Short-term interest rates of government bonds and the budget
balance
University of Groningen Faculty of Economics and Business
Department of Economics, Econometrics and Finance MSc Business Administration with the specialization Finance
Author: Kirsten Kottkamp Student number: 2007339
Email: Kirstenkottkamp@hotmail.com
Supervisor: dr. J.O. Mierau
1
Short-term interest rates of government bonds and the budget
balance
Kirsten Kottkamp September 2012
Abstract
The relationship between budget deficits and interest rates of government bonds shows ambiguous results in the theory and empirical literature. The research of this relationship is relevant due to the new developments of the economic crisis in the Economic and Monetary Union. The sample contains twelve Economic and Monetary Union countries with data from 1990 to 2010. Using the Generalized Method of Moments no significant results were found, so there exists no relationship. When including only the countries and years with a high debt level the results show a positive significant relationship.
JEL classification: C23, E43, H62, H63
Keywords: Interest rate, Budget balance, Budget deficit, Generalized Method of Moments
I. Introduction
The relationship between budget deficits and interest rates of government bonds has been a topic of interest for many years and is again of interest due to the economic crisis. Countries like Ireland, Greece, Spain and Italy got problems with (re-)financing their debt. They sold their government bonds with high interest rates which led to more costs which increased their budget deficit. A high interest rate indicates a less “safe” investment which results in more careful investors concerning government bonds of those countries. This could lead to higher interest rates than before.
2
repaid in the future with taxes. To pay these future taxes, desired private saving will increase, this will not affect desired national saving (Barro, 1989). The interest rate does not increase because there is no change in desired national saving, the balance between desired national saving and investment demand is not affected. The Ricardian Equivalence theorem does not hold with finite lives and tax distortions. The Neo-classical theorem and the Keynesian theorem state that the relationship does exist, which has the effect that an increasing budget deficit raises the interest rate. The Neo-classical theorem states that people plan their consumption for their total life time, when a budget deficit occurs their consumption increases. An increase in consumption may lead to a decrease in savings, so to balance the capital markets the interest rates should increase to attract more savings (Bernheim, 1989). The Keynesian theorem state that such a relationship exists because restoring the equilibrium will not work with adjusting prices or wage reductions due to the existence of underemployment of resources (supply is larger than demand). The government should increase their spending to stimulate the demand, this increases the budget deficit (Heijdra, 2009). Which increases the government debt, so there will be more debt than capital, this results in a rise of the interest rate (Engen and Hubbard, 2004).
In previous empirical literature most studies are conducted with a sample of the United States, only a few added some European countries. Aisen and Hauner (2008) added a variety of 60 countries from emerging and advanced economies including the eleven of the twelve countries1
which in 2002 started to use the euro.
An interesting aspect of the short-term interest rates is its convergence when the third stage of the Economic and Monetary Union (EMU) in 1999 came closer (see appendix A figure A2). Those short-term interest rates remained fairly similar until in 2007 the first signs of a credit crisis showed. The convergence of the short-term interest rate might affect the relationship between the interest rate and the budget deficit. Before the third stage of EMU became a fact some new rules about the budget balance and debt-levels were introduced in the Maastricht Treaty. It is possible that in this specific region with the start of this monetary payment system and the new rules the relationship between budget deficits and interest rates is affected.
The empirical literature shows diverging results and state that varying theories are true. Researchers like Gale and Orszag (2004) and Aisen and Hauner (2008) found a positive and
1 The twelve countries are Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg,
3
significant effect. Although when Aisen and Hauner split their sample in advanced economy countries and emerging economy countries, the significance disappears for the advanced economy countries. Also Evans (1987a and 1987b) could not find a significant relationship. The inconclusiveness of the theories and the empirical literature about the relationship between budget deficits and interest rates drives me to investigate the relationship. The research question that fits is:
Do government budget balances influence short-term interest rates of government bonds? To find the answer to this research question a sample is constructed of the twelve EMU-countries with a time-period from 1992 to 2008. The short-term interest rate will be used, because Aisen and Hauner (2008) used term interest rates and in the third stage of EMU the short-term interest rate was similar for several countries. The methodology used to deshort-termine the relationship is the Generalized Method of Moments as developed by Arellano and Bond (1991). This methodology has the option to correct for some problems that may occur. The Generalized Method of Moments one-step estimator will be used, because of the high amount of instruments which are necessary to overcome the problem of endogeneity. Furthermore this method has the option to correct for small samples. The results of this research should show if the relationship between the budget deficit and the short-term interest rates exists. This result will be compared with the theories and the empirical literature.
The thesis will be structured as follows: the next section will present the literature review, which includes the theories and the empirical literature. Followed by the empirical model and the methodology in section ΙΙΙ. In section ΙV the data is presented with the descriptive statistics to get a feeling of the data. The results of the estimation can be found in section V. And in the final section VI states the conclusion. Further on appendices and the references could be found.
II. Literature review
4 A. Theory
The Ricardian Equivalence theorem states that changes in the budget deficit do not affect the interest rate. An increase in the budget deficit may occur by decreasing taxes and holding government spending constant. The budget deficit should be repaid by future taxes, which will increase desired private saving (it does not affect desired national saving) therefore the interest rate will not increase (Barro, 1989). Investment, consumption and output are not affected by a decrease in taxes either (Heijdra, 2009). The Ricardian Equivalence theorem does not always hold. The first problem is that it assumes that people do not die (infinite lives). Since people do not have infinite lives, they are unwilling to save for future higher taxes (Barro, 1989 and Heijdra, 2009). This is what Bernheim (1987) and Seater (1993) call Non-Ricardian people, they transfer the higher tax burden to the next generation. The other type of people they identified are the Ricardian people these people save money to pay future taxes and do not want to transfer these to the next generation (Bernheim, 1987 and Seater, 1993). The second problem is distortion of tax (Bernheim, 1987, Heijdra, 2009 and Seater, 1993). The income out of taxes depends on factors like income, wealth and spending (Bernheim, 1987). This affects labour supply decisions, a tax cut results in a higher income as a smaller amount of their income is used to pay taxes, so to earn the same amount of money they will need to work less hours (Heijdra, 2009). This change in labour supply may affect income and the level of net household wealth, when it affects consumption and investment the Ricardian equivalence no longer holds (Heijdra, 2009).
The Neo-classical theorem states that a change in the budget deficit influences the interest rate (Heijdra, 2009). The Neo-classical theorem assumes that people will plan their spending of their entire life cycle (Bernheim, 1989). An increase in the budget deficit will increase the consumption (spending) over their total life cycle. To repay the debt resulted from the deficit taxes should be raised now or in the future. By raising taxes in the future the debt could be repaid by a next generation (Bernheim, 1989). A tax cut now leads to an increase in consumption, which may lead to a decrease in savings when in the long-term the budget deficit is persistent. A decrease in savings leads to an unbalanced capital market which will be restored by an increase in interest rates which motivates people to start saving (Bernheim, 1989).
5
the Keynesian theorem it is assumed that the nonexistence of the equilibrium can only be restored by fiscal policy (Heijdra, 2004). This can be done by increasing government spending, which increases the budget deficit and thereby increases the interest rate (Engen and Hubbard, 2004).
So far real economic theories have been discussed while the monetary economic theory may suggest another relationship. Where the Keynesians believe that fiscal policy can restore the equilibrium the monetarists don’t (Heijdra, 2009). Monetarists believe that monetary policy has an influence on the economy, the problem is that policy makers recognize problems too late, thereby the actions taken are too late and might accentuate the business cycle fluctuations (Heijdra, 2009). The monetarists state that the budget deficit could be financed by money creation and that surpluses could lead to money retirement, so the quantity of money is adjusted (Friedman, 1948). A part of the change in de budget balance is the effect of changes in interest rates and changes in community assets (Friedman, 1948). So if changes in the quantity of money are made to cover changes in the budget balance, the interest rate is possibly one of the reasons why the budget balance has changed.
Due to these diverging statements of the theories on the relationship between budget deficits and interest rates on government bonds, no conclusion can be made at this point. The empirical literature could give more insight in which theory holds.
B. Empirical literature
The results of the empirical literature are diverging from positive and negative significant results to insignificant results. The results of the empirical literature and the methods used will be discussed below. Starting with the extended research of Gale and Orszag (2004) which includes a review of 66 studies. Followed by the studies of Aisen and Hauner (2008), Correia-Nunes and Stemitsiotis (1995) and Feldstein (1986), Laubach (2003), Evans (1987a), Evans (1987b), Kim and Lombra (1989), Cebula (1991), Plosser (1987), Quayes and Jamal (2007) and Seater (1993).
6
Orszag’s (2004) own research are positive and significant and show an increase in the interest rate by 40 to 70 basis points if the expected budget deficit increases by 1%. Gale and Orszag (2004) used the expected budget deficit, because current interest rates are forward looking but current budget deficits are not (Feldstein, 1986). This is confirmed by the review of the studies (Gale and Orszag, 2004) which shows positive and significant results when the expected budget deficit is used and varying results when current deficits are used.
The results of Aisen and Hauner (2008) show a positive significant effect of a 1% increase in the current deficit raises the interest rate by 44 basis points. The result of the advanced countries in the sample is negative and insignificant. The interest rate could be influenced by other factors like the mobility of capital and the financing method of the deficit. According to Aisen and Hauner (2008) the mobility of capital could be of influence on the interest rate. The change in the mobility of capital will be adjusted by the exchange rate which influences the interest rate. The exchange rate will influence the prices and subsequently output, consumption and the interest rate. Budget deficits could be domestic or foreign financed, when domestic financed it may influence the interest rate (Aisen and Hauner, 2008). Aisen and Hauner (2008) state that the degree of Ricardian Equivalence is low, when including the sample as a whole. The Ricardian Equivalence does hold, when including only the advanced economy countries.
Correia-Nunes and Stemitsiotis (1995) studied the relationship between long-term interest rates and high deficits. The result is positive and significant, a 1% increase in the budget deficit over one year results in an increased long-term interest rate by 21 to 79 basis points. According to these results the Ricardian Equivalence theorem does not hold. This can also be concluded from the research of Feldstein (1986). Feldstein (1986) proves that budget deficits influence exchange rates, therefore a measure of the exchange rate should be included to test this relationship.
The study of Laubach (2003) shows that with annual projected data, a 1% increase in the projected deficit results in a significant increase of the long-term interest rate by approximately 25 basis points. The increase in the government debt of 4 basis points as a result of a 1% increase of the projected deficits, is also significant. Laubach (2003) states that the results are consistent with the Neo-classical growth model, so the Ricardian Equivalence theorem does not hold.
short-7
term interest rates and quarterly data, show no significant results. In another study of Evans (1987a) with a sample containing the United States and monthly data, he also found no significant results.
Kim and Lombra (1989) state that their results are consistent with the Ricardian Equivalence theorem. The reason for their research is that high interest rates possibly have an effect on private investment, this would decrease private capital (or at least a smaller growth of private capital) and economic growth. According to Kim and Lombra (1989) the interest rate is only affected by unexpected government debt (this could be an increase or decrease in debt).
Cebula (1991) wrote a note on “federal budget deficits and the term structure of real interest rates in the United States”. His conclusion is that the classical theorem holds, the quarterly data of 1971 to 1985 gives a positive and significant result of budget deficits on interest rates. The article of Plosser (1987) is an extension of “Government financing decisions and asset returns” by Plosser (1982). And the results of the new article confirmed the previous results, that there is little or no association between interest rates and deficits.
Quayes and Jamal (2007) researched the influence of an increase in budget deficit on the interest rate for corporate bonds. They find that an increase in the budget deficit leads to a higher interest rates for corporate bonds, they do not mention the influence of interest rates on government bonds. This confirms the intuition behind their model that suggests that an increase in the government budget deficit is not exactly met by an increase in savings and therefore should increase the government debt level. This increase in government debt results in an excess supply of government bonds which should increase the interest rate on those bonds.
8 C. Differences in the empirical literature
In the empirical literature, not all studies used the same kind of data or frequency of the data, these factors and some others might explain the differences in the results. Correia-Nunes and Stemitsiotis (1995) discuss some of these factors. The first factor is the type of interest rate, this could be long-term or short-term. According to Correia-Nunes and Stemitsiotis (1995) long-term interest rates result in significant results and short-term interest rates result in insignificant results. When looking at the empirical literature it is still unclear if it holds, as Plosser (1982) and Aisen and Hauner (2008) prove against their statement. Some others (including Evans, 1987a and 1987b) show results that do seem to agree with the statement of Correia-Nunes and Stemitsiotis (1995). Aisen and Hauner (2008) and Plosser (1982) used short-term interest rates and had positive significant results, while Evans (1987a and 1987b) showed insignificant results with short term interest rates. The second factor is the frequency of the data, this differs among studies from monthly to annual data. Correia-Nunes and Stemitsiotis (1995) state that data with a high frequency (monthly and quarterly) could lead to insignificant results. In this case empirical literature shows (again) multiple different results. Bovenberg (1988), Kim and Lombra (1989) and Plosser (1982) used annual and semi-annual data, quarterly data and quarterly data, respectively, all found significant results. Evans (1987a and 1987b) used monthly and quarterly data and did not find significant results. Evans (1987a and 1987b) and Plosser (1982) used high frequency data and short term interest rates which could also be of influence on the results. These diverging results and the combination of the type of interest rate and the frequency of the data are possibly of influence on the results. Because with these combinations of factors it is hard to tell which factor is of influence on the result and therefore which factors I should take into account when evaluating my results. The third factor is the type of budget deficit. Feldstein (1986) states that expected budget deficits should be used, because current interest rates are forward looking and current budget deficits are not. The review of 66 studies by Gale and Orszag (2004) shows that using expected budget deficits leads to positive and significant results and using current deficits leads to varying results.
D. Other factors of interest
9
The main part of the examined empirical literature uses the United States as region for their research, except for Aisen and Hauner (2008), Correia-Nunes and Stemitsiotis (1995) and Evans (1987b) who expanded their region under investigation. The relatively new Economic and Monetary Union (EMU) should catch up as a region on the topic. And therefore will be included in the sample of this study
The government debt levels vary across countries and time, as can be seen in appendix A figure A3. If a country reaches a certain government debt level does that change the influence budget deficits have on interest rates? A certain government debt level could be a very high level, more than 75% of national GDP (gross domestic product), or a very low level, less than 25% of national GDP.
With the EMU as region of interest it might be that results differ before the third stage of EMU in 1999 and after Greece joined the EMU in 2001 (European Central Bank, 2004). With the start of the third stage the euro is introduced2 and the countries willing to join the EMU should comply with
the Maastricht Treaty. The Maastricht Treaty states that every country joining the EMU should have a government debt level of 60% or less of national GDP and the budget deficit should be 3% or less of national GDP (European Central Bank, 2004). All countries that join the EMU should stick to these rules also after they joined the EMU. The government debt levels of Belgium, Italy and Greece have never been below the 60%3 (see appendix A figure A3).
E. Conclusion on the theory
The three theories (Ricardian Equivalence, Neo-classical and Keynesian theorem) are inconclusive about the influence budget deficits have on interest rates of government bonds. The Ricardian Equivalence theorem states that there is no influence while the Neo-classical and Keynesian theorem state that the budget deficit does influence the interest rate. The empirical literature does not solve this problem, as shown in the paper of Gale and Orszag (2004), the review of 66 studies shows different results. So both the theoretical as the empirical literature show ambiguous results, this empirical research should clarify the actual relationship between budget deficits and interest rates.
10
III. Empirical model and methodology
Following the empirical literature an empirical model is developed in this section, first. The second part of this section introduces the methodology of how to test the empirical model.
A. Empirical model
The theories and the empirical literature of the literature review remain inconclusive about what the correct theory should be. Therefore it is practically impossible to state hypotheses, so an empirical model will be developed. The three theories, the Ricardian Equivalence, Neo-classical and Keynesian theory, use similar variables, so I expect to find a relationship between those variables. The main variables as discussed in the theory are the interest rate, the budget deficit, the government debt and the taxes. In the empirical literature some other variables are added. Aisen and Hauner (2008) added a variable for country risk, while Correia-Nunes and Stemitsiotis (1995) state that country risk can be measured by the stock of public debt. For this study I will use a country risk factor as used in Aisen and Hauner (2008). And I will add the variable government debt to confirm that government debt has no influence on the interest rate as stated by Gale and Orszag (2004). Aisen and Hauner (2008) also added a variable for broad money (M3) this variable will not be added in this study, because the region of interest is the twelve EMU-countries. When these countries joined the third stage of EMU their broad money should be identical or similar. Feldstein (1986) and Correia-Nunes and Stemitsiotis (1995) also added expected inflation as a variable. Expected inflation may reflect uncertainty (Feldstein, 1986) when a government is unable to control the inflation rate now it may not be able to control the inflation rate in the future. Uncertainty may influence the interest rate and is therefore incorporated as a variable. Another variable that will be used in the research is depreciation of a currency, because it may affect consumption, import and export of a country. Consumption will not be affected if the Ricardian Equivalence theorem holds, in that case there will be no effect of depreciation of a currency. Another reason to add expected depreciation as a variable is because in the absence of capital mobility the interest rate is influenced by expected depreciation and foreign interest rate as stated by Aisen and Hauner (2008). The foreign interest rate may influence the domestic interest rate if import and export is high. According to Correia-Nunes and Stemitsiotis (1995) the business cycle on investments and consumption of durable goods is measured by GDP growth.
11
Starting with the Ricardian Equivalence theorem, it states that an increase in the budget deficit does not affect the interest rate but it affects the government debt or the taxes (Heijdra, 2009). So an increasing interest rate cannot be explained by a budget deficit. According to the Ricardian Equivalence theorem the coefficient of the budget deficit should be zero and insignificant. This is identical to the result of Evans (1987b) who states that the Ricardian Equivalence theory holds. If the Neo-classical or Keynesian theory holds the budget deficit does have an influence on the interest rate and has a positive significant influence (in this research the budget balance is used instead of the budget deficit and this should result in a negative significant influence). This is shown by Aisen and Hauner (2008), Gale and Orszag (2004) and some others (see literature review).
In formula form the empirical model is as follows
(1)
Where is the independent variable, the nominal interest rate. For each variable with subscripts or or the subscripts indicate respectively the variable of country at time or country or time . The dependent variables are the lagged nominal interest rate, , the foreign interest rate, , the budget balance, , the government debt, , the tax burden, , the expected depreciation, , the expected inflation, , the GDP growth, and the risk-factor,
. Also two fixed effects are incorporated, country fixed effects, and time fixed effects, . And stands for the coefficients of the variables.
The significance of the variables is hard to predict due to the diverging results of the theories. If the Ricardian Equivalence theorem holds, the budget balance variable will be insignificant and the government debt and tax burden variables should be significant (one or both), because the change in the budget balance should be captured by the government debt or tax burden. When the Neo-classical theorem or Keynesian theorem holds, the budget balance variable should be negative and significant. A decrease in the budget balance indicates that the budget is moving to a deficit or is increasing the deficit, which increases the interest rate as has been seen in Gale and Orszag (2004).
12
level (below 25% of GDP) and their influence of the budget deficit on the interest rate. The expectation is that countries with a high debt level at a certain point in time show a significant negative result on the influence of the budget balance on the interest rate and countries with a low debt level show no significant result or a significant result but with a smaller coefficient.
B. Methodology
This research is started with Ordinary Least Squares (OLS) with fixed effects.4 OLS may not be
the perfect method, due to some arising problems. Problems with non-normality, autocorrelation and stochastic independent variables may occur. Non-normality in the sample is very likely, due to the high number of dependent variables. To test for the presence of autocorrelation a test is developed for autocorrelation by Baltagi and Wu (1999). This test shows that a correction for autocorrelation is necessary.5 Although there are no exact critical values specified in the literature,
it is assumed that the statistic is too high.6 Another problem that occurs when having a dynamic
panel is the Nickell-bias (Nickell, 1981). When applying fixed effects and using a lagged independent variable as dependent variable this dependent variable becomes endogenous, because the disturbance term has also cross-section and time effects. Due to the added lags of the dependent variables these are stochastic this might result in problems with OLS (Brooks, 2008), although this is only a minor issue when already having determined the Nickell-bias. The problems with non-normality, autocorrelation and stochastic independent variables can be overcome by using OLS with robust standard errors.
With economic variables endogeneity7 is present this might be the result of the omitted
variables. The exogenous variables are foreign nominal interest rate, time and country dummies. And the endogenous variables are expected depreciation, political country-specific risk spread, expected inflation, budget balance, GDP growth, central government debt and tax burden.
Since endogeneity is present in the sample OLS with robust standard errors no longer gives correct estimates. This problem could be overcome by using the least square dummy variable method. The command in Stata for this method is XTLSDVC which is a corrected least squares dummy variable (Bruno, 2004). The least square dummy variable is appropriate with a small N (Bruno, 2005). XTLSDVC calculates the corrected least squares dummy variable estimators for the
4 According to the Hausman test fixed effects should be used. 5 The Baltagi-Wu LBI statistic: 1.841.
6 As has been assumed by Kpodar (2007).
7 Which has also be concluded in previous studies by Aisen and Hauner (2008), Correia-Nunes and
13
standard autoregressive panel data model using the bias approximations (Bruno, 2004). The specifications used in Stata are as the estimator a one-step Arellano and Bond estimator with no intercept, the accuracy of the approximation is set at the highest level (3: { } { }
)and 3000 repetitions to calculate a bootstrap variance-covariance matrix. The results of this method could still lead to a bias, because there is no correction possible for the problems of non-normality, autocorrelation and the lag dependent variable in the sample. This disadvantage of XTLSDVC, not having the option to use robust standard errors, could be overcome by XTABOND2 as described by Roodman (2009). XTABOND2 is an approach based on the Generalized Method of Moments (GMM) as developed by Arellano and Bond (1991). It is developed for situations in which the sample has a small time period (T) and a large number of individuals (N); a linear functional relationship; a dynamic left hand side variable; not strictly exogenous independent variables; fixed individual effects and within individuals; heteroskedasticity and autocorrelation. The sample has a smaller N than T, so XTABOND2 could give downward biased standard errors due to the smaller sample (Windmeijer, 2005). To solve the problem the Windmeijer correction could be applied. For this research a one-step difference GMM is applied with the Windmeijer correction for small samples. With the options of robust standard errors and some endogenous and exogenous variables included. Of which the endogenous variables are expected depreciation, political country-specific risk spread, expected inflation, budget balance, GDP growth, central government debt and tax burden, the lags of these variables are used as instruments. The other variables are the exogenous variables.
So eventually the best method to test the empirical model is the XTABOND2 method (Roodman, 2009) with the small sample correction of Windmeijer (2005). The other methods are estimated as well for comparability and informative reasons.
IV. Data and descriptive statistics
In this section the data will be described. The content of the sample in means of time and cross-section and why it is chosen. Where the data is collected and what it represents. And a table with descriptive statistics will be presented.
14
2010. The sample is constructed by using DataStream, within DataStream different resources are used,8 and to determine political risk the International Country Risk Guide (ICRG) is consulted.
From 1990 to 1992 there are some missing values of the nominal interest rate also due to using the lag of the nominal interest rate. And after 2008 the political country-specific risk variable was not available. These missing values make of the sample an unbalanced panel. To prevent having problems with an unbalanced panel in the regression the time period from 1992 to 2008 will be used. This results in 204 observations for every regression, so the sample becomes a balanced panel. The lagged variables that will be used as variable or as instrument result in a dynamic panel.
In the following paragraphs the variables will be introduced, starting with the dependent variable followed by the independent variables. With the descriptive statistics in table Ι.
The nominal interest rate (abbreviated with NIR) is the dependent variable, this is the three months interbank rate. The National Central Banks store these statistics and DataStream provides this data. Appendix A figure A2 shows how this short-term rate has developed over time. From 1990 to about 1995 interest rates in most countries (except Greece) had fallen from 8% and higher to about 3%. From 2000 forwards the interest rate of the countries moved together at almost the exact same rate due to being in the EMU. The maximum of the nominal interest rate (17%) is an outlier compared with the average of 4.8% and the minimum of 2.1% and with a standard deviation of 2.828. The reason why this maximum is so far above the average could be because there are no regular negative interest rates. The countries with a low interest rate are safer investments than those with high interest rates, because the mean is relatively low compared to the maximum most countries in the sample should have a low interest rate. The scatter plot in Appendix A figure A10 shows that for the first ten years it is a linear, negative sloping and weak pattern, which indicates a negative correlation which is close to minus one. And for about the last ten years it is non-linear, fluctuating and has a strong pattern indicating a correlation close to zero.
The independent variables are budget balance, foreign nominal interest rate, expected depreciation, political country-specific risk, expected inflation, real GDP growth, central government debt and tax burden.
The budget balance is a measure of the budget deficit and budget surplus. The OECD provided the data of the cyclically adjusted budget balance, this is measured as a percentage of potential GDP
8 World Bank, International Monetary Fund, National Central Banks and Organisation for Economic
15
of a country and is adjusted for the influence of business cycles on the budget balance. According to the Maastricht Treaty a country is not allowed to have a budget deficit higher than 3% of GPD (European Central Bank, 2004). In appendix A figure A8 can be seen that around the introduction of the third stage (1999) all countries complied with this rule. Before 1996 the largest budget deficit was 6% of Germany. In 2006 and later budget deficits rose to 23% for Ireland in 2010. Other countries with risen budget deficits are Spain, Greece and Portugal with deficits of 8%, 8% and 6% respectively in 2009. The budget balance has an average surplus of 0.984% which lies between the minimum of -7% and the maximum of 7% which have been found in the sample. In appendix A figure A11 the scatter plot of the cyclically adjusted budget balance could be found. The scatter plot shows a non-linear, fluctuating and weak pattern, which could also indicate that there exists no relationship between the countries, indicating a low correlation which is close to zero.
The foreign nominal interest rate is the one-year United States Treasury bill rate, which is collected from the International Financial Statistics (IFS) of the International Monetary Fund (IMF). This variable is for every country identical. The foreign nominal interest rate fluctuates from 2000 and forwards more than the nominal interest rates of the EMU-countries as can be seen in appendix A figures A2 and A4. Although the foreign nominal interest rate fluctuates more, the mean (3.761%) of the foreign nominal interest rate is lower than the mean of the nominal interest rate (4.751%) as can be read from table Ι. The reason for this could be that the nominal interest rates (16%) at the beginning of the period (1990 and forwards) are far above the foreign nominal interest rates (8%) (see appendix A figures A2 and A4).
16
mean of 0.113% (as showed in table Ι). These deviations from the mean are more than two times the standard deviation (3.503), so the minimum and maximum could be labelled outliers. The expected depreciation takes also into account changes in price levels. Therefore the expected depreciation is not identical for all countries with the start of the third stage of EMU. As can be seen from appendix A figure A5 the minimum and maximum are in the time period before the third stage of EMU started. The minimum of -16.4% of Finland in 1993 could be explained with the great depression of Finland from 1990 to 1993 (Kauppi, Lassila and Teräsvirta, 1996). The other outlier the maximum is Italy’s expected depreciation in 1996. After the downturn of expected depreciation in 1992/1993, the expected depreciation rose in 1996 to 10.7%, the reason for this increase could be due to tagging the Lira to the German Mark and a devaluation of the Lira.
The political country-specific risk variable is a measure of the political risk in a country. This data is computed and collected by the International Country Risk Guide (ICRG). The risk is a weighted sum of the different types9 of risk that make the political country-specific risk together.
This risk spread ranged from a minimum of 69.329 to a maximum of 97.164 with a mean of 85.345 as can be seen from table Ι.
The expected inflation is a measure of price changes in goods and services households consume. It is an index with as basis year 2005 and the prices are calculated based on harmonized prices. Based on this index the change in each year is calculated (by using (index of year index of year )/ index of year ). The data is collected from the World Economic Outlook of the International Monetary Fund. Appendix A figure A6 shows that there is a difference in movement before and after the introduction of the euro. Before the introduction of the euro France and Portugal had in 1990 an expected inflation of 27% and 13% respectively. And for all countries the expected inflation moved towards 2% in 1999 in which the third stage of EMU started. In the third stage of EMU the European Central Bank (ECB) aims at influencing interest rates to keep price stability with an inflation rate of 2% or lower. With the start of the third stage almost all countries remained their inflation rate between 0% and 5%. The mean expected inflation rate is 2.759% which is above the stated 2%, but this includes a period before the third stage of EMU. The maximum outlier is 15.9% of France in 1992 (see table Ι) is a downward going trend to 1999. In 1999 the countries which wanted to join the third stage of EMU had to comply to the goal of 2%
9 These different types of risk and their weights in parentheses: government stability (12); socioeconomic
17
inflation, which was reached by France. France had already a high inflation, in the 1980s it started to reduce its inflation to the inflation level of Germany.
Real GDP growth is a measure of the year on year growth of GDP in percentages. The source of this data is the World Bank. Appendix A figure A7 shows that in 1991 the GDP growth of Finland falls below the main stream of GDP growth, but recovers totally in 1993. Furthermore from 1993 Ireland’s GDP growth is higher than the main stream and drops back to the main stream in 2001, but then still is in the upper stream. As can be read from table Ι the mean GDP growth is 2.793% which lies closer to the minimum of -6% than to the maximum of 11.497%.
Appendix A figure A3 states the Central government debt of the sample countries. The central government debt is collected from the OECD and is in percentage of GDP. The figure shows that Belgium, Greece and Italy do not comply with the rule of the Maastricht Treaty that government debt is not allowed to be higher than 60% of GDP (European Central Bank, 2004). They did not comply before and after the start of the third stage of EMU. The mean (57.05%) is below the 60% of GDP set by the Maastricht Treaty, but the maximum is well beyond the 60% with 118% government debt of GDP.
18
Table Ι
Descriptive statistics
In this table the descriptive statistics of the whole sample are shown. The nominal interest rate is the short-term (3-month) interest rate on government bonds. The foreign nominal interest rate is the short-term (one-year) interest rate of the United States on their government bonds. The expected depreciation represents the real effective exchange rates. The Political Country-specific risk spread is the political risk of countries based on several factors. The expected inflation is is a measure of price changes in consumer products. Real GDP growth is year on year growth of the gross domestic product of a country. The budget balance is the balance of a country at the national level and cyclically adjusted. The Central government debt is the debt level measured in percentage of GDP. The tax burden variable is the amount of taxes as a percentage of GDP.
Nominal interest rate Budget balance Foreign nominal interest rate Expected depreciation Political Country-specific risk spread Expected inflation Real GDP growth Central government debt Tax Number of observations 204 204 204 204 204 204 204 204 204 Mean 4.751 0.984 3.761 0.113 85.345 2.759 2.793 57.051 21.952 Minimum 2.100 -7.013 1.013 -16.423 69.329 0.140 -6.000 0.821 10.440 Maximum 17.000 7.096 5.839 10.678 97.164 15.884 11.497 118.300 30.756 Standard deviation 2.828 2.599 1.536 3.503 6.112 1.886 2.305 31.754 5.093 Source National Central Banks
19
Aisen and Hauner (2008) used in their study the control variable broad money supply (M3) this variable will not be included in this research, because the broad money supply for the countries in the EMU-area is identical. So this will not result in an interesting outcome if it was included in the regression. And the measures for broad money supply should for each country be identical which is difficult to find, because each country administrates this differently.
To get a better idea about the sample it is split into two samples one before 1999 and one after 2001. The sample is split at these years because in 1999 the third stage of the EMU was introduced and in 2001 Greece joined the EMU. In the years between 1999 and 2001 not all countries in the sample are in or out the third stage of EMU therefore there are no summary statistics of these years separately. The summary statistics of before 1999 are given in appendix B table BΙ and the summary statistics of after 2001 are given in appendix B table BΙΙ. The differences between the whole sample and the sub samples are the biggest in foreign nominal interest rate, the expected depreciation, the budget balance and the nominal interest rate. The foreign nominal interest rate is higher before 1999 and lower after 2001 compared to the whole sample. This indicates that in the first sample importing products is expensive and saving money is encouraged and in the second sample people should spend their money by importing foreign10 products. The
expected depreciation is in the whole sample about zero. While before 1999 it is negative and after 2001 it is positive. The budget balance had decreased after 2001 compared to the whole sample and the period before 1999. This indicates that the surpluses became smaller and even possibly deficits occurred or became higher. The Maastricht Treaty states that they prefer a budget balance above zero (a surplus), so the budget balances of the countries are becoming worse in the eyes of the Maastricht Treaty. The nominal interest rate is in the period before 1999 higher than the period after 2001 and the sample. This could possibly relate to some other variables which are different in that period.
V. Estimation results
The results of the estimations are discussed below and presented in table ΙΙ. The results are discussed in the order of the methods as described in the methodology section: first ordinary least squares with fixed effects followed by ordinary least squares with fixed effects and robust standard errors. Thirdly the results with the XTLSDVC method and finally the XTABOND2 method and the results of the regression with dummy variables.
20 A. Main results
The method ordinary least squares with fixed effects results in a significant and positive foreign nominal interest rate (1% significance level), lagged nominal interest rate (1% significance level) and expected inflation (5% significance level). All other variables (including the constant) are insignificant. The results of this method are in column (1) of table ΙΙ. These results indicate that only the foreign nominal interest rate, the lagged nominal interest rate and the expected inflation have an influence on the nominal interest rate. And the budget balance has no influence at all with a small and negative coefficient. This could indicate that the Ricardian Equivalence theorem holds and the Neo-classical and Keynesian theorems do not hold. The presence of autocorrelation may affect the results, as the Baltagi and Wu (1999) statistic is too high. To correct for autocorrelation, non-normality and stochastic independent variables robust standard errors in combination with ordinary least squares should be used.
Ordinary least squares with robust standard errors and fixed effects do lead to different results. The biggest difference is that expected inflation becomes insignificant, furthermore the foreign nominal interest rate is still significant but at a 5% level. The lagged nominal interest rate is still significant at the same level. Column (2) of table ΙΙ states the results of this regression. The standard errors (in parentheses in the table) have changed compared to ordinary least squares with fixed effects. The results still indicate that the Ricardian Equivalence theorem might hold and that the Neo-classical and Keynesian theorems do not hold.
Since endogeneity is present in samples with economic variables, OLS is not the correct method to study the relationship between budget deficits and interest rates of government bonds. A method that does correct for endogeneity is the least square dummy variable (XTLSDVC) method. The disadvantage of this method is that it is not possible to use robust standard errors to correct for autocorrelation. The result of the least square dummy variable method is in column (3) of table ΙΙ and are positive significant for the foreign nominal interest rate and the lagged nominal interest rate at the 1% level. And as in the ordinary least squares method with fixed effects the expected inflation is significant, but now at the 10% level instead of the 5% level. Due to autocorrelation it could again be that the least square dummy variable approach is not appropriate, so Generalized Method of Moments (XTABOND2) should be applied.
21
and Keynesian theorems do not hold, because the budget balance has no significant effect on the interest rate. And the Ricardian Equivalence theorem possibly does hold, because of the insignificant budget balance. In column (4) of table ΙΙ it can be noticed that the standard errors compared to the standard errors of the least square dummy variable (XTLSDVC) method have become smaller. Also when comparing the standard errors of the ordinary least squares regression and the XTABOND2 regression of most variables the standard errors of the XTABOND2 regression are smaller. This indicates that the results of the XTABOND2 method represent the reality more precisely than the least square dummy variable approach (XTLSDVC) and ordinary least squares.
22
Table ΙΙ
Estimation results
The results of a regression with OLS with fixed effects indicated by (1), OLS with fixed effects and robust standard errors indicated by (2), least square dummy variable method (XTLSDVC) indicated by (3) and Generalized Method of Moments (XTABOND2) indicated by (4). The abbreviations of the variables stand for the following: FNIR – foreign nominal interest rate, ED- expected depreciation, PCSRS – political country-specific risk spread, EI – expected inflation, GDP – GDP growth, BB – budget balance, CGD – central government debt, TAX – taxes as a percentage of GDP, LagNIR – the lagged with 1 year nominal interest rate.
Dependent variable:
nominal interest rate (1) (2) (3) (4)
BB -0.0230 -0.0230 -0.0180 -0.0108 (0.0579) (0.0474) (0.0625) (0.0358) LagNIR 0.705*** 0.705*** 0.760*** 0.727*** (0.0395) (0.0710) (0.0422) (0.0401) FNIR 0.284*** 0.284** 0.242*** 0.283*** (0.0799) (0.0925) (0.0863) (0.0797) ED 0.0231 0.0231 0.0234 0.0182 (0.0302) (0.0492) (0.0324) (0.0437) PCSRS -0.0515 -0.0515 -0.0523 -0.0269 (0.0362) (0.0320) (0.0383) (0.0197) EI 0.188** 0.188 0.141* 0.111 (0.0776) (0.125) (0.0818) (0.0762) GDP 0.0441 0.0441 0.0672 0.0216 (0.0634) (0.0704) (0.0646) (0.0443) CGD -0.00471 -0.00471 -0.00547 -0.00192 (0.0111) (0.0158) (0.0117) (0.00228) TAX 0.0412 0.0412 0.0281 0.0163 (0.0751) (0.0668) (0.0794) (0.0206) Constant 3.177 3.177 1.617 (3.795) (2.336) (1.399) Observations 204 204 204 204 Number of Country 12 12 12 12
23 B. High debt dummies and before EMU-dummies
In table ΙΙΙ the columns (1) and (2) contain regressions with dummy variables by using the XTABOND2 method. The first column uses dummies for high debt and low debt and the second column uses dummies for a period of before the third stage of EMU started (so before 1999).
The dummies for high debt and low debt are such that when a country has a debt level of more than 75% of GDP it will be in the group of high debt countries (high debt dummy). And when a country has a debt level of less than 25% it will be in the group of low debt countries (low debt dummy). The result shows that the high debt dummy is positive significant with a coefficient of 1.164, as expected. When countries have a high debt level the interest rate will increase 1.16% more than countries with a normal debt level. When debt levels become higher it is riskier to invest in those specific countries, because it might become more difficult to repay the debt or to pay the interest. The dummy of the countries with a low debt level is insignificant with a small negative coefficient (-0.456). If this would have an influence it will be such that the interest rate decreases if a country has a low debt level.
24
Table ΙΙΙ
Estimation results of regressions with dummies
The results of a regression with the Generalized Method of Moments (XTABOND2) with the dummies for high and low debt levels is indicated by (1) and the Generalized Method of Moments (XTABOND2) with the dummy of before the third stage of EMU is indicated by (2). The abbreviations of the variables stand for the following: FNIR – foreign nominal interest rate, ED- expected depreciation, PCSRS – political country-specific risk spread, EI – expected inflation, GDP – GDP growth, BB – budget balance, CGD – central government debt, TAX – taxes as a percentage of GDP, LagNIR – the lagged with 1 year nominal interest rate, HighDebt – dummy for countries with a debt level higher than 75% of GDP, LowDebt – dummy for countries with a debt level lower than 25% of GDP, NoEMU – dummy for the time period before 1999 (before the third stage of EMU started).
Dependent variable:
nominal interest rate (1) (2)
BB -0.0434 -0.0150 (0.0354) (0.0335) LagNIR 0.718*** 0.766*** (0.0470) (0.0537) FNIR 0.314*** 0.331*** (0.0761) (0.0501) ED 0.0147 0.0191 (0.0409) (0.0435) PCSRS -0.0172 -0.0209 (0.0160) (0.0202) EI 0.0981 0.114 (0.0854) (0.0690) GDP 0.0236 0.0175 (0.0373) (0.0370) CGD -0.0166* -0.00151 (0.00841) (0.00194) TAX 0.0129 0.0153 (0.0168) (0.0196) HighDebt 1.164*** (0.333) LowDebt -0.456 (0.431) NoEMU -0.415 (0.306) Constant 1.458 0.901 (1.255) (1.457) Observations 204 204 Number of Country 12 12
25 C. Robustness checks
To see if the results still hold if slight changes are made, some robustness checks are performed. In appendix C table CΙ these results are shown. In the columns (1) up to and including (6) show the regression outcomes when the least significant variable is removed from the regression until two variables are left (excluding the lagged nominal interest rate). In the following columns (7) up to and including (14) regressions are performed with all variables separately. The appendix shows that removing the least significant variable does not lead to other results and GDP growth is the second least significant variable. When putting the variables separately into the regression the constant is always significant unless the foreign nominal interest rate is added. The variable that were insignificant in the regressions as in column (4) of table ΙΙ are still insignificant if the variables are added separately, except for GDP growth. The significance of GDP growth could mean that it does explain a part of the interest rate. Although the coefficient of GDP growth is still very low (0.0864%) and when excluding the least significant variable GDP growth is already excluded in the second regression (see column (2) of appendix C table CΙ). I expect that GDP growth is not of influence on the interest rate due to the previous results. As Kim and Lombra (1989) state the economic growth should decrease, in this case the coefficient of GDP growth is positive. This result contradicts Kim and Lombra (1989) and could be further investigated in future research.
VI. Conclusion
In this research the relationship between the budget deficit and the interest rate is investigated. The theory is inconclusive about what the result should be as well as the empirical literature, therefore it is hard to say what the possible relationship could be. The budget deficit is measured by using the budget balance (if negative than it is a deficit) and the interest rate is the short-term government interest rate. To control for other possible influences control variables are added, these are foreign nominal interest rate, expected depreciation, political country-specific risk spread, expected inflation, government debt and tax burden. The sample consists of twelve countries, the countries which were in the third stage of EMU in 2001 and which adapted the Euro as their currency in 2002. The sample period is from 1990 to 2010, due to unavailability of data the sample starts in 1992 and ends in 2008.
26
standard errors (in Stata XTABOND2). So the best method is the XTABOND2 method as developed by Arellano and Bond (1991), which corrects for non-normality, autocorrelation, endogeneity and for small samples. The endogeneity problem is solved by using lagged variables as instruments and for the small sample is corrected by applying the Windmeijer correction.
The result of all types of methods shows that there is no relationship between the budget balance and the interest rate. The XTABOND2 method shows that the foreign nominal interest rate and the lagged nominal interest rate are significant. This indicates that there does exist a relationship between the foreign nominal interest rate, the lagged nominal interest rate and the nominal interest rate. The nominal interest rate is determined by the foreign nominal interest rate (the interest rate of the United States) and the lagged nominal interest rate (the nominal interest rate of the previous year). The difference between the XTABOND2 and the least square dummy variable (XTLSDVC) approach is that it gives smaller standard errors, so a better representation of the relationship (results).
The reason for this research is that there are different conclusions about the relationship in the theory and the empirical literature. The results of my research indicate that the Ricardian Equivalence theorem holds that there does not exist a relationship. This also means that the Neo-classical and Keynesian theorems do not hold. In the empirical literature my results agree with the results of Evans (1987a and 1987b). Aisen and Hauner (2008) split their sample in the countries with emerging economies and countries in advanced economies. The sample as a whole (emerging and advanced economy countries together) gives a significant positive result, while the advanced economy countries as a subsample give an insignificant result. This insignificant result is similar as the result found in this study. The twelve EMU-countries can all be considered as in an advanced economy, only the sample size of the research of Aisen and Hauner (2008) included more advanced countries (9 more than eleven11 countries of this sample). Gale and Orszag (2004) do find a
relationship and also indicated that in the empirical literature many studies exist of which a part also finds a relationship between interest rates and budget deficits.
Due to some assumptions and decisions about the sample of the research there does exist a limitation. This limitation is a possible statistical bias due to a small sample which could be solved by expanding the countries or the time period. The problem with expending the time period and countries is that the availability of data declines by reaching further in the past and using countries which entered the EMU several years later (than 2001). Another method to solve this problem is by
27
adding countries of other world regions and different economic developments. The disadvantage is that there is no longer a result of the twelve EMU-countries separate from the whole sample (unless dummies are used).
Current developments in interest rates of government bonds make it necessary to reconsider this research with new data. The current developments are rising budget deficits and decreasing interest rates for countries of which investors think are “safe” investments. This could also result in that countries get money when they give out bonds, because of a negative interest rate. This “phenomenon” could influence the results of this research. Countries which had negative interest rates on short term government bonds are Germany, Netherlands, France and Belgium.12
Correia-Nunes and Stemitsiotis (1995) state that current or expected budget deficit, frequency of the data and short or long-term interest rates have an influence on finding a relationship between the budget balance and the interest rate. This has not been proven, because many regressions should be performed where at start only one aspect is of such an influence that there does exist a relationship, followed by testing the following aspect separately and finally all aspects of not finding a relationship should be added together. Another advice for future research is therefore to confirm or contradict the statements of Correia-Nunes and Stemitsiotis (1995).
28
Appendix A Figures
Figure A1
Long-term interest rates
29
Figure A2
Short-term interest rates
30
Figure A3
Central government debt
31
Figure A4
Foreign nominal interest rate
35
Figure A8
Cyclically adjusted government budget balance
37
Figure A10
Scatter plot of the short-term interest rate
38
Figure A11
Scatter plot of the cyclically adjusted budget balance
39
Appendix B Additional descriptive statistics
Table BΙ
Descriptive statistics of the sample before the year 1999
In this table the descriptive statistics of the sample before the year 1999 are shown. The nominal interest rate is the short-term (3-month) interest rate on government bonds. The foreign nominal interest rate is the short-term (one-year) interest rate of the United States on their government bonds. The expected depreciation represents the real effective exchange rates. The Political Country-specific risk spread is the political risk of countries based on several factors. The expected inflation is is a measure of price changes in consumer products. Real GDP growth is year on year growth of the gross domestic product of a country. The budget balance is the balance of a country at the national level and cyclically adjusted. The Central government debt is the debt level measured in percentage of GDP. The tax burden variable is the amount of taxes as a percentage of GDP.
Nominal interest rate Budget balance Foreign nominal interest rate Expected depreciation Political Country-specific risk spread Expected inflation Real GDP growth Central government debt Tax Number of observations 84 84 84 84 84 84 84 84 84 Mean 6.757 0.969 4.606 -0.839 85.183 3.184 2.762 58.795 22.283 Minimum 2.800 -6.227 3.019 -19.423 69.329 0.601 -6.000 1.282 10.599 Maximum 17.000 6.566 5.513 10.678 95.370 15.884 11.497 118.300 29.501 Standard deviation 3.305 2.920 0.833 4.309 5.564 2.658 2.570 33.896 5.035 Source National Central Banks
40
Table BΙΙ
Descriptive statistics of the sample after the year 2001
In this table the descriptive statistics of the sample after the year 2001 are shown. The nominal interest rate is the short-term (3-month) interest rate on government bonds. The foreign nominal interest rate is the short-term (one-year) interest rate of the United States on their government bonds. The expected depreciation represents the real effective exchange rates. The Political Country-specific risk spread is the political risk of countries based on several factors. The expected inflation is is a measure of price changes in consumer products. Real GDP growth is year on year growth of the gross domestic product of a country. The budget balance is the balance of a country at the national level and cyclically adjusted. The Central government debt is the debt level measured in percentage of GDP. The tax burden variable is the amount of taxes as a percentage of GDP.
Nominal interest rate Budget balance Foreign nominal interest rate Expected depreciation Political Country-specific risk spread Expected inflation Real GDP growth Central government debt Tax Number of observations 84 84 84 84 84 84 84 84 84 Mean 3.059 0.401 2.535 1.618 85.138 2.477 2.354 54.867 21.304 Minimum 2.100 -7.013 1.013 -2.515 75.058 0.140 -2.972 0.821 10.440 Maximum 4.967 5.474 4.722 10.446 96.267 4.725 6.639 110.617 29.671 Standard deviation 0.916 2.129 1.440 2.180 6.454 0.891 1.875 30.284 5.171 Source National Central Banks
41
Appendix C Robustness check
Table CΙ
The estimation results of the robustness checks
The results of the robustness checks are made with the following regressions with the Generalized Method of Moments (XTABOND2) indicated by numbers (1) – (6) at each regression the least significant variable is deleted from the regression formula, numbers (7) – (14) are Generalized Method of Moments (XTABOND2) regressions with each variable separately. The abbreviations of the variables stand for the following: FNIR – foreign nominal interest rate, ED- expected depreciation, PCSRS – political country-specific risk spread, EI – expected inflation, GDP – GDP growth, BB – budget balance, CGD – central government debt, TAX – taxes as a percentage of GDP, LagNIR – the lagged with 1 year nominal interest rate.
Dependent variable:
nominal interest rate (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
BB -0.0212 -0.0180 -0.0191 -0.0151 -0.0194 -0.0134 (0.0337) (0.0344) (0.0338) (0.0330) (0.0296) (0.0232) LagNIR 0.730*** 0.725*** 0.722*** 0.726*** 0.733*** 0.731*** 0.751*** 0.774*** 0.778*** 0.763*** 0.794*** 0.782*** 0.778*** 0.781*** (0.0383) (0.0348) (0.0385) (0.0369) (0.0349) (0.0366) (0.0316) (0.0268) (0.0294) (0.0349) (0.0319) (0.0282) (0.0319) (0.0294) FNIR 0.277*** 0.296*** 0.281*** 0.282*** 0.279*** 0.274*** 0.265*** (0.0774) (0.0589) (0.0327) (0.0328) (0.0317) (0.0362) (0.0333) ED 0.0173 0.0201 -0.0285 (0.0443) (0.0423) (0.0328) PCSRS -0.0197 -0.0181 -0.0176 -0.0146 -0.0102 (0.0165) (0.0163) (0.0154) (0.0125) (0.00981) EI 0.0908 0.0895 0.0897 0.0841 0.0811 0.0979 0.0853 (0.0592) (0.0620) (0.0623) (0.0577) (0.0559) (0.0613) (0.0504) GDP 0.0274 0.0864*** (0.0401) (0.0275) CGD -0.00101 (0.00167) TAX 0.0103 0.0112 0.0108 -0.00589 (0.0167) (0.0158) (0.0153) (0.00563) Constant 1.082 0.954 0.989 0.970 -0.296* -0.329* -0.120 0.761*** 1.605* 0.569*** 0.413** 0.725*** 0.787*** 0.846*** (1.146) (1.084) (1.039) (1.028) (0.149) (0.167) (0.0727) (0.126) (0.844) (0.122) (0.180) (0.124) (0.134) (0.140) Observations 204 204 204 204 204 204 204 204 204 204 204 204 204 204 Number of Country 12 12 12 12 12 12 12 12 12 12 12 12 12 12
42
References
Aisen, A. and Hauner, D., 2008. Budget deficits and interest rates: a fresh perspective. International Monetary Fund, Working paper WP/08/42.
Arellano, M. and Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The Review of Economic Studies 58, 277-297.
Baltagi, B.H. and Wu P.X., 1999. Unequally spaced panel data regressions with AR(1) disturbances. Econometric Theory 15, 814-823.
Barro, R.J., 1989. The Ricardian approach to budget deficits. The Journal of Economic Perspectives 3, 37-54.
Bernheim, B.D., 1987. Ricardian Equivalence: an evaluation of theory and evidence. NBER Macroeconomics Annual 2, 263-304.
Bernheim, B.D., 1989. A Neoclassical perspective on budget deficits. The Journal of Economic Perspectives 3, 55-72.
Bovenberg, A.L., 1988. Long-term interest rates in the United States: an empirical analysis. International Monetary Fund Staff Papers 35, 382-390
Brooks, C., 2008. Introductory Econometrics for Finance, edition 2. Cambridge University Press, Cambridge.
Bruno, G.S.F., 2004. Approximating the bias of the LSDV estimator for dynamic panel data models. 10th London Stata User Group Meeting June 2004.
Bruno, G.S.F., 2005. Estimation and inference in dynamic unbalanced panel-data models with a small number of individuals. The Stata Journal 5, 473-500.
Cebula, R.J., 1991. A note on federal budget deficits and the term structure of real interest rates in the United States. Southern Economic Journal 57, 1170-1173.
43
Engen, E.M. and Hubbard R.G., 2004. Federal government debt and interest rates. NBER Macroeconomics Annual 19, 83-138.
European Central Bank, 2004. The Monetary Policy of the ECB. Frankfurt.
Evans, P., 1987a. Interest rates and expected future budget deficits in the United States. Journal of Political Economy 95, 34-58.
Evans, P., 1987b. Do budget deficits raise nominal interest rates? Evidence from six countries. Journal of Monetary Economics 20, 281-300.
Feldstein, M.S., 1986. The budget deficit and the dollar. NBER Macroeconomics Annual 1, 355-392.
Friedman, M., 1948. A monetary and fiscal framework for economic stability. The American Economic Review 38, 245-264.
Gale, W.G. and Orszag, P.R., 2004. Budget deficits, national saving, and interest rates. Brookings Papers on Economic Activity 2, 101-187.
Heijdra, B.J., 2009. Foundations of Modern Macroeconomics, edition 2. Oxford University Press, Oxford.
Kauppi, E., Lassila, J. and Teräsvirta, T., 1996. Short-term forecasting of industrial production with business survey data: experience from Finland’s great depression 1990-1993. International Journal of Forecasting 12, 373-381.
Kim, S. and Lombra, R.E., 1989. Why the empirical relationship between deficits and interest rates appears so fragile. Journal of Economics and Business 41, 241-251.
Kpodar, K., 2007. Why has unemployment in Algeria been higher than in MENA and transition countries? International Monetary Fund, Working paper WP/07/210.
Laubach, T., 2003. New evidence on the interest rate effects of budget deficits and debt. Board of Governors of the Federal Reserve System, FEDS Working Paper No. 2003-12, 1-21.
Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica 49, 1417-1426.
44
Plosser, C.I., 1987. Fiscal policy and the term structure. Journal of Monetary Economics 20, 343-367.
Quayes, S., and Jamal, A.M.M., 2007. Budget deficits and interest rates: The US evidence since 1946. The Singapore Economic Review 52, 191-200.
Roodman, D., 2009. How to do XTABOND2: An introduction to “difference” and “system” GMM in Stata. The Stata Journal 9, 86-136.
Seater, J.J., 1993. Ricardian Equivalence. Journal of Economic Literature 31, 142-190.
