1
The effect of terrorism on economic growth, government
spending and private investment: a quantitative analysis of 35
European countries
Max Kuipers – 10659110
BSc Economics and Business
Specialisation: Economics and Finance
Thesis supervisor: dhr. dr. E. Westerhout
2
Statement of Originality
This document is written by Max Kuipers who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Acknowledgements
I would like to thank dr. E. Westerhout, my thesis supervisor, for his guidance on writing this paper and providing the necessary feedback and comments.
4
Abstract
As terrorism is an issue with many consequences, this thesis analyzes the effect of terrorism on several macroeconomic indicators for 35 European countries in a time frame from 1997 to 2011. The data used in this research is retrieved from the Global Terrorism Database and the World Bank. By using panel regressions with time and country fixed effects and other control variables on these data, the effect of the number of terrorist attacks per million persons is tested on growth of GDP per capita, the investment ratio and the government spending ratio.. The results suggest a weak significant negative effect of terrorist attacks on economic growth, as well as a strong significant adverse effect on the investment ratio. Regarding the government spending ratio, a strong significant positive effect is detected, indicating a shift of private
investment towards government spending related to terrorism.
Key words: terrorism, economic growth, investment ratio, government spending
5
Contents
1. Introduction ... 6
2. Literature Review ... 8
2.1 Definition of terrorism ... 8
2.2 Theory of terrorism and macroeconomic indicators ... 8
2.3 Existing literature ... 9 3. Methodology ...12 3.1 Data specification ... 12 3.2 The models ... 13 3.3 Hypotheses ... 15 4. Results ...16 4.1 Descriptive statistics... 17
4.2 Analysis of effect on growth of GDP per capita... 18
4.3 Analysis of effect on investment ratio. ... 19
4.4 Analysis of effect on government spending ratio. ... 21
4.5 Additional regressions ... 22
4.6 Summary of results ... 23
5. Discussion and limitations ...24
6. Recommendations ...25
7. Conclusion ...26
8. Bibliography ...27
6
1. Introduction
In light of the recent terrorist attacks, terrorism seems to be a modern issue growing faster. Attacks in Paris (13-11-2015 and 07-01-2015), Brussels (22-03-2016) and most recently in the United States (12-06-2016) are just a few of the numerous attacks occurring every year. The Global Terrorism Database (GTB) of the University of Maryland (2013) registered approximately 16,800 attacks worldwide in 2014 alone. However, terrorism is a phenomenon that has been present for several decades. The GTB has registered terrorist attacks since 1970. Subsequently, the GTB also illustrates - in graph 1- a significant increase of terrorist attacks for 35 European countries of approximately 231 per cent from 2004 to 2011.
Terrorism can take many forms, like bombings, hijacking, shootings and so on. This great variety of terrorist attacks causes a diversity of consequences; the human fatalities, the wounded, the increasing fear of terrorist attacks and many sorts of physical damage. Aside from these horrific consequences, there are possible economic consequences that are of importance as well.
On this topic, the impact of terrorism on a macroeconomic level, rather little has been written. Some literature exists on the macroeconomic influence of conflict, yet most of this literature and theory focuses on the economic consequences of war as the specification of conflict. This however, does not include terrorism, even though it is a vastly growing reality.
The literature that exists regarding the macroeconomic consequences of terrorism are investigations on a global level, where terrorism significantly and negatively affected macroeconomic indicators. However, little to no research was conducted on the macroeconomic effects of terrorism in Europe, whereas Europe is often a target of terrorism. Additionally, there is literature on the countries being most often targeted by terrorism. The results suggest targeted countries of terrorism often have high income, high democracy and more openness.
Looking at these results, a possible macroeconomic effect of terrorism could be detected for Europe as well, which is of importance for the determination on the measurements taken against terrorism. As a result, the following central research question is introduced: What is the effect of terrorism on economic growth, government spending and private investment for European countries?
This paper hypothesizes the effect of terrorist attacks per million persons on economic growth to be negative, by defining several channels through which terrorism effects economic growth of gross domestic product (GDP) per capita. Secondly, with respect to the investment ratio, this analysis hypothesizes a negative effect too. Finally,
7 the effect of terrorist attacks on government spending ratio is tested on a positive effect.
A secondary database is constructed by combining the GTB database on terrorist attacks in Europe and the database of the World Bank regarding macroeconomic
parameters and variables. Subsequently, a panel regression with fixed effects is performed on these 35 countries over a timeframe of 15 years, particularly 1997 to 2011. In addition, for comparison reasons, an Ordinary Least Squares (OLS) regression and a panel regression with only time fixed effects are included as well. Besides these regressions, the same panel regression with fixed effects is performed without possible outliers as a check for robustness.
This research will firstly review the existing literature on terrorism and the economic consequences. Then the research method is comprehensively described in section 3, the methodology. Afterwards, an analysis of the results retrieved from the regression will be provided, followed by the limitations and discussion regarding the interpretations of these results and recommendations for further research with respect to the economic consequences of terrorism. Ultimately, the interpreted results will be used to determine the effect of terrorism on macroeconomic variables in the conclusion.
Graph 1: Number of terrorist attacks between 1997 and 20111.
Note: Number of attacks per year is the total number of attacks in a sample of 35 European countries used in this analysis.
1 For exact numbers, see table 7 in the appendix.
1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 N u mb e r o f te rr o ri st a tt a ck s 1995 2000 2005 2010 Year
8
2. Literature Review
In this section, the literature regarding to the subject of this research will be discussed. First of all, the definition used in this analysis for terrorism is constructed. Secondly, the theory of the effect of terrorism on macroeconomic indicators and other determinants for economic growth will be elaborated. Finally, the existing literature on the
macroeconomic consequences of terrorism and the findings will be discussed.
2.1 Definition of terrorism
When determining the effect ofterrorist events on macroeconomic indicators as growth of GDP per capita, government spending and private investment, it is of importance to define terrorism first. Using the Global Terrorism Database (2013) of the University of Maryland, they describe their definition of terrorism as followed: “the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation”. Nonetheless, the GTB (2015) acknowledges the ongoing debate regarding the true definition of terrorism.
For instance, Blomberg et al. (2004) derive a similar definition from Mickolus et al. (2002), yet they exclude economic, religious and social purposes. Ruby (2002) distinguishes terrorism from other forms of violence using three criteria: it is politically motivated, it is directed at non-combatants and the perpetrators are subnational groups or clandestineagents.
To distinguish terrorism from war, Blomberg et al. (2004) use two criteria for the consideration of a war; there has to be a mobilization of more than 1000 individuals as well as the occurrence of at least 100 fatalities. For the conventional reason that the GTB is used, the definition of the GTB, added by the two distinguishing criteria of war, will be used.
2.2 Theory of terrorism and macroeconomic indicators
With terrorism now being defined, the channels through which terrorist events affect macroeconomic indicators are to be characterized. Abadie and Gardeazabal (2008, p.21) state that terrorism may cause substantial shifts of capital across countries. Their results argue that terrorist risks damp net foreign investment positions and thus foreign direct investment. Blomberg and Mody (2005, p. 18) point out that violence, including terrorism, adversely affects foreign direct investments and even discourages trade with economic significance.
Not only foreign direct investment is affected by terrorist attacks and risks, but domestic investment as well. Blomberg et al. (2004) empirically show, in their research
9 covering 177 countries, a shift of private investment towards government spending is correlated with the increase of transnational terrorism. This redirection is also stated by Gaibulloev and Sandler (2008), whereas they distinguish the effects of domestic and transnational terrorism on private investment. They show a transfer of growth-enhancing private investment towards less growth-promoting expenses by the government on for example defence and counterterrorist actions.
Another channel is the impact of terrorism on major key sectors as trade, transportation and tourism. Blomberg and Mody (2005) suggested a discouragement of trade with an increase of terrorist risk. Another study by Abadie and Gardeazabal (2003) researching the influence of terrorist attacks on Basque country, a high
terrorism risk area, indicates a negative performance of Basque stocks relative to non-Basque stocks at the end of cease-fire. Enders et al. (1992) demonstrate that, for European countries, terrorism daunts tourism revenues of the targeted country and even neighbouring nations, described as a generalization effect. The transportation sector may incur inefficiency due to legislation in order to reduce risk of terrorism. Palac-McMiken (2005) states the assumption that the risk of terrorism reduces productivity in transportation due to the diversion of resources towards security.
Some theory regarding the determinants of economic growth needs to be
considered as well. Barro (2003) provides some factors influencing real GDP per capita growth. He refers to Mankiw, Romer and Weil (1992) who empirically show the positive effect of the investment ratio on growth of real GDP. Fischer (1993) confirms that the rate of investment is, among others, an explanatory variable in the standard mixed regression for economic growth. Moreover, Barro (2003) provides a significant positive effect of trade openness, determined as the sum of the import and export ratio.
Another factor that is of importance in determining the growth of GDP per capita is the initial level of GDP per capita. The concept of convergence is the reason for this factor to be of influence. In the Solow-growth Model it is derived that every country eventually converges to its long-run balanced path of growth (Romer, 2012). This convergence is influenced by the initial level of GDP per capita. Romer, Mankiw and Weil (1992) suggests that poorer nations grow faster relative to richer countries, which is defined as conditional convergence. This convergence is also incorporated in the models used by Blomberg et al. (2004) and Gaibulloev and Sandler (2008), where they both use the logarithmic value of the initial level of GDP per capita.
2.3 Existing literature
10 GDP per capita, there is some literature. Blomberg et al. (2004) researched the impact for 177 countries in both cross-country and panel regressions. Their findings suggest that there is a significantly negative effect of terrorism on growth of GDP per capita. Nonetheless, they found a smaller insignificant negative effect of terrorism on countries part of the Organisation for Economic Co-operation and Development (OECD) or advanced economies. Whereas Blomberg and Hess (2005) indicate that the incidence of terrorism increases for targeted countries with democratic institutions, high income and more openness. Considering OECD and European countries rank high on these
characteristics, the insignificant effect found for OECD countries by Blomberg et al. (2004) can be considered somewhat contradictory.
Gaibulloev and Sandler (2008) used a similar model to research the difference in effect of domestic and transnational terrorism on 18 Western European countries, and point out that transnational terrorism has a greater marginal effect on the growth of GDP per capita than domestic terrorism. This effect is rather modest and they refer to Enders and Sandler (2006) who assert that mature economies are sufficiently
diversified to endure intermediate terrorist events without considerable economic effects.
Meierrieks and Gries (2012) discuss the effect of terrorism on economic growth (and vice versa) by performing F-statistic tests on the causality. They conclude, with a strong significance level (p-value<0.01) a causal effect of terrorism on the growth rate of real GDP per capita, particularly for African and Islamic countries with low levels of political institution development, major political instability and persistent terrorist activity. The significant causal relationship between terrorism and the growth rate of GDP per capita is considered destructive for the period of 1992 to 2007, but they do not indicate the size of this negative effect.
Some case studies can be found as well. For example, Eckstein and Tsiddon (2004) researched the impact of terrorism in Israel and suggested that the GDP per capita would be 10-15% higher in the third quarter of 2003 with the absence of
terrorism in the three years before. They predict a further decrease per capita output of 3-5% relative to 2003 for the two consecutive years with the presence of intensive terrorism.
Another case study by Abadie and Gardeazabal (2003) shows a 10 per cent average gap over twenty years between Basque GDP per capita and a similar area with the absence of terrorism. However, both of these country-specific researches should be viewed critically, as some may argue Israel or Basque country as nations at war, defining the violence as an international conflict and not terrorism.
11 With the literature regarding terrorism, economic growth and its relationship now being discussed, the next section, methodology, will implement these theories in a framework to construct three models for this research.
12
3. Methodology
This section will discuss the research method conducted for the macroeconomic consequences of terrorism. First, there will be a specification of the data and an
elaboration on the collection of the data used in this research. Secondly, the framework, upon which this research is built, will be introduced. With the framework elaborated, three models can be constructed to determine the effect of terrorism on economic growth of GDP per capita, the investment ratio and government spending ratio. Furthermore, the regressions that are performed and its assumptions are discussed. Lastly, the hypotheses of this research will be discussed, followed by the expected conclusions.
3.1 Data specification
By using the definition of terrorism as described in the literature review, the Global Terrorism Database (GTB) of the University of Maryland can be used, as it implements this definition as well. The GTB registered every terrorist attack worldwide and adds, among others, a characteristic “doubt”. This is incorporated in the database using a dummy variable (doubtter [0,1]), which equals 1 when there is doubt about the
terroristic nature of the attack, and zero otherwise. Within this analysis, the only attacks used in our database will be those attacks without doubt of terroristic nature (doubtter =0).
The data regarding the macroeconomic indicators originated from the World Bank database. Among these indicators are growth of gross domestic product (GDP) per capita, private investment ratio, government spending ratio, total population, export and import as percentage of the GDP and the GDP per capita. GDP per capita and the initial level of GDP per capita are denoted in US dollars with constant prices in 2005. Unfortunately, the data for some countries in the sample are incomplete before 1996, as some of the countries where still part of former Yugoslavia. As a result, the chosen time frame is between 1997 and 2011.
In addition, an attempt was made to gather the data regarding security spending by the government (defined as public order and safety) to investigate the possible positive effect of terrorism on this variable. Sadly, the dataset was incomplete, as not all 35 countries were listed for that database and several years for the available countries were missing as well.
Combining the two datasets and implying minor modifications resulted in the database of this research consisting of 35 European countries within a timeframe from
13 1997 to 20112. The dataset is a balanced panel, as all variables have observations for all 35 countries and all 15 years. Some of the modifications are the calculations of terrorist attacks per million persons and several logarithmic values of macroeconomic indices.
Nonetheless, there is a point of critique regarding the variable
. Namely, it does not indicate the intensity of the attacks. In the database used for this research, there is a great variety in type of attacks, number of fatalities, wounded and target types. These differences in terrorist attacks also have differences in the size of the economic impact.
For example, a possible measurement base could be the number of deaths. This, however, imposes the problem of a terrorist attack in the GTB of for instance arson on a factory without fatalities. As there were no fatalities, our measurement would imply a smaller value for the terrorism variable, maybe even zero, whereas the economic impact could be large. This shows the complications that arise with the implementation of this variable in the models, as the coefficient could be biased due to neglecting or over- or undervaluing some terrorist attacks
Unfortunately, it was not possible to process these discrepancies into the models. For conventional reasons, the denotation of terrorist attacks per million persons is used, as Blomberg et al. (2004) and Gaibulloev and Sandler (2008) incorporate this specification of terrorism in their models as well.
3.2 The models
The influence of terrorism will be investigated on three economic indicators. Firstly, the effect on economic growth will be researched, more specifically the effect on growth of GDP per capita. Other indicators that will be investigated are the government spending and the private investment, both as a percentage of GDP, as the literature suggests a redirection of private investment towards government expenditure associated with terrorism.
Most of the framework for this research is provided by Barro (2003). He discusses the determinants of economic growth, some of which will be incorporated as control variables in this analysis. As a result, three models can be constructed similar to the previously mentioned research by Blomberg et al. (2004) by using several
regressions on panel data with fixed effects:
( ) ( )
(1)
14
(2)
(3)
Where is the growth of GDP per capita, is the natural logarithm of initial level of GDP per capita, is the private investment as percentage of GDP (I/GDP), is the natural logarithm of export and import (or trade) as a percentage of GDP (Trade/GDP), is the number of terrorist attacks per million persons and is the government spending ratio (G/GDP), all for every country i and time t. Other parameters are added to implement the fixed effects, where respectively, and represent the entity fixed effects for every country i and , and represent the time fixed effects for every year t (Stock and Watson, 2015). The rationale behind the choice for fixed effects rather than random effects is based on the fact that Blomberg et al. (2004) and Gaibulloev and Sandler (2008) use it as well for their models on GDP per capita growth, government spending and private investment.
In addition, this analysis also includes an OLS regression and a time fixed only regression. However, the OLS regression ignores country and time differences, whilst the time fixed regression ignores country-specific discrepancies (Stock & Watson, 2015). Finally, in all three models, an error term ( , , ) is included for every country i and every year t.
Stock & Watson (2015) describe panel data as data for n different entities (35 countries) observed at T different time periods (1997-2011, 15 years). The fixed effects regression with both time and entity effects is a method to control for omitted variables in panel data, when some omitted variables vary across the entities but do not change over time and others are constant over countries but vary across time. Moreover, a regression using panel data with fixed effects is subject to four assumptions (Stock & Watson, 2015):
1. have a conditional mean zero: E( | ) = 0.
2. ( ) are independently and identically distributed draws from their joint
distribution.
3. Large outliers are unlikely: have nonzero finite fourth moments. 4. There is no perfect multicollinearity.
If these assumptions hold, then the distribution of the fixed effects OLS estimator of the sample is normal for large values of N and the standard error can be used for t-values. Consequently, these standard errors are implemented in the test for statistical
15 interference, by using a joint hypothesis for the F-statistic (Stock & Watson, 2015). This research performs this OLS regression on panel data with fixed effects under these assumptions.
Additionally, there are two other types of regressions included in this research. Firstly, an OLS regression, without time and fixed effects introduced. However, Stock & Watson (2015) state that by using panel data, omitted variable bias for the coefficients could occur. Secondly, a panel regression is shown with only time fixed effects by introducing t-1 dummy variables ( ) to the aforementioned OLS regression. Where equals 1 at year t and 0 otherwise. The rationale behind adding these other regressions is that some comparisons can be made with respect to the panel regression with both entity and fixed effects. Finally, the regressions with fixed effects are performed without possible outliers – mentioned in section 4.5 - for all three models as well, as a check for robustness.
3.3 Hypotheses
With the models and the form of regressions now being defined, the hypotheses can be formulated. Looking at the impact of terrorism on the macroeconomic indicators, the coefficients will be tested using a one-sided simple test. Put simply, a t-test will examine if the coefficients in the (OLS) panel regression are significantly different from zero, to be able to conclude any effect of terrorism.
Accordingly, the literature suggests a negative effect of terrorism on growth of GDP per capita. As a result, the one-sided hypothesis 1 can be formulated:
(H1)
Secondly, section 2 indicates a negative effect of terrorism on private investment, resulting in the one-sided hypothesis 2:
(H2)
Finally, the theory argues a positive effect of terrorism on government spending, implying the one-sided hypothesis 3:
(H3)
16 For every hypothesis, a significance level of 0.1, 0.05 and 0.01 will be used, indicating weak, standard and strong significance, respectively.
17
4. Results
This section provides the results of the OLS regression with fixed effects for the panel data of 35 countries spread over 15 years, namely 1997 to 2011. Firstly, some basic statistics regarding the data are displayed and interpreted. Thereafter, these findings will be interpreted and linked to the literature and hypothesis introduced in section 2 and 3, respectively. Each subsection will review one model (1, 2 and 3) from section 3, its estimations and results. Finally, the results for the additional regressions without outliers are discussed followed by a summary of the results.
4.1 Descriptive statistics
Table 1:Summary of used variables.
Variable N Mean Std. Dev. Min Max
525 2.91 4.10 -14.56 14.22 525 24.37 4.81 8.87 41.65 525 19.07 3.38 9.45 28.06 525 98.79 46.21 39.27 348.39 Population 525 20.50 30.0 0.42 148.00 525 23178 19434 570 87773 525 0.37 1.64 0 31.16
Notes: is annual growth of GDP per capita, is the initial level of GDP per capita and is denoted in
US dollar prices at 2005. , are denoted in percentages. Population is denoted in
millions. is the number of terrorist attacks per millions persons.
For a general overview of the data, table 1 summarizes the main variables in this analysis and provides some insights on economic indicators, population and terrorism. It shows an average of 0.371735 terrorist attacks per million persons in one year per European country and a maximum of 31.16367 attacks per million persons (Macedonia FYR, 2001).
Even though the sample consists of only European countries, some big differences are detected as well. For example, the minimum value of initial GDP per capita is approximately 570 dollar (Moldova, 2000) and the maximum is 87773 dollar (Luxembourg, 2008), with a standard deviation of 19434.65. Likewise a rather large value for the maximum of the sum of export and import ratio (348.39) is found (Luxembourg, 2008). These rather large values could violate assumption 3 of a panel regression mentioned in the methodology and will be dealt with in section 4.5.
18 Subsequently, table 2 presents correlations between some of the main variables. As the literature suggests, terrorism influences private investment. Since this analysis imposes both terrorism and private investment in model 1, the problem of perfect multicollinearity could arise. Nonetheless, table 2 shows that the correlations of
terrorism ( ) and the other control variables in the models are rather small (between -0.12 and 0.00). This implies that assumption 4 of a regression with fixed effects on panel data is not violated, as there is no perfect multicollinearity (Stock & Watson, 2015).
Table 2: Correlations of used variables.
1 0.40 1 -0.20 -0.27 1 0.10 0.09 -0.13 1 -031 -0.28 0.10 0.28 1 -0.14 -0.12 0.00 -0.11 -0.08 1
Note: correlations are shown between variables used in models 1, 2 and 3.
4.2 Analysis of effect on growth of GDP per capita.
Table 3 shows the results of several regressions on the dependent variable . Model 1.3, with both time and entity fixed effects, indicates strong (***) significant positive effects of 0.53 for the investment ratio ( and 10.74 for the logarithm of the trade ratio ( ), as Barro (2003) suggested. On the other hand, Model 1.3 indicates a strong significant negative coefficient of 13,00 for the logarithm of the initial level of GDP per capita ( ), also in line with the theory. Additionally, the model includes
a constant 65.62, which is strongly significant.
Looking at the variable of interest for this research, a weak significant (p-value = 0.089) negative effect can be detected. As indicated in the section 3.2, is terrorist attacks per million persons. With the coefficient being -0.15, the effect of 1 terrorist attack per millions persons is a 0.15 percentage-point drop in the growth of GDP per capita. Observing Model 1.2, with only time fixed effects, the effect is larger (-0.26) than model 1.3, with a stronger significance level of 0.01.
When interpreting these results, the null hypothesis 1(H1) can be rejected: there is a negative effect of terrorism on GDP per capita, albeit rather small. Moreover, the is rather low (0.16), which illustrates that the model explains only 16% of the variance
19 of . Finally, the F-statistic is significant (7.70***), implying the model is fit for
estimating the data.
Table 3: Results Model 1 Dependent variable Model OLS without fixed effects 1.1
Time fixed effects regression 1.2
Time and country fixed regression 1.3 0.27*** (0.03) 0.21*** (0.03) 0.53*** (0.04) 0.76** (0.40) 0.91*** (0.32) 10.74*** (1.25) -0.92*** (0.14) -0.86*** (0.11) -13,00*** (0.92) -0.27*** (0.01) -0.26*** (0.08) -0.15* (0.09) Constant 1.90 (2.38) 1.44 (2.03) 65.62*** (7.69) (Overall) 0.23 0.54 0.16 Fixed effects
time/country No/No Yes/No Yes/Yes
N 525 525 525
F 40.69*** 33.52*** 7.70***
df 4, 520 18, 506 4, 486
Notes: Standard errors are displayed between parentheses. *, **, *** represent significance levels of 0.01, 0.05 and 0.10, respectively. are denoted in percentages. Model 1.1 is a basic OLS
regression. Model 1.2 includes time fixed effects by adding t-1 dummy variables , for every year t. Where equals 1 in year t and 0 otherwise. Model 1.3 includes both time and country fixed effects and for every country i and every year t, processed by statistic software(STATA). shows the fraction of the sample variance of that is explained by the model. An F-statistic is included to test for significance of
the model. The degrees of freedom for F are given by df (K-1, N-K). 4.3 Analysis of effect on investment ratio.
Table 4 displays the results from multiple regressions on the dependent variable
. Model 2.3 shows a non-significant negative effect of 1.32 for the logarithm of the
trade ratio and a strong significant(***) negative effect of 6.34 the logarithm of the initial level of GDP per capita. Also, a constant with a value of -30.05 is added to Model 2.3, with a strong significance level.
20 As the literature suggests, a strongly significant (p-value = 0.001) negative effect is detected for in model 2.3, however again rather small. For every attack per million
persons, the investment ratio declines with 0.33 or 0.33 percentage point. Model 2.2, with only time fixed effects, shows similar results for the influence of terrorism (-0.33) on the investment ratio with a strong significance too. Model 2.1 suggests approximately the same effect for (-0.35). Also, variable does not lose its significance in model 2.3, whereas this is the case for the regressions for model 1.
These results reject the null hypothesis 2 (H2) and show a negative effect of terrorism on investment ratio, although it is again rather small. Furthermore, the is 0.06 and the F-statistic is significant for a 0.01 significance level.
Table 4: Results Model 2 Dependent variable
Model
OLS without fixed effects
2.1
Time fixed effects regression 2.2
Time and country fixed regression 2.3 1.92*** (0.50) 1.58*** (0.48) -1.32 (1.43) -1.02*** (0.17) -1.06*** (0.16) 6.34*** (1.01) -0.35*** (0.12) -0.33*** (0.12) -0.33*** (0.10) Constant 25.52*** (2.81) 26.14*** (2.88) -30.05*** (8.70) (Overall) 0.10 0.22 0.06 Fixed effects
time/country No/No Yes/No Yes/Yes
N 525 525 525
F 19.95*** 8.32*** 19.54***
df 3, 521 17, 507 3, 487
Notes: See table 4. In addition, model 2.1 is a basic OLS regression. Model 2.2 includes time fixed effects by adding t-1 dummy variables , for every year t. Where equals 1 in year t and 0 otherwise. Model 2.3 includes both fixed effects and for every country i and every year t, processed by statistic software (STATA).
21
4.4 Analysis of effect on government spending ratio.
As indicated below, table 5 provides the results for the various regressions on the dependent variable . Model 3.3, with both time and fixed effects, shows a strong
significant negative effect of the trade as percentage of GDP. Nonetheless, there is no significant effect of the initial level of GDP per capita. A constant is included in the model as well, with a value of 23.33 and a strong significance level.
The variable of interest, , has a strong significant positive effect (p-value = 0.004) on the government spending ratio of 0.14 which was pointed out in section 2. This coefficient is again fairly small. Nevertheless, it shows that for every increase of 1 terrorist attack per million persons, the government spending ratio increases by 0.14 percentage points. Interestingly, compared to models 3.2 and 3.1, the results of models 3.1 and 3.2 show a smaller nonsignificant effect of on .
Table 5: Results Model 3 Dependent variable
Model
OLS without fixed effects
3.1
Time fixed effects regression 3.2
Time and country fixed regression 3.3 -0.67** (0.36) -0.07** (0.37) -1.77*** (0.68) 0.58*** (0.12) 0 .55*** (0.12) 0.39 (0.48) 0.01 (0.09) 0.01 (0.09) 0 .14*** (0.05) Constant 16.62*** (2.03) 17.16*** (2.20) 23.33*** (4.15) (Overall) 0.05 0.07 0.03 Fixed effects
time/country No/No Yes/No Yes/Yes
N 525 525 525
F 8.50*** 2.33*** 5.98***
df 3, 521 17, 507 3, 487
Notes: See table 3. In addition, model 3.1 is a basic OLS regression. Model 3.2 includes time fixed effects by adding t-1 dummy variables , for every year t. Where equals 1 in year t and 0 otherwise. Model 3.3 includes both fixed effects and for every country i and every year t, processed by statistic software (STATA).
22 With these results, the null hypothesis 3 is rejected. There is a negative effect of terrorism on the government spending as a percentage of GDP. Combined with the results in section 4.3 -a decrease of the investment ratio due to terrorism- a redirection from private investment towards government spending associated with terrorism can be detected, which is in line with previous findings from Blomberg et al. (2004) and Gaibulloev and Sandler (2008).
Finally, the variance of explained by model 3.3 is only 2.78 per cent and the F-statistic shows that the model is significant.
4.5 Additional regressions
In section 3, assumption three of the regression with fixed effects on panel data states that large outliers are unlikely. However, as already stated in section 4.1 several
possible outliers are detected. Therefore, as a check for robustness, an unbalanced panel regression, with time and country fixed effects, but without these outliers, for all three models is performed as well. The results are displayed in table 6 below:
Table 6: Fixed effects regressions without outliers on models 1.3, 2.3 and 3.3. Model Dependent variable 1.3 2.3 3.3 0.52*** (0.04) Ln( 10.82*** (1.25) (1.43) -1.46 -1.79*** (0.69) Ln( ) -13.05*** (0.92) 6.13*** (1.01) 0.35 (0.49) -0.37** (0.17) (0.19) -0.77 (0.09) 0.07 Constant 66.00*** (7.71) -27.21*** (8.73) 23.77*** (4.19) (Overall) 0.16 0.06 0.02 Fixed effects
time/country Yes/Yes Yes/Yes Yes/Yes
N 523 523 523
F 81.33*** 21.40*** 2.77**
df 3, 485 3, 485 3, 485
Notes: All models are unbalanced panel regressions with country and time fixed effects, respectively, and , , . Standard errors are displayed between parentheses. *, **, *** represent significance levels of 0.01, 0.05 and 0.10, respectively. are denoted in percentages. shows the fraction
of the sample variance of that is explained by the model. An F-statistic is included to test for
significance of the model. The degrees of freedom for F are given by df (K-1, N-K).
The results for these extra regressions for the variable suggest a larger
23 stronger significance (**). For the investment ratio, the effect of (-0.77 against -0.33)
is again larger, with a strong significance. Regarding the government spending ratio, a smaller effect of 0.07 – against 0.14 – was detected, with no significance (p-value = 0.464). Nonetheless, as there is no clear boundary for values to be outliers, these regressions without outliers are solely included for comparison reasons and as a check for robustness.
4.6 Summary of results
Looking at these results, a strong significant negative effect of terrorism on economic growth of GDP per capita can be found for the regressions with no fixed effects and time fixed effects. Whereas the model of interest, model 1.3, only suggests a weakly
significant negative effect (0.15) of terrorist attacks per million persons on economic growth. As a result, an estimate with respect to the growth of GDP per capita without terrorist attacks can be made. Knowing the average of the 35 European countries is 0.37 terrorist attacks per million persons per year, the annual growth of GDP per capita is estimated to be 0.06 higher, on average, with the absence terrorist attacks.
The effect of terrorism with respect to the investment ratio is strongly significant and negative for all three regressions, with a coefficient of -0.33 for the regressions with time and country fixed effects. Using this estimator, this analysis estimates an average 0.12 percentage point higher investment ratio if no terrorist attacks occurred in these 35 European countries.
Finally, the effect on government spending is strongly significant positive for the time and country fixed effects regression (0.14), whereas models 3.1 and 3.2 show no significant relationship. The annual government spending ratio would therefore be estimated to be 0.05 percentage point lower on average with the absence of terrorist attacks for the 35 European countries.
24
5. Discussion and limitations
When interpreting these results, there are some points of criticism and caution is demanded. In this section, these comments and critiques will be discussed.
First of all, the variable terrorist attacks per millions ( has to be viewed with prudence, as it does not indicate number of terrorist attacks, but the number of terrorist attacks per million persons. For example, for the variable to equal one, there need to be (
=16693074/1000000) 16.69 terrorist attacks in
the Netherlands in the year 2011 alone, for an effect of -0.15 percentage points on growth of GDP per capita. The effect itself is rather small, but the amount of terrorist attacks to attain this small effect is rather large, as the mean of terrorist attacks per million persons is 0.371735 for European countries. These interpretations are also valid for models 2 and 3, as they use the same valuation of the variable .
By using the fixed effects for countries and years, the possible statement that there have been economic crises in 2000 and 2007 or other economic situations in different years for the countries, and consequently a bias, is somewhat eliminated. As stated before in section 3.2, the time effect is incorporated by and the country effect by .
Nevertheless, the results indicate a lower significance for variable in model 1, when adding country fixed effects. This could suggest that the estimated coefficient for this variable could be biased, as this country fixed effects capture a portion of the effect of terrorist attacks.
Another possible bias in the estimator for the coefficient of , is the indirect effect of terrorism on economic growth via private investment. The literature and the results indicate a significant negative effect of terrorism on the investment ratio, whereas the investment ratio is included in model 1 as well. Although the correlation between the variables and is low (-0.12), some indirect effect cannot be ruled out
25
6. Recommendations
As this research provides some insights regarding the influence of terrorism on GDP per capita, government spending ratio and private investment ratio, there is still research to be done with respect to the macroeconomic influence of terrorism.
As discussed in section 5, there are limitations on the research with respect to the measurement of every terrorist attack, as there are discrepancies between the economic impact of every attack. Nevertheless, this is indeed of importance in the analysis of the macroeconomic consequences of terrorism. More research can be done on the influence of terrorism with a more precise measurement of every terrorist attack. With these more accurate measures of the severity of the terrorist attacks, somewhat similar research can be conducted to extract more accurate results.
This analysis provides evidence on the redirection of private investment towards government spending, aligned with findings of the studies by Blomberg et al. (2004) and Gaibulloev and Sandler (2008). However, a possible area to research is a more detailed research concerning this redirection. For instance, the influence of terrorism on the expenditure of the government on public order and safety or even defence, with the war on terror as one of the examples. The other part of this redirection, the private investment decrease, is a possible field to investigate too.
Finally, as mentioned in section 5, an indirect effect of terrorism on GDP per capita growth is suggested by the literature and the results. This indirect effect, due to the relationship between terrorism and private investment, is another possible area to investigate, to prevent bias in the estimators.
26
7. Conclusion
Using the secondary data of terrorist attacks and economic parameters for 35 European countries over a timeframe from 1997 to 2011, this research analyses the effect of terrorism on macroeconomic indicators. By performing a panel regression with fixed effects for both country and time, this study examined the consequences of terrorist attacks per million persons on economic growth, private investment ratio and the government spending ratio.
The hypothesized negative effect of terrorism on economic growth – denoted in growth of GDP per capita – is accepted, although only with a weak significance. The findings suggest a rather small negative effect of 0.15, which represents a drop of 0.15 percentage points of the growth of GDP per capita for every increase of 1 terrorist attack per million persons. Using this estimator and the average of 0.37 terrorist attacks per million persons per year in a country, this analysis suggests on average a decrease of 0.06 percentage points in annual growth of GDP per capita.
Two other models were tested, implementing the same panel regression with fixed effects, to determine the effect of terrorist attacks per million persons on the private investment ratio and the government spending ratio. A strong significant, but again small, negative effect of 0.33 indicates a decrease of 0.33 percentage points of the private investment ratio for every increase of 1 terrorist attack per million persons. On average, the investment ratio, of one of the 35 European countries, is estimated to be 0.12 percentage point higher if no terrorism occurred.
The expected positive effect of terrorism on government spending is confirmed with strong significance too. The results present a coefficient of 0.14, implying an increase of 0.14 percentage points for every increase of 1 terrorist attack per millions persons of the government spending ratio. This confirms the suggested theory of a redirection of private investment towards government spending. Estimating the effects on government spending ratio with the absence of terrorist attacks, this analysis suggests a 0.05 percentage point drop in government spending as percentage of GDP.
Although all the models suggest a significant effect of terrorist attacks, some criticism is needed regarding the interpretations of the coefficient. As stated before, the variable terrorism was denoted as terrorist attacks per millions persons. As the findings are rather small and number of terrorist attacks needed for the increase or decrease, the effect of terrorism on these macroeconomic indicators is to be viewed with caution.
Other research can be directed at the precise shift of private investment towards government spending. Furthermore, an accurate measure of terrorism - indicating the severity of terrorism- is required for a more detailed analysis of this problem.
27
8. Bibliography
Abadie, A., & Gardeazabal, J. (2003). The Economic Costs of Conflict: A Case Study of the Basque Country. The American Economic Review, 93(1), 113-131.
Abadie, A., & Gardeazabal, J. (2008). Terrorism and the world economy. European Economic Review, 52, 1-27.
Barro, R. J. (2003). Determinants of Economic Growth in a Panel of Countries. Annals of Economics and Finance, 4, 231-274.
Bird, G., Blomberg , S. B., & Hess, G. D. (2008). International Terrorism: Causes, Consequences and Cures. The World Economy, 31(2), 225-274.
Blomberg, S. B., & Hess, G. D. (2005). The Lexus and the Oliva Branch: Globalization, Democratization and Terrorism.
Blomberg, S. B., & Mody, A. (2005). How Severly Does Violence Deter International Investment?
Blomberg, S. B., Hess, G. D., & Orphanides, A. (2004). The macroeconomic consequences of terrorism. Journal of Monetray Economics, 51, 1007-1032.
Eckstein, Z., & Tsiddon, D. (2004). Macroeconomic consequences of terror: theory and the case of Israel. Journal of Monetary Economics, 51, 971-1002.
Enders, W., & Sandler, T. (2006). The Political Economy of Terrorism. Cambridge University Press.
Enders, W., Sandler, T., & Parise, G. F. (1992). An Econometric Analysis of the Impact of Terrorism on Tourism. Kyklos, 45(4), 531-554.
Gaibulloev, K., & Sandler, T. (2008). Growth Consequences of Terrorism in Western Europe. Kyklos, 61(3), 411-424.
Johnston, R. (2004). Terrorism and Transportation Policy and Administration: Balancing the Model and Equations for Optimal Security. Review of Policy Research, 21(3), 263-274.
Meierrieks, D., & Gries, T. (2012). Causality between terrorism and economic growth. Journal of Peace Research, 50(1), 91-104.
Mickolus, E., Sandler, T., Murdock, J., & Peter, F. (n.d.). International terrorism: attributes of terrorist events (ITERATE). Vinyard Software, Codebook.
National Consortium for the Study of Terrorism. (2013). The Global Terrorism Database. Retrieved from https://www.start.umd.edu/gtd/:
https://www.start.umd.edu/gtd/search/
Palac-McMiken, E. (2005). Economics Costs and Benefits of Combating Terrorism in the Transport Sector. Asian-Pacific Economic Literature, 19(1), 60-71.
28 Romer, D., Mankiw, N. G., & Weil, D. N. (1992). A Contribution to the Empirics of
Economic Growth. Quarterly Journal of Economics, 107(2), 407-437.
Ruby, C. L. (2002). The Definition of Terrorism. Analysis of Social Issues and Public Policy, 9-14.
Stock, J. H., & Watson, M. W. (2015). Introduction to econometrics (Updated 3rd ed.). Harlow: Pearson Educated Limited.
The World Bank. (2016, 6 3). World Development Indicators. Retrieved from http://data.worldbank.org/:
http://databank.worldbank.org/data/reports.aspx?source=world-development-indicators
29
9. Appendix
Table 7: Number of terrorist attacks in 35 European countries
Year
Number of terrorist
attacks Year terrorist attacks Number of
1996 577 2004 86 1997 496 2005 154 1998 181 2006 133 1999 208 2007 116 2000 321 2008 310 2001 377 2009 336 2002 194 2010 388 2003 178 2011 285
Table 8: Countries used in the sample.
Albania Czech Republic Hungary Netherlands Slovenia
Austria Denmark Ireland Norway Spain
Belarus Estonia Italy Poland Sweden
Belgium Finland Latvia Portugal Switzerland
Bulgaria France Lithuania Romania Ukraine
Croatia Germany Luxembourg Russian Federation Macedonia, FYR
