1
The effectiveness of IMF stabilization
programs in Latin American countries
Empirical analysis of 26 countries between 2000 and 2014.
Name: Maria Fernanda Tituana Vasquez Student number: 10831770
BSc Economics and Business Specialization: Economics
Subject: Bachelor thesis and thesis seminar (Economics) Thesis Supervisor: Stan Olijslagers
2
STATEMENT OF ORIGINALITY
This document is written by Maria Fernanda Tituana Vasquez 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
Abstract:
In order to find out whether fund-stabilization programs supported by the IMF (International Monetary Fund) had led to an increase in economic growth in Latin American countries from 2000 to 2014, this paper uses a panel data of 26 countries covering the period 2000-2014. Half of the countries in the sample belong to Latin America and the other half comes from other regions. Furthermore, a clear explanation of two advanced panel data methods is provided. The Fixed effects and Random effects methods are described and analyzed. The study finds that IMF program participation has an overall negative effect in economic growth. However, Latin American countries experience a higher output cost of IMF program participation.
4
Table of Contents
Chapter 1: Introduction ... 5
Chapter 2: Literature review ... 6
2.1 Alternative approaches to program evaluation ... 6
2.2. Program participation effectiveness ... 8
Chapter 3: Data ... 9 3.1 Data description ... 9 3.2 Panel data…………...………..………..9 3.3 Dependent Variable ... 10 3.4 Explanatory Variables ... 10 3.5 Control Variables: ... 11 Chapter 4: Methodology ... 12 4.1 Estimation Method: ... 12 4.2 Hypothesis: ... 13
4.3 Fixed effects method ………..………13
4.5 Method………..………..………...…..16
4.6 Diagnostic tests………..…..………...…..17
Chapter 5: Empirical Results ... 17
5.1 Descriptive Statistics ... 17
5.2 General Results description ... 17
5.3 Pooled OLS Results ... 17
5.4 Comparison between Pooled OLS, Fixed effects and Random effects results ... 20
5.5 Fixed effects vs Fixed effects with stadard robust clustered errors results ... 20
Chapter 6: Discussion ... 21
Chapter 7: Conclusion ... 22
Chapter 8: Appendix ... 23
5
Chapter 1: Introduction
The International Monetary Fund (IMF) is an organization of 189 countries, who works to promote financial stability, international trade facilitation and sustainable economic growth achievement through loan programs (IMF, 2018). Each country member must give a contribution to the IMF called “quota”. This amount will depend on the size of the member’s economy, each member can freely draw up to 25% of its quota to address balance of payment deficits, but to draw more than 25% of its quota the IMF requires a special agreement (Przeworski, A., & Vreeland, J. R. 2000).
In its 73 years of existence, the IMF has been criticized for its institutional structure and lending programs. It is argued that its fund-supported stabilization programs are ineffective and create moral hazard among its most frequent borrowers (Evrensel, A. Y.2002). The short-term goals of the IMF credit programs include the current account and balance of payments stability; however, its ultimate goal is to provide long run and sustainable economic growth to its members (Przeworski, A. & Vreeland, J. R. 2000). Given the average low rates of economic growth and the wide reach of IMF programs in Latin America, it is essential to evaluate the program effectiveness in the region. Due to, its history of exceptionally high inflation rates, macroeconomic instability and unsuccessful monetary and fiscal stabilization efforts, making them the region with most frequent users of IMF fund supported stabilization programs (Hutchison, M. M., & Noy, I. 2003).
This paper focuses on the effectiveness of fund stabilization programs in Latin America and answers the question: “Have fund-stabilization programs supported by the IMF (International Monetary Fund) led to an increase on the economic growth of Latin American countries from 2000 to 2014?”. This thesis answers the research question by testing a hypothesis through a panel data that covers 15 years for 26 countries while using advanced panel data methods such as Fixed effects and Random effects. This research uses the annual percentage growth rate of GDP as the dependent variable, IMF program participation as the explanatory variable and other control variables. Additionally, this research provides an extra explanatory variable: IMF program participation in Latin American countries.
An examination of the effectiveness of these programs in structurally different regions in recent years can lead to a better understanding of how the effects of these programs vary according to the location and the economic situation of different regions.
Following the introduction, this paper is structured as follows: Chapter 2 covers the literature review, chapter 3 describes the data used and chapter 4 covers the methodology in which the panel data methods are clarified. Chapter 5 covers the empirical results, chapter 6 gives a short discussion, chapter 7 draws a conclusion from the findings, chapter 8 contains
6 the appendix and chapter 9 shows the reference list.
Chapter 2: Literature review
The first subsection of this chapter describes the alternative approaches, methods, and critiques to program evaluations and the second subsection analyses the results of IMF program participation effectiveness from different papers.
2.1 Alternative approaches to program evaluation
The effectiveness of IMF fund supported stabilization programs has been a subject of many kinds of research for a long time. However, most of the scientific papers found are not focused on a specific region meaning that they analyze developing countries in general. Furthermore, they choose different approaches, methods and explanatory variables to explain the IMF effectiveness to achieve its main objectives such as sustainable economic growth, current account and balance of payment equilibrium, which has led to a variety of findings.
According to Evrensel (2002), there are three alternative approaches for the evaluation of programs effectiveness:
The outcome vs. alternative outcome approach, which compares the outcome of the IMF program participation relative to the outcome of an alternative program such as a fund supporting program from a private organization that has a similar degree of adjustment (Evrensel, A. Y. 2002). The main critique and one of the reasons for not using this approach are related to the significant difficulties with estimating a robust alternative model and the provision a meaningful definition of a “similar or different degree of adjustment” (Evrensel, A. Y. 2002).
A second way is called the outcome vs. target approach which determines whether the program objectives have been achieved when comparing it to specific targets, this approach is difficult to implement due to the IMF reluctance to publish the content of individual adjustment programs (Evrensel, A. Y. 2002). However, a more general type derived from this approach is called the output vs. purpose approach. This approach is based on the fact that the fund purpose in adjustments programs is to lead to a sustainable economic growth and reduction of current account and balance of payments problems, in this case, the program has achieved its objectives if visible data can corroborate this assumption (Evrensel, A. Y. 2002).
The studies that used the outcome vs. purpose approach used diverse research methods such as the logit or probit, pooled regressions, and the before and after model to analyze the program effect. The logit and probit method discriminate program years from non-program years. Its main difficulty is to find truly exogenous explanatory variables that are not
7
simultaneously determined to distinguish program years from non-program years. On the other hand, the before and after method takes the difference in the mean values of evaluation variables of program years and non-program years as the program effects, this can be misleading because it implies that non-program years represent the counterfactual macroeconomic performance of program years (Evrensel, A. Y. 2002).
The third way of evaluating adjustment programs is the outcome vs. counterfactual approach which describes the program effect as the difference between the actual performance observed in a program and the performance that would have been taken place in the absence of the program (Evrensel, A. Y. 2002). Goldstein and Montiel (1986), Hutchison and Noy (2003) and Barro and Lee (2005) have used this approach by taking developing country members that have used very few or have not used IMF credit programs at all as the control group (Evrensel, A. Y.2002).
There are critiques of this approach because it has been applied to a program country vs. non-program country comparison, where non-program countries represent the control group. This can be misleading because viewing the control group as the counterfactual indicates that the macroeconomic performance of non-program IMF members represent the macroeconomic performance of program countries in the absence of an IMF program, but being a program or a non-program country is a self-selected attribute that is determined by a specific country situation (Evrensel, A. Y. 2002).
The outcome vs. counterfactual approach used by Goldstein and Montiel (1986) take pooled regressions with a program year dummy variable as the method to capture the program effect on country economic growth for program and non-program countries through the sign and the significance of the program dummy. There are critiques made to this method, first, it is suggested that the dependent and independent variables do not have a clear direction of causality and second, there is also an implied assumption that program years and non-program countries represent the counterfactual for non-program years and countries.
The discussion of approaches and evaluation methods showed that all methods and approaches have several critiques, which means that the existence of a perfect approach and method have not been discovered yet, however it is necessary to know the critiques and problems for a correct method selection and results interpretation. This paper uses the outcome vs. counterfactual approach with a fixed effects regression method, first applied by Barro and Lee (2005) for the program evaluation.
2.2 Program participation effectiveness
Ayse Y. Evrensel (2002) took a sample of 91 developing countries for the period 1971-1997, to check whether domestic credit creation, budget deficit, domestic borrowing,
8
inflation rate, current account and capital account deficit are reduced due to participation in IMF fund stabilization programs. The evaluation method is based on the outcome vs. purpose approach, using the observation of relevant variables during pre-program, program and post-program years method. The results of the analysis suggest that fund-stabilization post-programs certainly improve the current account and reserves deficits during the program period. However, it seems that stabilization programs only give a short-term balance of payment relief, and variables such as domestic credit creation, domestic borrowing, inflation rate and budget deficit are not significantly affected during the program years.
Przeworski and Vreeland (2000) took 79 random countries between 1951 and 1990 to check if the IMF participation programs lead to economic growth and analyze why countries enter and stay under IMF programs. The approach used is the outcome vs counterfactual approach, where they compared the performance of countries that had participated vs. countries that did not participate in the programs under the same conditions. The results from the analysis indicate that countries that do not enter in into IMF programs tend to grow faster than those that enter, even if both groups are facing a high domestic crisis or foreign reserves crisis.
Barro and Lee (2005) take data from 130 random countries for the period 1975-2000. It focuses on the relation between a country political economic situation and the IMF’s lending decisions. After the corresponding analysis, they found that IMF loans were larger when the country had a bigger quota and if there was an evident political and economic proximity to the major Western European countries and the United States (Barro, R. J., & Lee,
J. W. 2005). The approach used in this case was the outcome vs counterfactual approach and the evaluation method used is the Fixed effects regression method with a program year dummy variable. The research concludes that the IMF loan participation rate has a significantly negative influence on economic growth.
Hutchison (2003) focused their studies on Latin America, the data corresponds to 67 countries from 1975 to 1997. The study investigates whether the IMF fund-supported stabilization program participation has macroeconomic effects in Latin America, focusing particularly on output growth and balance of payment adjustment. Additionally, a relation between low program completion rates and high IMF program activity in the region is studied. The approach used is the outcome vs counterfactual approach and the evaluation method used is the General Evaluation Estimator (GEE) plus a matching procedure to control for selection biases. The study indicates that IMF programs in Latin America are associated with very small positive output effects in the 1990s, but large negative effects during 1975-1989.
From the data mentioned above, we can conclude that a new study of the fund supported stabilization programs in Latin American countries in the period 2000 and 2014 can give an additional view and fulfill the absence of recent work.
9
Chapter 3: Data
Firstly, there is a general description of the data in 3.1. It is followed by an explanation of the usefulness of panel data in this paper. Then the third, fourth and fifth subsections give a detailed description of the dependent variable, explanatory variables, and control variables.
3.1 Data description
This paper takes program participation as whether a country is under an approved agreement or not, it does not make any differentiation between the agreement type. A panel data set is conducted for 26 countries of which 13 are Latin American and the other 13 come from other regions. From these 13, 8 have 3 or less than 3 approved program arrangements and the rest have more than 3 approved program arrangements within the 15 years of analysis from January 2000 until January 2014. The countries used as a sample for Latin America and other regions are listed in Table 1 in the appendix. These countries have been chosen to have the same number of approved programs and an equivalent number of countries with the high and low rate of IMF program participation. The data is collected on a yearly basis.
3.2 Panel Data
Among the data that has a two-dimensional structure, we have panel data and pooled cross-sectional data. Panel data refers to samples of the same cross-sectional units that are observed at multiple points in time while pooled cross-sectional data observe a random set of countries per year (Wooldridge, J. M. 2009). Wooldridge (2009), describes the advantages of panel data over pooled cross-sectional data. Among them we have that panel data control for individual heterogeneity, meaning that it can control country and time-invariant variables, it has more informative data, variability, and degrees of freedom which gives lower multicollinearity (Wooldridge, J. M. 2009). Because of the advantages mentioned above, this thesis makes use of panel data. Table 2 in the appendix show the descriptive statistics of the panel data.
3.3 Dependent Variable
The dependent variable is the annual percentage growth rate of GDP at market prices based on constant local currency, this variable has a lag of 1 year in order to have a better appreciation of the effects of the explanatory variables. In this research, GDP is defined as the sum of gross value added by all resident producers plus any product taxes, minus subsidies that are not included in the value of the products. This calculation does not make deductions for depreciation or depletion of assets and degradation of natural resources. Data on growth of per capita real GDP is retrieved from the World Bank database.
Figure 1 from the appendix shows the growth rate of GDP per capita over time from 2001 to 2015. From the sample, the average annual growth rate of GDP per capita among the 390 annual observations for the 26 countries is 3.33%. Latin American countries have an average
10
annual growth rate of GDP per capita of 2.67% while the average annual rate of economic growth from other regions in the sample is 4.00% giving a difference of - 1.33%.
Figure 2 from the appendix shows the GDP per capita growth rate for Latin American countries with high and low rate of IMF program participation measured as countries with more than 3 agreements and countries with 3 or less than 3 agreements respectively, over 2001-2015. Their average annual growth rate of GDP per capita is 2.58% for high rate program participation, and 2.72% for the low rate of program participation, giving a difference of -0.133%.
Figure 3 from the appendix shows the GDP per capita growth rate for Other regions with high rate of IMF program participation which obtained 2.88% while Other countries with low rate of IMF program participation obtained 4.70% giving a difference of -1.83%.
3.4 Explanatory Variables
IMF program participation
The participation in IMF programs is taken as the agreement approval in the respective year for each country. Data is obtained from the International Monetary Fund Annual Report. Latin American region
The differentiation between IMF program participation elsewhere and in Latin America allows to clearly appreciate the uniqueness of the IMF intervention in Latin America compared to other regions (Hutchison, M. M., & Noy, I. 2003). Hutchison & Noy (2003) found that a program approval in Latin America over the entire sample led to a significant average reduction of output growth of 1.1%-1.5% points during 1975-1989, but it also led to a small positive output effect in the 1990s. The participation in IMF programs for Latin American countries is taken as the agreement approval in the respective year for each country. The data is retrieved from the International Monetary Fund Annual Report.
3.5 Control Variables
Openness to trade
Barro & Lee (2005) defined the openness to trade variable as the ratio of exports plus imports to GDP, the difference is taken because exports have a positive effect on economic growth which does not contribute with too much information as the one acquired when imports are also been taking into account. A significantly positive effect of openness to trade and economic growth was found by Barro & Lee (2005). Papers that do not investigate the effect of IMF programs on economic growth but mainly investigate the effect of other variables on economic growth as Yanikkaya (2002) found that a 1% increase in trade volume leads to an average GDP per capita increase of 0.18% annually. This research takes openness to trade as the
11
sum of exports and imports of goods and services measured as a share of GDP. Data is obtained from the World Bank’s databank.
Figure 4 from the appendix shows openness to trade rate as a share of GDP over time from 2000 to 2014. From the sample the average annual Openness to trade among the 390 annual observations for the 26 countries is 80.79%. Latin American countries have an average annual Openness to trade of 78.48% while the average annual openness to trade of Other regions in the sample is 89.31% giving a difference of – 10.83%. Additionally, Figure 5 from the appendix shows the Openness to trade rate for Latin American countries with high rate and low rate of IMF program participation. High rate countries have 48.72%, and low rate countries have 97.80% giving a difference of -48.36%. Figure 6 from the appendix show that Openness to trade rate for Other regions with high rate and low rate of IMF program participation. High rate countries obtained 62.89% while low rate of IMF program participation countries obtained 105.82% giving a difference of 42.93%.
Government Consumption
Barro & Lee (2005) found that real government consumption expenditure to GDP ratio has a negative effect on with economic growth, due to its indirect effect on private productivity. Additionally, it is stated that the impact of government expenditure will depend on how it is financed. For instance, government expenditure that comes from taxation finance can lead to a decline in output, investment, and consumption (Ludvigson, S. 1996). Government consumption is taken as final consumption expenditure which includes all government current expenditures for purchases of goods and services as a share of GDP. Data is retrieved from the World Bank’s databank.
Inflation
According to Fisher (1993), inflation has a negative effect on economic growth, it is used as an indicator to measure whether the government has a prudent control of the overall management of the economy. The loss of economic control lead to an unsustainable macroeconomic framework by reducing the efficiency of the price mechanism due to greater price fluctuation and the reduction of investment rate due to high levels of uncertainty (Fischer, S. 1993). Hutchison & Noy (2003) used inflation as a control variable to evaluate the IMF program effects on the economic growth in Latin America and concluded that the increase in inflation significantly reduce output growth. Additionally, Barro and Lee (2005) found that the estimated coefficient of inflation is significantly negative to the growth rate of GDP per capita.
Here, inflation is measured as the change in consumer price index which reflects the annual percentage change in the cost to the average consumer of acquiring a basket of goods and services. Data is retrieved from the World Bank’s databank.
12
At the beginning of this section, the research method is described, and the hypothesis is formulated. This is followed by a description of advance panel data methods, then, there is a description of the method selection. At the end of this section, the diagnostic tests are assessed to check the validity of the model and the fulfillment of the method assumptions.
4.1 Estimation Method
This paper evaluates fund supported stabilization programs based on the outcome vs. counterfactual approach. Therefore, to analyze the effect of IMF stabilization programs on GDP per capita growth rate in Latin American countries the following model is constructed and tested: YYi(t+1) = β0 + β1 *Xopit + β2 *Xnrit + β3 *Xgcit + β4 *Ddit +β5*DLAit +μ i +mμ it (1)
Where, the growth rate of GDP per capita for each country is considered at time t+1 in order to appreciate the effects of the explanatory variables one year after they have been applied, (YYit+1) is explained by openness to trade (Xopit ), inflation rate (Xnrit ), government consumption
(Xgcit), the participation on IMF supported programs (Ddit ) which is a vector of dummy variables
each equal to unity if an IMF program has been approved that year and zero otherwise, the participation on IMF programs in Latin America equal to unity if an IMF program agreement has been approved and if it is a Latin American country and zero otherwise ( DLAi), β0 is the intercept,
(μ i ) and (mμ it) are both error terms. Where (μi) is the country specific error term independent of
time, it represents the effects of all the time invariant variables that have not been included in the model, and (mμ it) is the idiosyncratic error term which varies for each country at each point in time.
4.2 Hypothesis
To understand how the central question will be answered the following hypothesis is tested: H0: β4 + β5 = 0 H1: β4 + β5 ≠ 0
The null hypothesis test whether the overall effect of IMF stabilization programs participation on economic growth in Latin American countries is zero, implying that IMF fund stabilization programs do not have a significant impact on economic growth in Latin American countries. The alternative hypothesis suggests that the coefficient differs from zero, meaning that the overall IMF participation programs have a positive or negative effect on economic growth in Latin American countries.
4.3 Fixed effects method
In order to the perform the Fixed effects method and test the hypothesis for panel time series data, several assumptions need to be made (Wooldridge, J. M. 2009).
13 Assumptions:
1. There is a random sample from cross section
2. Each explanatory variable change over time and there are not perfect linear relationships existing among the explanatory variables.
3. For each t, the expected value of the idiosyncratic error given the explanatory variables in all time periods and the unobservable effect is zero: 𝐸 (mμ 𝑖t|𝑋i, μ i) =0
Under these 3 assumptions, the fixed effects estimator is unbiased and consistent with a fixed T as N → ∞.
4. Var (mμ 𝑖t|𝑋it, μ i) = Var (mμ 𝑖t) = 𝜎2𝑚μ, for all t = 1……, T.
5. For all t ≠ s, the idiosyncratic errors are uncorrelated (conditional on all explanatory variables and μi ): Cov (mμ 𝑖t, mμ 𝑖s | 𝑋i, μ i ) = 0
Under assumptions 1 until 5, the fixed effects estimator is the best linear unbiased estimator. 6. Conditional on 𝑋i and μi, the idiosyncratic errors (mμ𝑖t) are independent and identically
distributed as normal (0, 𝜎2𝑚μ).
Model description:
The Fixed Effect model control for omitted variables in panel data when they are time invariant variables with time invariant effects (Stock and Watson, 2008). To see what this method involves, consider the fixed effect regression model with a single explanatory variable:
Yit = β1 *Xit + μ i +mμ it (2)
Where i=1……, n; t=1….,T; Xitis the value of the first regressor for entity i in time period t, and μi (μi =β0 + β *Xi ) are entity specific intercepts (Stock and Watson, 2008). A method used
to eliminate the fixed effect (μi) is called the fixed effects transformation, where demeaning
variables are used for this purpose (Stock and Watson, 2008).
The process starts by taking the average equation for each i over time:
Y̅i = β1 ∗ X̅i + μi + mμ̅̅̅̅i (3)
The next step in the demeaning process is that the entity specific average is subtracted from each variable. As μ𝑖 is fixed over time, when equation (3) is subtracted from equation (2) the unobserved effect μ𝑖 has disappeared:
14
This step basically gets rid of the unobserved time invariant variables variability and leaves only the entity demeaned variables to analyze (Wooldridge, J. M. 2009). In the final step, the regression is estimated using entity demeaned variables. To sum up, with a fixed effects model, we are analyzing what causes the dependent variable values to change across time so that the estimated coefficients of the fixed effects model cannot be biased due to omitted invariant characteristics (Stock and Watson, 2008).
A general time-demeaned equation for each i will be:
Ÿit = β1Ẍit … … + mμ̈ it (5) The main characteristic of the fixed effect estimator is that it allows for arbitrary correlation between μ i and the explanatory variables in any time period, because of this, any explanatory
variable that is constant over time for all (i) gets step away by the fixed effects transformation (Wooldridge, J. M. 2009).
4.4 Random effects method
Assumptions:
1. Random sample from cross section
2. There are no perfect linear relationships among the explanatory variables
3. For each t, the expected value of the idiosyncratic error given the explanatory variables in all time periods and the unobservable effect is zero: 𝐸(mμ 𝑖t|𝑋i, μ i) =0
4. The expected value of μ i given all explanatory variables is constant: 𝐸(μ 𝑖|𝑋i) =𝛽0
This assumption rules out the correlation between the unobserved effect and the explanatory variables, which is the key distinction between fixed effects and random effects methods. As it is assumed that μ𝑖 is uncorrelated with all elements 𝑋it, then it is possible to include
time-constant explanatory variables in the model (Wooldridge, J. M. 2009). 5. Var (mμ 𝑖t|𝑋i, μ i) = Var (mμ 𝑖t) = 𝜎2𝑚μ, for all t = 1……, T.
6. For all t ≠ s, the idiosyncratic errors are uncorrelated (conditional on all explanatory variables and μi ) : Cov (mμ 𝑖t, mμ 𝑖s | 𝑋i, μ i ) = 0
7. The variance of μ i given all explanatory variables is constant: Var (μ 𝑖|𝑋i) = 𝜎2μ
Under the mentioned random effects assumptions, the random effect estimator is consistent and asymptotically normally distributes as N gets large for fixed T (Wooldridge, J. M. 2009).
15
In the Random effects model the unobserved variables are assumed to be random and uncorrelated with all observed variables of the model (Stock and Watson, 2008). As it is assumed that (μi) is uncorrelated with the explanatory variables in all time periods, then
using a transformation to eliminate (μi) as it was done in the fixed effect method will result
in inefficient estimators (Wooldridge, J. M. 2009).
Model description:
Consider an unobserved effects model which assumes that the unobserved effect (μ𝑖)is uncorrelated with each explanatory variable:
Yit = β0 + β1 *Xit + μ i +mμ it (6)
Then the composite error term is defined as ( 𝜔𝑖𝑡 = 𝑢𝑖 + 𝑚μ𝑖𝑡), therefore it is possible to
write the regression as:
Yit = β0 + β1 *Xit +ωit (7)
Because (μi ) is included in the composite error term in each time period, the 𝜔it are serially
correlated across time, this positive serial correlation in the error term can be substantial (Wooldridge, J. M. 2009). Pooled OLS standard errors ignore this correlation, which leads to obtaining inefficient coefficient estimates (Wooldridge, J. M. 2009). To solve the problem of serial correlation a generalized least square transformation that eliminates serial correlation in the errors is used:
θ = 1 − [σmμ2/(σmμ2+ T σμ2)]1/2 (8)
Which value (θ ) is between zero and one. Then, the transformed equation turns out to be: (Yit – θ Y̅i) = β0 (1- θ ) + β1 *(Xit - θ X̅i ) + (ωit - θ ω̅it) (9)
Where the overbar denotes the time averages, it involves quasi-demeaned data on each variable. When compared with the fixed effects transformation, the random effects transformation only subtracts a fraction of the time average, where the fraction depends on 𝜎2mu, 𝜎2u, and the
number of periods (t). This transformation allows for explanatory variables that are constant over time (Wooldridge, J. M. 2009).
Equation (9) allows to relate the Random effects, Fixed effects and Pooled OLS methods (Wooldridge, J. M. 2009). Pooled OLS is obtained when 𝜃 = 0 and FE is obtained when 𝜃 = 1. In practice 𝜃̂ is never zero or one, if it is close to zero, it means that RE estimates will be close to the pooled OLS estimates (Wooldridge, J. M. 2009). This happens when the unobserved effects
16
(𝑢𝑖) are relatively unimportant. However, it is more common to observe that 𝜃̂ is close to unity, which makes RE and FE estimates similar (Wooldridge, J. M. 2009).
To summarize, as mentioned before the main difference between the two models is that random effects require 𝐸(μ 𝑖|𝑋i) =𝛽0. In this paper, this implies that openness to trade, inflation
rate, government consumption level, IMF program participation and IMF program participation in Latin American countries cannot be correlated with the unobserved effects such as geographical, cultural, and political aspects, however, this is not always reasonable (Wooldridge, J. M. 2009).
4.5 Method
This paper uses the fixed effects model because it effectively accounts for time invariant unobserved factors that are correlated to the independent variables. In the sample used in this paper those unobserved effects can be related to geographical, political and cultural aspects.
To check if this model is suitable for our data, the assumption that independent variables vary from year to year is assessed in Figures 7, 8, 9, and 10 from the Appendix, it shows all independent variables against time for each country. In these figures it is possible to observe that all independent variables vary along the sample time period which is in line with the assumption.
Hence, the econometric model used after demeaning the observations and therefore get rid of the unobserved effects (μi) is specified as:
YŸit+1 = β1 *Xop̈ it + β2 ∗ Xnr̈ it + β3 *Xgc̈ it + β4 *Dd̈it +β5*DL̈Ait + mμ̈ it (10)
4.6 Diagnostic tests
A set of diagnostic tests for the baseline model are performed to justify the choice of methodology. A summary of all tests is included in the Appendix Table 4.
First, a Breush and Pagan Lagrarian multiplier test is performed to check if panel data models are preferred over pooled OLS regressions, the null hypothesis states that
H0: Var (μ 𝑖|𝑋i) = 0, which violates the assumption (7) of the RE method. After testing the data, we obtained that Prob > chibar2 = 0.0000 which means that the variance of the random effect is significantly different from zero, and a pooled OLS regression is not appropriate for the data. Therefore, panel methods need to be considered.
Second, a Hausman test is performed, this test basically assesses whether the unobserved variables are correlated with the regressors which correspond to the assumption (4) in the random effect model described above. The null hypothesis states that 𝐸(μ 𝑖|𝑋i) =𝛽0. The results from the test show that random effects are preferred over fixed effects which means that the unobserved variables are not correlated with the independent variables of the model. Although this result is obtained, this thesis carries fixed effects method as it is not possible to treat this sample as a
17
random sample from a large population, especially as the unit of observation is a large geographical unit (country). Therefore, it makes sense to think of (μi) as a separate intercept to estimate for each cross-sectional unit.
Third, the Modified Wald test for GroupWise heteroscedasticity is performed where the null hypothesis is homoscedasticity. From the test results, we obtained that Prob>chi2 = 0.0000, which implies that the hypothesis of homoscedasticity is rejected.
Furthermore, a Wooldridge test for autocorrelation in panel data is taken. This test assumption (5) in the Fixed effects method, the null hypothesis states that the idiosyncratic errors are uncorrelated overtime. The results from the test are Prob > F = 0.0077, which suggest that there is serial autocorrelation in the idiosyncratic error terms over time.
Normality of residuals is tested by Jarque-Bera test, the null hypothesis of the test is normality of error distribution, from the test results we can observe that Prob > F = 0.0000, so we reject the null hypothesis.
To mitigate the violations mentioned above, Barro and Lee (2005) suggest using fixed effects method with robust and clustered by country standard errors. Barro and Lee (2005) mention that the reason for adjusting standard errors for clustering is that unobserved components in outcomes for units within clusters are correlated. After considering the characteristics of the panel data, there is sufficient evidence to infer that fixed effects with robust and clustered standard errors are a plausible estimation method. The method can mitigate some issues that were discussed above, but, the non-normality of the errors is beyond the scope of this research and will be disregarded.
Chapter 5: Empirical Results
5.1 Descriptive Statistics
Table 2 from appendix gives a summary statistic from the panel data collected. It can be observed that the mean for the growth rate of GDP per capita is around (3.19%), the minimum and maximum values of GDP growth are achieved by Estonia (-14.56%) and Azerbaijan (33.03%), respectively. We can observe that both countries do not belong to Latin America. The mean of IMF program participation is shown to be approximately (0.17%) and the mean for IMF programs participation in Latin America is around (0.08%).
Moreover, from the control variables section, we can observe that the government consumption mean is around (14.47%), openness to trade is around (83.07%) and the inflation rate mean is around (6.21%). However, an important fact that can be observed in the table is that the maximum levels of government consumption and inflation rate belong to Suriname (37.49%) and
18
Ecuador (96.09%) respectively, the last one occurred in a currency crisis period. The country with the highest openness to trade is Malaysia with (220.41%) and the country with lowest openness to trade is Argentina with (21.85%).
From the statistics tables, we can conclude that countries with the highest levels of government consumption and inflation rate belong to Latin America while the highest levels of openness to trade and growth rate of GDP per capita are represented by other regions.
Table 3 shows correlations between all the variables in the model. The growth rate of GDP per capita exhibits a negative correlation with IMF program participation, IMF program participation in Latin America, inflation rate and government consumption but it has a positive correlation with openness to trade. Additionally, IMF program participation has a negative correlation with openness to trade and government consumption, but a positive correlation with the inflation rate. The same applies for IMF participation programs in Latin American. Another fact that can be observed in the table is that the correlation between IMF program participation in Latin America and the growth rate of GDP is more negative than the general IMF program participation and the growth rate of GDP.
5.2 General results description
In this section, the results of the performed regressions are shown, analyzed and linked to the existing literature.
When applying the Fixed effects and Random Effects methods, it is informative to compute the pooled OLS estimates as well. Comparing the three cases helps to determine the nature of the biases caused by leaving the unobserved effects (μi) entirely in the error term as it is done in the Pooled OLS or partially in the error term as it is done in the random effects method (Wooldridge, J. M. 2009).
Table (5) from the appendix shows the regression results. Column (1) represents the outcome of the model estimated by a pooled OLS, column (2) shows the estimation results of a regression under Fixed effects method without fixing for heteroscedasticity and autocorrelation problem, column (3) shows the estimation results of a regression with Random effects method and column (4) shows estimation results from the Fixed effects method with robust and clustered standard errors.
5.3 Pooled OLS results
As mentioned before, Pooled OLS standard errors ignore the positive serial correlation in the composite errors of the model (𝜔𝑖𝑡 = 𝑢𝑖 + 𝑚μ𝑖𝑡), therefore its standard errors and test statistics are incorrect. The effect of IMF program participation on economic growth in Latin American countries taken as: (β4 + β5) from column (1) is negative (-1.625), which is not in line
19
decreases the GDP per capita growth rate but there is even more decrease for Latin American countries, however, these coefficients are individually statistically insignificant.
It can also be observed that the coefficient estimator of IMF program participation is negative (-0.306) and it is not statistically significant at 1%, 5%, and 10% significance levels. These results are in line with Przeworski and Vreeland (2000) and Barro and Lee (2005) whom suggest that IMF participation programs have negative effects on economic growth.
The results of the control variables coefficient estimators suggest that 1% increase in government consumption causes a decrease of 0.18% point per year of the growth rate of GDP per capita which is in line with Ludvigson, S. (1996). While, 1% increase in openness to trade causes an increase of 0.017% point per year on the growth rate of GDP per capita, which is in line with Yanikkaya (2002). Both estimators are statistically significant at the 1% and 10% significance level respectively.
5.4 Comparison between Pooled OLS, Random effect and Fixed effect results
From table (5) in the appendix, it is possible to observe that the difference between column (1) and column (3) or the Pooled OLS method and the Random effects method is that the pooled OLS standard errors underestimate the true value of the standard errors, as they ignore the positive serial correlation. Additionally, the coefficients of the explanatory variables: IMF program participation and IMF program participation in Latin America are similar for both methods but they are still statistically insignificant. Furthermore, from Table (6) it can be seen that pooled OLS t-values are higher than the t-values in the random effects model.
When comparing the Fixed effects model regression in column (2) with the random effects models in column (3), it is possible to observe few dissimilarities. For instance, Fixed effects method suggest that IMF program participation has a negative effect on the growth rate of GDP in Latin American countries (β4 + β5) of (-1.41) while the random effects method suggest
that it has a negative effect of (-1.50). In both cases the estimated model coefficients of the dependent variables are not statistically significant. From table (5) in the appendix, it is possible to observe that another difference between column (2) and column (3) is that the Fixed effect standard errors are smaller than the Random effects standard errors.
Government consumption coefficient estimates are -0.336 and -0.257 respectively, however, both are statistically significant at the 1% significance level. Furthermore, openness to trade coefficients are positively related to the growth rate of GDP per capita, by 0.0310 and 0.022 respectively, and they are statistically significant at the 5% significance level.
5.5 Fixed effect vs Fixed effect with standard robust clustered errors results
After assessing the fixed effects method with the Wooldridge test for autocorrelation in panel data, the null hypothesis of no correlation between the error terms over time was rejected.
20
Additionally, when testing for heteroscedasticity with the Modified Wald test the null hypothesis of homoscedasticity was rejected as well. The violation of these assumptions in the fixed effects model can bring several consequences, to mitigate this consequences Barro and Lee (2005) suggest using the fixed effects method with robust clustered standards errors in order to correct for autocorrelation and heteroscedasticity. The results of this correction in the fixed effects method can be observed in column (4).
When comparing the fixed effects estimated coefficients and the estimated coefficients of the Fixed effect method with robust clustered standard errors, we observe that the estimated coefficient parameters do not change. This is true by definition because correcting for autocorrelation and heteroscedasticity does not affect the estimates, only change tests, standard errors and p-values.
From table (6) in the appendix, it is observed that the t-values of column (2) are generally larger than those in column (4). Furthermore, Table (5) from the appendix shows that the standard errors of column (4) are larger than those in column (2). Column (4) coefficient estimates are not statistically significant, but, in column (2) government consumption and openness to trade coefficients are statistically significant at the 1% and 5% significance level respectively.
The estimation of robust clustered standard errors is justified if there are several different covariance structures within the sample data that vary by a certain characteristic (Wooldridge, J. M. 2009).
Chapter 6: Discussion
The methodology choice could have been a major shortcoming of this paper, which could lead to insignificant and inconclusive results. However, this paper compares two main methods available for panel data (Fixed effects and Random effects) to identify the differences in the coefficient estimators, tests statistics, standard errors and p-values.
Wooldridge (2009) explains the benefits of using the Fixed effects method which is widely thought to be a more convincing tool for estimating ceteris paribus effects. However, in the case that one of the explanatory variables were to be constant over time, it is not possible to use this method to estimate its effect on the dependent variable. On the other hand, if a time constant variable is included among the explanatory variables, then Random effects method is preferred. However, in most cases, the regressors are themselves outcomes of choice processes and likely to be correlated with individual preferences and abilities as captured by (μ i) which lead to the use of
the Fixed effects method (Wooldridge, J. M. 2009).
An alternative methodology widely used in previous papers is the Instrumental Variable (IV) regression, this method is able to mitigate issues arising from panel data regression
21
assumptions. The difficulty of implementing the Instrumental Variable method lies in the finding of variables that only affect the program participation variables, and not affect the growth rate of GDP. Due to the limited time span and data availability, it was not possible to identify enough Instrument Variables for this method.
Chapter 7: Conclusion
Although a large number of studies have been carried out about the effectiveness of the participation in the IMF programs, limited research has been done for the program participation in Latin American countries. The findings in this study and the literature surveyed in previous sections suggest that IMF support stabilization programs in Latin America are usually unsuccessful.
The main research question is: Have fund-stabilization programs supported by the IMF (International Monetary Fund) led to an increase on the economic growth of Latin American countries from 2000 to 2014?”. According to the theoretical evidence presented in the literature review and theoretical framework the overall effect of IMF program participation in the growth rate GDP per capita is negative. But this negative effect is even larger for Latin American countries. This does not necessarily imply that IMF programs are poorly designed. Evrensel (2002) identified the shortcomings of using the output vs counterfactual approach that its used in this paper. It is stated that the fact that the non-program IMF country members represent the macroeconomic performance of countries that are not participating in the programs can be misleading as the country needs to be under certain conditions to access to the program. However, this paper uses a sample of countries with similar economies in order to have a better appreciation of the program effects.
Considering the results of this research and past researches, it is recommendable that future research focus on the finding of an appropriate estimation methodology accompanied with a longer time horizon and larger sample. This will contribute to obtain more conclusive results that lead to a better understanding of the effects of IMF program participation in the growth rate of GDP per capita.
22
Chapter 8: Appendix
Table 1: Country Description
Latin American Countries Program # Chile 0 French Guiana 0 Puerto Rico 0 Costa Rica 1 Panama 1 Paraguay 1 Ecuador 2 El Salvador 2 Bolivia 4 Peru 4 Argentina 5 Mexico 5 Colombia 7
Other Countries Program # Malaysia 0 Thailand 0 Azerbaijan 1 Estonia 1 Indonesia 1 Greece 2 Lithuania 2 Poland 3 Croatia 4 Cameroon 5 Turkey 5 Kenya 6 Romania 6
23
Figure 1. GDP growth rate of Latin American countries vs Other Countries from year 2001 to 2015. Figures on the Y-axis are in percentage, and the x axis represents the time period.
Figure 2 . GDP growth rate of Latin American Countries with low rate of IMF programs vs Latin American Countries with high rate of IMF programs from year 2001 to 2015.
Figures on the Y-axis are in percentage, and the x axis represents the time period. -4 -2 0 2 4 6 8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 GDP Growth Rate
Latin American countries vs Other Countries
LA GDP Growth Rate OC GDP Growth Rate -5 0 5 10 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 GDP Growth Rate
Latin American Countries with low rate of IMF programs vs Latin American Countries with high rate
of IMF programs
LA without IMF
24
Figure 3 . GDP growth rate of Other Countries with low rate of IMF programs vs Other Countries with high rate of IMF programs from year 2001 to 2015.
Figures on the Y-axis are in percentage, and the x axis represents the time period.
Figure 4. Openness to trade rate of Latin American countries vs Other Countries from year 2000 to 2014. Figures on the Y-axis are in percentage, and the x axis represents the time period.
-4 -2 0 2 4 6 8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 GDP Growth Rate
Other Countries with low rate of IMF programs vs Other Countries with high rate of IMF programs
OC without IMF OC with IMF 0 20 40 60 80 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Openess to Trade rate
Latin American countries vs Other Countries
OC Trade LA Trade
25
Figure 5. Openness to trade rate of Latin American countries with low rate of IMF programs vs Latin American Countries with high rate of IMF programs from year 2000 to 2014.
Figures on the Y-axis are in percentage, and the x axis represents the time period.
Figure 6. Openness to Trade Other Countries with low rate of IMF programs vs Other Countries with high rate of IMF programs from year 2000 to 2014.
Figures on the Y-axis are in percentage, and the x axis represents the time period.
0 20 40 60 80 100 120 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Openess to Trade rate
Latin American Countries with low rate of IMF programs vs Latin American Countries with high rate of IMF
programs LA with IMF LA without IMF 0 20 40 60 80 100 120 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Openess to Trade
Other Countries with low rate of IMF programs vs Other Countries with high rate of IMF programs
OC with IMF OC without IMF
26
Figure 7 . Independent variable vs Time : Government Consumption from year 2000 to 2014.
Figure 8 . Independent variable vs Time: Inflation rate from year 2000 to 2014.
1 0 2 0 3 0 4 0 1 0 2 0 3 0 4 0 1 0 2 0 3 0 4 0 1 0 2 0 3 0 4 0 1 0 2 0 3 0 4 0 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 X g c it Year Graphs by group(Country) 0 5 0 1 0 0 0 5 0 1 0 0 0 5 0 1 0 0 0 5 0 1 0 0 0 5 0 1 0 0 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 X n ri t Year Graphs by group(Country)
27
Figure 9 . Independent variable vs Time: Openness to trade from year 2000 to 2014.
Figure 10. Independent variables vs Time : IMF Participation program and Government Consumption from year 2000 to 2014. 0 2 0 0 0 2 0 0 0 2 0 0 0 2 0 0 0 2 0 0 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 0 1 0 2 0 3 0 4 0 0 1 0 2 0 3 0 4 0 0 1 0 2 0 3 0 4 0 0 1 0 2 0 3 0 4 0 0 1 0 2 0 3 0 4 0 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 Xgcit DLAit Year Graphs by group(Country)
28
Table 2 Summary Statistics Panel Data
DEPENDENT VARIABLE
Variable | Mean Std. Dev. Min Max | Observations
Yyit overall | 3.186896 4.286544 -14.55986 33.03049 | N = 390 between | 1.933808 .1301621 10.29637 | n = 26 within | 3.843103 -16.41796 25.92101 | T = 15
INDEPENDENT VARIABLES
Variable | Mean Std. Dev. Min Max | Observations | | Ddit overall | .174359 .3799052 0 1 | N = 390 between | .1473643 0 .4666667 | n = 26 within | .3512738 -.2923077 1.107692 | T = 15 DLAit overall | .0820513 .2747954 0 1 | N = 390 between | .1350657 0 .4666667 | n = 26 within | .2406788 -.3846154 1.015385 | T = 15 CONTROL VARIABLES
Variable | Mean Std. Dev. Min Max | Observations | | Xnrit overall | 6.214161 8.61059 -1.312242 96.09411 | N = 390 between | 4.404021 2.248098 18.37499 | n = 26 within | 7.446132 -8.85443 89.13171 | T = 15 | | XXopit overall | 83.07414 40.34943 21.85242 220.4074 | N = 390 between | 38.78043 34.82794 180.5054 | n = 26 within | 13.35219 40.881 130.4153 | T = 15 | | Xgcit overall | 14.47276 4.180402 6.134905 37.49442 | N = 390 between | 3.596526 8.107003 20.36592 | n = 26 within | 2.237475 4.755609 32.12722 | T = 15 |
Note: N refers to the total number of observations for a given variable, n is the number of countries. Between and within standard deviations refer to the variation among the countries and within respectively.
Table 3: Correlation table
| Yyit Ddit Xnrit XXopit Xgcit DLAit
Yyit | 1.0000 Ddit | -0.1192 1.0000 Xnrit | -0.0212 0.1475 1.0000 XXopit | 0.1588 -0.2597 -0.1630 1.0000 Xgcit | -0.1418 -0.0033 0.0479 0.1775 1.0000 DLAit | -0.1332 0.6260 0.0522 -0.2461 -0.0452 1.0000
29
Table 4: Diagnostic tests
Diagnostic Tests 1) Breusch and Pagan Lagrangian multiplier
test for random effects
H0: Var (μ i ) = 0
Result:
Prob > chibar =0.0000
2) Hausman test H0: The unobserved variable (μ i) is
uncorrelated to the independent variables
Result: Prob>chi2 = 0.1146
3) Modified Wald test for group Heteroscedasticity
Result: H0:
Variance of error term 𝑚μit is constant Prob>chi2 =0.0000
4) Wooldridge test for autocorrelation in panel data
H0: Cov (mμ 𝑖t, mμ 𝑖s | 𝑋i, μ i ) = 0
Result:
Prob > F =0.0077
5) Jarque-Bera residual normality test
H0: Normality Result:
Chi2=0.0000
30
Table 5: Regression Analysis (Standard Errors)
Column (1) (2) (3) (4) Independent V. Yyit+1 Yyit+1 Yyit+1 Yyit+1
Ddit -0.306 -0.649 -0.484 -0.649 (0.728) (0.747) (0.723) (0.909) DLAit -1.319 -0.764 -1.015 -0.764 (0.996) (1.086) (1.034) (1.170) Xnrit 0.0113 0.000998 0.00361 0.000998 (0.0252) (0.0268) (0.0257) (0.0344) Xgcit -0.181*** -0.336*** -0.257*** -0.336 (0.0516) (0.0887) (0.0692) (0.198) XXopit 0.0176** 0.0310* 0.0221* 0.0310 (0.00561) (0.0148) (0.00860) (0.0272) _cons 4.430*** 5.641** 5.223*** 5.641 (0.863) (1.857) (1.254) (3.385) N 390 390 390 390 n 26 26 26 26 Method P. OLS FE RE FE RC
Note: The first number is the coefficient estimate, the value in the brackets corresponds to the standard errors.*indicates the significance of the coefficient at 0.05 level, ** indicates significance at 0.1 level and *** indicates the significance of the coefficient at 0.01 level. The explanatory variables of interest are presented in bold. N refers to the total number of observations for a given variable, n is the number of countries.
31
Table 6: Regression Analysis (t-values)
Column (1) (2) (3) (4) Independent V. Yyit+1 Yyit+1 Yyit+1 Yyit+1
Ddit -0.306 -0.649 -0.484 -0.649 (-0.42) (-0.87) (-0.67) (-0.71) DLAit -1.319 -0.764 -1.015 -0.764 (-1.32) (-0.70) (-0.98) (-0.65) Xnrit 0.0113 0.000998 0.00361 0.000998 (0.45) (0.04) (0.14) (0.03) Xgcit -0.181*** -0.336*** -0.257*** -0.336 (-3.50) (-3.79) (-3.71) (-1.70) XXopit 0.0176** 0.0310* 0.0221* 0.0310 (3.14) (2.09) (2.56) (1.14) _cons 4.430*** 5.641** 5.223*** 5.641 (5.13) (3.04) (4.16) (1.67) N 390 390 390 390 n 26 26 26 26 Method P. OLS FE RE FE RC
Note: The first number is the coefficient estimate, the value in the brackets is the t-value.*indicates the significance of the coefficient at 0.05 level, ** indicates significance at 0.1 level and *** indicates the significance of the coefficient at 0.01 level. The explanatory variables of interest are presented in bold. N refers to the total number of observations for a given variable and n is the number of countries.
32
Chapter 9: Reference List
Baltagi, B. (2008). Econometric analysis of panel data. John Wiley & Sons.
Barro, R. J., & Lee, J. W. (2005). IMF programs: Who is chosen and what are the effects?.
Journal of monetary Economics, 52(7), 1245-1269.
Bird, G., Hussain, M., & Joyce, J. P. (2004). Many happy returns? Recidivism and the IMF. Journal
of International Money and Finance, 23(2), 231-251.
Evrensel, A. Y. (2002). Effectiveness of IMF-supported stabilization programs in developing countries. Journal of International Money and Finance, 21(5), 565-587.
Fischer, S. (1993). The role of macroeconomic factors in growth. Journal of monetary economics,
32(3), 485-512.
Goldstein, M., & Montiel, P. (1986). Evaluating Fund stabilization programs with multicountry data: Some methodological pitfalls. Staff Papers, 33(2), 304-344. Hausman, Jerry A., and William E. Taylor. "Panel data and unobservable individual effects."
Econometrica: Journal of the Econometric Society (1981): 1377-1398.
Hutchison, M. M., & Noy, I. (2003). Macroeconomic effects of IMF-sponsored programs in Latin America: output costs, program recidivism and the vicious cycle of failed stabilizations.
Journal of International Money and Finance, 22(7), 991-1014.
International Monetary Fund (IMF) (2017).
Ludvigson, S. (1996). The macroeconomic effects of government debt in a stochastic growth model.
Journal of Monetary Economics, 38(1), 25-45.
Przeworski, A., & Vreeland, J. R. (2000). The effect of IMF programs on economic growth. Journal
of development Economics, 62(2), 385-421.Retrieved from: http://www.imf.org/en/About/Factsheets/IMF-Lending
Stock, J. H., & Watson, M. W. (2012). Introduction to econometrics: Global edition. US:
Pearson Education.
Ul Haque, N., & Khan, M. S. (1998). Do IMF-supported programs work? A survey of the cross-country empirical evidence. A Survey of the Cross-Country Empirical Evidence.
33
Cengage Learning, Mason, OH)
Yanikkaya, H. (2003). Trade openness and economic growth: a cross-country empirical investigation.
