Unconventional monetary policy and the investment policy of
U.S. public pension funds
Master’s Thesis to obtain the degree
MSc Finance, specialization in Quantitative Finance
Faculty of Economics and Business Amsterdam Business School University of Amsterdam
Thesis supervisor: Dr. Aleksandar Andonov (University of Amsterdam) July 2021
Statement of Originality
This document is written by Student Egert-Gerret Kreek who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
UvA Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
This paper studies the relationship between unconventional monetary policy and changes in the investment policy of U.S. public pension funds. The author has extended a unique dataset of U.S. pension fund return expectations, using the data first collected by Andonov and Rauh (2020). As opposed to the existing empirical papers on pension fund investment, unconventional monetary policy is measured by changes in the Federal Reserve’s assets instead of changes in interest rates, to quantify the effects of central bank asset purchases more directly. The findings indicate that increases in the Federal Reserve’s assets (which measure the central bank’s asset purchases, i.e., quantitative easing) are associated with a decrease in the pension plans’ total expected rate of return through a decrease in the expected risk-free rate. Additionally, an increase in the central bank’s assets is associated with an increase in the pension funds’ target asset allocations to riskier and “alternative” assets, indicating that quantitative easing is associated with an increase in pension fund risk-taking.
Keywords: quantitative easing, institutional investors, return expectations, asset allocation, U.S.
public pension funds, unconventional monetary policy.
4 Table of Contents
1. Introduction 5
2. Literature review 10
2.1. Research on the return expectations of institutional investors 10 2.2. Research on the target asset allocations of institutional investors 12 2.3. Research on the effects of quantitative easing programs 15
3. Hypotheses and methodology 19
4. Data and descriptive statistics 24
4.1. Data collected and merged 24
4.2. Summary statistics 27
5. Results 30
5.1. The effect of expansive monetary policy on return expectations 30 5.2 The effect of expansive monetary policy on components of the total return expectation 34 5.3. The effect of expansive monetary policy on target allocations 38
6. Robustness checks 42
7. Conclusion 46
5 1. Introduction
It has been noted that since the quantitative easing (QE) programs were first introduced by the world’s largest central banks as an “unconventional” alternative to conventional monetary policy, as the policy rates had already been lowered to their effective lower bounds following the financial crisis in 2008, central bank balance sheets have become the main policy instrument as central banks attempt to inject money and liquidity into the economy through purchasing securities on the open market (Hausken and Ncube, 2013; Gambacorta et al., 2014). Quantitative easing programs have been found to depress the yields on government bonds (Joyce, Lasaosa et al., 2011;
Koijen et al., 2021), have a positive effect on the stock market (Lima et al., 2016), and result in a rise in economic activity and the price level (Joyce, Tong et al., 2011; Hausken and Ncube, 2013;
Gambacorta et al., 2014). Quantitative easing has recently become more relevant as a sizable QE program was announced in March 2020 to ensure the stability of the financial system in the wake of the coronavirus pandemic (Board of Governors of the Federal Reserve System, 2020).
Several recent papers have examined the effect of unconventional monetary policy on the risk-taking of U.S. public pension funds, finding that the persistently low interest rates have resulted in increased pension funds’ asset allocations to equity assets (Boubaker et al., 2018) and to alternative assets, mainly to private equity and real estate (Ivashina and Lerner, 2018).
Additionally, as U.S. public pension funds can discount their liabilities by their long-term investment return assumption, the pension funds are incentivized to invest more in risky assets to improve their funding status, as explained by Andonov et al. (2017). Based on empirical data, also presented in Figure 1, the trend of increasing pension fund target allocations to risky and alternative assets seems to coincide with the increases in the Federal Reserve’s (Fed’s) balance sheet, which can be considered a measure of quantitative easing.
However, perhaps surprisingly, the body of literature has yet to examine the relationship between the changes in the Federal Reserve’s balance sheet and the investment policy of U.S.
public pension funds, although quantitative easing has become an important policy tool for the Federal Reserve. The total actuarial assets of the 164 major U.S. public pension plans included in the Public Plans Database (by the Center for Retirement Research at Boston College) amounted to 3.69 trillion U.S. dollars at the end of financial year 2019. Pension assets are therefore significant in size and changes in pension funds’ investment policy (return expectations and target allocations)
could have a significant effect on financial markets. For example, Greenwood and Vissing- Jorgensen (2018) have found that the demand for long-dated assets by pension and insurance companies drives the long end of the yield curve. However, the effect of QE on the investment policy of U.S. public pension funds has not been empirically quantified.
This thesis attempts to answer two research questions. The first goal is to examine the effect that the quantitative easing programs have had on the return expectations of U.S. public pension funds. The second goal is to examine, to which extent has the shift in asset allocations towards riskier assets been influenced by the quantitative easing programs. The discussion contributes to the literature on the determinants of U.S. pension fund return expectations by also considering the effect that changes in monetary policy could have on return expectations through changes in the risk-free rate, risk premium, inflation rate, and real return expectations. It also contributes to the research on the pension fund risk incentives in relation to unconventional monetary policy by considering the changes in Federal Reserve’s balance sheet as a monetary variable, whereas Boubaker et al. (2018) used changes in Treasury yields to account for monetary policy shocks.
Accordingly, two hypotheses are set. The first hypothesis is that the quantitative easing programs have reduced the total long-term return expectations of U.S. public pension funds through a decrease in the risk-free rate expectation. As QE programs have been found drive down the long-term government bond yields (Joyce, Lasaosa et al., 2011; Koijen et al., 2021), which are often used as the risk-free rate expectation, then the total return expectations of the pension funds should also decrease, unless the QE programs enhance the pension funds’ risk premium expectation significantly to offset the decrease in the expected risk-free rate. The second hypothesis is that the quantitative easing programs have a statistically significant positive effect on the target allocations towards riskier assets. The assumption is, that increases in the Federal Reserve’s assets, through asset purchase programs, have a similar effect on pension fund risk- taking as lower interest rates, which have been found to incentivize pension fund risk taking (Ivashina and Lerner, 2018; Lu et al., 2019).
To test these hypotheses, a novel and unique dataset of U.S. public pension funds’ return expectations and target asset allocations is used. The data was initially collected from the annual reports of U.S. public pension funds by Andonov and Rauh (2020) and has been extended by the author to include more recent data, up to the financial year 2020. The full dataset includes data on
222 pension plans (“pension plan” and “pension fund” are used interchangeably in this thesis) for the period 2014-2020. Additional data on pension funds is extracted from the Public Plans Database by the Center for Retirement Research at Boston College, including annual data on 200 U.S. public pension plans, with historic target allocations available for the period 2001-2020. The monetary and macroeconomic variables are extracted from the Federal Reserve Economic Data database.
The relationship between the Federal Reserve’s assets and pension fund return expectations is determined by OLS regressions with the expected return as the dependent variable and the monetary policy variable as the regressor, controlling for pension plans’ past return, size, funding ratio, discount rate assumption, the federal funds rate, and the target allocations to risky asset classes. Regressions are estimated separately for pension funds reporting in arithmetic and geometric terms. As a robustness check, the regressions are also estimated without the target asset allocations as explanatory variables, as QE is expected to have an indirect effect on return expectations through changes in target allocations. Additional regressions are estimated to determine the relationship between QE and changes in the components of the total expected return:
the inflation rate expectation, the real return expectation, the risk-free rate expectation, and the risk premium expectation.
The relationship between the pension funds’ target allocation to risky assets and the Federal Reserve’s assets is also determined by OLS regressions, with the target allocation to risky assets as the dependent variable and the monetary policy variable as the regressor, with pension funds’
size and past investment return, and the effective federal funds rate included as controls.
Additionally, the regression is estimated with target allocation to alternative assets (excluding allocations to public equities) as the dependent variable. In some regressions, the effective federal funds rate is included as a regressor to estimate the relationship between target allocations and a measure of conventional monetary policy.
The results indicate that a 1 percent increase in Fed’s assets results in a 0.64 basis point decrease and a 0.80 basis point decrease in pension plans’ total expected rate of return for plans reporting in arithmetic and geometric terms, respectively. The magnitude of the negative relationship is 11 to 17 percent larger when accounting for the indirect effect that the QE potentially has on the return expectation through changes in the target allocations to riskier assets.
When accounting for this indirect effect, a quarterly one-standard-deviation increase in the natural logarithm of Federal Reserve’s total assets is associated with a 0.50 percentage point and a 0.67 percentage point decrease in pension plan’s total expected rate of return for plans reporting in arithmetic and geometric terms, respectively.
This effect of a one-standard-deviation increase in the QE variable is rather large, considering that the mean expected rate of return of the pension plans in the sample is 8.05 percent and 7.39 percent in arithmetic and geometric terms, respectively, over the sample period, as can be seen in Table 1. The results also indicate a negative relationship between the level of Fed’s assets and the risk-free rate expectation, and a positive relationship between the level of Fed’s assets and the risk premium expectation. However, the QE’s negative effect on the risk-free rate expectation is larger in magnitude compared to the positive effect on risk premium, resulting in a negative overall relationship between quantitative easing and the total expected rate of return of U.S. public pension funds. Based on these results, the first hypothesis can be accepted.
Focusing on changes in target allocations, the results suggest that a 1 percent increase in Fed’s assets is associated with a 2.7 basis point increase in the target allocation to risky assets (omitting allocations to cash and fixed income) and a 10.7 basis point increase in the target allocation to alternative assets (omitting allocations to cash, fixed income, and public equity). The second hypothesis can also be accepted as the positive relationship is empirically confirmed. As could have been expected, the relationship between the federal funds rate and the pension funds’
target allocations to risky assets is negative, suggesting that an interest rate hike by the Fed is associated with reduced risk-taking by the pension funds.
In the context of the existing literature, the negative relationship between the level of Fed’s assets and the risk-free rate is in line with previous findings by Joyce, Tong et al. (2011) and Koijen et al. (2021) which estimated that quantitative easing results in lower government bond yields, although their analysis focused on the UK and the euro area, respectively. The finding that quantitative easing is associated with an increase in the risk premium expectation, is in line with Lu et al. (2019), which estimated that U.S. pension funds increase their risk premium when Treasury-bill rates fall, which could be considered a consequence of QE.
However, the finding that quantitative easing is associated with an overall decrease in the total expected rate of return of U.S. pension plans, is novel and substantial, while confirming
empirically a relationship that could have been theoretically predicted as existing literature indicated that QE could affect pension funds’ return expectations through a reduction in their risk- free rate expectation. The basis for this thesis is data and methodology first applied by Andonov and Rauh (2020), while the results are not comparable as their paper did not consider the effects of monetary policy and instead focused on the cross-sectional differences between the funds to measure the pension funds’ reliance on past performance for setting return expectations.
Similarly to existing literature on the relationship between unconventional monetary policy and pension fund risk-taking, the results indicate that an increase in the level of Fed’s assets is associated with an increase in the pension funds’ target allocation to risky and alternative assets.
This in line with the findings of Boubaker et al. (2018), Ivashina and Lerner (2018), and Lu et al.
(2019), although the use of changes in the Fed’s balance sheet to measure quantitative easing is novel, whereas the aforementioned papers found that an increase in pension fund risk-taking was associated with lower interest rates, which is not a good measure of unconventional monetary policy as interest rates do not measure Fed’s asset purchases directly, although asset purchases are expected to impact interest rates.
The findings of this thesis imply that changes in the Federal Reserve’s balance sheet are associated with changes in the investment policy of U.S. public pension funds, which are substantial participants in the financial markets, based on the amount of assets under management.
Considering that quantitative easing has become the main policy tool used by the largest central banks in the world, as new rounds of QE were recently introduced to ensure stability in the financial markets during the coronavirus pandemic, these findings could prove useful for the central bankers who are responsible for changes in monetary policy, as quantitative easing potentially stimulates the economy through several channels, also through changes in pension investments.
The remainder of this thesis proceeds as follows. Section 2 discusses the relevant literature on the return expectations and target asset allocations of institutional investors as well as on the effects of QE. Section 3 describes the hypotheses and methodology. Section 4 describes the dataset used. Section 5 analyzes the results, with additional robustness checks presented in Section 6.
Section 7 concludes.
10 2. Literature review
2.1. Research on the return expectations of institutional investors
The two main assumptions set by institutional investors, on which this thesis also focuses, are the expected rates of returns of asset classes and the target asset allocations. These two assumptions can be used together to form the expected portfolio return. There is a considerable body of literature discussing the discount rate assumptions of U.S. public pension funds which are closely related to portfolio expected returns as according to GASB Statement No. 25 (Governmental Accounting Standards Board, 1994), the discount rate “… should be based on an estimated long-term investment yield for the plan, with consideration given to the nature and mix of current and expected plan investments …”. That is, U.S. public pension funds are obligated to report an investment return assumption, based on the expected portfolio return, which is then used to discount the liabilities of the defined benefit pension plans. The papers discussing this peculiarity point out, that using the investment return assumption as a discount rate goes against the logic of financial economics, as the discount rate used to discount the liabilities does not reflect the riskiness of the liabilities, if the discount rate is based on the expected investment returns, as discussed by Brown and Wilcox (2009), Novy-Marx and Rauh (2011), and Rauh (2017).
Brown and Wilcox (2009) point out that U.S. public pension plans can reduce their funding obligations by investing in riskier securities, while Mohan and Zhang (2014) confirm empirically that this relationship exists, finding that public pension plans in the U.S. assume more risk if they are underfunded as this justifies using a higher discount rate, which reduces the reported value of pension liabilities. Therefore, there is an incentive for “risk-shifting”, that is, transferring underfunded current pension obligations to future taxpayers, as shown by Mohan and Zhang (2014). Brown and Wilcox (2009) suggest that the strong constitutional protections make the U.S.
public defined benefit pension fund obligations virtually risk free, and an application of an appropriate discount rate would reveal that the state and local pensions are more underfunded than the reporting suggests.
Novy-Marx and Rauh (2011) use Treasury rates as an approximation for an appropriate default-free rate, discounting pension liabilities using the zero-coupon Treasury yield curve. Rauh (2017) later applied the same methodology and found that as of the financial year 2015, U.S. public pension plans had unfunded accumulated benefits of 3.846 trillion dollars under Treasury yield
discounting. This is 2.8 times more than is reported in the government disclosures as the liability- weighted average discount rate chosen by the pension plans in 2015 was significantly higher than the Treasury yield – 7.6 percent – with this reported figure implying that investments are expected to double approximately every 9.5 years, according to Rauh (2017).
Most of the literature on setting return expectations focuses on the accuracy of the expectations and the relationship between return expectations and past performance. Greenwood and Shleifer (2014) analyze investor expectations of future stock market returns and find that the return expectations of investors are positively correlated with past stock market returns as well as the level of the stock market, while higher expectations of returns result in low future market returns on average. In other words, investors expect high stock market returns after experiencing higher stock returns, once the stock market is already at a high level. Greenwood and Shleifer (2014) find that at such a high level, expected return measures such as the dividend price ratio correctly predict that future returns will be low, while investors wrongly predict high future returns. There is however evidence, that market participants can be good forecasters in the case of inflation expectations, as Ang et al. (2007) have found that surveys among consumers beat time- series, Phillips curve, and term structure forecasts.
According to Malmendier and Nagel (2011), individuals do not incorporate all available historical data when forming beliefs about risky outcomes as standard models would suggest and instead rely more on past personal experiences. Empirical results imply that the next 12 month expected stock market return of the surveyed individuals can be 0.5-0.6 percentage points higher for each 1 percentage point higher weighted-average market return experienced (Malmendier and Nagel, 2011). Similarly, the one-year inflation expectation is found to be 0.67 percentage points higher for each 1 percentage point higher inflation experienced (based on a learning algorithm for updating the experience), according to Malmendier and Nagel (2016).
Several empirical papers have found that more recent return experiences have stronger effects, especially for younger individuals (Greenwood and Nagel, 2009; Malmendier and Nagel, 2011; Malmendier and Nagel, 2016). An example of this are younger mutual fund managers who, at the peak of the technology bubble of the late 1990s, exhibited trend-chasing behavior, increasing technology holdings following quarters in which technology stocks generated high returns, as noted by Greenwood and Nagel (2009). According to empirical papers, younger individuals tend
to learn and extrapolate from the little data they have observed over their lives, updating their expectations more strongly since recent experiences account for a greater share of their lifetime history, while not properly adjusting for the small sample of data at hand (Greenwood and Nagel, 2009; Malmendier and Nagel, 2016).
Applying these ideas on a dataset of U.S. public pension plans’ portfolio expected returns, Andonov and Rauh (2020) find that institutional investors rely on past performance in setting return expectations so that each additional percentage point of past return in excess of other pension funds in the same year raises the portfolio expected return by 30 basis points. More specifically, past performance has a positive effect on portfolio expected return through the real return assumption (Andonov and Rauh, 2020). The relative level of unfunded liabilities is also positively related to portfolio expected return in the regression models of Andonov and Rauh (2020).
In conclusion, there is substantial empirical evidence on the factors which U.S. public pension funds may consider when setting expected rates of return for asset classes, such as the level of unfunded liabilities, the past performance of the fund, and the assumed discount rate.
However, the current body of literature has not examined the relationship between pension fund return expectations and quantitative easing or monetary policy. This thesis attempts to fill this gap in the research on pension fund return expectation determinants by testing whether changes in monetary policy are associated with changes in pension fund return expectations. More broadly, there is literature which has examined the relationship between quantitative easing and target asset allocations, which is the other major component of pension funds’ investment policy, and which is also considered over the course of this thesis.
2.2. Research on the target asset allocations of institutional investors
There is a consensus among empirical papers that the U.S. public pension funds which are more underfunded, have changed their target asset allocations towards riskier assets in order to justify higher discount rates which are used to discount pension liabilities in accordance with GASB guidelines (Mohan and Zhang, 2014; Rauh, 2017; Andonov et al., 2017; Lu et al., 2019;
Andonov and Rauh, 2020). However, an earlier empirical analysis by Rauh (2009) found that after a funding status of a pension plan worsens, the assets tend to be invested more in safe assets. The more recent paper by Mohan and Zhang (2014) calculates the pension funding ratio based on the
market value of pension liabilities, using the Treasury rate as the discount rate (based on Novy- Marx and Rauh (2011) methodology), as opposed to Rauh (2009) which used the reported value of liabilities. By using this improved variable for funding status, Mohan and Zhang (2014) find that pension funds tend to take higher investment risks when funding ratios are lower. The results by Mohan and Zhang (2014) imply that public pension plans practice risk-transfer, i.e., shifting the pension obligations to future taxpayers, instead of risk management, as suggested by Rauh (2009).
Ivashina and Lerner (2018), looking into the pension investment trends of private and public pension funds in 23 developed economies and 16 emerging markets between 2008 and 2017, found that the pension funds in developed markets on average increased the share of alternatives (private equity, private debt, real estate, hedge funds, infrastructure, and natural resources) in their portfolios from 7.2% of assets under management to 11.8%. Rauh (2017) explains that the U.S. pension funds have been incentivized to invest into alternative riskier assets as these provide higher expected returns which can be used for discounting pension liabilities as public sector accounting assumes that the risky returns will be achieved with certainty. The shift towards alternative investments has been similar for both public and private pension funds and for funds of different sizes, while investments are largely concentrated in real estate and private equity, according to Ivashina and Lerner (2018).
U.S. public pension plans with higher past performance expect higher risk premium when investing in risky assets, while each percentage point of higher past return increases the target allocation to risky assets by two percentage points, as has been found by Andonov and Rauh (2020). Conversely, Mohan and Zhang (2014) found that U.S. public pension funds undertake more risk if the investment returns in previous years have been lower, which could be due to the fact that Mohan and Zhang (2014) used 1-year, 3-year, and 5-year past returns, while Andonov and Rauh (2020) used 10-year past returns.
In addition to the pension funding status and past returns, Andonov et al. (2017) have found that another factor that increases the risk-taking of U.S. public pension funds is the number of politicians and elected plan participants serving on the pension fund’s board. According to Mohan and Zhang (2014), pension funds in states facing financial constraints, and pension funds with higher pension discount rate assumptions also allocate more assets to equity and have higher
portfolio beta. Moreover, Mohan and Zhang (2014) find that there is also a tendency among U.S.
public pension funds to mimic the influential CalPERS pension fund in regards of equity allocation or portfolio beta.
Several empirical papers, such as Ivashina and Lerner (2018), Boubaker et al. (2018), and Lu et al. (2019), have also found that monetary policy has had an effect on the risk-taking of public pension funds. Lu et al. (2019), focusing on the period from 2001 to 2016, suggest that U.S. public pension funds reach for yield by taking more investment risk in a low interest rate environment, as lower risk-free interest rates alter the risk premium and may also result in lower funding ratios, while both of these factors may incentivize the pension plans to take more risk. Ivashina and Lerner (2018) also found that the shift in public and private pension fund allocations towards riskier alternative assets is more pronounced for nations with lower long-term interest rate environment, determining that a 50 basis points decrease in the five-year average natural rate (real short-term interest rate) is associated with a 0.25 percentage point increase in the average annual change in the allocation to alternatives. Boubaker et al. (2018) finds that a 5 percent decline in the 10-year Treasury yield over the period 1999-2014 is associated with a 17 percent increase in the allocation to equity assets and a 18 percent decrease in the allocation to bond securities.
Furthermore, Lu et al. (2019) find that lower funding ratios of U.S. public pension funds had a more pronounced effect on the risk-taking behavior when interest rates were relatively low.
This could mean that the underfunded pension funds, which are incentivized to report a lower actuarial value of liabilities, will take on more risk to justify a higher discount rate in a low-rate environment. The trends of interest rates and pension fund risk-taking over the period from Q4 2004 to Q4 2020 are illustrated by Figure 2, which shows that the mean pension discount rate has more or less remained at the same level, although the 10-year Treasury bond yield has significantly decreased. As the yields on safer assets have decreased, the pension funds have increased their mean allocation to risky assets (excluding cash and fixed income), perhaps in order to avoid changing the discount rate assumption. Decreasing the discount rate assumption would result in an increase in the actuarial value of the pension liabilities, which will increase the amount of unfunded benefits and worsen the funding status of pension plans. Based on Figure 2, it also seems that expected investment return (discount rate) is set to a considerably high level in comparison to the actual past investment returns.
Boubaker et al. (2018) examines the relationship between unconventional monetary policy and the risk incentives of U.S. public pension funds, focusing on changes in Treasury yields, finding that the shift from bonds to equities has been greater since the unconventional monetary policy measures were launched by the U.S. Federal Reserve. Based on a counterfactual scenario analysis for the period 1999-2014, if the 10-year Treasury yield had been higher by 100 basis points, the investment return on bond securities would have been 7.19% instead of 6.56%, suggesting that the low interest rates and the unconventional monetary policy affected the risk- taking behavior of U.S. public pension funds (Boubaker et al., 2018).
To summarize, the literature on public pension fund target asset allocations has empirically identified several factors which are associated with risk-taking, such as the funding ratio of the pension fund, the past return of the pension fund, and changes in monetary policy. However, the discussion on the effects of unconventional monetary policy and low interest rates on pension fund risk-taking has focused mostly on the changes in risk-free interest rates (e.g., how have target asset allocations changed as a result of decreasing Treasury yields). This thesis contributes to the existing literature by focusing on the Federal Reserve’s open market purchases of Treasury securities (i.e., quantitative easing programs) and its effects on the investment policy of institutional investors by including the level of Federal Reserve’s total assets as an explanatory variable.
2.3. Research on the effects of quantitative easing programs
In the U.S., the federal funds rate has historically been the main policy tool used by the central bank to conduct monetary policy, while over a three-year period from 1979 to 1982 the Fed briefly switched its target from the federal funds rate to a measure of money growth, according to Laopodis (2013). The federal funds rate is a target rate or range, set by the Federal Open Market Committee for trading in the federal funds market (Federal Reserve Bank of New York, n.d.). In the wake of the financial crisis that began in 2007, some of the world’s largest central banks, the U.S. Federal Reserve (Fed), the Bank of England (BoE), the Bank of Japan (BoJ), and the European Central Bank (ECB) turned to unconventional monetary policies since the interest rates had declined to near zero and there was very limited possibility to combat the global crisis by further monetary easing through lower policy rates (Joyce et al., 2012; Hausken and Ncube, 2013).
For example, the BoE cut the bank rate (UK’s policy rate) in a sequence of steps from 5 percent in October 2008 to 0.5 percent in March 2009 (Joyce, Lasaosa et al., 2011).
As the interest rates were already set to the effective lower bound, the central bank balance sheets became the main policy instrument (Gambacorta et al., 2014). By using unconventional policies, the central banks could ease monetary conditions further through quantitative easing (QE) programs of asset purchases financed by the issuance of central bank reserves (Joyce, Lasaosa et al., 2011). The phrase “quantitative easing” was first applied to Japan which started implementing QE in 2001 (Joyce et al., 2012; Hausken and Ncube, 2013). Such “open market operations” are not entirely new as under conventional monetary policy, central banks also use buying or selling securities from the banking system to influence the level of reserves held by banks, to achieve the desired changes in interest rates (Joyce et al., 2012). The goal of quantitative easing is to prevent financial instability, while driving up asset prices and removing deflationary forces (Joyce et al., 2012; Hausken and Ncube, 2013). During QE, the central bank makes money available for banks to borrow by purchasing government or other bonds on the open market, thereby expanding the amount of money circulating in the economy, which in turn results in lower long-term interest rates, according to Hausken and Ncube (2013). The quantitative easing policies may also include elements of “credit easing”, that is, changing the composition of the central bank’s balance sheet by shifting between short and longer-maturity government bonds or by shifting into riskier private assets (Joyce, Lasaosa et al., 2011).
The Federal Reserve purchased mortgage-backed securities and longer-term Treasury securities in three major rounds of quantitative easing: QE1 in 2008, QE2 in 2010, and QE3 in 2012 (Hausken and Ncube, 2013). The latest QE program was announced in March 2020 as a response to the coronavirus pandemic (Board of Governors of the Federal Reserve System, 2020).
Additionally, the Fed implemented “Operation Twist” in 2011, which focused on lowering long- term interest rates without expanding the monetary base by simultaneously selling short-term government bonds and buying long-term bonds (Joyce et al., 2012; Hausken and Ncube, 2013).
There is consensus in the literature on the main channels through which quantitative easing might potentially have a stimulating effect on the economy. Firstly, QE works through the
“signaling channel” as market participants will expect policy rates to remain low as central banks commit to keep the rates low as they would incur losses on the assets purchased if the rates would
rise (Joyce, Tong et al., 2011; Hausken and Ncube, 2013). Secondly, QE works through the
“portfolio rebalancing channel” as the sellers of bonds will attempt to rebalance their portfolios by buying other similar assets as substitutes, which reduces the spread of longer-term interest rates over the expected policy rates and the risk premium in general (Joyce, Tong et al., 2011; Hausken and Ncube, 2013). Finally, QE also works through “liquidity premia channel” by increasing liquidity, although there is disagreement on whether the reduced premium for illiquidity lowers yields (Joyce, Tong et al., 2011) or raises yields (Krishnamurthy and Vissing-Jorgensen, 2011).
Hausken and Ncube (2013) conduct a counterfactual analysis using Bayesian vector autoregression models, assuming a no-QE scenario with higher Treasury yields, and conclude that in the USA, the QE programs implemented starting from 2008 prevented the CPI inflation from being more negative, prevented the unemployment rate from being higher, raised expectations towards future inflation, kept the S&P500 index at a higher level, but only stimulated industrial output, while failing to stimulate overall real economic growth. Similarly, Joyce, Tong and Woods (2011) find that in the UK, the BoE’s QE programs may have raised the level of GDP by 1.5 to 2 percentage points and the CPI inflation by 0.75 to 1.5 percentage points, mostly through the portfolio rebalancing channel, while Hausken and Ncube (2013) estimate that in the U.S. the signaling channel was dominant for the Fed’s QE programs. Furthermore, the increase in central bank balance sheets led to a temporary rise in economic activity and consumer prices in Canada, the euro area, Japan, Norway, Sweden, Switzerland, the UK, and the USA without major differences across countries, according to Gambacorta et al. (2014).
Quantitative easing, through the “portfolio rebalancing” channel, is expected to reduce the required return on risky assets relative to risk-free assets (Joyce, Tong et al., 2011), which should result in an increase in prices of risky assets. Lima et al. (2016) confirmed empirically that the QE programs by the central banks in the USA, Japan, and the UK have had a positive impact on the local stock markets, using money supply as an explanatory variable describing stock index returns.
Focusing on the period prior to QE programs, Laopodis (2013) found that there is no consistent relationship between the U.S. stock market and the federal funds rate set by the Fed.
Through the “portfolio rebalancing” channel, QE programs also increase the demand for other bonds, which are bought as a substitute for government bonds, resulting in reduced long- term yields, with asset purchases by the BoE depressing medium to long-term government bond
yields by about 100 basis points in the UK, according to Joyce, Lasaosa et al. (2011). Looking at the major Fed’s balance sheet increases, Krishnamurthy and Vissing-Jorgensen (2011) find that QE1 and QE2 programs significantly lowered nominal interest rates on Treasuries, Agencies, corporate bonds, and mortgage-backed securities. Similarly, Koijen et al. (2021) found that the euro are QE program from 2015 to 2017 decreased government bond yields by 65 basis points on average.
In summary, the empirical results indicate that the QE programs, through lower long-term interest rates, have raised the level of GDP as well as inflation, as opposed to a scenario without QE. Also, the central bank asset purchases have been found to increase the prices of bonds, stocks, and potentially of other risky assets. It is therefore reasonable to assume that as QE affects asset prices, it could also affect the investment policies of institutional investors, which is the focus of this thesis. Furthermore, pension and insurance company assets have been found to have a strong effect on the long end of the yield curve by Greenwood and Vissing-Jorgensen (2018), which indicates that pension fund investments could also be highly dependent on monetary policy which is implemented by targeting long-term yields. This thesis contributes to the existing literature by examining the relationship between QE and the changes in investment assumptions (e.g., expected rates of returns and target asset allocations) set by institutional investors.
19 3. Hypotheses and methodology
Based on the existing literature on the investment policy of pension funds and on the effects of quantitative easing, two major hypotheses are formed and tested in this thesis. The first hypothesis is that the quantitative easing programs have reduced the total long-term expected return of U.S. public pension funds through a decrease in the risk-free rate expectation. As QE programs increase demand for bonds, through central bank asset purchases and the “portfolio rebalancing” channel, the bond prices are driven up, resulting in depressed medium to long-term government bond yields, i.e., a lower risk-free rate (Joyce, Lasaosa et al., 2011; Krishnamurthy and Vissing-Jorgensen, 2011; Koijen et al., 2021). On the other hand, lower risk-free rates alter the risk premium, incentivizing pension plans to take more investment risk in order to reach for higher yields (Lu et al., 2019), which would indicate that QE could increase the total expected return of U.S. public pension funds. However, the QE’s negative effect on pension fund return expectations through the decrease in the risk-free rate is expected to be larger in magnitude compared to the QE’s effect through an increase in the risk premium, as one of the main goals of QE is achieving lower policy rates, i.e., risk-free rates (Joyce et al., 2012; Hausken and Ncube, 2013).
The second hypothesis is that the quantitative easing programs have a statistically significant positive effect on the target allocations towards riskier assets. This hypothesis is in line with previous findings, especially with Boubaker et al. (2018), which found that unconventional monetary policy in the USA affected the risk-taking behavior of U.S. public pension funds as lower 10-year Treasury yields contributed to the pension funds’ shift from bonds to equities. Lower interest rates have also been found to incentivize pension fund risk-taking by Ivashina and Lerner (2018) and Lu et al. (2019). This hypothesis assumes that increases in the central bank’s assets, through asset purchase programs (QE), have a similar effect on pension fund risk-taking as a decrease in Treasury yields, which is also one of the measures closely related to quantitative easing and expansionary monetary policy.
To test these hypotheses, a unique set of data, collected by Andonov and Rauh (2020) for their paper examining the relationship between past returns and U.S. public pension fund investment policy (return expectations and target asset allocations), is extended with more recent observations, and used as the source on pension fund expectations. Additional descriptive data on
U.S. public pension funds, used to control for differences among different pension funds, is available through the Public Plans Data (PPD) dataset by the Center for Retirement Research at Boston College. The monetary variables which measure unconventional monetary policy are available through the Federal Reserve Economic Data (FRED) database.
As this thesis sets the U.S. public pension funds’ total return expectations and target asset allocations as dependent variables, similarly to Andonov and Rauh (2020), similar econometric regression models are applied to test the hypotheses. All regression models that are estimated are ordinary least squares (OLS) models. The first model, used to examine the relationship between pension funds’ total expected return and unconventional monetary policy, is illustrated by equation (1).
𝐸𝑅𝑖𝑡 = 𝛼𝑖 + 𝛽𝐺𝑒𝑜𝑚𝑒𝑡𝑟𝑖𝑐𝑖𝑡+ 𝜃′𝜔𝑖𝑡+ 𝛾𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝛿𝑄𝐸𝑡+ 𝜀𝑖𝑡 (1) In equation (1), Geometric is a dummy variable which indicates whether a pension plan reports in geometric terms, as opposed to reporting in arithmetic terms. To improve the regression model, the Geometric variable is later dropped as pension plans reporting in geometric and arithmetic terms are regressed separately. Similarly to Andonov and Rauh (2020), ω is a 5-vector of allocations to public equity, real assets, private equity, hedge funds, and other risky assets, while fixed income and cash are the omitted asset categories. These variables control for the fact that reporting in geometric terms mathematically results in lower return expectations (a negative regression coefficient) and the fact that the total return expectation is calculated by multiplying nominal expected rates of return of each asset class with the target allocation for each asset class and summing the products.
The controls include PF size, which is the natural logarithm of pension fund assets under management, found to have a negative regression coefficient, and Past return, which is found to have a positive regression coefficient by Andonov and Rauh (2020), although in this case, the Past return is extracted from the PPD dataset in gross terms, while Andonov and Rauh (2020) used net investment income for calculating the past returns, which may result in differences in regression coefficients. Other control variables, used to improve the regression model, are PF assets/liabilities, which represents the funded ratio of the pension fund and is expected to have a negative relationship with expected returns as well-funded pension funds are not incentivized to
set high return expectations as opposed to significantly underfunded pension funds (e.g., Mohan and Zhang (2014)), and Pension discount rate, which is the long-term investment return assumption set by the pension fund and is expected to have a positive relationship with expected returns as higher expected returns justify a higher pension discount rate (e.g., Andonov et al.
The QE variables of interest, included in the regression models, are the natural log of the Federal Reserve’s total assets, ln(Fed’s assets), measuring the amount of assets purchased by the Fed, and the Effective federal funds rate, measuring the funds rate set by the Fed. The first variable is expected to be negatively related to the dependent variable and statistically significant, as per the first hypothesis, while there is no clear expectation for the second variable as the federal funds rate has not been the main policy tool of the central bank since interest rates have been near zero, according to Hausken and Ncube (2013).
When conducting robustness checks, ω is excluded from the regression model to examine, to which extent could quantitative easing have an effect on pension funds’ total expected return through the changes in target allocations. To further examine the relationship between unconventional monetary policy and pension funds’ total expected return, the expected return is decomposed in two ways, similarly to Andonov and Rauh (2020): into the inflation rate assumption Eπt and the real return assumption Ert, and into the risk-free rate assumption Erft and the risk premium assumption E(Rt – rft).
Equations (2) and (3) have the inflation rate assumption and the real return assumption as dependent variables, respectively. The same QE variables are included, whereas an increase in Fed’s assets is expected to increase the inflation rate assumption as QE has been found to raise the level of CPI inflation by Joyce, Tong et al. (2011), Hausken and Ncube (2013), and Gambacorta et al. (2014). There is no expectation on the relationship between QE and the real return assumption, based on the existing literature.
𝐸𝜋𝑖𝑡 = 𝛼𝑖+ 𝛽𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝛾𝑄𝐸𝑡+ 𝜀𝑖𝑡 (2) 𝐸𝑟𝑖𝑡 = 𝛼𝑖+ 𝛽𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝛾𝑄𝐸𝑡+ 𝜀𝑖𝑡 (3)
Equations (4) and (5) have the risk-free rate assumption and the risk premium assumption as dependent variables, respectively. The same QE variables are included and an increase in the Fed’s assets is expected to decrease the risk-free rate assumption and to increase the risk premium assumption as this is the basis of the first hypothesis, based on prior literature (e.g., Joyce, Lasaosa et al. (2011), Krishnamurthy and Vissing-Jorgensen (2011), Joyce et al. (2012), Hausken and Ncube (2013) and Koijen et al. (2021)).
𝐸𝑟𝑓𝑖𝑡 = 𝛼𝑖 + 𝛽𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝛾𝑄𝐸𝑡+ 𝜀𝑖𝑡 (4) 𝐸(𝑅𝑖𝑡− 𝑟𝑓𝑖𝑡) = 𝛼𝑖 + 𝛽𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝛾𝑄𝐸𝑡+ 𝜀𝑖𝑡 (5) The regression models based on equations (2), (3), (4), and (5) include different control variables, based on which variables improve the model and have an economic reasoning behind the relationship with the dependent variable. The control variables in these models include Pension discount rate, PF assets/liabilities, and PF size, similarly to models based on equation (1), but also additional controls such as 30y Treasury rate (expected to be associated with the expected risk premium, the expected risk-free rate, and the expected inflation rate) and Past inflation (expected to be associated with the expected inflation rate and the expected risk-free rate).
To test the second hypothesis, the model used to examine the relationship between pension funds’ target asset allocations and unconventional monetary policy, is illustrated by equation (6).
The dependent variable Target allocation is firstly set to the target allocation to risky asset classes (excluding cash and fixed income) and later set to the target allocation to alternative asset classes (excluding cash, fixed income and public equity), by aggregating the target allocations in the PPD dataset. Based on the second hypothesis, an increase in Fed’s total assets is expected to be positively related to the dependent variable and statistically significant, as Boubaker et al. (2018), using the same PPD dataset on U.S. public pension fund target allocations, found that unconventional monetary policy has contributed to the shift from bonds to equities.
𝑇𝑎𝑟𝑔𝑒𝑡 𝑎𝑙𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛𝑖𝑡 = 𝛼𝑖 + 𝛽𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝛾𝑄𝐸𝑡+ 𝜀𝑖𝑡 (6) The control variables included in the regression models based on equation (6) are Past investment return (Investmentreturn_5yr in PPD dataset) and ln(actuarial assets) to control for size (the natural logarithm of ActAssets_GASB in the PPD dataset). Based on Andonov and Rauh
(2020), which ran similar regressions, the past return control variable is expected to have a positive relationship with the target allocations to risky assets, while there is no clear relationship between pension plan size and the target allocations to risky assets. Regression models, based on equation (6), are also used for robustness checks to determine, whether changes in Fed’s assets are associated with changes in the pension funds’ target allocations to each of the risky asset classes, while these models include Past return and PF assets/liabilities as controls based on the findings of Mohan and Zhang (2014) and Andonov and Rauh (2020).
In all of the regression models, fixed effects are applied. Similarly to Andonov and Rauh (2020), reporting-month fixed effects are included in all models to control for the fact that pension plans have different fiscal year endings and report their expectations based on different information available. In the regressions in which the plans reporting in arithmetic and geometric terms are regressed separately, and in regressions in which the target allocation is the dependent variable, pension plan fixed effects are also applied to examine changes within a pension plan over time, eliminating other differences which significantly improves the explanatory power of the regression models. However, this differs from Andonov and Rauh (2020) which used year fixed effects focusing on the cross-sectional differences between the pension funds. As this thesis focuses on changes in pension funds’ investment policy over time in response to changes in the monetary policy, year fixed effects would absorb the entire variation in the monetary variables and therefore cannot be applied in the OLS models.
Additionally, in all regression models, standard errors are clustered by pension plan to account for the fact that observations from the same pension plan are not independent from each other (i.e., controlling for heteroskedasticity and autocorrelation). Also, the inclusion of several control variables reduces the standard errors as the controls take away a part of the residual variance in the OLS models. Finally, the explanatory variables included in the models are not correlated to each other to a high degree. By applying this methodology, the OLS estimates of the QE variable of interest are expected to be unbiased and consistent.
24 4. Data and descriptive statistics
4.1. Overview of the data collected
Two main datasets are used to examine the relationship between unconventional monetary policy and the investment policy of U.S. public pension funds. Both are based on publicly available Comprehensive Annual Financial Reports (CAFRs) and their accompanying statements outlying the funds’ financial position as well as the expectations for the future, which are reported at an annual frequency.
Firstly, the unique and novel dataset of pension fund return expectations collected by Andonov and Rauh (2020) is extended by the author to also include the more recent period of 2018 to 2020. As stated by Andonov and Rauh (2020), the U.S. Governmental Accounting Standards Board Statement 67 (GASB 67) requires U.S. public pension plans to report long-term expected rates of return for (and the target allocations to) each asset class, beginning in the 2014 fiscal year.
The pension plans’ assumptions regarding the target asset allocation, the expected long-term returns of asset classes, the inflation rate, and the long-term investment return, which is used to discount actuarial liabilities, are collected from the CAFRs or GASB 67 statements.
While the initial dataset used by Andonov and Rauh (2020) included data on 228 state and local U.S. government pension plans for the period 2014-2017, the dataset was extended by collecting data for 222 pension plans (see Table 15 for an overview of the sample) for the whole period of 2014-2020. More specifically, the dataset covers the period from Q2 2014 to Q3 2020.
It should be noted that in several cases, different pension plans (e.g., the Public Employee Retirement Fund Base Plan, the Firefighters’ Retirement Fund, the Judges’ Retirement Plan) in the same region (e.g., Idaho) are managed and reported as a whole. However, there are some differences in reporting between the pension plans in the sample, which must be addressed. For most of the pension plans, the financial year ends in the end of June and the annual reports for the 2020 financial year were published at the end of 2020. For other plans in the sample, the financial year ends in March, August, September, or December, which should affect the cross-sectional differences between the return expectations of the pension plans as these plans set their assumptions at different times. Also, in cases where the financial year ends in December, the latest report available during data collection was for the financial year 2019.
Another important factor that must be addressed in the analysis is the fact that the pension plans report their expected rates of return as either arithmetic or geometric and this differs across pension plans and, in some cases, within one plan over time. Approximately 65% of the observations are in arithmetic terms and the other 35% in geometric terms. The return expectations could also be reported as real or nominal returns. However, as pension plans also report an inflation assumption, all observations in the dataset can be converted into nominal expected rates of return.
As pension plans report their expectations at different levels of detail, the target allocations are aggregated to seven major asset classes – public equity, fixed income, cash, real assets, private equity, hedge funds, and other risky assets (e.g., commodities and managed futures) – and the return expectations for each major asset class are formed as weighted averages accordingly, similarly to Andonov and Rauh (2020).
Each pension plan’s total expected rate of return on all investments, labeled Portfolio ER in the rest of the analysis, is calculated as a weighted average return of the seven major asset classes, i.e., multiplying nominal expected rates of return with the target allocation for each asset class and summing the products.
Additionally, the Portfolio ER is decomposed in two ways, similarly to Andonov and Rauh (2020). Firstly, the return expectation can be decomposed into the expected inflation rate Eπt
(reported by the funds) and the expected real return Ert, so that ERt = Eπt + Ert. Secondly, the Portfolio ER can be decomposed into the expected risk-free rate Erft, which can be calculated as the weighted average return on cash and fixed income, similarly to calculating Portfolio ER, and the expected risk premium, so that ERt = Erft + E(Rt – rft).
The second main dataset used is the Public Plans Data dataset produced by the Center for Retirement Research at Boston College in partnership with the Center for State and Local Government Excellence and the National Association of State Retirement Administrators, which includes detailed annual data on 200 U.S. public pension plans (both state- and locally-run) over the fiscal years 2001-2020 (at the March 30, 2021 update).
The PPD dataset includes several variables of interests that will be used as controls in the panel regressions, such as the pension plans’ funding ratios, average return over the past ten years, etc. The dataset also includes the target allocations which go back even further than the dataset
with return expectations, although the allocations are aggregated, slightly differently, into nine major asset classes: public equity, fixed income, real estate, private equity, hedge funds, commodities, miscellaneous alternatives, cash, and other investments.
The variables describing the U.S. monetary policy, as well as some other macroeconomic control variables are extracted from the Federal Reserve Economic Data database at quarterly frequency using end-of-quarter values. The main variable describing “conventional” monetary policy is the Effective Federal Funds Rate, which is influenced by the Federal Open Market Committee (FOMC) which establishes the target rate, or range, for trading in the federal funds market (Federal Reserve Bank of New York, n.d.). The main variable describing “unconventional”
monetary policy is the Total Assets of the Federal Reserve, which mostly reflects the amount of U.S. Treasury bills and notes, as well as mortgage-backed securities purchased on the open market and held by the Fed (according to the Consolidated Statement of Condition of All Federal Reserve Banks by FRED). The data for past inflation is extracted from the database of U.S. Bureau of Labor Statistics (BLS).
The Fed’s total assets is included in the regressions as the natural logarithm of the dollar level of assets, i.e., the natural logarithm of the Fed’s total assets in millions of U.S. dollars. The variable is not normally distributed over the 2014-2020 period which includes “outliers” as the Fed’s total assets increased significantly over the last three quarters of the observation period (from Q1 2020 to Q3 2020). When looking at a longer period (from Q4 2003 to Q3 2020), there is another spike in Fed’s assets with the total assets significantly increasing over four quarters: from Q3 2008 to Q2 2009. The quarter-end figures of macroeconomic variables are matched to pension plan annual data based on the quarter closest to the pension fund’s fiscal year end.
The Portfolio ER dataset and the PPD data are easily merged as the pension plans overlap and both datasets are based on the same annual reports. For merging the quarterly FRED data while accounting for the different financial year ends, each reporting month (March, June, August, September, or December) is matched with the nearest quarter and then merged. It is appropriate to match the reporting months with the nearest quarters as can be illustrated by the following example. If the pension plan’s fiscal year ends in June, it reports its long-term return expectations together with its financial position at the end of June, but as the annual report is published several months after the fiscal year end, the plan managers already consider the macroeconomic data for
Q2 (i.e., end of June), and the Q2 data is already incorporated in the pension funds’ long-term return expectations.
4.2. Summary statistics
The merged dataset including Portfolio ER, PPD and FRED/BLS data is declared as panel data with observations grouped by fiscal years and pension plans. The summary statistics of the merged dataset can be seen in Table 1 which differentiates between pension plans setting assumptions in arithmetic and geometric terms. Panel A reports data on pension funds’
assumptions regarding target allocations to specific asset classes, expected rates of returns (ER) for the respective asset classes, expected inflation rate, and the long-term investment return assumption which is used by pension funds to discount actuarial liabilities. Real return is the difference between Portfolio ER and the inflation rate assumption, whereas Risk premium is the difference between Portfolio ER and weighted average expected return on cash and fixed income.
Other pension plan specific variables are presented in Panel B. The pension plans in the sample which report in geometric terms are, on average, slightly larger in size (i.e., based on net fiduciary position) and less underfunded.
The differences in average allocations and return expectations between arithmetic and geometric pension plans are illustrated by Figure 3. The sample period mean Portfolio ER is 8.06 percent and 7.39 percent for arithmetic and geometric plans respectively, as the mean expected returns of all asset classes are lower for plans reporting in geometric terms, which can be partly explained by the fact that due to the compounding effect, the geometric mean of a dataset is always lower than the arithmetic mean.
As can be seen from Figure 3, average target allocation (for the sample period of 2014- 2020) of the pension funds is largest for equity (ca 46%), fixed income (ca 25%), and real assets (ca 8-11%). The mean expected return (ER) is highest for private equity, followed by public equity and other risky assets (e.g., commodities and managed futures), while cash and fixed income have the lowest expected returns.
Each pension fund reports a long-term investment return assumption annually, which is then used as the Pension discount rate for discounting actuarial liabilities, and which should be equal to or similar to the Portfolio ER. This is because, as a consequence of the GASB regulations
for U.S. public pension funds, the funds are incentivized to set the Portfolio ER to a level which justifies the discount rate chosen for discounting the value of the liabilities (Andonov et al. 2017).
As can be seen from Figure 4, the Pension discount rate and Portfolio ER are positively correlated, and based on the means reported in Table 1, also similar.
Pension discount rate is also positively correlated with the target allocation towards riskier assets (not cash or fixed income), i.e., higher long-term investment return assumption is positively correlated with a target allocation towards riskier assets, based on Table 2. Based on other correlation coefficients in Table 2, the level of Federal Reserve’s total assets is negatively correlated with the 1-year, 10-year, and 30-year Treasury bond yields. This relationship is expected as the Federal Reserve increases its assets mostly by purchasing U.S. Treasury securities on the open market, increasing the demand for Treasury securities, resulting in an increase in the prices which reduces the Treasury bond yields.
Furthermore, the 10-year Treasury yield is highly positively correlated with the 1-year and 30-year Treasury yields, while the correlation coefficient between the 1-year and 30-year Treasury yields is lower. The Effective federal funds rate is highly and positively correlated with the 1-year and 10-year Treasury yields, while highly and negatively correlated with past inflation, and the changes in Federal Reserve’s total assets.
The dynamics between the 10-year Treasury yield, the federal funds rate and the Fed’s asset purchases can also be seen in Figure 1. Starting from Q4 2003, the Federal Reserve has significantly increased its total assets (by open market purchases of Treasury securities) and lowered the effective federal funds rate to near the zero lower bound twice: first as a response to the Great Recession in 2008-2009 and again in 2020 as a response to the recession caused by the COVID-19 pandemic. In both cases, the central bank has managed to reduce the yield to maturity on the 10-year Treasury bond, which can be considered as the risk-free rate, to incentivize borrowing, investment, and economic activity (the policy targets suggested by Hausken and Ncube (2013)).
In Table 3, summary statistics are presented separately for periods before and after the approximate start of the Fed’s unconventional monetary policy (QE). Panel A includes variables from the PPD dataset, which differentiates between nine major asset classes and gives an overview of the changes in target allocations over a longer period.
In the second period, the mean target allocation to risky assets (%Risky, which excludes allocations to cash and fixed income) has increased to 74.3%, up from 69.9% in the first period.
More interestingly, the mean target allocation to alternative assets (%Alternative, which excludes allocations to cash, fixed income, and equities) has increased by ca 14 percentage points, from 12.4% to 26.0%, mainly driven by increased target allocations to hedge funds (ca 5.1 pp increase), private equity (ca 3.5 pp increase), and commodities (ca 2.6 pp increase). Surprisingly, the mean target allocation to equities, which is a more traditional risky asset class, has decreased by ca 9.2 percentage points on average, compared to the first period.
The significant increase in the mean target allocation to alternative assets is also illustrated in Figure 5 which plots %Risky and %Alternative target allocations as well as the Fed’s assets in trillions of U.S. dollars over time. The most significant increase in the target allocation to alternative assets visually coincides with the sharp increase in the Fed’s balance sheet around the 2008-2009 period, following the Great Recession.
Panel B in Table 3 reports the summary statistics for several macroeconomic variables of interest. After the start of the Fed’s unconventional monetary policy, the yields have been lower for the 1-year, 10-year, and 30-year Treasury bonds, on average. Also, the mean Effective federal funds rate has dropped to near zero, while the mean Fed’s total assets have increased significantly.
Additionally, the mean Past inflation has decreased in the second period.
Panel C in Table 3 includes summary statistics on two pension-plan specific variables that will be included in regressions as controls. The first one, past 5-year geometric mean annual investment return, has decreased slightly in the second period. The second variable, a control for pension plan size based on total assets, has slightly increased in the second period.
In the next section, regressions are first estimated with Portfolio ER as the main dependent variable and later with %Risky and %Alternative as dependent variables. Most of the pension plan specific and macroeconomic variables are used as controls, while Effective federal funds rate and ln(Fed’s assets) will be used as measures of conventional and unconventional monetary policy, respectively.