE equals 1 when the interest rate is either negative or zero. The control variables consist of the inflation rate per country, the GDP growth rate per country and the total amount of assets as a proxy for size. A crisis dummy that is equal to 1 when the year is 2008 is included for each country in regression (5). The dataset consists of 31 European countries with data over the period 2005-2015.

(20)

19 securities in the portfolios of insurance companies in the year before seems to have the biggest influence on the percentage of fixed income securities in the portfolios of insurance companies in the current year. This effect is significant at a 99% confidence level. As predicted, the official interest rate of the corresponding county has a negative coefficient and is significant at a 95% confidence level. This means that the lower interest rate, the higher the allocation to fixed income securities in the portfolios of insurance companies. This corresponds to the majority of the described literature. The adjusted R2 shows that 40.2% of the variation in the percentage of fixed income securities in the aggregated portfolios of insurance companies is explained by this model.

Regression (2) adds time varying effects to the model, in the form of the interaction term 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 ∗ 𝐼𝑁𝑇𝑖,𝑡 to test whether the effect of interest rates on the percentage of fixed

income securities in the portfolios of insurance companies is different when the interest rate is negative. The results show that at a 90% confidence level there is no evidence to infer that the effect of the interest rate is different when the interest rate is negative, therefore I cannot reject the null hypothesis: Negative interest rates do not have a significant impact on the percentage of fixed income securities in the aggregated portfolios of European insurance companies. Similar to regression (1) the percentage of fixed income securities in the portfolios of insurance companies in the previous year has a positive effect on the percentage of fixed income securities in the portfolios of insurance companies in the current year and is significant at a 99% level. The interest rate has a negative effect on the percentage of fixed income securities, thus the lower the interest rate, the higher the percentage of fixed income securities, all else held constant. This effect is significant at a 95% confidence level.

Next, I add time fixed effects to the model to capture the influence of time trends. Regression (3) shows no major changes in the results of the percentage of fixed income securities in the portfolios of insurance companies in the previous year and the interaction term 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸 ∗ 𝐼𝑁𝑇𝑖,𝑡 . The first is still significant at a 99% confidence level, the latter is still not significant

at any of the usual confidence levels. However, in contrast to the earlier described regressions, the effect of the interest rate is no longer significant when I account for time fixed effects. This result is in line with Andonov et al. (2017) who find no evidence that declining interest rates have an influence on the percentage of risky assets in the portfolios of pension funds.

(21)

20

6.3 Robustness

I will first discuss all robustness tests, as described in the methodology section, for pension funds and then I will present the robustness tests for insurance companies.

6.3.1 Pension funds

6.3.1.1 Change in fixed income

As described earlier, a problem with the results above is that the sample size is too small for the interest rate to be stationary. By taking the first differences of the interest rate, the variable becomes stationary. This also bridges the problem that the result might be insignificant, because institutional investors invest in long term securities. In the course of a year they cannot change their entire portfolio. Therefore, this section shows the change in the percentage of fixed income securities in the portfolios of pension funds regressed on the change in the interest rate and the change in the control variables, as described in section 5. Table 7 below shows the results of this regression.

Table 7: Robustness test first differences pension funds.

The table displays the coefficients of the variables listed on the left side of the table. The robust standard errors are displayed in parentheses. * means that the coefficient is significant at a 10% level, ** means that the coefficient is significant at a 5% level and *** means that the coefficient is significant at a 1% level. INTi,t stands for the official interest rate of a certain country. NEGATIVE * INTi,t stands for the interaction term of the dummy NEGATIVE and the interest rate. The dummy NEGATIVE equals 1 when the interest rate is either negative or zero. The control variables consist of the change in the percentage of retired members, the change in the number of total members per country, the change in the inflation rate per country, the change in the GDP growth rate per country and the change in the average coverage ratio per country. A crisis dummy that is equal to 1 when the year is 2008 is included for each country in regression (5). The dataset consists of 25 European countries with data over the period 2004-2016.

(1) (2) (3) (4) (5) VARIABLES Official interest rate Official interest rate Time fixed effects Control variables Control variables + Crisis dummy

INTi,t - INTi,t-1 -0.003 -0.002 0.002 -0.007 -0.010

(0.004) (0.004) (0.009) (0.010) (0.013)

NEGATIVE * (INTi,t - INTi,t-1) -0.043 -0.045 -0.027 -0.028

(0.061) (0.058) (0.082) (0.088)

Country fixed effects Yes Yes Yes Yes Yes

Time fixed effects No No Yes Yes Yes

Control variables No No No Yes Yes

Crisis dummy No No No No Yes

Constant 0.004*** 0.002 -0.031 -0.036 -0.034

(0.001) (0.003) (0.019) (0.028) (0.029)

Observations 233 233 233 167 167

Number of countries 25 25 25 17 17

(22)

21 The change in the official interest rate of the country in respect is not significant at any of the usual confidence levels. Also, the interaction term of the dummy 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸, that is equal to 1 when the corresponding interest rate is either negative or zero, and the change in the interest rate does not show any significant outcomes. The null hypothesis cannot be rejected as there is not enough evidence that the interest rate being negative has any impact on the change in the percentage of fixed income securities in the portfolios of pension funds. The Hausman test shows that the assumptions of the random effects model hold in this regression, in essence, there is no endogeneity in this model. The random effects model could thus be used to test the hypothesis. However, this does not change the outcome of this paper. In addition, adding time fixed effects, control variables or the effect of the crisis to the model, as described in the methodology section, does not change the results of this test as can be seen in table 7 above.

6.3.1.2 Euro countries

This section shows the results of testing on only countries that have the Euro as a currency for pension funds. The interest rate for this set of countries is thus the deposit facility rate of the European Central Bank. The results will show whether the effect is different when excluding non-Euro countries. The regressions are displayed in table 12 in appendix C.

The effect of the interest rate is similar to the full sample, when I assume the effect of the control variables is equal to zero. However, when I account for the effect of the control variables the effect of the interest rate becomes significant. This does not change when I include the effect of the crisis to the model. A 1-percent decrease in the interest rate increases the percentage of fixed income securities in the aggregated portfolios of European pension funds on average with 0.130%, all else held constant. This effect is in line with the studies of Lian et al. (2018) and Célérier & Vallée (2017). This result, combined with the result of the original sample, suggests that interest rate in non-Euro countries does not affect the percentage of fixed income securities in the aggregated portfolios of the pension funds in those countries.

With regards to interaction term, I find no evidence that negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European pension funds. These findings support the conclusion from the original sample.

6.3.1.3 Outliers

Section 4 shows that the dataset contains a number of observations with an amount of total assets that is much higher than the median value. By excluding the observations with the lowest and highest 10% of total assets, the results will show whether the results are influenced by the outliers. This section presents the results for this test. The regressions are displayed in table 14 in appendix D.

(23)

22 for the effect of the crisis and/or the effect of other control variables the results show that the interest rate significantly lowers the percentage of fixed income securities in the aggregated portfolios of European insurance companies. This is consistent with the results of the subsample of Euro countries.

Similar to the full sample and the other robustness tests, the interaction term is insignificant at any common confidence level. This confirms that there is no evidence that negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European insurance companies, besides the effect of a lower interest rate itself.

6.3.2 Insurance companies

6.3.2.1 Change in fixed income

As described in the methodology section, a problem of the main test is that insurance companies might be slow to adjust their portfolio as they are long-term investors. A way to get around this problem is by taking the first differences. Table 8 below shows the results of these regressions.

Table 8: Robustness test first differences insurance companies.

The table displays the coefficients of the variables listed on the left side of the table. The robust standard errors are displayed in parentheses. * means that the coefficient is significant at a 10% level, ** means that the coefficient is significant at a 5% level and *** means that the coefficient is significant at a 1% level. INTi,t stands for the official interest rate of a certain country. NEGATIVE * INTi,t stands for the interaction term of the dummy NEGATIVE and the interest rate. The dummy NEGATIVE equals 1 when the interest rate is either negative or zero. The control variables consist of the change in the inflation rate per country, the change in the GDP growth rate per country and the change in the amount of total assets. A crisis dummy that is equal to 1 when the year is 2008 is included for each country in regression (5). The dataset consists of 31 European countries with data over the period 2004-2016.

(1) (2) (3) (4) (5) VARIABLES Official interest rate Official interest rate Time fixed effects Control variables Control variables + Crisis dummy

INTi,t - INTi,t-1 -0.012*** -0.011*** -0.010** -0.010* -0.020***

(0.003) (0.003) (0.005) (0.005) (0.006)

NEGATIVE * (INTi,t - INTi,t-1) -0.011 -0.029 -0.061 -0.049

(0.026) (0.047) (0.054) (0.055)

Country and enterprise fixed effects

Yes Yes Yes Yes Yes

Time fixed effects No No Yes Yes Yes

Control variables No No No Yes Yes

Crisis dummy No No No No Yes

Constant 0.002** 0.001 -0.011 -0.008 0.003

(0.001) (0.001) (0.008) (0.008) (0.009)

Observations 836 836 836 815 815

Number of entities 94 94 94 91 91

(24)

23 The change in the official interest rate of the country in respect has a negative coefficient and is significant at at least a 90% confidence level for all of the regression, in contrast to the original regressions. When I account for time fixed effects, control variables and the effect of the crisis in the first difference regression the interest rate is even significant at a 99% confidence level. The size of the coefficient is larger than in the main test, this could mean that the original regression underestimates the effect of the interest rate.

The interaction term between the dummy 𝑁𝐸𝐺𝐴𝑇𝐼𝑉𝐸, which is equal to one when the interest is either negative of zero, and the change in the official central bank interest rate is not significant at any of the common significance levels. This shows that there is not enough evidence to conclude that the change in the percentage of fixed income securities in the portfolios of insurers is impacted by the fact that the interest rate is negative. Adding control variables to the regression does not change the outcome. This confirms the findings of the main test.

6.3.2.2 Euro countries

In this section I will discuss the results of testing on only countries that have the Euro as a currency for insurance companies. The official interest rate for this set of countries is the European Central Bank deposit facility rate. The results will show whether the effect is different in Euro countries compared to in non-Euro countries. The regressions are displayed in table 13 in appendix C.

When regressing the percentage of fixed income securities in the aggregated portfolios of insurance companies in a certain country on the deposit facility rate of the ECB, accounting for the percentage of fixed income securities in the previous period, the effect of the interest rate is significant at a 99% confidence level. A 1-percent increase in the interest rate decreases the percentage of fixed income securities in the current year with 0.011 on average, all else held constant. In the full sample this effect was 0.005. This does not change when I account for time-varying effects. However, similar to the full sample, when I account for time fixed effects, control variables and/or the crisis dummy the effect is no longer significant at any of the common levels.

The interaction term is insignificant at a 90% confidence level, just like in the original sample. This confirms that there is no evidence that negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European insurance companies, besides the effect of a lower interest rate itself.

6.3.2.3 Outliers

(25)

24 Similar to the full sample the interest rate significantly affects the percentage of fixed income securities in the aggregated portfolios of European insurance companies at a 99% confidence level, accounting for the percentage of fixed income securities in the previous period. A 1-percent increase in the interest rate decreases the 1-percentage of fixed income securities in the current year with 0.006 on average, all else held constant. In the full sample this effect was 0.005. This does not change when I account for time-varying effects. However, similar to the full sample, when I account for time fixed effects, control variables and/or the crisis dummy the effect is no longer significant at any of the common levels. I can therefore conclude that the effect of the interest rate is not determined by the outliers in the data.

The time-varying effect, as measured by the interaction term is insignificant at any common confidence level. This is in line with the results of the main test and the other robustness tests. This confirms that there is no evidence that negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European insurance companies, besides the effect of a lower interest rate itself.

7. Conclusions

In this study, I investigate the effect of a negative interest rate on the percentage of fixed income securities in the aggregated portfolio of European pension funds and insurance companies. I can conclude that there is not enough evidence to infer that negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European pension funds. Furthermore, the interest rate in itself does not have a significant effect on the percentage of fixed income securities. My findings are in contradiction with the life cycle theory followed by, among others, Duijm and Bisschop (2017), the theory states that institutional investors are often likely to act as market stabilizer in the sense that they generally sell assets that have a relatively high price, increasing the supply of this asset class, while they buy assets with a relatively low price. However, consistent with my findings, they show that this strategy is not followed by pension funds in practice. My results are also consistent with Andonov et al. (2017) who find that there is no significant relation between the interest rate and the asset allocation of pension funds.

Interest rates have trended downwards during my sample period. Furthermore, Bams et al. (2016b) show that pension funds are slow in incorporating changes in their portfolios. This could cause the results to be insignificant, while in fact interest rates might affect the percentage of fixed income securities in the aggregated portfolio of European pension funds and insurance companies. Therefore, a robustness test is set up to test the effect of the change in interest rate on the change in the percentage of fixed income securities, instead of testing on the percentage fixed income securities itself. However, the results of the regression do not change the outcome of this study. Based on this dataset, the null hypothesis cannot be rejected. There is not enough evidence to infer that negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European pension funds.

(26)

25 fixed effects the interest rate is actually significant at a 95% confidence level. This is consistent with the findings of Lian et al. (2018) and Célérier & Vallée (2017). However, there is no evidence that the effect of the interest rate on the percentage of fixed income securities in the aggregated portfolios of European pension funds is different when the interest rate is negative. A subsample of only Euro countries confirms these findings. Altogether, I cannot reject the null hypothesis: Negative interest rates do not have a significant impact on the percentage of fixed income securities in the aggregated portfolios of European pension funds.

Regarding insurance companies, I find no evidence that negative interest rates significantly lower the percentage of fixed income securities in the aggregated portfolios of European insurance companies. Consistent with the findings of Lian et al. (2018), Célérier & Vallée (2017), Duijm & Bisschop (2017) and Niedring (2015) I find that when testing for the unconditional relationship between the official interest rate and the percentage of fixed income securities, the interest rate is significant at a 95% confidence level. However, when I take time fixed effects into account this effect is no longer significant. Therefore, there is not enough evidence to infer that interest rates negatively affect the percentage of fixed income securities in the aggregated portfolios of European insurance companies. This is consistent with the results of Andonov et al. (2017).

Testing for the relation between the change in the interest rate and the change in the percentage of fixed income securities in the aggregated portfolios of European insurance companies confirms the above results. The same applies to both the results of the regressions without outliers as the regressions without non-Euro countries. Altogether, I do not find enough evidence to reject the null hypothesis: Negative interest rates do not have a significant impact on the percentage of fixed income securities in the aggregated portfolios of European insurance companies.

(27)

26

References

Andonov, A., Bauer R.M.M.J. & Cremers, K.J.M. (2017). Pension fund asset allocation and liability discount rates. The Review of Financial Studies, 30(8), 2555–2595.

Anson, M.J.P. (2004) Strategic versus tactical asset allocation. The Journal of Portfolio Management, 30(2), 8-22.

Antolin, P., Schich, S. & Yermo, J. (2011) The economic impact of protracted low interest rates on pension funds and insurance companies. OECD Journal: Financial Market Trends, 2011(1), 237-256.

Banca Naţională a României (n.d.) Monetary policy and standing facilities interest rates. - Available at: https://www.bnr.ro/Monetary-Policy--3318.aspx [Accessed 31 May 2019]

Bank of England (2019) Interest rates and bank rates. - Available at:

https://www.bankofengland.co.uk/monetary-policy/the-interest-rate-bank-rate [Accessed 28 April 2019]

Bams, D., Schotman, P.C. & Tyagi M. (2016a). Pension fund asset allocation in low interest rate environment. Netspar Discussion Paper, 3, 1-40.

Bams, D., Schotman, P.C. & Tyagi M. (2016b). Asset allocation dynamics of pension funds. Netspar Discussion Paper, 3, 1-35.

Bendrich, D. & Bergström, J. (2015) Impact of asset allocation on insurance companies’ performance: A study of the European economic area. - Available at http://www.diva-portal.org/smash/get/diva2:844037/FULLTEXT01.pdf [Accessed 18 March 2019]. Umeå University.

Berends, K., McMenamin, R., Plestis, T. & Rosen, R.J. (2013) The sensitivity of life insurance firms to interest rate changes. Economic Perspectives, 37(2), 47-78.

Bodie, Z., Light, J.O., Morck, R. and Taggart, R.A. (1987) Funding and asset allocation in corporate pension plans: an empirical investigation. In: Bodie, Z., Shoven, J.B. and Wise A. Issues in Pension Economics. University of Chicago Press, pp. 15-48.

Boubaker, S., Gounopoulos, D., Nguyen, D.K. & Paltalidis, N. (2018) Reprint of: Assessing the effects of unconventional monetary policy and low interest rates on pension fund risk incentives. Journal of Banking & Finance, 92, 340-357.

Bouvatier, V. & Rigot, S. (2013) Pension funds' allocations to hedge funds: an empirical analysis of US and Canadian defined benefit plans. Applied Economics, 45(26), 3701-3710. Célérier, C. & Vallée, B. (2017) Catering to investors through security design: headline rate and complexity. Quarterly Journal of Economics, 132(3), 1469–1508.

Central bank of Iceland (n.d.) Key interest rates. - Available at: https://www.cb.is/other/key-interest-rates/ [Accessed 28 April 2019]

Referenties

GERELATEERDE DOCUMENTEN

Als leerlingen door middel van observerend leren (X1) creatieve teksten schrijven (X2), waarbij de nadruk ligt op het toepassen van non-core-vocabulaire (X3), dan zullen zij

official interest rate cuts give significant results for the Euro zone. The medium and

The table provides the results of the fixed effects model regressing the financial-debt-to-book value of total assets on the ten year treasury rate.. All data is recorded annually

This local peak is caused by local flow acceleration and is strongly coupled to the impinging velocity profile, which has to be of uniform type in order to generate an increasing

We conclude that compared to FeraSpin particles, these microfabricated particles saturate in fields as low as 12 mT, the shape of the hysteresis loop is relatively independent on

In the literature review, none of the methods found were able to comprehensively combine the market and the technology change likelihood into PSS roadmapping. Table II relates

Cumulative failure distributions of RFID tags exposed to ever- increasing high temperatures in vacuum for a given amount of soak time t s.. Sample size is 30 RFID tags per

Far away from the ground, pressure forces (u g , v g ) are balanced by the local velocity, this is the geostrophic equilibrium, known as the Taylor-Proudman theorem [30, 31]. Close