DEMOGRAPHIC INTEREST Exploring the effect of population dynamics on interest rates

University of Groningen Faculty of Economics and Business

MSc International Economics and Business (IE&B)

Master Thesis:

DEMOGRAPHIC INTEREST

Exploring the effect of population dynamics on interest rates

Author: Gerko Dwarshuis

Student number: 2206293

E-mail address: K.G.Dwarshuis@student.rug.nl Supervisor: Dr. A.C. Steiner

Co-Assessor: Dr. M.J. Gerritse

2

A

To shed light on the cause of the current underperforming economy, this paper relates the development in age distributions of several industrialized economies to the development in their interest rates. Using a fixed effects panel regression, it was found that economic agents affect the real interest rate on 10-year maturing government bonds depending on their stage in life. Up to the ages of 30 to 39, agents are found to gradually increase the interest rate due to participation in the labor force, procurement of loans, and increased consumption. When these agents age even further, they cause a decrease in the long-term real interest rate due to two separate channels. First, aging economic agents start to increase the share of their income they safe for retirement and in doing so expand the supply of loanable funds. The second and more complex channel works through the current and future decreased spending of these economic agents, and the profitability of new investments linked to it. When this profitability of an investment decreases due to decreased (future) consumption, the demand for investment funds falls as those simply cannot be used profitable.

3

T

C

1. Introduction ... 4

2. Literature review ... 5

2.1 The Secular Stagnation Theory ... 5

2.2 Several causes of stagnation ... 7

2.3 Demographic forces on interest rates ... 9

2.3.1 The natural rate of interest ... 9

2.3.2 Objections to the decline in natural rates ... 10

2.3.3 Demographics and real interest rates ... 11

3. Hypotheses development ... 14

4. Methodology and data collection ... 17

4.1 Objections to current research methods ... 17

4.2 Analytical perspective ... 18

4.3 Statistical approach ... 19

4.4 Data collection and examination ... 22

5. Analytical analysis ... 25

5.1 Exploring the effect in more stable economies ... 28

5.2 Using the World Interest Rate ... 32

6. Discussion and conclusion ... 34

6.1 Hypothesis confirmation and research limitations ... 36

6.2 Literature corroboration and further research recommendation ... 37

6.3 Conclusion and policy recommendation ... 38

7. Bibliography: ... 39

4

1.

I

The American mortgage-backed security crisis that ravaged the world economy lies nearly a decade behind us. But despite what might be heard from overoptimistic politicians, the current economic development still pales in comparison to the growth achieved in the years leading up to the crash. Attempting to explain this underwhelming economic recovery, Lawrence Summers, former head of the World Bank and advisor under the Clinton administration, refers back to the Secular Stagnation Theory as proposed by Alvin Hansen in 1939. Summers (2014, 2015, 2016) argues that an excessive propensity to save has lowered total demand, which in turn reduced growth and pushed inflation below optimal levels. The increased propensity to save in combination with lower levels of investment also decreased real interest rates, constraining monetary policy due to the zero lower bound restriction on interest rate setting. The question which then remains is why this lower interest rate would not entice economic agents to cut back on their savings and either spent their money on consumer goods or start investing, as standard economic theory dictates. Answering that question however requires a deeper understanding of the interest rate and determinants thereof. To contribute to that discussion, this paper relates developments in age distributions over time to fluctuations in real interest rates. Important to note in this regard is that, as Bloom, Canning, and Graham (2003) point out in their research on life-cycle savings, demographic forces predominantly enter into economic models through the size and growth of a population. They prove however that, while growth and size are important, the often neglected age distribution within an economy should also be treated as an influential macroeconomic variable.

5 Figure 1: USA GDP; Source: Lawrence Summers 2014

2.

L

ITERATURE REVIEW

2.1

T

S

S

T

At the start of the last century, Alvin Hansen published a paper in which he expressed his concerns on America‟s declining population growth. According to his statistics, the period from 1920 to 1930 yielded a population growth of approximately 16.000.000 people; something he characterized as a growth that was in excess of any previous decade in American history (Hansen, 1939). This growth however was not sustained over time, as the 1930‟s only showed population growth figures of half the level of the previous decade, and forecasts for the upcoming decade showed a population growth that was only one third of what it was during the post-World War I period.

Hansen worked under the premise that, absent of governmental policy stimulating consumption, full employment depends on the level of investment. Investment in turn is not merely determined by the level of interest to be paid, but predominantly by the potential profit rate of the investment. And as the profit rate of investments is dependent on the development of the economy as a whole, Hansen concluded that factors that underlie economic growth are also dominant determinants of both investment and employment. In accordance with his research, Hansen poses that slightly under half of the three percent for Europe and slightly over half of the four percent of economic growth for the USA in the post-war period simply has to be attributed to the increase in labor supply. The problem he therefore underlines is that due to declining population growth, it will become increasingly difficult to keep investment levels at a rate supporting full employment. He foresaw that the decreasing birthrates would eventually lead to a decrease in labor force growth, which would temper economic development. This decline in economic growth would in turn decreases the profit rate of investments, causing a decline in new projects and lead to increasing levels of unemployment. Overcoming this future problem would, according to Hansen, require large increases in governmental investments or further developments in technology opening new fields for investments and employment. The Second World War however, starting only a few years after Hansen made his prediction, did not only draw attention away from his work but also caused such economic turbulence that Hansen‟s predicted effect was drowned out. And over the years, his thesis was forgotten. A full 75 years later it was Lawrence Summers, speaking in front of the National Association for Business Economics in early 2014, who brought the term Secular Stagnation back into circulation.

Just as Hansen did before him, Summers argued that under the prevailing circumstances in the USA and other industrial economies, it would become increasingly difficult to minimize the output gap whilst maintaining financial stability. In his speech Summers posed that, in the years following the financial crisis,

economic growth remained well below

6 Summers also did not see any signs of a return to the long run economic potential; rather, it seemed as if the crisis negatively affected the estimates of future economic performance. And this, according to Summers, is not just to be attributed to a slower growth of factor productivity or a decreasing pace of technological development. Instead, the two most prominent reasons he sees for the stagnation are decreasing capital investments and declining levels of labor input. Together, he argues, these two factors account for an approximate 5% decline in potential economic output (Summers, 2014).

Summers holds that a decline in the equilibrium real interest rate, or the natural rate, lies at the heart of the decrease in capital investment. The natural rate is here understood as the Wicksellian interest rate, and can shortly be described as the return on new capital or the real profit rate.1 And when this profit rate is not sufficiently higher than the interest and other costs associated with an investment to be paid, the level of investments and with it the demand for funding decreases. The decline in natural rate thus puts downward pressure on the already low interest rates. This results in a situation where it becomes increasingly difficult to support economic growth and achieve full employment, as the zero lower bound restricts monetary intervention. A last effect Summers underlines is that the decrease in interest rates increases the risk-seeking behavior of investors, causing financial instability.

A study by Laubach and Williams (2001) supports Summers‟ thesis of declining natural rates. They find that the assumption by central banks of relatively stable natural rates over time is largely unfounded and show that there have been significant fluctuations in the natural rate in recent years. In a later paper by Holston, Laubach, and Williams (2017), the same Laubach-Williams methodology is used for the finding that large declines in trend GDP growth occurred during the period where the natural rate of interest decreased. Their result thus seems to strengthen the claim that there is comovement between the natural rate of interest and (expected) economic performance. This comovement of economic development and the natural rate is in line with Wicksell‟s theory as economic decline signals a decreased return on investment, and an expected decline in return on investment can deter agents from engaging in projects and growing the economy.

The work of Holston et al. (2017) also dictates that, since the economic crisis, the natural rate of interest for the USA is estimated to be slightly above zero percent. This would partially explain the lack of investment despite the availability of cheap funding, as the natural rate or expected return still lies below the amount of interest and other costs to be paid. The researchers further stress that this decrease in natural rate is an international phenomenon, and therefore stems for a large part from developments common to many countries rather than coming from idiosyncratic developments. Another paper by the Bank of England also substantiates the claim of declining natural rates. This paper by Rachel and Smith (2015) tracks the long-term real interest rates (as proxy for the natural rate) over time and also observes the downward trend Summers hypothesized. Over a period of 30 years, the long-term real interest rates across the world have declined with roughly 450 basis points. The authors claim that, following their quantitative analysis, they can account for 400 points of this decrease.

7 It could be argued though that this decline in real interest rates is not due to a decrease in the natural rate, but is caused by for example monetary policy. The work of Rachel and Smith states however that, as the decline in real rates happened during a period where the economy generally was not seen as “overheating” and featured low and stable levels of inflation without much government intervention, the drop in long-term real interest is to be seen as a symptom of a decline in the natural rate of interest. Also, their observed decline in natural rate already began long before the financial crisis and its interest rate distorting aftermath. Bernanke (2015) adds to this discussion by saying that the contemporary low levels of real interest rates are not necessarily caused by the interest rate setting of the Federal Reserve. Although the Fed heavily influences the nominal short-term interest rate, their influence over the long-term real interest rate is limited and depends on a wide range of economic factors. The low short-term interest levels did not cause the decline in the natural rate of interest, Bernanke argues, rather the opposite is the case. The declining natural rate forces the Federal Reserve to set short-term interest at such low levels, as without these low rates there would not be any investment at all.

The decline in real rates and underperformance of the economy is therefore not exclusively to be attributed to the monetary authorities, but, amongst other factors, to the decline in the natural rate of interest. And according to Summers (2016), it is this decline in the natural rate of interest that causes the inability to equate the equilibrium of savings and investment with economic growth, full employment, and financial stability. Or with other words, it is the decline in the natural rate of interest that plays a pivotal role in explaining the underperformance of the current economy.

2.2

S

8 growth figures, sluggish innovation, and the stabilizing “catch-up” growth of developing economies (Rachel & Smith, 2015).

The effect of a growing population on aggregate output has also been studied by Simon Kuznets. He argued that, in the absence of resource constraints, evidence suggest that rapid population growth increases capital accumulation and hence supports economic growth (Kuznets, 1960). However, as the developed economies now stand on the brink of a steep demographic fall in both total population figures as well as old age ratios, the reverse of his theory might now become true. Decreasing populations would then lead to declining levels of aggregate output, which in turn lead to a decreased level of investment following both standard economic thinking and Hansen‟s reasoning. Robert Gordon (2016) underlines the same development in his book “The Rise and Fall of American Growth” where he states that the American economy has come to a slowdown which is likely to be permanent. Gordon argues that innovations are having a smaller impact on the economy than they used to have, and that there are so called “headwinds” that include the rise of inequality, plateauing education levels, and an aging population that temper economic growth (Gordon, 2016). These headwinds mentioned by Gordon are found throughout the economic literature as causes for the current stagnating economy. The aforementioned research by Rachel and Smith‟s (2015) substantiates Gordon‟s claim of headwinds as their estimates show that increasing levels of inequality caused a 45 basis point decline in the real interest rate since 1980, and the demographic transition another 90.2 There are however several other papers that support the idea of demographics affecting the economy. Jinil Kim (2016) for example argues that an aging population played a role in the recent decline of economic growth. She estimates in a quantitative analysis on 18 OECD countries that a 1% shift of the population from the 40-64 bracket to the 65+ bracket reduces the GDP growth rate with 0.47% (Kim, 2016). A study by the IMF on the same topic yields corresponding results. Using a wider country set, they find that a 1 basis point increase in the share of the working-age population (15-64) would increase real GDP per capita growth with 8 basis points. Conversely, a 1 basis point increase in the elderly population of a country (65+) would decrease economic growth with 4 basis points (Callen, Batini, & Spatafora, 2004).

The researcher now most associated with the term Secular Stagnation, Lawrence Summers, argues that, next to the connection between demographics and interest rates that will be explored more intensively later on, there are several other elements that have caused decreasing interest rates across the world. When it comes to the capital market, Summers argues that there has been a reduction in demand for debt-financed investments as large companies hold enormous amounts of cash on hand. The creation of new (tech-) companies also comes with lower and lower capital requirements, and the relative price of capital goods and business equipment has declined over the years. These developments have reduced the demand for investment funds, affecting the equilibrium of interest rates (Summer 2014).

2 _{Inequality leads to decreasing interest rates through decreased human capital accumulation, a decrease in}

9 On the other hand, the propensity to save and the availability of investment funds with it have increased over the years. Summers argues that this increase in savings is due to rising inequality within countries and higher savings by Asian governments following the Asian crisis. These developments are further strengthened by an incline in the holding of central bank reserves and an increase in corporate-retained earnings, resulting in a higher supply of loanable funds (Summers, 2014).

Summing up, industrialized economies around the world have recovered from the economic crisis with a significantly slower pace than economist had expected. If the American economy had performed as the Congressional Budget Office had predicted in 2009, US GDP would be more than 1.3 trillion dollar higher than it is today. And if economist would have known that the G-7 countries would collectively expand the balance sheet of their central banks with over 5 trillion dollar, they would have sworn that we were heading for a period of inflation. But in all major economies, inflation rates generally stayed below two percent. And while government debt-to-GDP ratio‟s exploded in the last decade, the rate on long-term government bonds in for example Germany and Japan are 0.5 and 0.2 percent respectively. Indicating that markets don‟t expect inflation or increases in real interest rates for the foreseeable future (Summers, 2016).

For reasons already mentioned, countries and consumer increased their propensity to save, and the demand for investment funds has significantly decreased. This propensity to save has decreased overall demand, reducing growth and tempering inflation. Furthermore, the increased supply of loanable funds and the decreased demand for those funds has lowered real interest rates across the world. This caused the long-term real interest rate to settle on a low equilibrium that makes it hard to stimulate the economy through monetary interest rate manipulation. To make things worse, these low interest rates increase the risk-seeking behavior of investors causing financial instability, resulting in an unstable low-growth economic situation where the hands of the monetary authorities are bound by the inability to artificially decrease interest rates any further.

2.3

D

2.3.1THE NATURAL RATE OF INTEREST

10 problem we face today is that the natural rate of interest (or at least the approximation thereof) has fallen significantly over the last decades. And with that fall, the ideal interest rate for sustained growth absent of inflation has fallen. As mentioned previously, researchers from the Bank of England estimated that the long-term real interest rates (which they use as proxy for the natural rate of interest3) fell with 450 basis points since 1980 (Rachel & Smith, 2015). Concerning the cause of the decline they argue that this development only partially stems from a decline in economic growth, and also point to, amongst other factors, the increased level of savings by aging populations causing a 90bps decline. Barsky, Justiniano, and Melosi (2014) who use a rich DSGE model to study natural rates come to a much starker conclusion then Rachel and Smith. While they also find that the natural rate has fallen significantly, their analysis shows that since the recession in 2008, the natural rate has been negative.

Feng Zhu, a researcher working at the Bank for International Settlement also conducted a similar research as the one performed by Rachel and Smith. He finds that for several Asian economies, the real interest rate has dropped by 4 percentage points between 1950 and 2014 (Zhu, 2016). His research further uncovers that there is a weak correlation between both asset price developments or credit-to-GDP ratio‟s and the equilibrium real interest rates, but a much stronger correlation between the natural rate and trends in global and demographic factors.

2.3.2OBJECTIONS TO THE DECLINE IN NATURAL RATES

There are however economist who are more reserved about the idea of declining natural rates. Prior to the study of Laubach and Williams in 2003, most work concerning the effect of monetary policy was directed at estimates of potential GDP or measuring future inflation. But with their work, Laubach and Williams sparked interest in studying the real-time measurement of the natural rate and paved the way for several other studies on the same topic. The renowned economist Jon Taylor however criticizes the methodology used in their research, and the methodology of the academic work it sprouted. Together with Volker Wieland he argues that the models used in the measurement of the natural rate of interest are all more or less flawed in their own way (Taylor & Wieland, 2016). They hold that these models, in essence, relate the difference in potential GDP and achieved GDP to the difference between the real rate and the natural rate. They stress however that there is a good to fair change that a variable omitted in these models can influence this relationship. They thus argue that when real GDP is lower than expected GDP, this does not necessarily have to stand in direct relation with a discrepancy between the real and natural interest rate. It could for example also be caused by the (omitted) effect of costly regulations that lower the demand for investment given a certain real interest rate (Taylor & Wieland, 2016).

Next to the potential omission of variables as described above, they argue that the decline in real rates -what is seen by others as caused by a declining natural rate, may just as well be caused by monetary policy. Such policy actions have always made it hard to perceive the natural rate of interest, as this rate can also be seen as the rate that would prevail without disturbances such as governmental intervention. The significant amount of monetary policy undertaken in the last decade would thus make it increasingly difficult to correctly estimate

3

11 the natural rate. Taylor and Wieland also show that the recommendations made to adjust monetary policy to incorporate the changed natural rate are incomplete and inaccurate. Their work concludes that the real-time estimates of the natural rate of interest are still in their infancy, and have to be further developed before they are useful for application in monetary policy.

Taylor and Wieland (2016) further disagree with Summers (2014) and Rachel and Smith (2015) in that they do hold that the economy was overheating before the crisis, which they substantiate by saying that unemployment was exceptionally low and growth was higher than expected. This is important because an overheating economy with actual levels of GDP higher than estimated would suggest that the natural interest rate was higher than the real rate, and thus caused this overheating economy. And when the real rate in the years before the crisis in the USA ranged between 4 and 6 percent, it indicates that the natural rate was even higher at that point in time. The research by Rachel and Smith (2015) on the other hand holds that the economy was not overheating, which causes them to think that the decline in real interest rates must be due to declining natural rates. A potential explanation for this difference is that, in the years before the crisis, there were other factors that can explain the high interest levels despite decreasing natural rates. One such explanation can be the USA government actively stimulating the American people to take out loans and thus increase the demand for mortgages and pushing up the price of money (Horowitz & Boettke, 2009).

When it comes to the post-crisis period, Taylor and Wieland concede that there has indeed been a slow recovery and that real GDP remained well under potential GDP. This forecast error in turn gives economists supporting the Secular Stagnation theory reason to believe that the natural rate has declined. However, Taylor and Wieland hold that this decline can also stem from, for example, regulations negatively affecting investments or policies negatively affecting consumption (2016).

Eichengreen is another economist who is doubtful on the idea of Secular Stagnation, and rather sees inapt policy as the main problem. He holds that the current stagnating American economy is caused by the failure to invest in for example infrastructure, education, and training, and can be solved by concrete policy actions (Eichengreen, 2014). A third rebuttal to the idea of declining natural rates is given by Hamilton, Harris, Hatzius, and West (2015). Like Taylor and Wieland, Hamilton et al. maintain that there is an enormous uncertainty around the estimates of the equilibrium real interest rate, and that these estimations come with enormous confidence intervals. They find little support for the idea of a permanent decline in real interest rates, and point to temporary factors such as debt deleveraging, shocks in personal discount rates, and financial regulation as potential causes for the decline in interest rates (Hamilton et al., 2015).

2.3.3DEMOGRAPHICS AND REAL INTEREST RATES

12 current economic situation. One such study on the influence of demographics on the prevailing interest rate is performed by Favero, Gozluklu and Yang (2016). These authors argue that spot rate calculations are a combination of a persistent long-term expected value and a mean-reverting component. Normally, the persistent component of the term structure is calculated by extending the one-period risk free rate over a longer time horizon. Favero et al. argue however that this causes a high level of persistency that makes future estimations inherently inaccurate. The researchers therefore propose a model in which the (equilibrium) real rate is slowly evolving as a function of changes in demographic variables. Including this alteration in the standard modeling of the term structure for nominal rates yields them results that far better describe the development in interest rates than standard models do.4

A paper by Carvalho, Ferrero and Nechio (2016) also underlines the relationship between demographics and interest rates. Replicating the economy in a life-cycle model they find that, in response to demographic shifts, the real interest rate fell 1.5 percentage points between 1990 and 2014, and is expected to fall another 50 basis points in the upcoming 20 years. This fall is attributed to two main mechanisms, of which increasing life expectancy plays the most dominant role. The longer life span acts as a preference shock to save more in all stages of life to finance the consumption over a longer time horizon, which lowers the interest rate as agents accumulate more savings.

The second mechanism is the result of two conflicting effects of slowing population growth. On the one hand decreases the declining population growth the pool of workers, which, in their model, influences the capital to labor ratio as more workers get replaced by machines. This decreases the marginal product of capital as a whole and slows down economic growth, which in turn decreases the real rate.5 The declining population growth on the other hand also increases the ratio of retirees to workers. And as economic agents that are already retired tend to spent their money rather than continue saving for the future, the amount of savings declines, which increases the interest level. The overall effect of these two mechanisms is negative, so that the current declining demographic growth and old-age ratios sums to a negative effect on real interest rates. And although the effect of a declining population on interest rates might be a relatively temporary shock, the increasing life expectancy will continue to have an effect on real interest rates for the upcoming decades (Carvalho, Ferrero, & Nechio, 2016).

Also interesting to mention is that in a time much different from the one we are in right now, Howard Howe and Charles Pigott (1991) performed research into the ever increasing real interest rate. Since the mid-1970‟s, the US, together with Japan, France, Germany, and the UK, witnessed real interest rates significantly departing from their historical averages. And in a similar vein as Holston et al. (2017), Howe and Pigott suggest that this upward trend in real interest rates is caused by internationally occurring phenomena as the development has been

4 _{Important to note is that for the incorporation of these demographic trends into the calculation of nominal rates,}

Favero et al. created a demographic variable for which they take the ratio of middle-aged (40-49) to young (20-29) Americans.

5 _{Decreasing availability of workers leads to a greater reliance on capital and machinery, depressing its marginal}

13 remarkably similar across countries. They further conclude that the changes in real interest rate stem from relative permanent changes in the underlying natural rate, and are not simply caused by the effect of monetary policy.

To explain the increasing rates, Howe and Pigott look at the rate of return on capital, the risk premium, and changes to the financial structure. They conclude that capital productivity has increased, and the increase in availability and use of financial instruments together with deregulation of the market increased private sector debt. Higher private debt in turn increased the risk of default and caused an increase in interest rates. And although this sounds as a reasonable explanation, its still seems odd that only 25 years later we see economist studying the exact opposite development in real interest rates who also point at developments in the underlying natural rate. And if the methodology used by Howe and Pigott is sound, we should therefore witness declining returns on capital, decreasing risk premiums, and changes in the financial structure that would decrease the interest rate. And for a part, this is corroborated by the fact that the return on capital has declined rather than increased the last years, and that quantitative easing possibly distorted the natural equilibrium of interest.

It might also be the case however that both the increasing rates of the 70‟s and 80‟s as well as the decreasing rates in the current economic situation are caused by the same underlying demographic factor. Howe and Pigott even unknowingly provide reason to believe that population developments may have played a role in this process as they state that the markets probably did not anticipate the rapid rise in inflation during the mid-1970‟s. These high levels of inflation can however be explained (next to other sources of inflation such as oil-crises) by the rapid increasing population and Baby Boomers entering the workforce (Dent, 2013; Bobeica, Nickel & Sun, 2017). So despite the exclusion of a demographic variable in the original work of Howe and Pigott, it might be prudential to consider the development of this macroeconomic variable as it seems to align with both the increasing natural rate several decades ago as well as its hypothesized decrease nowadays.

To summarize the literature review it can be said that, although there have been substantial studies into the current economic situation, it is not altogether clear whether the natural rate of interest rate has declined per se or that the decline in economic activity stems from governmental intervention or the omission of other relevant variables. However, the idea of a declining natural rate and the Secular Stagnation theory are becoming increasingly acknowledged in the economic literature since the reintroduction of the concept by Summers in 2014. Taylor and Wieland (2016) also do not outrightly reject the idea of declining natural rates, but assert that the models that have been used thus far have been flawed and need to become more advanced.

14 Figure 2: Japelli & Modigliani’s life-cycle hypothesis representation

policy or caused by quantitative easing. This would in turn substantiate the claim that real interest rates have gone down due to reasons other than monetary policy, and for reasons that are seen in the literature as relating to the Secular Stagnation theory and the decline of the natural rate of interest. Besides, the decline in natural rate is assumed to have begun before the large interest rate disturbing events after the 2008 recession (Rachel & Smith, 2015).

3.

H

The life-cycle hypothesis holds that economic agents save in order to accumulate resources to spend later on in life. This entails that savings are positive during the years which agents spent working, and negative during the time of their retirement. When you combine this with the notion that in their early years, economic agents take out loans to support themselves, you derive at a hump-shaped model of savings and wealth (Jappelli & Modigliani, 1998). Figure 2 depicts this development and shows that, in their younger stages of life, economic agents borrow to finance their consumption. When their income curve starts to rise, they repay those loans and begin saving for retirement. Eventually, the economic agent quits his or her job and lives from its savings.

Thus if the life-cycle hypothesis is valid, the consumption smoothing of economic agents would entail that people move from net borrowers in their youth, to net savers in their working years, and to net dissavers in their elderly years (Yoon, Kim, & Lee, 2014). And these stages of saving and dissaving affect the economy when the age distribution of a population is skewed. Equal population spreading over age groups on the other hand would minimally affect the economy when those groups age over time.

15 Because this generation, known as the Baby Boomers6, was larger than any other population cohort, they have (had) a significant effect on the economy. Consumer Expenditure Surveys from the U.S. Bureau of Labor Statistics provides us with valuable data in this regard. If you know at what ages which purchases are made, you can create a “consumer life-cycle” which maps the purchases of the average economic agent throughout its lifespan. These surveys tell us for example that the average family borrows the most when the parents are age forty-one, which corresponds with the age where consumers typically make their largest home purchase (Dent, 2013).

These surveys further tell us the age at which an economic agent on average enters into the workforce, gets married, buys its first home, purchases a better car, or spends his or her money on cruise ships, prescription drugs, and eventually nursing homes. They also tell us that the age at which an economic agent reaches its peak spending lies around 46, as can be seen in figure 3. Important to note is that, following the work of Dent (2013), consumption is thus not as straight as assumed in line C of figure 2.

And due to the sheer size of the Baby Boom cohort –almost 75 out of 216 million Americans in 1975, they account for a major share economic activity (Colby & Ortman, 2014). Combining this widely available data on demographics with information on spending behavior over time and the life-cycle hypothesis of Modigliani, it is estimated that the peak spending potential for the American economy was in the years 2003 through 2007, and was highly affected by the Baby Boom generation. For Japan this was between the years 1989 and 1996, for Germany between 2010 and 2013, and for China this will be between 2015 and 2025. After those years, the Baby Boom generation will have passed their peak spending level, and aggregate demand will decrease (Dent, 2013).

As the Baby Boomers were never followed by an even bigger or an at least equally sized generation, we are now witnessing declining population figures and increasing old-age ratios. This problem is further aggravated by the decline in fertility rates and increasing life expectancies, resulting in both a contracting as well as aging population. And with this development, Alvin Hansen‟s theory on Secular Stagnation made its reintroduction in our

6 _{Defined as the sudden increase in fertility rate after the Second World War. Although the duration of this}

period differs slightly amongst countries, it is often classified as the period between 1946 and 1964 (Colby & Ortman, 2014)

16 current economic situation. We witnessed an excessive population growth right after the Second World War which was not sustained over time. And just as hypothesized in 1939, this was eventually followed by declined economic performance, a decreased real interest rate, and possibly a decreased natural rate.

The original work of Hansen in which he expressed his growing conviction that “the

combined effect of the decline in population growth, together with the failure of any really important innovations of a magnitude sufficient to absorb large capital outlays, weighs very heavily as an explanation for the failure of the recent recovery to reach full employment” still

seems to hold value after all those years (Hansen, p.230, 1939). His theory encompasses both the work of Robert Gordon (2016) who stated that the American economy has come to a slowdown due to a lack of revolutionary innovations, as well of that of Summers (2014, 2015, 2016) who blames the declining rate of population growth and increasing loanable funds supply. And to briefly summarize the literature review, not only Summers himself but also economists as Kuznets (1960), Callen et al. (2004), Kim (2016) and the authors of the IMF World Economic Outlook (2004) relate these economic developments to changing demographic factors.

What is more, the decline in real interest rates which is connected to the decline in economic performance (Summers 2014, 2015, 2016; Holston et al., 2017), seems to be affected by this declining demographic trend (Bloom et al., 2003; Carvalho et al., 2016; Favero et al., 2016; Zhu, 2016). The demographic development of a country thus seems to play a pivotal role in its economic situation. So, as aging life spans have led to a higher propensity to safe at all stages in life (Callen et al., 2004; IMF, 2004; Zhu, 2016), and the decreasing fertility rates have declined both the amount of young economic agents and the need for new loans (Yoon

et al., 2014), the economy has come to a demographic-induced standstill. And although the

increasing old-age ratio observed in our current demographic structure has a slight interest rate increasing effect through the increased dis-savings of the elderly, this effect is offset by an even larger effect of decreasing returns on capital (Carvalho et al., 2016), further strengthened by an increasing amount of cash-on-hand by corporations and decreasing costs for start-up enterprises (Summers, 2014).

There thus seems to be ample support for the idea that changes in the demographic composition of a country play an important role in its economic performance and interest rate. And to gain further insight in this relationship, this paper relates the demographic composition of several economies over time to the development in their real interest rates. The first, more general, hypothesis that follows from the literature review tests whether there is a relation between age distributions and interest rates at all. This is done under the following hypothesis: H1: The demographic composition of an economy has an effect on the development of the real interest rate.

To further tailor the empirical analysis to our current economic position, the second hypothesis will specifically take the effects of an aging society on real interest rates into consideration. The formulation of the second hypothesis will therefore be as follows:

17

4.

M

4.1

O

It seems as if the research that has been conducted on the effect of demographics on real interest rates, although important, has had its flaws. The paper by Holston, Laubach, and Williams (2017) for example holds that the decline in the natural rate of interest must be an international occurring phenomenon as they observe the same trend in multiple countries. However, the countries they use in their analysis (being the USA, the Euro area, Canada, and the UK) show highly similar demographic developments. African countries for example are in a completely different demographic phase, showing decreasing old-age ratios as their populations are rapidly growing. Including countries with an opposite development in demographic structure might create a more complete understanding of the effect of demographics on interest rates, and will at least generate a less biased view.

The paper by Favero et al. (2016), which shows that there is a significant effect of incorporating a demographic variable in the calculation of interest rates, also uses an unsound methodology. Only focusing on the ratio of 40-49 to 20-29 year olds discards enormous parts of the population. And as the Baby Boomers now range between the age of 53 and 71, they completely omit this influential group of people. Employing a ratio that looks at the age distribution in its entirety might therefore give the demographic transition an even more pronounced effect on interest rates. What is more, looking at the entire age distribution is also justified when you consider that we are now in a time period were many of the Baby Boomers are on the brink of transitioning from being employed to being retired. This transition will have a noticeable effect on the economy, and might even lead the industrialized Western world to follow a path similar to Japan‟s lost decade.

Other papers relating the effect of demographics to changing interest rates tend to start at the year 1990 (Carvalho et al., 2016 and Holston et al., 2017). A probable reason for doing so is that data on interest rates before the 1990‟s is poorly registered, and only a few countries have records dating further back in time than the 1980‟s. However, as I will argue later on, demographics might have had a pronounced effects even before 1960 as this is when the disruptively large baby boom cohort made its introduction. Taking interest rate developments before 1990 into account is therefore reasonable. Another shortcoming in the literature is providing statistics for the difference between the working population and the non-working population, or any other hard separation between groups. Callen et al. (2004) for example find that a 1 basis point increase in the share of the working-age population (15-64) would increase real GDP per capita growth with 8 basis points. However, they do not take the gradual transition of the population into account. In my opinion it would be better the look at the shifting age distribution and the amount of people in multiple age groups over time, and relate that gradual transition to the development of the real interest rate.

18 pointed out. Those models also have problems with correctly representing the data on demographics provided to them by the United Nations. The paper by Carvalho et al. (2016) for example models a dependency ratio that is constantly a few percentage points above the ratio provided by the UN. In the period between 2020 and 2060 they are even “under-modeling” the dependency ratio with 10 to 15 percent, which distorts the effect of demographic developments on the real interest rate.

Models also create a more homogenized view of countries. In the same model by Carvalho et

al., a country like Japan with a decreasing population and a life expectancy above 80 years

finds itself in the same condensed representation of reality as the USA with a slowly increasing population and life expectancy below 80 years (Carvalho et al., 2016). Even the models used for macro-economic policy are prone to mistakes when it comes to demographic variables. These models incorporate demographic factors through the growth and size of a population, but not through the development of the age distribution (Bloom et al., 2003). Similarly, the standard representative agent model operates under the assumption that all household members work, which allows for variation in the total population. It however does not include variation in the ratio of workers to non-workers, which is related to the age of the economic agent (Ikeda & Saito, 2012). And although it comes with its own set of limitations, relating actual data on the development of demographic trends to actual data on real interest rates through panel data analysis might circumvent the problems that are found in the use of models.

4.2

A

To observe the change in real interest rates stemming from demographic forces, it is essential to understand how this rate is constructed. And this distinction between factors brings us back to the Wicksellian natural rate of interest. The importance of this distinction lies in the fact that, according to Wicksell himself, the interest rate consists out of two components. First, there is the natural rate of interest which is where the market rates will go to over time in absence of disturbances. This natural rate is equal to the rate of return on new capital, or the real profit rate of an investment (Anderson, 2005). Secondly, there are the disturbances themselves. These disturbances cause the discrepancy between the natural rate and the market rate, or the financial rate as Wicksell called it. In his framework, Wicksell recognizes that the actual interest rate only gradually adjust to changes in the underlying natural rate as asset supply and demand respond slowly to altered conditions (Howe & Pigott, 1991). The long-term real interest rate is therefore a combination of a slowly developing natural rate and temporary events that cause deviations from this rate.

19 is substantial reason to assume that there is a link between demographics and the natural rate (Bloom et al., 2003; Carvalho et al., 2016; Favero et al., 2016; Zhu, 2016).

This is helpful in the construction of the empirical analysis as it tells us not to put too much emphasis on factors that are to be seen as temporary disturbances. Interest rate shocks due to for example the Brexit, or shifting rates after announcements from the Federal Reserve are important in the day to day assessment of interest fluctuations. But these events play a more limited role in the long-term development of the (natural) interest rate. This idea is corroborated by the aforementioned statement by Bernanke who poses that the low short-term interest levels set by the Federal Reserve did not cause the decline in the natural rate of interest, but that the declining natural rate itself forced the Federal Reserve to set short-term interest at such low levels (Bernanke, 2015).

The research by Howe and Pigot (1991), which also adhered to the difference between changes in the natural rate and the determinants of fluctuations around this trend, comes to a similar conclusion. They find that the development in the money market intervention rate (such as the federal funds rate for the USA and the call money rate for Japan) probably will have influenced short term rates, but are not to blame for the sustained development of long-term rates. Their argument aligns with the argument made by Rachel and Smith (2015) who also state that government policy was certainly not the only factor to play a role in the trend development of the (equilibrium) real interest rate. So despite the recent developments in the market that might disturb the current real interest rate, it is assumed that there is also a significant share of variation caused by the decreasing natural rate. And this part that is not due to monetary policy and quantitative easing therefore can be related to demographic transitions.

4.3

S

Testing for the effect of demographics on real interest rates requires performing a statistical analysis in which the development of real interest rates over time functions as dependent variable. As there are various sources that influence both the underlying natural trend as well as the disturbances amongst this trend per country over time, a fixed effect panel regression will be used. The use of a fixed effect model is further corroborated by the idea that the interest lies at the within variation of a country, the sample is not randomly drawn from a greater population, and this method only looks at variables that change over time.7

However, there is still a chance of residual confounding when there are omitted variables that change over time that influence the dependent variable and are moderately correlated with demographic development. Two such variables are the level of savings and investment, as both variables are influenced by the demographic transition of a country. So as to account for the fact that levels of saving and investment influence the interest rate but are also correlated with the population distribution itself, these levels will be included as control variables. To use of savings and investment to GDP is also recommended by the work of Orr, Edey, and Kennedy in 1995. These authors list several conditions in their article to which a study into long-term interest rates should adhere. The main point they make is that an empirical analysis

7

20 should ideally make a distinction between low and high-frequency factors as determinant of interest rates. High-frequency determinants are matters like monetary policy and local shocks to inflation that temporarily influence the interest rate and differ amongst countries. Low-frequency variables are the more slowly developing factors and are expected to be relatively equal amongst comparable economies. This overlaps with the idea of trend development and determinants amongst this trend expressed in chapter 4.2 as proposed by Wicksell.

Orr et al. (1995) find influences on the low-frequency determinants to be the structural shifts in the rate of return on capital, risk premia development, and the levels of savings and investment relative to GDP. Interesting to note is that Orr et al. in listing the factors that influence savings and investment already mention longer-term demographic factors by pointing to the effect of aging populations and the rise in dependency ratios. They also underline the global integration of capital markets as a low-frequency determinant for interest rates. Due to globalization and the possibility of capital flows between countries, a global capital market has emerged. Companies and investors therefore no longer rely on their own country for the supply of loanable funds but can get their money abroad.

And not only does this influence the development of interest rates as Orr et al. mention, open capital markets also limit the role of population dynamics. As capital markets become more and more open, the country specific demographic trends play less of a role in the determination of its interest rate, as the interest rate is now also dependent on the rate in other countries. This also means that the demographic development in a country with a strong influence on the rate in other countries, such as the USA, matters more than the demographic development of a country like Burkina Faso.8

To (partially) account for the openness of capital markets, the world real interest rate as estimated by King and Low (2014) will be incorporated in the regression analysis. In their research, King and Low use real rates on 10-year maturing government bonds to construct an average long-term real interest rate. With the exclusion of Italy which has had its bond rates influenced by their chance of bankruptcy and exit from the European Monetary Union, their research uses data from the G7 countries. Their chosen approach creates a GDP weighted average of the return on government bonds issued with inflation protection expressed in US dollars, while accounting for exchange rate developments. They show that bond rates are highly correlated with the rates in other countries, and argue that there is reason to talk about a “world” interest rate (King & Low, 2014). From the year 1985 onwards, they provide yearly estimates for this world rate which will be included in the regression.

The return on capital and risk premia development mentioned by Orr et al. (1995) are not used in the regression. The return on capital per country and year is hard to come by or identify empirically and is therefore not included in the regression. Also, this research is interested in the return on new capital, while the normal ratio looks at the ratio between current profits and capital in its entirety. The risk premia is excluded as this research uses the interest rate on government backed bonds. However, to more conclusively distill the effect of demographics, other variables will also be controlled for. In a report from the (central) Bank of Canada, Brigitte Desroches and Michael Francis (2010) also find evidence that suggest that

21 the real long-term interest rate in G-7 countries share a common global component. These countries are, due to open capital markets, integrated with the world market to such an extent that their interest rate reflects the global savings and investment equilibrium. Their research further list several other factors that determine long-term real interest rates. The most important determinant of the real interest rate they mention is the development in labor force growth, which is evidently linked to demographic transitions. From the perspective that every employee needs a certain amount of equipment, the decrease in labor force growth would entail a decreased need for capital. This development coincides with the increase in interest rates when the Baby Boomers reached the age of labor participation and the decrease nowadays.

Their work concludes that the relative decrease in investment demand has been more important than the increase in global savings for explaining the decline in global rates. They further argue that the two most influential factors, being the labor force growth and age distribution (which they measure using dependency ratios), adjust slowly over time and reflect long-term trends. Desroches and Francis (2010) further mention variables that were also mentioned in the literature review such as the world distribution of income (Rachel and Smith, 2015), increases in corporate retained-earnings (Summers, 2014), and financial openness of markets (King and Low, 2014). The former two variables will however not be included as it goes beyond the scope of this research. Another variable that, following the literature review, influences the interest rate and correlates with the demographic variable is GDP growth. This variable will therefore also be included in the regression.

The aforementioned work of Orr et al. (1995) concludes that an empirical analysis into the determinants of long-term real interest rates should abide to a certain list of standards. First, the research should ideally make a distinction between low and high-frequency determinants. Additionally, it should identify the fundamentals determining the low-frequency developments in real rates such as savings and investment to GDP. The regression should also incorporate a variable that identifies the process by which real rates move together internationally. In the case of this research, the inclusion of a demographic variable fulfills this last role as it is hypothesized that the demographic development that is common to multiple industrialized economies lies at the heart of the decline in interest rates. And lastly, the analysis should allow for the influence of real rate developments in larger countries on the rates in smaller countries, which is done by the inclusion of the World Interest Rate as provided by King and Low (2014). Abiding to these demands regarding the empirical analysis of long-term real interest rates yields the following equation:

Real interest rate it

i t

22 level of investment is predicted to have a positive effect on interest levels as higher investment increases the demand for capital. GDP growth is expected to be positively correlated to interest rates, as a growing economy would entail the profitability of investments. The world rate is included to capture the effect of open capital markets limiting the effect of the demographic composition within a country. And because of the already found relationship by King and Low of interest rates in G7 economies affecting the rest of the world, this variable is expected to be positive.

One shortcoming of this regression is that taxation differs amongst countries, which affects the effective (after-tax) rates (Orr et al., 1995). Another shortcoming is that the focus of this regression lies with the low-frequency or trend development of interest rates, and does not include variables for the disturbances amongst this trend. This choice is rationalized by the idea that this research is interested in the effect of demographics on long-term real interest rate development and not the disturbances amongst this trend. Also, as expressed in chapter 2.1 and the last paragraph of chapter 4.2, even despite the recent developments that might disturb the real interest rate it is believed that there is also a significant share of variation that is caused by the decline in natural rate. However, the exclusion of such high-frequency variables does slightly decrease the reliability of the results as the effects of these omitted variables are now partially attributed to the variance in demographics. One way to minimize these effects and increase the focus on the trend or low-frequency development is smoothing the interest rate variable. Smoothing this rate will partially filter out the high-frequency disturbances while leaving the trend development intact.

4.4

D

Data for the demographic variable is retrieved from the United Nations Database, which provides data on several indicators for age distributions. One such way of modeling the age distribution within a country is by use of dependency ratios. These ratios show the amount of people per 100 working-age inhabitants who are “dependent” due to either young or old age. The hypothesized cause of low interest rates, being the age distribution, can be depicted through this dependency ratio as done in figure 4.

23 The specific use of G8 economies in this graph is rationalized as the work of Orr et al. (1995), King and Low (2014), and Desroches and Francis (2010) tells us that these are the countries with the greatest impact on the prevailing world rate, and thus have the greatest effect on interest rates in other countries. Looking at the demographic development for these countries in specific is therefore more interesting than looking at the development of eight randomly picked economies.

Because the total dependency is the combination of both young and old per hundred economic agents in their working ages, the line roughly follows a u-shape. From 1950 to around 1965, the ratio goes up for most G8 countries as the Baby Boom generation makes it introduction and ads to the amount of young people. Several years later, these Baby Boomers reach the age where they move from the young to the working part of the population, thus reducing the dependency ratio. For the United States for example this is shown by a rate of 73 dependents per 100 workers in 1950, 94 in 1965, and then down again to 66 in 2010.

These days, the Baby Boom generation is starting to move from the working-age to the 65+ part of the population. This causes the amount of dependent people in the US to rapidly increases from 66 in 2010 to 81 in 2030 and 88 in 2060. And although it can be argued that the dependency ratio was even higher in 1965, it should be noted that during those days this figure was high due to the amount of young people who would grow the economy in the upcoming years. Nowadays, the ratio is high simply due to the share of old people in society. The effect of the Baby Boom generation on population distributions is further strengthened by the increasing life span and the decrease in fertility rates (see appendix section D). As mentioned in chapter 3, the increase in life span not only causes an increase in savings in all stages of life, but it also simply increases the years of retirement for an economic agent. The decrease in fertility rates exacerbates the impact of aging Baby Boomers as there are fewer young economic agents that can balance out the amount of old people. This shifting age distribution becomes even more noticeable when only the ratio of elderly people to the working-age population is observed, as can be seen in figure 5.

24 All G8 countries start off with fewer than 20 old-age dependents per 100 persons in the working-age population in 1950. Due to longer life spans and decreasing fertility rates, this ratio slowly increases to around 25 per 100 in 2000. From here on out however, the increase becomes more rapidly. The ratio of old people in Japan for example increases from 25 per 100 in 2000 to 55 in 2025. And although Japan is an extraordinary case as they had an earlier and steeper Baby Boom curve, the development in the seven other countries follows the same trend.

These developments most certainly have an impact on the economy as a whole, as already found by Summers (2014, 2015, 2016), Kuznets (1960), Callen et al. (2004), Kim (2016) and the authors of the IMF World Economic Outlook of 2004. And due to the ever decreasing fertility rates and increasing life spans, demographic transitions might thus have a profound impact on the economy and real interest rates.

To continue with the data collection process, the interest rates have been found in two separate ways. Using the OECD Database, the long-term nominal interest rates on government bonds maturing in 10 years were found. The use of government bonds minimalizes the impact of omitting a variable for risk premia, as all loan repayments are guaranteed by governments. For non-OECD countries however such data was hard to be found. Instead, data from the World Bank on the real lending rate for the medium-term financing needs of the private sector were used. This World Bank dataset al.so contained medium term interest rates for OECD countries. A third variable was therefore constructed that contains the government bond interest rates for OECD countries, and medium term private sector rates for the rest of the world. These nominal government rates where then transformed into real rates following the same method the World Bank uses to transform nominal into real rates. This constructed variable was then smoothed as mentioned in chapter 4.3.9

The use of the private medium-term World Bank data is not optimal as it is more influenced by high-frequency developments and has a higher risk premium. Also, medium-term loans do not have the same relation to the natural rate of interest as long-term interest rates do. For the interpretation it should thus be taken into account that the data in this constructed variable comes from separate sources, which inherently compromises its reliability.

Data on savings and investment to GDP also comes from the World Bank database. Savings to GDP is measured as gross national income minus total consumption plus net transfers. Investment to GDP consists of the gross capital formation for additions to the fixed assets of the economy plus net changes in the level of inventories. The variable for GDP growth is the annual per capita GDP growth in percentages measured in dollars, and also retrieved from the World Bank. Lastly, data on world interest rates comes from the aforementioned paper by King and Low (2014) and is included from the year 1985 onwards up until 2013.

9 _{Transforming nominal into real rates was, for consistency, done by using the same formula as the World Bank:}

25

5.

A

Several different regressions are ran throughout the empirical analysis depending on which perspective is being explored. And all these perspectives will be regressed under increasing amounts of independent variables. The first regression will only incorporate to proxy for the demographic transition to see the effect of demographics in its entirety. Consecutive test, shown in the same figure, will control for other macro-economic variables that determine the interest rate and are moderately correlated with population dynamics. The first test concerns the countries on which richer data is available, as the measurement and registration of real interest rates started at different times amongst economies. Most countries start this practice around 1985; there are however some OECD countries that recorded long-term interest rates starting several decades earlier. These countries, being Belgium, Canada, France, the Netherlands, South Africa, the UK, and the USA, will be used in a first regression using data starting in 1961. This is done as using a longer time period in the regression could give greater clearance on the effect of changes in demographic trends over time.

The output in table 1 shows that, first of all, the variables used in the models do have some explanatory power as reflected by the significant F-tests. To be more precise, the change in the total-dependency ratio over time accounts for 19.31% of the change in the smoothed interest rate. Including all variables yields and even higher 69.45% explained variation. This amount of explained variation is rather high, which can be attributed to the inclusion of both the normal determinants for the cost of money, and the use of the smoothed interest rate. The same regression ran on the “un-smoothed” rates yields a lower 13.52% and 60.24% of within variation depending on either including only the total-dependency ratio or all variables. This is an important difference as it shows the distorting effect of high-frequency disturbances on the empirical process. Artificially excluding those disturbances therefore gives the slow moving variance in demographics the chance to overlap with developing interest rates.