Estimating the Impact of IMF Lending programs using fixed effects regression

Abstract

The data used in this paper was panel data spanning from 1985-2002. The time series data is that of the collective labor rights date and the cross section data is the country. All of the economic and population indicators are taken from the World Bank’s World Development Indicators database (2012). Blanton et al. utilized the conditional mixed-process recursive estimator (CMP), which is a two equation model that has a correlated error structure. The seemingly unrelated regressions (SUR) estimator allows for different types of response variables to be used, in this case continuous (labor rights) and binary (IMF/World Bank assistance). The regression utilizes multiple control and binary variables. The authors control for labor covariates (international trade, poverty/poor economic conditions, large populations) and democracy (political regime influence on labor rights). To account for the effects of conflict on labor rights, a civil war variable is included, the data sourced from the Peace Research Institute Oslo’s (PRIO) Armed Conflict Data Base. Other variables include a political regime binary variable (communist or leftist government), an indicator for the presence of international nongovernmental organizations (INGOs) within a state, past labor standards, and five regional dummy variables.

Blanton et al. found that IMF programs have a significant and negative impact in relation to labor rights, where the longer the programs were in place, the greater the negative effect. This was consistent across the other two models (labor laws and labor practices) as well. The requirements for IMF programs (privatization, labor flexibility, and a decrease in the public sector) reduce free association and collective bargaining rights, which then undermine labor laws and any protection of labor rights currently in practice. These negative effects of IMF programs were also found to not just be a short term problem, where the decrease in the labor rights is substantial after fifteen years under an IMF program.

Fidrmuc and Kostagianni (2015) explore the effect of IMF loans on countries that have sought assistance, however they take into account previous criticisms of IMF loans (and the analyses of the loans effectiveness) such as endogeneity bias and the delay that occurs between the implementation of the program and its economic effect. To accomplish this, Fidrmuc and Kostagianni do not focus on economic indicators (such as indebtedness or probable interest rates on borrowing), which would illustrate the need for the assistance and low economic growth thus violating endogeneity. Therefore, Fidrmuc and Kostagianni concentrate on non-economic instrumental variables, namely non-permanent membership of the United Nations Security Council and the country’s degree of democracy (Fidrmuc & Kostagianni, 2015).

The authors used time series and cross sectional data from 213 countries over 38 years (though unbalanced, as not all countries reported data for every year). The analysis was performed by estimating an augmented Solow model of growth (Fidrmuc & Kostagianni, 2015). The dependent variable in this model was the growth rate of the GDP per capita. The explanatory variables were population growth, investment (gross fixed capital formation to GDP per ratio, and an IMF loan dummy (having a value of 1 every year the country had a loan and 0 otherwise). Fidrmuc and Kostagianni focused on three major IMF programs: Stand-By Arrangements, Extended Fund Facility, and Poverty Reduction and Growth Facility Arrangement. The differences in the model are the length of the repayment period, the interest rates and eligibility criteria, and a dummy variable which obtains a value of 1 if the loan program was in effect for at least five months of a given year (Fidrmuc & Kostagianni, 2015). 

To account for possible reverse causality between the dummy variable for IMF involvement and the growth rate, the authors discussed three main counters to this event. The first is that the instruments have to be uncorrelated with the error term. To do this, Fidrmuc and Kostagianni used a Sargan statistic, of which an insignificant result would suggest that the instruments can be excluded from the main regression (Fidrmuc & Kostagianni, 2015). The second was that the instruments had to be uncorrelated with the economic hardship, in order to avoid the endogeneity assumption of self-selection bias. The final counter was that the regression would focus on instruments that reflected institutional and/or political conditions rather than those that were economic. 

Fidrmuc and Kostagianni found that in general, the effect of IMF assistance on the economic growth is insignificant and the other two explanatory variables, investment and population growth, are significant at the 1 percent level (with a positive and negative effect respectively). To find the how the growth rate behaved over time, in relation to IMF aid, the IMF program dummy was lagged between one to three years. The dummy variable was then found to be significant at the 1 percent level in all three of the cases, though the effect was low: average annual growth increased between 0.68 percent – 0.82 percent and the other variables remained unchanged. Therefore, the effect of aid on growth is determined to be at best zero (Fidrmuc & Kostagianni, 2015). The authors offer a counterargument to previous literature which determined the impact of lending programs to be insignificant by noting that the actual effects of assistance do not occur immediately and lag by a few years and that this is in large part due to the countries that seek aid self-select and this result in an endogeneity bias. Lagging IMF assistance allows for there to be a positive effect of IMF growth, where the longer the lag, the greater the positive effect. 

Butkiewicz and Yanikkaya (2005) did an empirical investigation into the growth effects of both IMF and World Bank lending programs. Included in the investigation are fundamental economic and political factors that affect growth, the impact of international finance institutions (IFIs) lending by country income level and the degree of democracy, and the effects of different IMF programs. The authors used a standard empirical growth model to examine the relationship between IFI lending and long-run economic growth. Specifically, the model used is an augmented Solow growth model, and includes physical and human capital. The country growth rate is measured with real GDP per capita in the given period, physical capital stock per person, human capital per person, and the most important characteristic of the model is that of convergence to a steady state (Butkiewicz & Yanikkaya, 2005). 

In steady state convergence, the growth of per capita income is attributed to technological progress (where technology is nonrival to all countries). Convergence also implies that countries with low levels of income grow rapidly (ceteris paribus) and countries with higher levels of income grow at a slower rate. From this implication the following are observed; the effect of a country’s initial level of income is negative, convergence occurs slowly (where factors affecting steady state level of income raise the growth rate over observable periods of time), physical and human capital only affect the level of income per capita and not the growth rate, and high level of human capital relative to physical capital may increase the growth rate (Butkiewicz & Yanikkaya, 2005). Since all variables are measured in per capita terms, an increase in population relative to the amount of physical and human capital reduces steady state equilibrium level of income per capita and high rates of population cause a reduction in the rate of economic growth. 

The data used by Butkiewicz and Yanikkaya are of 100 developing countries over five year intervals, though limitations reduce the panel data from 500 observations to 407 observations. There are multiple variables included in the analysis: growth of per capita real GDP, initial GDP per capita levels, approximate measures of physical and human capital, Gastil (1988) indices as a measure of democracy, fertility rates, government consumption, inflation rates, war (dummy), geographic (dummy), IMF and World Bank credits, and additional variables.

Butkiewicz and Yanikkaya found that log level of initial GDP implies that the convergence rate is 2 percent to 3 percent annually and openness to trade has a significant positive effect on growth. Factors that reduce growth are fertility, inflation, high government consumption, war, and geographic location (specifically Latin America and sub-Saharan Africa). The proxies for human and physical capital were not found to be statistically significant, though they were significant in estimates that included both developed and developing nations. The Gastil index was also not found to be statistically significant, at the conventional levels, but was close to significance in the case of most other estimates. IMF lending programs were found to have a small, but significant, negative effect on the real growth rate, including lagged values. The World Bank was also found to have a negative impact on real growth, however it was not statistically significant. In conclusion, IMF lending has negative effects on growth while World Bank lending can lead to an increase in growth in some cases (specifically in countries with low income and a low level of democracy). The effects occur most likely through aggregate investment: IMF lending reduces aggregate investment (especially that of public investment) and the World Bank lending may increase public investment in countries that are borrowing (Butkiewicz & Yanikkaya, 2005). 

Blanton et al. examined the effect of lending from the IMF and World Bank on labor rights, instead of the much more commonly examined economic growth, as Fidrmuc and Kostagianni and Butkiewicz and Yanikkaya did in their papers. Blanton et al. did have controls for their two-equation error correction models, running additional tests based on different models in order to check the validity of their models. Fidrmuc and Kostagianni only investigated the effect of IMF lending on economic growth and found it to be positive, especially given the longer the lending program was in effect. Fidrmuc and Kostagianni incorporated the idea of self-selection bias (and therefore endogeneity) into their model, which is postulated by Butkiewicz and Yanikkaya in their paper, but not conducted, so as to see what would happen to economic growth while accounting for the countries requesting assistance only once they are at a negative stage of growth. Butkiewicz and Yanikkaya performed an empirical analysis on the effect of IMF and World Bank lending on economic growth, where both had negative impacts though only the IMF lending was statistically significant. The augmented economic growth model utilized was supported and controlled for a multitude of variables, similar to those controlled by both Blanton et al. and Fidrmuc and Kostagianni, though this model had far more control variables than the other models. 

Fidrmuc and Kostagianni and Blanton et al. used almost identical data, pulling from the World Bank’s World Development Indicators (2012) and the Polity IV data in order to find population, real GDP, democracy, and other key variables. Butkiewicz and Yanikkaya used a variety of data sets, using the Gastil (1988) indices instead of the Polity IV measures as the former had more countries available in the data than the former. All of the papers had time series and cross sectional data (panel data) of multiple variables, countries, and spanning more than a decade. The limitations for the papers are all similar: the unobservables that are unaccounted for (though, this is mitigated by the amount of controls used in each of the models) and, most notably, the lack of data on human and physical capital measures from developing countries, therefore having to use estimators or proxies in the model.

While the studies over the years have found different results, this may be due to the unobserved bias of variables not considered or not included because of the lack of data. However, the IMF had a major policy reform in 2010, which has resulted in the IMF offering assistance to countries which had previously taken loans and may have ended up in a worse position (and in more debt), than when they started.

This paper will examine the effects of IMF and the World Bank lending programs for countries that received loans before, during, and/or after the financial crisis. Though it is similar to Blanton et al., this paper will be similar to the investigations conducted by Fidrmuc and Kostagianni and Butkiewicz and Yanikkaya in terms of methods of analysis and in data. While this paper is amongst the many examining the overall effectiveness of IFI and their loan programs, this paper will utilize fixed effects regression in order to compare countries that were allowed to borrow and those that were not, based on the provided loan criteria from both institutions.

For the country groups, the two main groups are “Advanced Economies” and “Emerging Market and Developing Economies.” Under “Advanced Economies” are the following categories: Euro Area, Major Advanced Economies (G7), Other Advanced Economies (Advanced economies excluding G7 and euro area), and the European Union. Under “Emerging Market and Developing Economies” are the following categories: Emerging and Developing Asia, ASEAN-5 (Indonesia, Malaysia, Philippines, Thailand, and Vietnam), Emerging and Developing Europe, Latin America and the Caribbean, Middle East and Central Asia, and Sub-Saharan Africa. It is notable that under the “Advanced Economies” heading, that some sub-groups have multiple countries that overlap and are thus being double counted. For the purposes of analysis, countries will be observed at the individual level, with 179 countries having observable data.

As this paper investigates the impact of IMF lending on countries, compared to those who did not receive a loan, the measure of economic growth used in the model was GDP per capita. This was calculated using the yearly average of market exchange rates in U.S. dollars, adjusted for Purchasing Power Parity in 2011 dollars, where exchange rate projections are provided from economists in emerging market and developing countries. It is noted that this is expenditure-based GDP, the total final expenditures at the price of the purchasers (IMF, WEO). 

Unemployment is, by the OECD harmonized definition, the number of unemployed people as a percentage of the labor force, where the labor force is the total number of those who are employed and unemployed. The population is determined from a country’s census (and thus only captures those that were participants of the latest census). From the IMF, the unemployment and population data are only available for advanced economies and some emerging and developing economies. If the data on unemployment for a country was not available from the IMF, the data was obtained from the World Bank. If no data was available, the country was dropped from the analysis.

A dummy variable was included indicating whether or not the country received a loan (the model may be adjusted to have a done for the different types of loan arrangements, due to the different conditions for each loan type) from the IMF. The dummy variable is ‘turned on’ (equal to 1) from the year the loan was first implemented to the end of the loan period, the dummy is 0 for all other time periods. 

Figures 5 and 6 illustrate the correlation between the GDP per capita and population and the unemployment rate. In Figure 4, with the exception of Latin America and the Middle East regions, which start out positive and end with a decline (though, an overall positive relationship), the Advanced economies, Emerging Asian, Emerging European, and South Saharan Africa economies all have positive relationships: an increase in population results in an increase in GDP per capita. In Figure 5, as the unemployment rate increases, the GDP per capita decreases in all countries. Emerging Asia has a stronger negative relationship then other regions. 

The fixed effects regression shows that all three variables, including the main variable of interest, loan, are statistically significant at the 1 percent level. The unemployment rate has a negative effect, where for every one percent increase in unemployment, there is a -4.4 percent decrease in the GDP per capita of a country. A change in population has a positive effect, increasing the population by one person, results in a 1.1 percent increase in the GDP per capita. If a country received a loan, then there was a positive impact on the country’s GDP, receiving a loan increase GDP per capita by 9.9 percent.