Public Debt and Economic Growth-Is There A Change in ThresholdEffect - Hu - Xinruo - Honor Thesis
Public Debt and Economic Growth-Is There A Change in ThresholdEffect - Hu - Xinruo - Honor Thesis
Public Debt and Economic Growth-Is There A Change in ThresholdEffect - Hu - Xinruo - Honor Thesis
Effect?
Xinruo Hu
Economics Department
May 10 2019
Abstract
My paper investigates the relationship between public debt and GDP growth for European countries and
the heterogeneous debt threshold effect before and after the financial crisis, as well as between countries
with high and lower credit rating, by using a large cross sectional panel data and two estimation methods.
My empirical results suggest that the short run impact of debt on GDP growth is positive and statistically
significant but decrease to negative and lose some significance at public debt and GDP ratio of 109%.
After the financial crisis, either the threshold effect vanishes or the threshold values increase based different
cohort and method. Furthermore, countries with higher credit rating have higher threshold value and more
1
1 Introduction
Majority of the literature from the last decade discovers a negative correlation or an inverted U-shaped
relationship between public debt and economic growth. When debt is low, increase in debt will encourage
economic growth; however, when debt reaches some threshold value, further increase in debt would lower
GDP growth. In the pioneer paper The Impact of High and Growing Government Debt on economic Growth:
An Empirical Investigation for the Euro Area, Reinhart and Rogoff estimated a threshold value of 90%
where above growth prospects are severely undermined. However, there remains debate regards to what the
In the aftermath of global financial maelstrom in 2008, soaring public debt in many European countries
have outdistanced the threshold of 90%. According to the debt threshold effect, those countries should
have significant decline in growth if the GDP growth rate if not negative. However, the economic growth
of European countries with high credit rating are not much different as what they have before the financial
crisis in 2008 and when their public debt to GDP are below or around the threshold of 90%. At the same
time, GDP growth rate of European countries with low credit rating are indeed hampered (See Figure 3.3
for details). There are two natural questions to ask: 1) Whether financial crisis in 2008 made European
countries more tolerance to debt, and 2) Are European countries with lower credit rating more responsive
to debt.
The remainder of the paper is organized as follows. In Section 2, review relevant literature regarding
public debt and economic growth rate with a focus on debt threshold. In Section 3, describe the data, the
set-up and the methodology. In Section 4, give estimation results and test results. In Section 5, provide
result from robustness checks, including check for influential countries and an analysis on impact of debt on
long-term interest rates. Section 6, conclusion and further research. Appendix 1, data description and data
2 Literature Review
Most of the theoretical literature conclude a negative relationship between public debt and economic
growth. In Dividend Policy, Growth, and the Valuation of Shares, Merton H. Miller and Franco Modigliani
(1961) conclude that the external debt leads to lower stock of private capital, which reduces flow of income
and lowers growth. Peter Diamond (1965), in National Debt in a Neoclassical Growth Model, argues that
when in the efficient case (interest rate is higher than growth rate), national debt negatively affects growth
2
by increasing tax burden and move interest rate way from the Golden Rule, and thus leads to a decrease
in savings and consumption of taxpayers, which then lowers capital stock. However, when growth rate is
higher than interest rate, indicating an over accumulation of capital, an increase in national debt improves
allocation, and thus is beneficial to growth. In Fiscal Policy in an Endogenous Growth Model, Saint-Paul
(1992) built on the growth model from Blanchard (1985) and Weil (1989) and analyzed the impact of fiscal
policy on growth and found that, in an endogenous growth model, an increase in national debt reduces
economic growth. David Aschauer (2000), in Do States Optimize? Public Capital and Economic Growth,
proposed a growth model in which public debt has a non-linear (inverted U shape) relationship with economic
growth and found that high level of national debt adversely affects growth rate. From the result of theoretic
literature on national debt and growth, a logical question to ask is whether growth always negatively affected
by debt or is there a debt threshold above which debt becomes a burden on growth. John Cochrane, in
Inlfation and debt The public debt can negatively impact growth because increases uncertainty or leads
to expectations of future confiscation, possibly through inflation and financial repression (see Cochrane,
2011a,b, for a discussion of these issues). In this case, higher debt could have a negative effect even in the
short-run.
However, this question is still in dispute in existing empirical literature. I will first review three influential
papers in detail. Then, I will discuss the disagreements from major literature regarding this topic.
The most influential paper regarding debt and growth is Growth in a Time of Debt published in 2010
written by Carmen Reinhart and Kenneth Rogoff. The paper used an empirically approach, using a panel
of 3700 annual observations of 44 countries spanning about two hundred years from 1790 to 2009. They
classified the regimes into four categories: advanced economies with low national debt to GDP ratio (below
30%), medium-low national debt to GDP ratio (between 30% and 60%), medium-high national debt to GDP
ratio (between 60% and 90%), and high national debt to GDP ratio (above 90%), and conclude that countries
with high public debt have significantly lower growth rate. Although the large data enables readers a have
panoramic view, the lack of incorporating different country’s specific factors such as political system and trade
openness leads to the issue of heterogeneity. Also without using econometric and instrumental variables,
the results did not prove any causal relationship between debt and growth. Nevertheless, their findings
have inspired many scholars to research regrading this topic and evaluate their results. Later researches on
this topic have used econometric and multivariate regression to control endogeneity, reverse causality, and
cross-country heterogeneity.
3
The Impact of High and Growing Government Debt on economic Growth: An Empirical Investigation for
the Euro Area written by Cristina Checherita and Philipp Rother in another influential paper on this topic
that focused on only developed European countries. They used data mainly from the European Commission
AMECO database of 12 developed European countries from 1970 to 2011. Thus, the issue of heterogeneity,
which is often problematic in growth regressions, is alleviated by using a relatively restricted cross-sectional
sample. They used an empirical growth model which is built on a conditional convergence equation that
relates the GDP per capita growth rate to the initial level of GDP per capita, the investment/saving-to GDP
rate, and gross government debt as a share of GDP. To control the issue of endogeneity, they instrument the
debt variable for each country through either its time lags (up to lag of order 5) or through the mean debt
levels of the other countries in the data. They conclude that the debt-to-GDP threshold of this inverted
U-shape relationship is roughly between 90% and 100% on average for the data.
A similar paper Debt and Growth: New Evidence for the Euro Area written by Anja Baum, Cristina
Checherita-Westphal, Philipp Rother (BCR) used a slightly different approach. They also used annual data
of 12 developed European countries but from period 1990 to 2010. They first used the dynamic panel
threshold model and 2SLS to find the threshold value that minimizes the residuals. Then they used the
benchmark model which regresses GDP growth on short term interest rate, debt to GDP ratio, openness
to trade, ratio of gross capital formation and EMU membership to test the relationship between debt and
growth and the significance their result. The issue of heterogeneity and endogeneity is control using the
method stated in the previous paper. They found that for high debt ratios (above 95%) the impact of
In The real effects of debt, Stephen Cecchetti, M Mohanty and Fabrizio Zampolli (2011), report a debt
threshold of 85% of GDP by using data from 1980 to 2010 of 18 OECD countries and perform a growth
regression on control variables such as debt to GDP ratio, gross saving, inflation, population growth and
openness to trade. Pier Padoan, Urban Sila, and Paul Noord (2012), in Avoiding Debt Trap: Financial
Backstops and Structural Reforms, using a panel of 28 OECD countries over 1960 to 2011 and similar
growth equation, discover a similar debt threshold of 90%. However, Balazs Egert (2015), in Public Debt,
Economics Growth and Nonlinear Effects: Myth or Reality?, using the dataset and the four-regime model of
RR (2010) and a bivariate regression, discovers a much lower debt threshold of 20% and 60% above which
GDP growth prospects are severely undermined. Nevertheless, Chang and Chiang (2009), in Transitional
Behavior of Government Debt Ratio on Growth: Case of OECD Countries, using data of 19 OECD countries
over the period 1993-2007 and a panel smooth transition regression model that controls heterogeneity across
countries, find a significantly positive relationship between one-year lagged government debt ratio and real
GDP across all regimes. However, Alfredo Schclarek (2004), Debt and Economic Growth in Developing and
4
Industrial Countries, reports only a negative relationship between debt and growth for developing countries
but not for developed countries using a panel of 54 emerging markets and 29 advanced economies between
1970 and 2002 and performing linear and nonlinear regression on growth. Similar result was found by Andrea
Pescatori, Damiano Sandri, and John Simon, in Debt and Growth: Is There a Magic Threshold?. They used
36 developed country, covering period from 1821 to 2011. Using method similar to RR’s, they conclude that
there is no debt ratio threshold above which an adverse relationship between debt and growth kicks in.
I contribute to the current strand of literature on public debt and economic growth by examine possible
change in threshold effect from financial crisis and difference in response to debt for European countries
cohort by credit rating. My result is robust as I controlled for omitted variable bias and reverse causality.
This paper makes a unique contribution to the debate on debt threshold by presenting new empirical evidence
on the heterogeneity across European countries cohort by year and credit rating based on a sizeable dataset.
3.1 Data
Data used in this paper was extracted from the European Commission Database (AMECO). The panel
data consists observations over year 1988-2018 including 30 European Countries (See Appendix 1 for details).
By using a relative short time span, my data is less prone to political and economic structural changes and
more comparable with today’s economic settings. Given the data encompasses large number of countries,
country fixed effect is used to control for cross-section heterogeneity. Because of the relatively unconstrained
cross-section data, my result can be susceptible to outlier countries with extremely low or high debt level.
Regression result with outlier countries removed is included in robustness check. In this paper, GDP growth
rate is the main dependent variable. Indicator of whether the country’s debt to GDP ratio excess some
threshold value and debt to GDP ratio are the two main variables of interest.
One should note that in section 3, same letter may be used to represent a coefficient in different equation.
This does not mean estimated coefficients are the same. The coefficients in each equation are estimated
5
3.2.1 Initial Estimation Equation and Possible Issue
Using variables of interest, the two initial estimation equations are in the form of
where GROW T Hi,t is the one year GDP growth rate of country i in year t, αi is a constant, I(dt > d∗ ) is
an indicator which equal 1 if the country i has debt to GDP ratio (di,t ) is greater than the threshold d∗ and
where di,t is the current debt to GDP ratio di,t −di,t−1 , di,t 2 is added to capture the possible non-linear effect
of debt on economic growth, other variables are the same as as from the first initial estimation equation
(equation (1)).
The above two equations clearly suffer from reverse causality and omitted variable bias. Reverse causality
happens when the dependent variable also effects independent variable, making the result not casual. To
deal with reverse causality, instrumental variable need to be used. Omitted variable happens when that
regression excludes a variable that is correlated with other covariates and the dependent variable. Consider
Y1 = α + β1 X1 + u (3)
Y1 = α + β1 X1 + β2 X2 + v (4)
Then, u = β2 X2 + v. Therefore,
cov(X1 , X2 ) cov(X1 , v)
βˆ1 = β1 + β2 × + (5)
var(X1 ) var(X1 )
Then if β2 6= 0 and cov(X1 , X2 ) 6= 0, βˆ1 will be biased. The direction of bias depends on the sign of β2 and
cov(X1 , X2 ). To deal with omitted variable bias, control variables need to be added.
6
3.2.2 Variable Selection and Construction
To mitigate the issue of reverse causality, a internal instrument of one year lagged debt to GDP ratio
(di,t−1 ) is used. One should be aware, however, this instrument does not eliminate reverse casualty. The
advantage of using this internal instrument compare to an external instrument is that the internal instrument
provides a more direct and interpretable result. To measure the possible non linear effect, squared one year
lagged debt to GDP ratio (di,t−1 2 ) is also included in the final estimation.
To avoid omitted variable bias, country fixed effect, population growth and size of government, approxi-
mated by total government expenditure to GDP ratio, and lagged GDP growth are construed and included
in the final estimation. Ideally, year fixed effect should be included to capture the possible difference in
economic condition across year. Interaction terms between dummies for year and one year lagged debt to
GDP ratio and between dummies for country and one year lagged debt to GDP ratio should also be added
to capture the heterogeneous effect of debt on economic growth across year and across country. However,
because my data includes 30 countries and spans 30 years, including above variables will introduce too many
coefficients which need to estimated relative to the sample size. This will introduce large variance and lower
estimation precision. Therefore, I shall not include year fixed effect, interaction terms between dummies for
year and one year lagged debt to GDP ratio and between dummies for country and one year lagged debt
to GDP ratio. By doing this, I am posing the assumptions: 1)Economic condition are similar across year
(exclude year fixed effect), 2)Responses to a increase in debt is the same across year (exclude interaction
between year dummy and debt), and 3) Countries have the same response to a increase in debt (exclude
interaction between country dummy and debt). However, it is possible that economic condition and response
to a increase in debt changes after financial crisis. Therefore, to capture the impact of financial crisis on the
effect of debt on growth, I included fixed effect of financial crisis by adding a dummy variable which equals
to 1 if year is greater than or equal to 2008 and 0 otherwise and interaction between this dummy and debt.
I shall not elaborate on the significance of country fixed effect, population growth and size of government
here (Support of this can be found from other literature). I will, however, explain the reason and possible issue
with including lagged GDP growth. If current GDP growth and current debt to GDP ratio are negatively
correlated. Then excluding lagged GDP growth will lead to a downward bias on the coefficient for debt To
find the optimal year of lags, Autoregressive model of the form below is used.
Xi,t = µi + βt + φi,t−1 Xi,t−1 + φi,t−2 Xi,t−2 + ... + φi,t−p Xi,t−p + Zi,t (6)
where Xi,t is GDP growth rate for country i at year t, µi and t are deterministic term (constant and trend),
and Zi,t is residual for country 1 in year t assumed to have zero mean and constant variance σ 2 . I choose
7
p = 1, 2, 3, 4, 5 because of the constrained time span of my data. The optimal level of lag p is chosen by: 1)
Model selection method such as Akaike information criterion(AIC) and Baysian Information criterion (BIC).
2) t-test on each lagged variable and F-test on all lagged variable. For method 1): for a selection of model
and
BICK = −2log(Lˆk ) + K × log(n), where Lˆk is the maximum likelihood of Mk and n is the sample size (8)
Both AIC and BIC choose a model with higher maximum likelihood while controlling for the number of
parameters (issue of overfitting). However, one should note, that BIC tends to select a model with fewer
parameters because of the larger punishing term k ×log(n) compare to AIC’s punishing term 2k. For method
2): Both t test and F test need the assumption that residuals are distributed as N (0, σ 2 ). T-test tests the
hypothesis that φi = 0 against its two sided alternative that φi 6= 0. F test tests the null hypothesis that
φi = φJ = ... = φm = 0 against its two sided alternative that the null hypothesis is not true. If p-values is
large (> 10%) resulting from t-test, then covariates corresponding to φi does not have significant predicting
power on the dependent variable, and thus can be exclude from the regression. Similarly, if p-values is large
(> 10%) resulting from t-test, then one or more of covariates corresponding to φi , φj , ..., φm jointly does
not have significant predicting power on the dependent variable, and thus some can be exclude from the
regression. For majority of countries, using just one year lagged GDP growth rate gives the lowest value of
AIC and BIC. Furthermore, for majority of countries, the only significant variable is one year lagged GDP
growth rate; p-value from F test increase significantly after adding pth order lag for k > 1. This means
that longer lags of GDP growth have relatively small predicting power on current change in GDO growth
compare to on year lagged GDP growth. Result using debt to GDP ratio as dependent variable (regress
on lagged growth rate of GDP) is analogues. I only include one year lagged GDP growth rate in the final
regression. One possible issue with including one year lagged GDP growth rate is if correlation between this
variable and change in debt to GDP ratio is extremely high, then the precision of the estimated regression
coefficients decreases. However, highly correlated covariates does not appear to be an issue.
8
3.2.3 Final Estimation Equations
The first estimation equation for the threshold value is in the form,
n−1
X
GROW T Ht = α + β1 I(dt−1 > d∗ ) + θi I(countryi ) + φ1 GROW T Ht−1 + φ2 P OP GROW T Ht
i=1 (9)
+ φ3 GOV SIZEt + φ4 I(P OST 2008) + φ5 I(P OST 2008) ∗ I(dt−1 > d∗ ) + Ut
where GROW T Hi,t is the one year GDP growth rate in year t, α constant, I(dt−1 > d∗ ) is vector of an
indicators which equal to 1 if one year lagged debt to GDP ratio is above the debt threshold and 0 otherwise,
I(countryi ) is vector of an indicators which equal to 1 if country is countryi and equal to 0 otherwise
(n is the total number of country in the data), dt−1 is a vector of one year lagged debt to GDP ratio,
P OP GROW T Ht is a vector of population growth at time t, GOV SIZEt is a vector of size of government at
time t, approximated by total government expenditure to GDP ratio, I(P OST 2008) is an vector of indicators
which equals to 1 if year is greater than or equal to 2008 and 0 otherwise, I(P OST 2008) ∗ I(dt−1 > d∗ ) is
a vector of interaction terms between a vector of indicators for post financial crisis and a vector of indicator
for debt to GDP ratio above the threshold, which intends to capture the different effect of debt threshold on
economic growth before and after the 2008 financial crisis, and Ut is a vector of residuals assumed to have zero
mean and jointly distributed as N (0, Σ). Threshold value(d∗ ) is chosen by two procedures: 1) ”Coefficient
Method”: Choose the threshold value such that itself has a negative coefficient, any smaller threshold value
has positive coefficient, and any large threshold value has a negative coefficients. 2) ”Minimizing RSS
Method” : Find the threshold value that minimizes the residual sum squared.
The second estimation equation for the threshold value is in the form,
n−1
X
2
GROW T Ht = α + β1 dt−1 + β2 dt−1 + + θi I(countryi ) + φ1 GROW T Ht−1 + φ2 P OP GROW T Ht
i=1
where dt−1 is one year lagged debt to GDP ratio, I(P OST 2008) ∗ dt−1 is a vector of interaction terms
between a vector of indicators for post financial crisis and a vector one year lagged debt to GDP ratio above
the threshold, which intends to capture the different effect of debt on economic growth before and after the
2008 financial crisis, and other variables are the same as from the the first estimation equation (equation
(8)). The threshold value is chosen by calculating the turning point from coefficients for one year lagged
debt to GDP ratio (dt−1 ) and squared one year lagged debt to GDP ratio (dt−1 2 ) in the following method
9
(solution to a polynomial of degree 2),
β1
turning point = − (11)
2 × β2
where β1 is the coefficient for dt−1 and beta2 is the coefficient for dt−1 2 .
One should note that in section 4, same letter may be used to represent a coefficient in different equation.
This does not mean estimated coefficients are the same. The coefficients in each equation are estimated
For the first estimation equation, I run regression for every integer value between 0 and 130 as the
threshold value (d∗ ) and obtained coefficient corresponds to each I(dt−1 > d∗ ) (Table 2.1). Using full set
of data, I do observe the threshold effect, where the coefficients have a decreasing trend as threshold values
increase and eventually become negative (Figure 3.4). The coefficient decrease dramatically from positive
to negative at debt to GDP ratio of 109% and further decrease as threshold value increase. However,
the coefficient become less negative when threshold value increase to 127%. The coefficient of I(d > d∗ )
corresponds to d∗ = 109% is −0.876. The result is significant at 5% level. This suggests a robust result
that when everything else is the same, country with debt to GDP ratio above the threshold value 109%
experience 0.876% lower growth rate than country with debt to GDP ratio below the threshold. However,
applying ”the minimizing RSS” method on the first equation gives my lower estimated threshold values (10
%-20%). This can be explained by the dramatic decrease in coefficient significant when threshold value is
large, which means that when debt is sufficient high, government debt has significantly lower predicting
power on economic growth. Therefore, minimizing RSS tends to give lower debt threshold value where
the predicting power on debt is still high. Estimated threshold value from minimizing RSS does not mean
country with debt to GDP ratio above this threshold will experience negative economic growth. Therefore,
the ”minimizing RSS method” can viewed just as a optimization method with less economic interpretation
compare to the ”coefficient method”. For this reason, I shall only consider ”the coefficient method” in the
following paper.
Based on the second estimation equation, one year lagged debt to GDP ratio and squared one year lagged
debt to GDP ratio are also variable of interest (Table 2.2). The estimated coefficient for the two variables
10
are 0.126 and 5.46×10−4 respectively. Both coefficients are significant at 0.1% level. This suggest that a 1%
increase in debt to GDP ratio will result in a 0.126% increase in economic growth rate in the following year.
However, because of the coefficient for squared one year lagged debt to GDP ratio is negative, the positive
effect of debt to GDP ratio on economic growth rate decreases as debt increase. The turning point can be
calculated using equation 11 which gives a estimate for the threshold value of 115%.
Both regressions give similar estimations for the threshold value which are higher than 90% found by
Reinhart and Rogoff. Both estimation of threshold value suggest that countries become more tolerance to
debt after the financial crisis. Using 90% as threshold value and run regression using equation 9, the estimated
coefficient is 0.512. This suggest that country with debt to GDP ratio higher than 90% experience 0.512%
more growth than country with debt to GDP ratio higher than 90%, which is contradicts to Reinhart’s
and Rogoff’s result. Furthermore, coefficient for (P OST 2008) and I(P OST 2008) ∗ dt−1 are significant at
1% level. Financial crisis in 2008 do seem to have impact on the effect of debt on economic growth rate.
Therefore, I will estimate threshold value before and after the financial crisis to make this impact clearer.
The two estimation equations for investigating the heterogeneity by year cohort are,
n−1
X
∗
GROW T Ht = α + β1 I(dt−1 > d ) + θi I(countryi ) + φ1 GROW T Ht−1 + φ2 P OP GROW T Ht
i=1 (12)
+ φ3 GOV SIZEt + Ut
and
n−1
X
GROW T Ht = α + β1 dt−1 + β2 dt−1 2 + + θi I(countryi ) + φ1 GROW T Ht−1 + φ2 P OP GROW T Ht
i=1
+ φ3 GOV SIZEt + Ut
(13)
where variables in equation 12 is the same as in equation 9 and variables in equation 13 is the same as in 10
but with I(P OST 2008) and I(P OST 2008) ∗ I(dt−1 > d∗ ) removed in equation 12, and I(P OST 2008) and
Before the financial crisis, the estimated threshold value from equation 12 and 13 are 71% and 76%
respectively (Figure 3.5, Table 2.1, Table2.2). The coefficient for I(dt−1 ) is -1.27 which is higher than the
estimated coefficient of -0.876 using the full data set. The estimate is also significant at % level. This
lower and significant threshold value combined with the large magnitude estimated coefficient suggest that
11
countries are more sensitive to debt before the financial crisis.
After the financial crisis, equation 12 does not give an estimation for threshold value because coefficients
on I(dt−1 ) are all positive for threshold value between 30 and 150 (Figure 3.6, Table 2.1, Table2.2). However,
I do observe a decreasing trend in economic growth rate as debt threshold increase. Using the threshold
value 109% found by equation 9, the estimated coefficient for I(dt−1 ) is 0.375. This suggest that keeping
everything else the same, after the financial crisis, country with debt to GDP ratio above 109% experience
0.375% higher growth rate than country with with debt to GDP ratio less than or equal to 109%. Equation
13 also does not give a practical estimation of threshold value. Furthermore, coefficients for variables of
interest from equation 12 and 13 are not significant at 10% level. This suggest not only country become
more tolerance to debt, but also debt has less predicting power on economic growth rate after the financial
crisis.
Credit rating can effect threshold value through its effect on interest rate. Higher credit rating is associ-
ated with lower risk, and therefore, lower risk premium. Then high credit rating country should theoretically
experience less increase in interest rate, and thus less impact on economic growth. To investigate the effect
of credit rating on threshold value, I ran regression using equation 9 and 10 on country with low credit rating
and high credit rating separately. Using S&P crediting rating, I classify a country as high credit rating if it
receives rating better or equal to AA, and country has low credit rating if it has rating worse than all equal
to BB. According to this classification method, low credit rating countries are Bulgaria, Croatia, Cyprus,
Hungary, Italy, and Romania. High credit rating countries are Austria, Germany, Sweden, Netherlands,
Norway, Luxembourg, Belgium, Denmark, Finland, and France. One should note that this classification
method will significantly reduce sample size. Also, because country with low credit rating tends to be less
developed, there are less data available. Therefore, the precision of estimates will be low for country with
high credit rating and even worse for low credit rating country.
For low credit rating country, the estimated threshold value from equation 9 is 60% (Figure 3.7). However,
estimated coefficient for I(dt−1 > d∗ ) is insignificant with a value of -0.0415 (See table 2.1 for detailed result).
For high credit rating country, the estimated threshold value from equation 9 is 115% (Figure 3.8). The
estimated coefficient for I(dt−1 > d∗ ) is -0.765 and significant at 10% level (Table 2.1). However, for low
credit rating country, equation 10 does not give a practical estimated threshold value (Table 2.2). However,
the result is not significant which may be due to limited sample size. For high credit rating country, equation
10 estimated a turning point at 119% debt to GDP ratio (Table 2.2). Because data is already limited using
years from 1990 to 2018, I shall not further investigate the effect of financial crisis on debt threshold for low
12
and high credit rating country.
Although the estimated threshold value may not be precise, the relative magnitude of threshold values
for low and high credit rating country can still shed light on the heterogeneity in debt tolerance. High
credit rating country have significantly higher threshold value than low credit rating country. In addition,
estimated threshold value for high credit rating country is also higher than the estimated threshold value
when using full data set. Therefore, one may conclude that high credit rating country are more tolerance to
5 Robustness Check
One should note that in section 5, same letter may be used to represent a coefficient in different equation.
This does not mean estimated coefficients are the same. The coefficients in each equation are estimated
Higher government debt is likely to be associated with higher long term interest rates because investors
often connects high debt level with high sovereign risk premium. Therefore, large increase in debt can
theoretically lead to higher long term interest rates which cause a decrease in private spending growth, and
thus reduce economic growth. To investigate the effect of high government debt on long interest rate, the
n−1
X
IN Tt l = α + β1 I(dt−1 > d∗ ) + θi I(countryi ) + φ1 GROW T Ht−1
i=1 (14)
+ φ2 IN Tt−1 s + φ3 I(P OST 2008) + φ4 GOV SIZEt + Ut
where IN Tt l is long term interest rate at year t, IN Tt−1 s is long term interest rate at year t-1, and other
Running regression using equation 14 with threshold value from 0% to 130%, the coefficient on I(dt−1
have a increasing trend from negative to positive as the threshold value increases (Figure 3.9, Table 2.3).
This means that if two countries both have low public debt then the country with higher public debt have
lower interest rate. The coefficient become positive at threshold value of 103% and experience a dramatically
increase at threshold value of 132%. This suggest that country with debt to GDP ratio above 103% will have
13
higher long term interest rate than country with debt to GDP ratio below or equal to 103%. One should note
that this threshold value is slight below the estimated threshold from equation 9. This does not discredit
the estimations because a increase interest rate does not necessarily lead to negative economic growth. The
increase interest rate has to be sufficiently large to bring economic growth negative. Consider the following
case, if a country’s debt to GDP ratio just went above the threshold of 103% estimated by equation 14, a
further increase in debt to GDP ratio is needed so that the increase in long term interest rate is sufficient
large to bring economic growth negative. Then the threshold value which correlates with negative growth
is higher than 103% estimated by equation 14. Then the discrepancy between threshold values estimated
from equation 9 and 14 is even smaller. The estimated coefficient on I(dt−1 > d∗ ) is 0.01 and insignificant
when threshold value of 103% is used. This small coefficient validates my argument above because a 0.01%
higher interest rate is too small to bring the economic growth negative. Using threshold value of 109% found
from equation 9, the estimated coefficient on I(dt−1 > d∗ ) is 1.56 and is significant at 0.1% level. This
suggest that country with debt to GDP ratio above 109% experience 1.56% higher long term interest rate
than country with debt to GDP ratio below 109%. This 1.56% higher long term interest rate is more likely
to have a negative impact on economic growth than 0.01%. However, further analysis needs to be done to
estimate a long term interest rate which brings economic growth negative.
Running equation on data before 2008, coefficients of I(dt−1 > d∗ ) have a increasing trend from negative
to positive as threshold value increase for threshold value below 40% (Figure 3.9). However, the coefficients
for threshold value over 40% fluctuates around 0.196 (sometimes decrease to negative) (Table 2.3). The
threshold value is hard to identify in this case. One could argue that the threshold value for which the
coefficient change from positive to negative for the first time is 33%. This threshold value is significant lower
than the threshold value estimated using equation 9 on data before 2008. The coefficient on I(dt−1 > d∗ )
associated with this threshold value is 0.0273%. This suggests that before the financial crisis, country with
debt to GDP ratio below or equal to 33% experience a 0.0273% higher long term interest rate. This is
relatively small compare to the coefficient of 1.56 estimated in previous subsection using threshold value of
109%. The small coefficient corresponds to threshold value of 33% and the pattern of coefficient for threshold
value above 40% may explain the extremely low threshold value here. Because difference in long term interest
rate for country with debt to GDP ratio above and below threshold value from 33% to 130% is only 0.197 on
average, the country need to significantly increase government debt so that the increase in long term interest
rate is sufficient enough to bring economic growth negative. If this is the case, the debt to GDP ratio for this
country would be a lot higher than 33%. The small fluctuating coefficients mean that, for country with debt
14
to GDP ratio above threshold 33%, an increase in debt to GDP ratio has small and ambiguous effect on long
term interest rate. This may suggest that an increase in debt to GDP ratio has small effect on risk premium.
Therefore, before the financial crisis, investors may be confident that even country with high debt has the
ability to pay its debt. One may think that my result for threshold before financial crisis contradicts my
finding in section 4.2.1 that threshold value is lower before financial crisis. Based on my finding, one would
expect debt threshold to be higher before the financial crisis because increase in debt does not translate to
large clear increase in long term interest rate, then large increase in debt is needed for a large increase in
long term interest rate so that a negative growth rate is possible. However, this discrepancy merely suggests
that there are other channels for which increase in debt effects economic growth.
Running equation on data after 2008, coefficients of I(dt−1 > d∗ ) have a increasing trend from negative to
positive as threshold value increase (Figure 3.11). The estimated threshold is 130% for which any threshold
value above have positive coefficients. This is largely in line with my result from section 4.2.2 that no
threshold value is found using data after 2008. The coefficient on I(dt−1 > d∗ ) using threshold value of
130% is 0.229 (Table 2.3). Although this coefficient is not large, the coefficient increase dramatically to 4.65
for threshold value of 132%. This explains the large threshold value because for country with debt to GDP
ratio above the threshold of 130%, any further increase in debt has a dramatic increase in interest rate that
Overall, my result from this section largely in line with my finding in section in 4.2.1. Estimated threshold
using equation 14 for data before the financial crisis is significantly lower than using data after the financial
crisis. This increase in threshold value suggests that countries became more tolerance to debt after the
financial crisis.
To avoid my result being driven by country with too high or low debt to GDP ratio, I removed 1) Country
outlier: country with both debt to GDP ratio and economic growth significant different from other countries
and 2) Country with high leverage (statistically): country with extremely high or low debt to GDP ratio.
Then I ran regressions using equation 9 and 10 excluding some countries identified by above method (Table
First, I excluded Estonia (country outlier) with on average extremely low debt to GDP ratio (5.18%)
and low economic growth (4.20%). Using equation 9, the estimated threshold is 110% for which any higher
debt to GDP ratio has a negative impact on economic growth. This threshold is slightly higher than 109%
estimated using question 9 on the full data set. However, the coefficient corresponds to I(dt−1 > 111%) is
15
-0.310 which is smaller than -0.876 estimated using question 9 on the full data set. This means that difference
in economic growth is very small for country with debt to GDP ratio below and above the debt threshold
value. This result seems reasonable because Estonia also has the highest economic growth rate on average.
Therefore, removing Estonia narrows the difference in economic growths between country with debt to GDP
ratio below and above the threshold. The threshold estimated using equation 10 is 117% which is higher
then the turning point of 115% estimated using equation 10 on the full data set. Overall, removing Estonia
Second, I excluded Greece (country outlier) with significantly high debt to GDP ratio (130.09%) and
low economic growth (0.7444%). The estimated threshold using equation 9 is 109% which the same as the
threshold value estimated using equation 9 on the full data set. This result seems somewhat surprising
because one would expect threshold value to decrease after removing Greece from the regression. However,
after taking a close look at Greece’s data, the estimations seems plausible. Greece experienced some slow
positive and some negative economic growth when it has a debt to GDP ratio above 109%. Because the
threshold is estimated by looking at the average growth for country below and above the threshold while
controlling for other factors, it is probable that the effect of removing Greece vanishes after averaging.
However if one construct more granular partition of threshold values, the effect of removing Greece may be
more obvious. The coefficient of I(dt−1 > 109%) is -0.322 which is smaller than -0.876 estimated in section
4. This can be explained by the negative growth associated with debt to GDP ratio above 109% removed
from regression. The threshold estimated using equation 10 is 107% which is lower then the turning point of
115% estimated using equation 10 on the full data set. This decrease in threshold value is reasonable because
equation 10 captures the linear and non-linear effect 1% increase in debt to GDP ratio. Therefore, equation
10 captures more information from data than equation 9 which looks at the average. One should note that
the two estimated thresholds are still higher than 90% which was estiamted by Reinhart and Rogoff. Overall,
Third, I excluded Estonia (country outlier) with on average extremely low debt to GDP ratio (5.18%)
and low economic growth (4.20%), Luxembourg with on average extremely low debt to GDP ratio (14.99%),
and Bulgaria on average extremely low debt to GDP ratio (21.41%). The estimated threshold using equation
9 is 128% for which any higher threshold value correspond to a negative coefficient. However, one should
know that, the coefficient is also negative at threshold 119%, however coefficients correspond to threshold
values between 119% and 128% are positive. The coefficient associated with threshold value 128% is -0.276
which is also lower than -0.876 estimated by equation 9 using the full data set. This means that difference in
economic growth is relatively small for country with debt to GDP ratio below and above the debt threshold
value. The estimated threshold using equation 10 is 117% which is higher then the turning point of 115%
16
estimated using equation 10 on the full data set. Overall, removing Estonia does not change my result
significantly.
Fourth, I excluded Greece (outlier) with significantly high debt to GDP ratio (130.09%) and low economic
growth (0.7444%), Italy (outlier) with significantly high debt to GDP ratio (113.14%) and low economic
growth (0.564%) and Belgium (high leverage) with significantly high debt to GDP ratio (103.57%). The
estimated threshold using equation 9 is 105% which is low than the threshold 109% estimated using equation
9 on the full data set. the coefficient correspond to I(dt−1 > 105%) is -0.374 which smaller than -0.876
estimated in section 4. This result is logical because Greece and Italy have the top 2 debt to GDP ratio
and the bottom 2 GDP growth rate. Therefore, the economic growth rate for remaining countries with
debt to GDP ratio above threshold 105% is not significantly lower than countries with debt to GDP ratio
below threshold 105%. One should note that this threshold value is still higher than 90% estimated by
Reinhart and Rogoff. The coefficient correspond to I(dt−1 > 90%) is 0.377, which means that on average,
with other factors the same, country with debt to GDP ratio above 90% experience higher growth than
country with debt to GDP ratio below 90%. The estimated threshold from equation 10 is 111% which is
lower than 115% estimated in section 4. However, this estimation is not significant. This may be due to lower
variance (spread) in debt to GDP ratio and GDP growth rate which makes the relationship less obvious.
In conclusion, removing Greece, Italy and Belgium do result in a lower threshold estimation, but it is still
Finally, I excluded all countries mentioned above. The estimated threshold value using equation 9 is 106%
which is lower than 109% estimated by using the full data. The estimated coefficient for I(dt−1 > 106%)
is -0.106 which is smaller than -0.876 estimated in section 4. This decrease in magnitude of coefficients is
logical because countries with too high or low GDP growth and debt to GDP ratio were excluded. Using
equation 10, the estimated threshold is 108%. However, both estimates are not significant which may be due
to smaller sample size and lower variance (spread) in debt to GDP ratio and GDP growth rate.
For the most part, running regression excluding influential counties still give consistent results. Therefore,
having outliers and data points with high leverage in my regressions do not incur substantial consequence;
GDP per capita can also be used to investigate the relationship between debt and growth economic
17
∆x/x ∂x/x 1 ∂x ∂lnx ∆lnx
≈ = = ≈ (15)
∆t ∂t x ∂t ∂t ∆t
Also,
∆ x/y x
= ∆ln = ∆lnx − ∆lny = g x − g y (17)
x/y y
Recall that variable P OP GROW T H in equation 9 and 10 is population growth rate, then estimates from
n−1
X
GDP CAP IT AGROW T Ht = α + β1 I(dt−1 > d∗ ) + θi I(countryi ) + φ1 Growtht−1
i=1 (19)
+ φ2 P OP GROW T Ht + φ3 GOV SIZEt + Ut
Similarly, estimates from equation below should give similar estimates as from equation 10.
n−1
X
GDP CAP IT AGROW T Ht = α + β1 dt−1 + β2 dt−1 2 + + θi I(countryi ) + φ1 Growtht−1
i=1 (20)
+ φ2 P OP GROW T Ht + φ3 GOV SIZEt + Ut
where GDP CAP IT AGROW T Ht is GDP per capita growth at year t and other variables the same as from
Using the full data set, the estimated threshold from equation 19 is 108% which is very similar to 109%
calculated from equation 9. The coefficient corresponds to I(dt−1 > 108%) is -0.734 which is also very close
to -0.876 estimated from equation 9 (Table 2.6). The result is significant at 5% level. However, the smaller
coefficient means that given other factors the same, the adverse effect of debt on growth of GDP per capita
is smaller than on output growth. The estimated threshold from equation 20 is 115% which is the same
as the threshold estimation of 115% from equation 10 (Table 2.7). Both difference may be explained by
measurement error.
18
Using the data before the financial crisis, the estimated thresholds from equation 19 is 71% which is the
same as my estimates from equation 9. The coefficient corresponds to I(dt−1 > 108%) is -0.985 which is
also very close to -1.27 estimated from equation 9 (Table 2.6). Similar as using the full data, the smaller
coefficient means that given other factors the same, the adverse effect of debt on growth of GDP per capita
is smaller than on output growth. The threshold estimation from equation 20 is 77% which is very close to
76% estimated by equation 10 (Table 2.7). Again, the difference may be explained by measurement error.
Using growth of GDP per capita as dependent variable, I observe a decrease in threshold value using data
Using data after the financial crisis, equation 19 fail to give an estimated threshold value because all
coefficient correspond to I(dt−1 > d∗ ) for integer valued d∗ between 30 and 150 are all positive. However,
similar as the result from section 4, coefficient does have a creasing trend as threshold value increase. Using
threshold value found in section 4 by equation 9, the coefficient of I(dt−1 > 109%) is 0.287 (Table 2.6).
Equation 20, similar to equation 10 fail to give a practical estimation of threshold (Table 2.7). Therefore,
using growth of GDP per capita as dependent variable, I can draw the same conclusion that after the financial
Overall, using growth of GDP per capita as dependent variable, I got very almost identical result as in
section 4. Therefore, my results are robust and should be free from calculation error.
6.1 Conclusion
Debt threshold a topic subject to heated academic debate in the last decade. What is the effect of public
debt accumulation on economic growth is an central question to policy maker when designing the optimal
fiscal policy. This topic gains increasing attention in the aftermath of financial crisis in 2008. This paper aims
to answer the question of the relationship between government debt and economic growth and if financial
crisis has an impact on the relationship, by using a large cross sectional data set.
I developed two estimation methods to capture any threshold effects while controlling for endogeneity.
Then I conducted formal statistical analysis of debt threshold effect on output growth by applying the two
estimation methods on a panel data which spans 30 years and consists 30 European countries, as well as on
two subgroups classified by credit rating. To find the impact of financial crisis on the debt threshold effect,
I splitted the panel data into two subgroups, data before and after financial crisis, and performed statistical
19
I found a inverted U shape relationship between debt and growth in some cases. Statistically significant
threshold effect was found using both estimation methods on the full data sets. However, my estimates
are not identical (109% and 115%). The short-term impact of debt on GDP growth become negative and
less significant once the public debt to GDP ratio reaches 109% or 115%. In addition, the threshold effect
is heterogeneous by crediting rating cohort. Countries with low credit rating are more vulnerable to debt
(estimated threshold of 60%) compare to countries with high credit rating (estimated threshold of 115%).
However, this result is less statistically significant due to data limitation. Furthermore, financial crisis does
have considerable impact on threshold effect. Countries become substantially more tolerance to debt. Before
the financial crisis, estimated thresholds from the two estimation methods are 71% and 76%, whereas after
the financial crisis, no thresholds were found. This means that after the financial crisis, there does not exists
a debt to GDP ratio for which any further accumulation in debt will have a reversal effect on growth.
My results show that the positive short term economic stimulus from additional debt decreases severely
when the initial debt level is high, and can become negative as debt increases. The adverse effect suggests
that when debt to GDP ratio is extremely high, reducing it would be beneficial to economic growth. On
the other hand, if the initial debt is low, increasing debt would have a positive effect on growth in the short
run, which in line with conventional Keynesian multipliers effect. Hence, my result supports the idea that
increasing debt stimulates the economy only when the initial debt level is below certain threshold. However,
one should be caution when putting these results in practice because every country is unique in the sense
Further research can be done to investigate mid-term and long term effect debt accumulation and growth.
Whether similar threshold effect exist in Asian and North and South Americas also worth study. The channel
which increase in debt effects growth should also be studied as policy maker than introduce policies through
these channels to counter the negative effect of debt on growth. The effect of foreign debt on growth is also
an intriguing topic.
20
Reference
Ardagna, S., F. Caselli and T. Lane (2007) “Fiscal Discipline and the Cost of Public Debt Service: Some
Estimates for OECD Countries,” The B.E. Journal of Macroeconomics: Vol. 7: Iss. 1 (Topics), Article 28.
Aschauer, D. A. (2000). “Do states optimize? Public capital and economic growth”. The Annals of Regional
Science, 34(3), 343-363.
Baum, A., Checherita-Westphal, C., & Rother, P. (2012). “Debt and Growth: New Evidence for the Euro
Area”. ECB Working Papers.
Cecchetti, S., Mohanty, M., & Zampolli, F. (2011). “The real effects of debt”. BIS working papers.
Chang, T., & Chiang, G. (2009). “ Transitional Behavior of Government Debt Ratio on Growth: Case of
OECD Countries”. Institute for Economics Forecasting.
Checherita-Westphal, C.,& Rother, P. (2012). “The impact of high government debt on economic growth and
its channels: An empirical investigation for the euro area”. European Economic Review, 56(7), 1392-1405.
doi:10.1016/j.euroecorev.2012.06.007
Chalk, N. & V. Tanzi (2004), “Public debt and economic growth. Channels of the longterm impact” in “The
behaviour of fiscal authorities: stabilisation, growth and institutions”, edited by M. Buti, J. von Hagen and
C. Martinez-Mongay.
Cohen, D. (1997), Growth and external debt: A new perspective on the African and Latin American tragedies,
Centre for Economic Policy Research Discussion Paper No. 1753.
Diamond, P. (1965). “National Debt in a Neoclassical Growth Model”. American Economic Association,
55(5), 1126-1150.
Égert, B. (2015). “Public debt, economic growth and nonlinear effects: Myth or Reality?”. Journal of
Macroeconomics, 43, 226-238. doi:10.1016/j.jmacro.2014.11.006
Hamilton, E.J. (1947), “Origin and Growth of the National Debt in Western Europe”, The American Eco-
nomic Review, Vol. 37(2), Papers and Proceedings of the Fifty-ninth Annual Meeting of the American
Economic Association, pp. 118-130.
Laubach, T. (2009), “New Evidence on the Interest Rate Effects of Budget Deficits and Debt”, Journal of
the European Economic Association, Vol. 7(4), pp. 858-885.
Meade, J. E. (1958), “Is the National Debt a Burden?” Oxford Economic Papers, New Series, Vol. 10(2),
pp. 163-183.
Miller, M., & Modigliani, F. (1961). “Dividend Policy, Growth, and the Valuation of Shares”. Chicago
Journals , 34(4), 411-433.
Padoan, P. C., Sila, U., & Noord, P. V. (2012). “Avoiding Debt Trap: Financial Backstops and Structural
Reforms”. Organisation for Economic Co-operation and Development working papers.
Pescatori, A., Sandri, D., & Simon, J. (2014). “Debt and Growth: Is There a Magic Threshold?”. IMF
Working Papers, 14(34). doi:10.5089/9781484306444.001
Reinhart, C., & Rogoff, K. (2010). “Growth in a Time of Debt”. NBER Working Papers. doi:10.3386/w15639
21
Saint-Paul, G. (1992). “Fiscal Policy in an Endogenous Growth Model”. The Quarterly Journal of Eco-
nomics, 107(4), 1243-1259.
Schclarek, A. (2004). “Debt and Economic Growth in Developing and Industrial Countries”. Department
of Economics Lund University working papers.
Smyth, D. & Hsing, Y. (1995), “In search of an optimal debt ratio for economic growth”, Contemporary
Economic Policy, 13:51–59
22
Appendix 1
Note: Countries included are Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech, Denmark, Estonia,
Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta,
Netherlands, Norway, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, and United Kingdom.
d 60.29 31.91
IN T l 4.39 2.35
IN T s 0.540 2.74
23
Appendix 2
24
(0.2974) (1.0385) (0.3323)
Country Dummy Included (30) Included (30) Included (30) Included (5) Included (9)
Note: The dependent variable is GDP Growth. The abbreviations for the explanatory variables are explained in Table 1.1, Appendix 1. Countries
included in complete data, before financial crisis, after financial crisis are Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech, Denmark, Estonia,
Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Poland, Portugal,
Romania, Slovakia, Slovenia, Spain, Sweden, and United Kingdom. Countries included in low credit rating are Bulgaria, Croatia, Cyprus, Hungary,
Italy, and Romania. Countries included in high credit rating are Austria, Germany, Sweden, Netherlands, Norway, Luxembourg, Belgium, Denmark,
Finland, and France. The table shows the estimated coefficients, standard errors (SE) which are in parentheses and their significance level (*10%;
**5%, ***1%). The main variable of interest is the indicator on debt to GDP ratio above threshold.
Table 2.2: Regression Result from Equation 10
Variable Complete Data Before Financial Crisis After Financial Crisis Low Credit Rating High Credit Rating
25
−4.542∗∗∗ −4.6208∗∗∗ 4.0658∗∗∗
I(P OST 2008)
(0.5862) (1.6624) (1.1471)
Country Dummy Included (30) Included (30) Included (30) Included (5) Included (9)
Note: The dependent variable is GDP Growth. The abbreviations for the explanatory variables are explained in Table 1.1, Appendix 1. Countries
included in complete data, before financial crisis, after financial crisis are Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech, Denmark, Estonia,
Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Poland, Portugal,
Romania, Slovakia, Slovenia, Spain, Sweden, and United Kingdom. Countries included in low credit rating are Bulgaria, Croatia, Cyprus, Hungary,
Italy, and Romania. Countries included in high credit rating are Austria, Germany, Sweden, Netherlands, Norway, Luxembourg, Belgium, Denmark,
Finland, and France. The table shows the estimated coefficients, standard errors (SE) which are in parentheses and their significance level (*10%;
**5%, ***1%). The main variable of interest is one year lagged debt to GDP ratio and squared one year lagged debt to GDP ratio.
Table 2.3: Long Term Interest Rate as Dependent Variable
Variable Complete Data Before Financial Crisis After Financial Crisis
∗
0.01002 0.0273∗ 0.2293
I(dt−1 > d )
(0.2035) (0.01992) (0.3288)
−1.6779∗∗∗
I(P OST 2008)
(0.1876)
Note: The dependent variable is long term interest rate. The abbreviations for the explanatory variables
are explained in Table 1.1, Appendix 1. Countries included in complete data, before financial crisis, after
financial crisis are the same as in table 2.2. The table shows the estimated coefficients, standard errors (SE)
which are in parentheses and their significance level (*10%; **5%, ***1%). The main variable of interest is
indicator on debt to GDP ratio over threshold.
Note: The dependent variable is GDP growth. Explanatory variables are the same as in table 2.1. Model 1
excludes Estonia. Model 2 excludes Greece. Model 3 excludes Estonia, Luxembourg, and Bulgaria. Model
4 excludes Greece, Italy, and Belgium. Model 5 excludes Estonia, Luxembourg Bulgaria, Greece, Italy, and
Belgium. The table shows the estimated coefficients, standard errors (SE) which are in parentheses and their
significance level (*10%; **5%, ***1%). The main variable of interest is indicator on debt to GDP ratio over
threshold.
26
Table 2.5: Regression Result Without Influential Data Using Equation 10
Variable Model1 Model2 Model3 Model4 Model5
0.06053∗∗ 0.08061∗ 0.06273∗ 0.09009 0.03440
dt−1
(0.03008) (0.04773) (0.04521) (0.07525) (0.05227)
2 0.0002587∗ 0.0002757∗ 0.0002681 0.0004058 −0.0001592
dt−1
(0.0001536) (0.0001688) (0.0001792) (0.0003111) (0.001555)
0.1536∗∗∗ 0.1655∗∗∗ 0.1177∗∗∗∗ 0.1201∗∗∗ 0.1351∗∗∗
GROW T Ht−1
(0.04468) (0.04453) (0.04378) (0.04371) (0.04874)
0.1497 0.3161 0.06979 0.2215 0.2433
P OP GROW T Ht
(0.1049) (0.3164) (0.1066) (0.3090) (0.3343)
−0.3160∗∗∗ −0.3040∗∗∗ 0.3993∗∗∗ −0.3271∗∗∗ −0.2658∗∗∗
GOV SIZEt
(0.04182) (0.04125) (0.04380) (0.04237) (0.04484)
−5.0439∗∗∗ −4.8130∗∗∗ 4.4910∗∗∗ 4.0728∗∗∗ −3.3903∗∗∗
I(P OST 2008)
(0.6660) (0.7880) (0.7871) (0.7489) (0.9786)
0.04303∗∗∗ 0.04191∗∗∗ 0.08859∗∗∗ 0.03539∗∗∗ 0.04191∗∗∗
I(P OST 2008) ∗ dt−1 )
(0.01456) (0.01593) (0.01544) (0.01462) (0.01859)
Country Dummy Included (29) Included (29) Included (27) Included (27) Included (24)
Debt Threshold 117% 115% 117% 111% 108%
Note: The dependent variable is GDP growth. Explanatory variables are the same as in table 2.2. Model 1
excludes Estonia. Model 2 excludes Greece. Model 3 excludes Estonia, Luxembourg, and Bulgaria. Model
4 excludes Greece, Italy, and Belgium. Model 5 excludes Estonia, Luxembourg Bulgaria, Greece, Italy, and
Belgium. The table shows the estimated coefficients, standard errors (SE) which are in parentheses and their
significance level (*10%; **5%, ***1%). The main variable of interest is one year lagged debt to GDP ratio
and squared one year lagged debt to GDP ratio.
Table 2.6 Growth of GDP per Capita as Dependent Variable Using Equation 9
Variable Complete Data Before Financial Crisis After Financial Crisis
∗ −0.7343∗∗ −0.9851∗∗∗ 0.2870
I(dt−1 > d )
(0.4521) (0.3592) (0.3777)
0.1687∗∗∗ 0.3626∗∗∗ 0.05553∗∗
GROW T Ht−1
(0.04787) (0.07115) (0.02615)
−0.4933∗∗∗ −0.2445∗∗∗ −0.3546∗∗∗
GOV SIZEt
(0.04272) (0.03220) (0.07112)
−3.8418∗∗∗
I(P OST 2008)
(0.3788)
∗ −1.7010∗
I(P OST 2008) ∗ I(dt−1 > d )
(1.0035)
Country Dummy Included (30) Included (30) Included (30)
Debt Threshold 108% 71% NA
Note: The dependent variable is GDP per capita growth. Explanatory variables are the same as in table
2.1. The table shows the estimated coefficients, standard errors (SE) which are in parentheses and their
significance level (*10%; **5%, ***1%). The main variable of interest is indicator on debt to GDP ratio over
threshold.
27
Table 2.7 Growth of GDP per Capita as Dependent Variable Using Equation 10
Variable Complete Data Before Financial Crisis After Financial Crisis
0.1335∗∗ 0.1243∗∗∗ −0.03548
dt−1
(0.03635) (0.02424) (0.03920)
2 0.0006181∗ 0.0008071∗ 0.8795∗∗∗
dt−1
(0.0003887) (0.0004928) (0.1962)
0.1599∗∗∗ 0.3178∗∗∗ 0.04729
GROW T Ht−1
(0.05643) (0.07985) (0.04534)
−0.4398∗∗∗ −0.2019∗∗∗ −0.3911∗∗∗
GOV SIZEt
(0.05329) (0.03918) (0.06774)
−4.3837∗∗∗
I(P OST 2008)
(0.4651)
0.03211∗∗∗
I(P OST 2008) ∗ dt−1
(0.01059)
Country Dummy Included (30) Included (30) Included (30)
Debt Threshold 115% 77% NA
Note: The dependent variable is GDP per capita growth. Explanatory variables are the same as in table
2.2. The table shows the estimated coefficients, standard errors (SE) which are in parentheses and their
significance level (*10%; **5%, ***1%). The main variable of interest is one year lagged debt to GDP ratio
and squared one year lagged debt to GDP ratio.
28
Appendix 3
70
Average Debt to GDP Ratio (%)
60
50
40
3
Average Economic Growth
−3
29
Figure 3.3 Average Economic Growth by Year
GDP Growth Rate Versus Debt to GDP Ratio (Cohort by Year and Credit Rating)
Low Credit Rating High Credit Rating
20
●
● ●●● ● ● ● ●
●
GDP Growth Rate
10 ● ●
● ● ● ●
●● ● ● ● ●
●
●
● ● ●
● ● ●● ●
●●●
● ● ●
● ●●●
● ●●
●
●
●
● ●
● ● ●
●●●● ● ●
● ●
● ● ● Before 2008
●
●●●●●●
● ●● ●●
● ●● ● ●
● ●●●●
● ●●
●● ●●●●●●● ●● ●● ●
●●●
● ●●● ● After 2008
●
●●●●
●●
●
●●
● ●
●
●●
●
●●●
●●●● ●● ●● ● ● ●● ● ● ● ●●●● ●
●● ●
● ●●● ● ●● ●
●
●●●
●●●
●●
●
●
●●●
●
●●
●●
●●
●●
●●
● ●
● ● ●
●
● ●
●●● ●
●● ● ● ● ●
●●
●●●
●●
●● ● ●●●
●
●●●● ●●●●● ●●● ●●● ●●
●● ● ●● ● ● ●
●●
●●● ●● ●
● ● ●● ●
● ●● ●●
●
●●
● ● ●●● ●
●●● ●
● ● ●●● ● ●
●● ● ●●
●
● ●● ●● ●●● ●●●
●● ●
●
● ●
●● ●
● ●
●●
● ● ● ●●● ●
●
●
● ●
●
● ● ● ●● ●
● ●●
●
● ●● ●●
● ●
●
●
●
● ●● ●● ●
● ● ● ●●● ●● ● ●● ● ●● ●● ● ● ●● ● ●
● ●●
●● ●
● ●
●
0 ● ● ●
● ●● ●
● ●
●
● ● ●● ● ●●● ●●● ●● ●
●●
●
●
●● ● ● ● ● ●
● ●● ● ● ● ● ●
● ● ● ● ●●● ● ● ●
●● ● ● ● ● ●
● ●● ●
●● ● ●●
● ● ● ● ● ● ●
●●
● ●● ●
●
−10
● ●●
1
Coefficients
−1
−2
50 100
Threshold Value
30
Figure 3.5 Coefficient On Debt Threshold (Before Financial Crisis)
1
Coefficients
−1
−2
25 50 75 100
Threshold Value
3
Coefficients
40 80 120
Threshold Value
31
Figure 3.7 Coefficient On Debt Threshold (Low Credit Rating)
0.1
Coefficients
0.0
0 40 80
Threshold Value
2
Coefficients
80 100 120
Threshold Value
32
Figure 3.9 Coefficient On Debt Threshold (Full data, Long Term Interest Rate as Dependent Variable)
Coefficients on Debt Thresholds
6
4
Coefficients
−2
40 80 120
Threshold Value
Figure 3.10 Coefficient On Debt Threshold (Before Financial Crisis, Long Term Interest Rate as Depen-
dent Variable)
Coefficients on Debt Thresholds Pre 2008
1
0
Coefficients
−1
−2
40 80 120 160
Threshold Value
33
Figure 3.11 Coefficient On Debt Threshold (After Financial Crisis, Long Term Interest Rate as Dependent
Variable)
2.5
Coefficients
0.0
−2.5
40 80 120 160
Threshold Value
Figures corresponds to estimations in section 5.2 and 5.3 are very similar to figure 3.4, figure 3.5 and
figure 3.6. One can refer to these figures for estimations in section 5.2 and 5.3
34