Japan : high debt, low yield

(1)

1

Japan: high debt, low yield

Jeroen Nobel 10273271

University of Amsterdam

Supervisor: S. Chan

(2)

2 Abstract

High public debt-to-GDP ratios and fiscal deficits usually result in high yield on long-term

government bonds. For Japan, however, this is not the case. This study examines the possible factors due to which the yield on Japanese government bonds has been low, and fairly stable despite their high and sharply increasing government debt. These factors have been regressed on the nominal yield on 10-year government bonds using an OLS regression. The results are that there are two important and significant factors in driving down the yield, namely: the gross domestic savings and the exchange rate as an indirect measure of home bias

(3)

3

1. Introduction

4 2. Literature review

5 2.1 Existing studies

5 2.2 Channels of impact

6 3. Japan

7 3.1 Debt comparison with other countries

7 3.2 Accumulation of Japanese debt

8

3.3 Causes for the low long-term interest rates

11 4. Data

14 5. Model

15 5.1 Estimated model

15 5.2 Pre-estimation tests

16 6. Results and analysis

17 7. Discussion

19 8. Conclusion

20 9. References

21

(4)

4

1. Introduction

During the last financial crisis, starting in 2008, many countries have run large fiscal deficits to overcome their small economic growth. These fiscal deficits have led to high debt-to-GDP ratios in the developed countries, in some cases even around 100%. Greece even has been bailed out twice, in 2010 and 2011. However, it is not uncommon that governments are having large accumulations of debt. Since the 1980s, debt levels are increasing due to the deficits that governments are running. These sharply rising fiscal deficits and government debt have raised many questions about their effects on long-term interest rates. It is not surprising that there have been done a lot of studies on the effect of both fiscal deficits and government debt on the yield of their long-term government bonds. Economic theory suggests that both increasing levels of government debt and budget deficits result in higher interest rates on long-term government bonds. In most advanced countries, this indeed has been the case. During the financial crisis that started in 2008, the countries which have witnessed high debt accumulation also witnessed sharp decreases in the prices of their government bonds, and thus increases in the yield on their government bonds. This is mainly due to the higher risk premiums these countries have to promise their investors.

When it comes to debt, deficits and the yield on long-term government bonds, Japan is a special case. Since the beginning of the 1980s, Japan has seen its debt-to-GDP ratio increase to above 215% in 2012. However, unlike what the theory would suggest, the yield has decreased from almost 7% to below 1% in 2012.

The main purpose of this thesis is to investigate what the reasons are for this low yield on Japanese government bonds. Therefore the main question of this thesis is: how can the yield on 10-year Japanese government bonds be that low, while the debt-to-GDP ratio is at this high level?

This rest of this thesis is organized as follows: Section 2 summarizes existing literature to show what previous studies of the effect of both government debt and fiscal deficit have pointed out. In section 3, the case for Japan is being discussed. After that, the data and model will be explained in respectively section 4 and 5. Section 6 explains the results and analysis of this model and section 7 and 8 contain the discussion and conclusion.

(5)

5

2. Literature review

2.1 Existing studies

Gale and Orszag (2003) have examined the relationship between term fiscal discipline and long-term interest rates. They have examined 42 studies, of which only 17 found a significant positive effect of the deficits on interest rates. The differences in these various studies possibly arise due to differences in model specifications, explanatory variables, sample and time periods. But the

differences could also arise because different budget variables are used. Gale and Orszag found that studies that incorporate deficit expectations in addition to the current deficits tend to find more significant linkages between deficits and the long-term interest rates. Of the 42 papers they examined, only 17 incorporated expected deficits of which 12 found statistically significant linkages between deficits and interest rates and 4 found mixed effects. The conclusion they find on these papers, is that an increase of 1 percent of GDP in the budget deficit results in an increase of long-term interest rates of about 30-60 basis points. Empirical studies of Hauner and Kumar (2006) and Tokuoka (2010) found similar results on empirical studies and concluded that an increase of 1 percent of GDP in budget deficits result in increases of respectively 13-24 basis points and 15-40 basis points in long-term interest rates. Both these studies used a panel of G7 countries to base their results on. Engen and Hubbard (2005) developed a theoretical framework in which the potential effects of government debt on interest rates are described. In this model, the interest rate is determined by the marginal product of capital. The level of the interest rate is determined by the level of the capital stock, and thus by the level of government debt. The change in the interest rate is determined by the change in the level of capital stock, or the change in the level of government debt. They add to this that since the change in government debt in a year is roughly equal to the government deficit of that year, the change in the interest rate is determined by the government deficit. The results of this framework indicate that an increase of 1 percent of GDP in government debt would result in an increase of 1.4-2.4 basis points in interest rates.

(6)

6

2.2 Channels of impact

According to Baldacci and Kumar (2010) there are a few important channels which impact the long-term interest rate through the fiscal balance and debt.

One of the main channels is via the national savings, which is the sum of public and private savings. If the budget deficit increases (which means there is a decline in public savings), national savings decline unless it is fully offset by private saving. Given the saving identity, which states that national saving must be equal to the sum of domestic investment and net foreign investment, a reduction in national saving relative to domestic investment and net foreign investment implies that there is a shortage of funds to finance these investments. Therefore firms (and the government) have to compete for their funds and offer higher interest rates (Gale & Orszag, 2003).

However, Mankiw (2000) says is that the government’s debt policy can have a short-run effect, but not a long-run effect on an economy and its interest rates. He has developed a model in which there are spenders and savers. If the government adapts a tax cut which is financed by a higher level of debt, the response of the savers is to do nothing. The savers are forward looking, and with the

Ricardian equivalence holding and this higher level of debt, they are expecting a tax hike in the future. Therefore they will save, to be able to pay the higher future taxes. On the other hand, the spenders will consume their higher disposable income which reduces investment and thus the interest rate. This higher interest rate stimulates the savers to save even more until the interest rate is back to its normal rate. So when assumptions about the agents their behaviors are a little bit modified, the results through this channel of impact may not fully hold.

Elmendorf and Mankiw say that fiscal deficits may increase aggregate demand in the short run. If a government creates a deficit by reducing their tax revenues with equal spending, the income of households will rise. This increase in income results in higher consumption and, thus, in higher aggregate demand. In the long run, this created fiscal deficit will reduce public savings and so national savings as well (1998). Both higher aggregate demand and reduced national savings result in an excess supply of government bonds, which in turn results in higher real interest rates (Baldacci & Kumar, 2010).

Another channel of impact is through the inflation expectations. Higher inflation expectations may increase the nominal yield because inflation premiums will be demanded by investors. In countries with a positive output gap, or countries where there are concerns about the monetization of their debts, these higher inflation expectations may also generate macroeconomic uncertainty which can lead to worsening fiscal solvency concerns and higher country risk premiums and government bond yields (Baldacci, Gupta & Mati, 2011).

A fourth important factor is through the short-term interest rate. The short-term interest rate is set by the Bank of Japan and is the rate at which short-term government securities are issued. The monetary policy and cyclical conditions of a country are reflected in the short-term rate.

(7)

7

Blanchard (1991) says that if actions are taken which will lead to large deficits in the future, the short-term interest rates are expected to increase in the future as well. Following the expectations theory, the future short-term interest rates influence the long-term interest rate now. So, large deficits and debt could raise the long-term rates through the short-term rates. Additionally, they could lead to concerns about the government’s ability to service its debt. These concerns can raise credit risk premiums and, just like the country risk premiums with the inflation expectations, raise the government bond yield. However, Mankiw (2000) thinks that government debt is neutral. According to Buiter (1988) debt neutrality means that, given public spending program, the real equilibrium of an economy is

independent on the way this public spending is financed. This proposition is also called the Ricardian equivalence. The Ricardian equivalence holds that agents are forward looking. So, if a tax cut is introduced, the agents will expect a tax hike in the future and therefore will save part of their additional income to overcome the future tax hike. Rebelein (2006) says that if the Ricardian

equivalence holds, the amount of saving depends on the ability of the child to manipulate the parents. He concluded that if the parents can be manipulated by the child and a tax cut will be induced, a bigger amount of the extra income will be saved for their children to let them overcome an eventual future tax hike than if the parents cannot be manipulated by their children.

In the end, the conclusions of all these existing studies come down to the fact that running (large) fiscal deficits and having a large government debt-to-GDP ratio negatively impact the long-term interest rate. The only thing that differs across them is the way, or the magnitude of the impact. This is the case because different methods, timeframes and countries are used. However, the magnitude of the impact of these fiscal indicators on long-term interest rates also depends largely on country specific factors.

3. Japan

3.1 Debt comparison with other countries

If the Japanese debt-to-GDP ratio is compared with the other countries of the G7 plus Greece and Ireland (which are known to have accumulated high debts during the crisis of 2008) it can be seen that Japan has experienced excessive growth in their stock of debt. It has been rising since the beginning of the 1990s and has been growing ever since, with exception of 2006 and 2007. In figure 1, the debt-to-GDP ratios of the G7 countries plus Greece and Ireland are shown. According to economic theory, high debt-to-GDP ratios usually result in a higher long-term interest rate. High debt can raise concerns about the government’s ability to service its debt, which in turn raises credit risk premiums

(8)

8

Also, with the government running deficits, decreases in national savings mean that there is an excess supply of government debt. So the government issues more bonds to finance the deficits (Baldacci & Kumar, 2010). This excess supply causes an increase in long-term interest rates. In Greece, Ireland and Italy the long-term interest rates increased indeed, because of their increasing debt and raising concerns about its solvencies.

Figure 1. Gross debt-to-GDP ratios of G7, Greece and Ireland

Data source: OECD, Economic outlook 95 database

3.2 Accumulation of Japanese debt

When it comes to the long-term interest rate, Japan is a special case. Since the 1990s Japan’s debt-to-GDP ratio has deteriorated sharply, as can be seen in figure 2. The net debt-to-debt-to-GDP ratio has

increased from 10.9% in 1991 to 137.5% in 2013, whereas the gross debt-to-GDP ratio has increased from 74.6% to 224.6% in the same timeframe. Ito (2011) says that in both measures the government would become insolvent. However, it is better to look into the gross debt-to-GDP ratio rather than the net debt-to-GDP ratio, since some of the government´s financial assets are held with non-Japanese government bond liabilities being behind it.

0,0% 50,0% 100,0% 150,0% 200,0% 250,0% 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 20 12 Canada France Germany Greece Ireland Italy Japan U.K. U.S.

(9)

9 Figure 2. Japanese Debt-to-GDP ratio

Data source: IMF, World Economic Outlook Database, April 2014

The Japanese debt grew so high because of their high spending. In the 1980s, the Japanese stock and asset markets were booming. Due to these booming markets, Japanese household and corporate savings increased and after years of budget deficits, the government finally ran budget surpluses. During these years, the domestic savings were in a large surplus. But at the beginning of the 1990s the asset price bubble burst and tax revenues decreased. The government increased spending to overcome the recession while their economic growth stayed behind. As can be seen in figure 3, the result was that the government was running deficits again and their debt-to-GDP ratio grew (Doi, Hoshi & Okimoto, 2011). However, these deficits could be offset by the high domestic savings in the years before. Household and corporate savings were invested in Japanese government bonds through the banking system and thanks to this; the Japanese government was able to finance their deficits. In 1997, the consumption tax rate increased from 3% to 5%, in order to decrease their fiscal deficits. However, three months later the Asian currency crisis started and four months after that the Japanese banking crisis also started. Before the bubble burst, banks lent cheap and easy loans to the corporate sector. After the bubble burst part of these debts could not be repaid, leaving the banks overly leveraged. The government tried to bail out these banks by giving capital infusions. These ongoing capital infusions made the Japanese government to run large deficits. From 2003 to 2007, a mild fiscal consolidation took place, due to economic recovery. But as of 2008, the fiscal balance weakened again, and deficits became even larger. This was mostly caused by the global financial crisis. Just like the rest of the world, Japan had to apply large fiscal stimuli to slow down the impact of the crisis. In 2010, the amount of tax revenues only covered less than half of the total expenditures. A large amount of new debt had to be issued, which was even larger than the amount of tax revenues. This worsened the fiscal situation in Japan even more (Ito, 2011). Due to these high fiscal deficits and worsening debt-to-GDP ratio, the 1990s and 2000s were labeled as “the lost two decades”.

0% 50% 100% 150% 200% 250% 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 20 12 Net Gross

(10)

10 Figure 3. Japanese budget deficit and government debt

Data source: OECD, Economic outlook 95 database

However, despite the large fiscal deficits and growing government debt, the long-term interest rates stayed low. Between 1989 and 1991, the 10-year nominal bond yield increased from 4.85% to 6.76%, whereas the debt-to-GDP ratio decreased slightly in this period. As from 1992 the debt kept

increasing. Nevertheless, after 1991, the government bond yield decreased. While the debt ratio increased to 225% in 2013, the yield came down to 0.69%. Even credit rating downgrades by

Standard & Poor’s (in 2011, from AA to AA-), Moody’s (also 2011, from Aa2 to Aa3) and Fitch (in 2012, from AA for foreign currency and AA- for local currency to A+) have not caused the long-term interest rate to increase.

Figure 4. Japanese debt-to-GDP ratio and 10-year nominal government bond yield

Data source: OECD, Economic outlook 95 database.

-12,00% -10,00% -8,00% -6,00% -4,00% -2,00% 0,00% 2,00% 4,00% 0,00% 50,00% 100,00% 150,00% 200,00% 250,00% 19 79 19 82 19 85 19 88 19 91 19 94 19 97 20 00 20 03 20 06 20 09 20 12 Debt Deficit 0,00% 1,00% 2,00% 3,00% 4,00% 5,00% 6,00% 7,00% 8,00% 0% 50% 100% 150% 200% 250% 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 20 12 Debt Yield

(11)

11

3.3 Causes for the low long-term interest rates

Figure 4 shows that the situation in Japan is not as would be expected according to the literature. As said, high levels of government debt usually result in high long-term interest rates due to different factors. However, this has not been the case in Japan. After the asset price bubble burst, the long-term interest rate decreased to below 2% in 1998, after which it stayed steadily under the 2% and even below 1% at some points. It seems that investors do not demand any risk premiums on Japanese government bonds, but that they can live with the high stock of debt. However, according to Ito (2011), even with these optimistic investors, an insolvency of the Japanese government cannot be avoided. Several Japan-specific factors have been pointed out which are probably important determinants for these low interest rates.

According to Hoshi & Ito (2014), an important factor is the high private savings together with the home bias that Japanese investors and institutions have. In the 1970s, the Japanese households had large saving surpluses. These household surpluses were financing the corporate and government borrowings, although the government was running only small deficits. In the 1980s and 1990s, however, the household savings declined, but in the 1990s the corporate saving surpluses were offsetting the declined household savings so that the private savings remained high. Because of these high private savings, the Japanese government did not have to borrow from abroad. The government budget deficits could be easily offset by the private savings, so the domestic savings (the sum of public and private savings) were always in a large surplus (Ito, 2011). However, there is a problem with the household savings. As the corporate savings were high, and kept the private savings up, the household saving rate declined since the early 1980s (with an exception for the bubble years). This could be a problem, since the Japanese population is aging rapidly. According to the life-cycle model, households smooth the consumption of their income over their entire lifetime. This means that they will save during their working years so that they can spend their savings after they are retired and suggests that when the ratio of retired population to the working-age population increases, the aggregate savings will decrease.

Official projections of the IPSS (Japan’s National Institute of Population and Social Security Research) indicate that the group of working-age population (15-64) will decrease from 63,8% in 2010 to 50,9% in 2060 while the group of old-age population (65+) will increase from 23% in 2010 to 39,9% in 2060. This means that the ratio of retired population to the working-age population will increase sharply, and aggregate household savings will decline. Then there also is the old-age dependency ratio, which is a ratio obtained by dividing the old-age population by the working-age population. This ratio is projected to increase from 36,1% in 2010 to 78,4% in 2060.

(12)

12

This means that in 2060 the group of old-age people will be as big as 78,4% of the working-age people, so there will be many more elderly households. Like the growing old-age population, this will also have a big influence on the saving rate. Most of the studies, examining the causes of the decline in the household saving rate, link the decline to the rapidly aging population (Walker, 2005)

Because the corporate savings were still high and the government did not have to borrow from abroad, most of the Japanese debt is held domestically. In 2013, only 4% of the outstanding government debt is held abroad, whereas almost 65% is held by private financial institutions. Table 1 shows the amount of outstanding government bonds, broken down by holders, in percentage of the total amount from 2009 to 2013. As can be seen, in 2013, the largest share (almost 65%) is held by private institutions and only 2,5% is held directly by Japanese households. However, just 4,1% of the total amount outstanding is held abroad which means that almost 96% is held by Japanese residents. According to Ito (2011), this high share of domestically held bonds means that the perspective of foreign investors does not really affect the yield on Japanese government bonds. Any capital in- or outflows or exchange rate concerns do not affect the Japanese government bond market.

Table 1. Government bond holders

2009 2010 2011 2012 2013 Bank of Japan 7,51% 8,30% 9,51% 11,62% 18,66% Private institutions 70,73% 71,38% 71,45% 70,80% 64,80% - Depository corp. 40,44% 41,54% 40,54% 38,97% 34,20% - Insurance funds 26,03% 25,59% 26,60% 27,50% 26,63% - Other 4,26% 4,25% 4,31% 4,33% 3,97% Nonfinancial corp. 0,98% 1,19% 1,53% 0,90% 1,26% Government 11,57% 10,31% 9,38% 8,83% 8,25% - Central _0,04% _0,04% _0,04% _0,03% _0,03% - Local _0,10% _0,10% _0,10% _0,09% _0,09% - Social security 11,42% 10,18% 9,25% 8,71% 8,13% Private nonprofit institutions 0,44% 0,44% 0,44% 0,44% 0,44% Households 5,05% 4,29% 3,63% 3,00% 2,50% Overseas 3,73% 4,09% 4,05% 4,41% 4,10% Source: Bank of Japan, Flow of funds statistics

However, these high private savings do not necessarily have to be invested in Japanese government bonds since these give a very low return. The savings could easily be invested in higher yielding bonds, for instance other government bonds, even with a lower risk.

(13)

13

The fact that they are invested in domestic government bonds has to do with the home bias, mostly of the Japanese institutional investors. As can be seen in table 1, a very small amount is held directly by households and nonfinancial corporations. Private savings are mostly deposited into the banks, which in turn buy Japanese government bonds with these deposits. Hoshi and Ito (2014) give some

explanations for this home bias. They say that the most important reason for this home bias is the aversion to currency risk. Since the currency risk on foreign bonds is historically high and investment in domestic government bonds does not imply currency risk, the banks and other financial institutions may find these bonds more attractive. Also, the capital adequacy requirements of Basel I, II and III make Japanese government bonds (and sovereign debts of other advanced countries) attractive for banks and other institutions. The capital requirement is expressed as the ratio of equity that must be held as a percentage of risk-weighted assets. These requirements were imposed to increase liquidity and decrease the leverage to ensure that banks and other financial institutions do not take on too much leverage and would become insolvent. Since government bonds are assigned zero-weight in

calculating the risk-weighted assets, which determine the amount of capital that must be held by the banks, they do not increase the amount of capital that must be held. Walker (2005) has studied the home bias of Japan and used the international capital asset pricing model as a reference. In the absence of a home bias, the mix of assets held by domestic investors will be equal to the mix of assets held by foreign investors. So the share of domestic assets in each portfolio will be equal to the share of the domestic market in the world market. However, empirical studies in the 1980s and 1990s have showed that this is far from true. Although a reduction in the degree of home bias is found more recently, it seems that investors prefer domestic assets over foreign assets. Walker also says that the aversion to currency risk, if it cannot be fully hedged, is an important factor causing the home bias. Another possible factor for the low yield is the low VAT (value-added tax), or consumption tax. The VAT was introduced in Japan at a level of 3%. Just before the beginning of the Asian crisis and the Japanese banking crisis in 1997, this tax rate was increased to 5% under the fiscal consolidation package. In 2012, a bill to double the tax rate passed the lower house of the Japanese National Diet. This bill increased the tax rate in April 2014 to 8% and will increase it to 10% by October 2015. However, this is still far below the level in most of the European countries, where they have VAT rates of more than 15% and some even 25%. Ito (2011) has estimated that the 44 trillion yen deficit of 2010 could be brought down to zero if the VAT was raised to 20%. Apparently investors do not demand extra risk premium since they believe that the government is able to cut the deficits and perhaps even reduce the government debt by increasing the VAT.

The final mentioned reason for the low long-term interest rate is the low inflation, and even deflation in some years as can be seen in figure 5. These low inflation rates made that the real yield was not far below the nominal yield, where the deflation even made the real yield higher than the nominal yield.

(14)

14 Figure 5. Inflation (consumer prices)

Source: OECD, Economic outlook 95 database

4. Data

This section describes the data in the model and explains where the data were obtained from. In order to find which factor is the most important one in the determination of the 10-year nominal bond yield, annual data between 1979 and 2012 has been used. This timeframe is chosen on the basis of the availability of the data. The variables are the same as in the model of Baldacci and Kumar (2010), who have used time series panel data to estimate the impact of fiscal deficits and public debt on long-term interest rates. However, since the debt and deficits do not explain the low yield on Japanese government bonds, two important variables for Japan are added in this regression.

The dependent variable is the 10-year nominal bond yield (yield). The data of the dependent variable were found on the OECD database.

The main independent variables are the level of gross general government debt in percent of GDP (debt), the squared level of gross general government debt in percent of GDP (debtsq), the fiscal balance in percentage of GDP (deficit), which are obtained from the OECD database, and the CPI inflation rate (inflation), which is obtained from the World Bank database. The added Japanese factors are the percentage of the Japanese government bonds held overseas (forinv), which is obtained from the flow of funds statistics of the Bank of Japan, the gross domestic savings (domsav) which is obtained from the World Bank database and the exchange rate in Japanese Yen per US Dollar (exchrate) to capture the home bias which is also obtained from the OECD database.

-2,0% -1,0% 0,0% 1,0% 2,0% 3,0% 4,0% 5,0% 6,0% 7,0% 8,0% 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 20 12 Inflation

(15)

15

There are also two control variables. The first one is the short-term nominal interest rate (shortterm) to control for the effects of monetary policy on the term structure and the second one is the output growth (growth) to control for Japan’s cyclical position. Both the variables were obtained from the OECD database. The timeframe 1980-2012 has been chosen since the data from the Bank of Japan were only available as from 1980, and for the World Bank data and debt and deficit only annual data were available.

The squared level of gross general debt is added since Ardagna, Caselli and Lane (2007) have found evidence for nonlinearities. They found that the relation between long-term interest rates and the stock of public debt depends on the level of debt and clarify this by saying that when the stock of debt is low and a government issues new debt, investors might switch into them from bad quality debt, which is less likely to be repaid, since they see this new issued debt as better quality. So the price goes up en yield decreases. However if the debt rises above a certain threshold, further increases are associated with higher interest rates.

The percentage of foreign debt holders and gross domestic savings are added since these are the main Japan specific factors explaining the low yield according to the theory, together with the home bias. There is not a clear measure to capture the home bias, but since one of the main factors causing this home bias is the currency risk, the exchange rate will be added to the model as an indirect measure of home bias.

5. Model

5.1 Estimated model

To set up this time series model, several tests had to done. First, the stationarity properties of all variables were examined using the DF-GLS test for unit roots. Since the null-hypotheses (the variables contain unit roots) could not be rejected (which can be seen in the appendix, table 1), the result of this was that the level of the time series was non-stationary. Thus, the next step is to determine whether the chosen variables are cointegrated or not. This requires several steps. The first thing to do was to determine the amount of lags in the underlying VAR. Stata recommended the use of 7 lags in the VAR representation of the model, as can be seen in the appendix in table 2.

Now that the amount of lags is known, the cointegrating rank of the VECM can be estimated using Johansen tests for cointegration. With 7 lags, the estimated cointegrating rank of the VECM is at least 10. In table 3 of the appendix, the cointegrating ranks are shown. Since the null hypothesis that there are 9 or less cointegrating ranks can still be rejected, there have to be at least 10. A problem emerged when estimating the VECM model.

(16)

16

The VECM model cannot be run with 10 cointegrating equations, so the used cointegrating rank is 9. However, the model could also not be clearly estimated using 9 cointegrating equations, probably due to the few amount of observations.

The VECM model cannot be estimated, but since the variables are cointegrated (there is at least one cointegrating equation since the null hypothesis that there are nine or less cointegrating equation can be rejected), they can be estimated using the OLS method.

The estimated model is:

In this model, the dependent variable (the 10-year nominal yield) is being explained by the explanatory variables mentioned above, using OLS estimates.

5.2 Pre-estimation tests

To obtain information about whether the residuals were normally distributed, homo- or

heteroskedastic or autocorrelated, several tests were done of which the results can be found in the appendix in respectively table 4, 5 and 6.

First the residuals have been tested for normality, using the Skewness/Kurtosis tests for normality. The adjusted chi-squared statistic of 3.33 that came out, gives a p-value of 0.1889 as can be seen in the appendix, table 4. With this probability, the null hypothesis that the residuals are normally distributed cannot be rejected, so the assumption that the residuals are normally distributed can be made.

After the normality test, the Breusch-Pagan / Cook-Weisberg test for heteroskedasticity was run to see if the residuals are homoskedastic or not. The chi-squared statistic that came out was 5.78 and the matching p-value was 0.0162. The result of this test can be found in table 5 of the appendix. Since the null hypothesis states that the variances of the residuals are equal among observations and the chi-squared statistic of 5.78 rejects this hypothesis, there is enough evidence to say that the variances of the residuals are heteroskedastic. Therefore the OLS regression with robust standard errors is used.

(17)

17

Third, there has been tested for autocorrelation and partial autocorrelation in previous lags using the Ljung-Box Q statistic. The null hypothesis for this test is that the residuals in the model have no autocorrelation. Since the found p-values are all above 0.05, there is not enough evidence to reject the null hypothesis. These results can be found in the appendix, table 6.

After these tests the regression can be done and the estimates can be estimated with the OLS method using robust standard errors.

6. Results and analysis

The results of the OLS regression are given in table 3.

Table 3. OLS estimates

As can be seen in the table, the estimates of most of the coefficients on the variables of interest, namely: debt, squared level of debt, deficit, gross domestic savings and exchange rate enter the model significantly at the 5% level. The coefficients on the other variables of interest, the inflation rate and the share of foreign investors, are not significant. Most of the variables enter the model with the expected sign.

Surprisingly, however, the coefficient on debt is negative. This suggests that an increase of one percentage of GDP in the debt level results in a decrease in the nominal yield of 10 percent. However, negative effect on the yield is partly offset by the positive coefficient on the squared level of debt. This means that if the initial debt level is higher, the effect of the squared level of debt will reduce the negative effect of debt.

_cons .226338 .0471837 4.80 0.000 .1289556 .3237204 exchrate .0000726 .0000326 2.22 0.036 5.25e-06 .0001399 domsav -.4161063 .1276008 -3.26 0.003 -.6794614 -.1527512 forinv .0756517 .0649217 1.17 0.255 -.0583401 .2096434 shortterm .3199908 .0781493 4.09 0.000 .1586986 .4812831 inflation -.0789182 .0755147 -1.05 0.306 -.2347729 .0769364 growth .056248 .0232504 2.42 0.023 .0082615 .1042346 deficit .1044533 .0314843 3.32 0.003 .0394729 .1694336 debtsq .0208036 .0052377 3.97 0.001 .0099935 .0316137 debt -.1084029 .0154461 -7.02 0.000 -.1402821 -.0765237 yield Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00318 R-squared = 0.9899 Prob > F = 0.0000 F( 9, 24) = 247.16 Linear regression Number of obs = 34

(18)

18

This seem to correspond with the findings of Ardagna et al. (2007), who suggest that if there is a low level of debt, investors might seem to find this level of debt more attractive than newly issued debt of countries which already have a high amount of debt outstanding. This appetite for the newly, more attractive debt increases demand and decreases the yield. If the initial debt is already at a high level, the opposite will occur and thus the yield increases.

The coefficient on the fiscal deficits has its expected sign. Fiscal deficits are usually financed by a tax increase or the issuance of new debt. If the government issues new debt, there will be an excess supply of debt and the government has to promise higher returns, so the interest rate on these bonds increases.

The domestic savings enter the model as expected, namely negative and significantly. As explained in the literature review, the public deficit could mostly be offset by the high private savings. Because of these high private savings the total domestic savings were high in the surplus, which could then finance the deficits in the years following. When the private savings are higher, the government can borrow easier to finance their deficits. Therefore the gross domestic savings were expected to be largely negative and significant.

Also, the coefficient on the exchange rate enter the models positively and significant and positive. This means that a depreciation of the Yen causes the yield to increase. A depreciation means that the Yen loses value in comparison with the U.S. Dollar and the interpretation in this model is that a depreciation increases currency risk. Since the home bias of Japanese investors is mainly based on the aversion of currency risk, a depreciation of the Yen implies that there will be less invested in Japanese government bonds. Therefore, this coefficient is as expected. Higher currency risk (a weaker Yen) increases the yield on their long-term government bonds.

The coefficients on the inflation rate and the share of foreign bondholders, however, are not significant. This can possibly be due to the small amount of observations in the model, and the fact that this share has been quite stable over the years. Therefore the model does not see a relation between the yield and the inflation rate and the share of foreign bondholders, even though theoretically there is a relationship.

(19)

19

7. Discussion

The results of this model may not be optimal. Since the model was not stationary at level, as the variables contained unit roots, at first an OLS regression did not seem to be optimal. At the first difference level though, the model was stationary since the first differences did not contain unit roots. Therefore, the best model to be used for this time series was the VECM model. However, this model could not be well estimated due to the small amount of observations. Since there were only 34 observations in the model, Stata could not estimate the coefficients. In the estimation of the VECM model though, there seemed to be cointegrating equations. Therefore an OLS regression is done and these estimates were used.

Most of the coefficients on the variables had the expected sign and were significant. Only the sign of the coefficient on debt was negative, were a positive sign was expected according to the literature. However, the negative effect of the debt on the yield in this model is partly offset by the positive coefficient on the squared level of the debt.

During the OLS regression, robust standard errors were used since the Breusch-Pagan / Cook-Weisberg test for heteroskedasticity showed that the residuals were heteroskedastic. Unlike as one would expect for such a small amount of observations, the standard errors were small. This can be due to the use of OLS while having unit roots. These underestimated standard errors result in high t-values and a high squared value. As a result, the model may give overestimated coefficients. The high R-squared value can also be due to the presence of unit roots and therefore may not be a right measure to see if the model fits well.

Since the used model is not optimal for the chosen timeframe, a larger timeframe or the use of quarterly data (if available) would be recommended for future research. With a larger set of observations, the VECM model can probably be estimated correctly and the estimated coefficients would be more reliable.

(20)

20

8. Conclusion

Earlier studies on the effect of both government debt and fiscal deficits on the long-term interest rates have shown that increases in both fiscal measures result in long-term interest rates through several channels. The most important channels are: national savings, aggregate demand, inflation

expectations and the short-term interest rate. Japan’s debt-to-GDP ratio increased from 44% in 1979 to above 215% in 2012 and the Japanese government has been running deficits for years However, the long-term interest rate, and thus the nominal yield on their 10-year government bonds, has decreased from almost 7% in 1985 to below 1% in 2012.

Several reasons have been pointed out for the low yield. The high private savings Japan had were offsetting the deficits that the Japanese government had to run to reduce the impact of different crises Japan has suffered. These private savings were invested in Japanese government bonds to finance the high debt that the Japanese government has built up. The main reason that these private savings were invested in domestic government bonds is the home bias of Japanese institutions and investors. The most important reasons for this home bias are the aversion to currency risk and the capital adequacy requirements of Basel I, II and III. Another important factor driving down the yield according to the literature is the high share of domestic bondholders. Since only 4% of the Japanese government bonds is held abroad, and the perspective of foreign investors does not really affect the yield on Japanese government bonds. Two more, less important factors, are the low VAT and the low inflation rate (in some years even deflation).

A model is set up in which the nominal yield on 10-year government bonds is explained by the debt-to-GDP ratio, the squared level of the debt-debt-to-GDP ratio, the fiscal deficit, the inflation rate (CPI), the share of foreign investors, the gross domestic saving rate and the exchange rate (as a measure for home bias). This model was estimated using OLS-estimates. The result showed that, in this model, the debt-to-GDP ratio negatively affects the yield. However, this negative effect is partly offset by the positive effect of the squared level of the debt. As expected, the gross domestic savings rate shows a negative effect on the yield. As gross domestic savings increase, more of the issued debt can be offset by private savings. Therefore, the government does not have to borrow from abroad but obtains cheaper capital from domestic investors, which drives down the yield on their bonds. Also, the exchange rate, as an indirect measure of home bias, enters the model positive and significantly. This can be interpreted as such: a depreciation of the Yen increases currency risk. Since the currency risk is an important factor of the home bias, a depreciation implies that less Japanese institutions and

investors will invest in Japanese government bonds and the Japanese government has to borrow from abroad. Borrowing from abroad is more expensive since foreign investors might demand a risk premium. Therefore a depreciation increases the yield on long-term government bonds.

(21)

21

However, the coefficients on the inflation rate and the share of foreign investors are not significant. This can be due to the small amount of observations, since there is a theoretical relationship between both these variables and the yield.

(22)

22

9. References

Ardagna, S., Caselli, F., & Lane, T. (2007). Fiscal discipline and the cost of public debt service: some estimates for OECD countries. The BE Journal of Macroeconomics, 7(1).

Baldacci, E., Gupta, S., & Mati, A. (2011). Political and fiscal risk determinants of sovereign spreads in emerging markets. Review of Development Economics,15(2), 251-263.

Blanchard, O. J. (1991, January). Current and anticipated deficits, interest rates and economic activity. In International Volatility and Economic Growth: The First Ten Years of The International

Seminar on Macroeconomics (pp. 361-390). Elsevier Science Publishers BV, 1991.

Buiter, W. H. (1988). Death, Population Growth, Productivity Growth and Debt Neutrality. Doi, T., Hoshi, T., & Okimoto, T. (2011). Japanese government debt and sustainability of fiscal

policy. Journal of the Japanese and international economies, 25(4), 414-433. Elmendorf, D. W., & Gregory Mankiw, N. (1999). Government debt. Handbook of

macroeconomics, 1, 1615-1669.

Engen, E. M., & Hubbard, R. G. (2005). Federal government debt and interest rates. In NBER

Macroeconomics Annual 2004, Volume 19 (pp. 83-160). MIT Press.

Gale, W. G., & Orszag, P. R. (2003). The economic effects of long-term fiscal discipline. Hauner, D., & Kumar, M. (2006). Fiscal Policy and Interest Rates: How Sustainable is the 'New

Economy'?.

Ito, T. (2010). Sustanability of Japanese sovereign debt. Assessment on the Impact of Stimulus,

Fiscal Transparency and Fiscal Risk. ERIA Research Project Report, 1, 29-76.

Kumar, M. S., & Baldacci, E. (2010). Fiscal deficits, public debt, and sovereign bond yields. International Monetary Fund.

Mankiw, N. G. (2000). The savers-spenders theory of fiscal policy (No. w7571). National bureau of economic research.

Rebelein, R. P. (2006). Strategic Behavior, Debt Neutrality, and Crowding Out.Public Finance

Review, 34(2), 148-172.

Tokuoka, K. (2010). The Outlook for Financing Japan’s Public Debt (No. 10-19). International Monetary Fund.

(23)

23

10. Appendix

Table 1. Unit root tests

1 -1.977 -3.770 -3.400 -3.058 2 -1.917 -3.770 -3.305 -2.970 3 -1.712 -3.770 -3.195 -2.866 4 -1.475 -3.770 -3.080 -2.754 5 -1.422 -3.770 -2.971 -2.643 6 -1.489 -3.770 -2.879 -2.542 7 -1.338 -3.770 -2.814 -2.460 8 -1.039 -3.770 -2.788 -2.406 9 -1.070 -3.770 -2.811 -2.388 [lags] Test Statistic Value Value Value DF-GLS tau 1% Critical 5% Critical 10% Critical Maxlag = 9 chosen by Schwert criterion

DF-GLS for yield Number of obs = 24

DF-GLS for debt Number of obs = 24

(24)

24 1 -1.808 -3.770 -3.400 -3.058 2 -2.311 -3.770 -3.305 -2.970 3 -1.925 -3.770 -3.195 -2.866 4 -1.791 -3.770 -3.080 -2.754 5 -2.367 -3.770 -2.971 -2.643 6 -2.060 -3.770 -2.879 -2.542 7 -2.276 -3.770 -2.814 -2.460 8 -1.528 -3.770 -2.788 -2.406 9 -1.254 -3.770 -2.811 -2.388 [lags] Test Statistic Value Value Value DF-GLS tau 1% Critical 5% Critical 10% Critical Maxlag = 9 chosen by Schwert criterion

DF-GLS for deficit Number of obs = 24

DF-GLS for growth Number of obs = 24

(25)

25 1 -2.568 -3.770 -3.400 -3.058 2 -1.461 -3.770 -3.305 -2.970 3 -1.716 -3.770 -3.195 -2.866 4 -1.204 -3.770 -3.080 -2.754 5 -1.649 -3.770 -2.971 -2.643 6 -1.200 -3.770 -2.879 -2.542 7 -1.392 -3.770 -2.814 -2.460 8 -1.067 -3.770 -2.788 -2.406 9 -1.379 -3.770 -2.811 -2.388 [lags] Test Statistic Value Value Value DF-GLS tau 1% Critical 5% Critical 10% Critical Maxlag = 9 chosen by Schwert criterion

DF-GLS for shortterm Number of obs = 24

DF-GLS for forinv Number of obs = 24

(26)

26

Table 2. Lags in underlying VAR model

DF-GLS for exchrate Number of obs = 24

Exogenous: _cons

domsav exchrate

Endogenous: yield debt debtsq deficit growth inflation shortterm forinv 8 8155.11 -376.56 100 . . -607.316 -603.693 -594.735 7 8343.39 134.7* 100 0.012 . -621.799* -618.176* -609.218* 6 8276.04 327.97 100 0.000 . -616.618 -612.995 -604.037 5 8112.05 -103.76 100 . . -604.004 -600.381 -591.423 4 8163.93 -18.165 100 . . -607.995 -604.372 -595.414 3 8173.02 . 100 . . -608.694 -605.071 -596.113 2 . . 100 . -2.e-108* . . . 1 799.763 557.2 100 0.000 7.6e-36 -53.0587 -51.526 -47.736 0 521.163 4.0e-30 -39.3203 -39.1809 -38.8364 lag LL LR df p FPE AIC HQIC SBIC Sample: 1987 - 2012 Number of obs = 26 Selection-order criteria

(27)

27

Table 3. Cointegrating ranks

Table 4. Skewness/Kurtosis tests for normality

Table 5. Breusch-Pagan / Cook-Weisberg test for heteroskedasticity

10 310 . 1.00000 9 309 . 1.00000 805.7268 3.76 8 306 . 1.00000 894.9368 14.07 7 301 . 1.00000 959.1630 20.97 6 294 . 1.00000 987.9472 27.07 5 285 . 1.00000 1039.0557 33.46 4 274 . 1.00000 . 39.37 3 261 . 1.00000 . 45.28 2 246 . 1.00000 . 51.42 1 229 . 1.00000 . 57.12 0 210 . . . 62.81 rank parms LL eigenvalue statistic value maximum max critical 5% 10 310 . 1.00000 9 309 . 1.00000 . 3.76 8 306 . 1.00000 . 15.41 7 301 . 1.00000 . 29.68 6 294 . 1.00000 . 47.21 5 285 . 1.00000 . 68.52 4 274 . 1.00000 . 94.15 3 261 . 1.00000 . 124.24 2 246 . 1.00000 . 156.00 1 229 . 1.00000 . 192.89 0 210 . . . 233.13 rank parms LL eigenvalue statistic value maximum trace critical 5%

Sample: 1982 - 2012 Lags = 3 Trend: constant Number of obs = 31 Johansen tests for cointegration

resid 34 0.1249 0.3986 3.33 0.1889 Variable Obs Pr(Skewness) Pr(Kurtosis) adj chi2(2) Prob>chi2 joint Skewness/Kurtosis tests for Normality

Prob > chi2 = 0.0162 chi2(1) = 5.78 Variables: resid

Ho: Constant variance

(28)

28

Table 6. Ljung-Box Q statistic for autocorrelation

15 -0.0528 0.0300 19.796 0.1799 14 0.1429 0.3193 19.616 0.1427 13 0.1724 0.1148 18.365 0.1441 12 0.0496 0.3401 16.634 0.1639 11 0.0737 0.1594 16.496 0.1237 10 0.1468 -0.1648 16.207 0.0939 9 0.1898 0.1518 15.109 0.0880 8 -0.2158 -0.2917 13.346 0.1005 7 -0.3786 -0.3117 11.154 0.1320 6 -0.1831 -0.2477 4.655 0.5888 5 -0.0209 -0.0938 3.1897 0.6708 4 0.0244 0.1209 3.1714 0.5296 3 -0.2321 -0.2002 3.1471 0.3695 2 -0.0818 -0.1196 1.0195 0.6006 1 0.1435 0.1545 .76353 0.3822 LAG AC PAC Q Prob>Q [Autocorrelation] [Partial Autocor] -1 0 1 -1 0 1

Japan : high debt, low yield