UNIVERSITY
OFGRONINGEN
Faculty of Economics and BusinessMASTER THESIS IN INTERNATIONAL ECONOMICS AND BUSINESS
Threshold Effect of Inflation on Economic Growth
Panel analysis on industrialised and developing countries
Author: Supervisor:
Ákos Kancsal dr. D. H. M. Akkermans
ABSTRACT
This paper is going to investigate the effects of inflation on economic growth devoting special attention to the threshold effect. The main idea of this concept is that there is a breaking point in the level inflation below which its effect on growth is positive or neutral, whereas above it inflation is adverse to the economic progress. Furthermore, this level is expected to be higher in case of developing countries. After conducting panel data analyses this thesis is going to prove the existence of these structural breaks and the idea that above them inflation is detrimental for economic progress.
Table of Contents
1. INTRODUCTION ... 4
2. LITERATURE REVIEW...7
2.1 FOUR PRINCIPAL PREDICTIONS REGARDING THE RELATIONSHIP BETWEEN INFLATION AND GROWTH...7
2.2 OTHER THEORIES RELATED TO INFLATION AND ECONOMIC GROWTH...9
2.3 EMPIRICAL RESULTS ABOUT THE EFFECTS OF INFLATION... 11
2.4 BREAKING POINTS IN THE LEVEL OF INFLATION... 12
2.5 DIFFERENT THRESHOLD FOR RICH AND POOR COUNTRIES... 14
3. DATA AND METHODS ...16
3.1 DATA SOURCES... 17
3.2 PANEL ANALYSIS... 17
3.3 THE REGRESSION MODEL... 18
3.3.1 Variables... 19
3.3.2 Dummy Variables... 24
3.4 OTHER IMPORTANT ISSUES... 26
3.4.1 Heteroskedasticity... 26
3.4.2 Endogeneity... 26
3.4.3 Robustness ... 27
3.5 THE STRUCTURE OF THE RESEARCH... 27
4. RESULTS ...29
4.1 PRELIMINARY TESTS... 29
4.2 MAIN REGRESSIONS... 33
4.2.1 Regressions on developed countries below the threshold... 34
4.2.2 Regressions on developed countries above the threshold... 36
4.2.3 Regressions on developing countries below the threshold ... 37
4.2.4 Regressions on developing countries above the threshold ... 38
4.3 TRANSPOSITION OF THRESHOLDS... 40
4.4 DISCUSSION... 42
5. CONCLUSIONS...43
6. REFERENCES ...46
1. Introduction
Today many countries’ – developing and industrialized as well – most important macroeconomic objections are to maintain economic growth and keep the rate of inflation at a low level. How can we find this possible level of inflation? Is the high level of inflation detrimental for a country’s growth rate, while low level is less adverse or maybe beneficial? These questions are in the focus of economists and political economists since the 1950s.
Inflation affects economic progress in many ways. One of them is that it makes the profitability of projects questionable in the future which may decrease the propensity of firms and individuals to make investments in a certain country or group of countries and this can result in slower economic progress (Gokal and Hanif, 2004). De Gregorio (1993, p. 272) lists some negative factors stemming from inflation. They are: pessimistic perceptions, capital flights or delays in investment decisions. Fischer (1993) calls inflation an indicator which shows how effectively a government can manage the economy.
Another ways through which inflation can affect economic progress are the international trade and the taxation system. In the former case high level of money devaluation can relatively increase the price of exported goods and influence the balance of payments, eventually it decreases the competitiveness of the country. Moreover, inflation can have an impact on growth through the taxation system, changing the decisions for lending and borrowing. However, De Gregorio (1993) argues that high rate of inflation may be the consequnce of the inefficient tax system of a country.
If we mention inflation we should think not only on the rate of it but its volatility also. Neither future’s rate of inflation nor its volatility can be anticipated perfectly which increase the uncertainty of projects or investments, making future-planning more expensive and difficult1. Bruno (1993) finds these two features of inflation as the most important determinants of the rate of return on investment. This is the other channel of inflation to influence economic progress and it is called investment channel.
In the 1950s and 1960s the effects of inflation on the growth rate were thought marginal, insignificant and sometimes slightly positive because it cannot be detected in the data. This view prevailed until the big oil crises in the 1970s when economies experienced serious shocks and persistent inflation. As more data were available researchers could detect the negative effects of money devaluation but in the 1970s and 1980s this coefficients had only small magnitude (Sarel, 1996). Today the mainstream is that inflation has adverse impact on economic progress and therefore, countries should completely avoid it. We can see that the common supposition changed drastically in the second part of the last century.
Andrés and Hernando (1999) claim that there have not been written any academic studies which find positive association between inflation and future income until their research. Furthermore, they argue that even low or moderate rate of inflation has negative effect on growth in the long run. This finding is in contrary with the „structuralist” tradition which says that moderate inflation can be good for economic progress since nominal wages are lagged behind prices (Temple, 2000).
According to Gokal and Hanif (2004) we can find consensus among economists about the positive relationship between macroeconomic stability, more precisely between low inflation rate and economic growth. Ghosh and Phillips (1998, p. 673) claim that low rate of inflation may be lubricator for the engine of economy, since „if prices exhibit downward rigidity then very low inflation rates may ossify the structure of relative prices, impending adjustment to real shocks.”
Why do central banks take inflation so serious and try to curb it? Barro (1995) suggests a very short but meaningful answer to this question: inflation is costly. Households and businesses perform poor if inflation is high or unpredictable and their
weak performance has very severe additive effects on the long run economic progress. Central bank announcements are quite regular today and Lin and Ye (2007, 2009) estimate that inflation targeting does not have any effect on developed countries inflation rate but in case of developing countries it resulted in a 3 percent fall in the rate. Furthermore, they emphasize the substantial credibility gain can be earned from such declarations.
Many authors suggest that the effects of inflation on economic progress is not linear. It implicates the presence of a breaking point above which the adverse effects of inflation can be captured or in others’ view the negative impact becomes more intense. Researchers tried to capture this point but their results are different since all of them carried out their researches with different group of countries, time period or methodology. More recently authors estimated different breaking points for distinct groups of countries.
The empirical results of many researches have a clear-cut evidence for policy-makers: if inflation affects economic growth negatively it will be beneficial to target zero inflation rate and preserve price stability as much as possible (Faria and Carneiro, 2001). It may be a good idea but countries are different. We cannot find two countries with the same endowments, characteristics, or macroeconomic values. Those differences implicate that there is not exist one possible good solution for industrialised or developing countries. However, rich states try to force developing ones to follow their example even if they pursued a completely different strategy during their enrichment (Chang, 2002).
In many articles researchers were trying to find different explanations for developing and industrialised countries during the last decades since countries and economies are not similar. Introduction of different structural breaks for these groups seems a reasonable idea because developing countries tend to reach higher level of inflation than rich states. Following this way of thinking, I would like to find answer for the following questions during my thesis:
“Where are these possible breaking points – if they exist – in the rate of inflation for industrialised and developing countries?”
“Does inflation affect economic progress positively below the structural break which is lower for developed states and higher for emerging ones?”
The contribution of this paper is to utilize those possible structural breaks above which inflation adversely affects economic progress. Since many authors calculated these possible levels, this paper may help to find which author or authors could estimate these thresholds most precisely. According to Drukker et al. (2005) the non-linearity of inflation is widely accepted, however, there are still debates about the rates of inflation which act as thresholds. Furthermore, this thesis can reflect on the prevailing policy of central banks which tries to preserve price stability. The results may justify whether it is reasonable policy to keep inflation as low as possible.
In the development to provide an answer to this question, I will take panel data analysis in which I will carry out a research on the effect of inflation on economic growth. In chapter two, I am going to highlight the relevant literature in short and describe the results of previous analyses. In chapter three I will describe the methodology of this paper. Chapter four is going to interpret the results of my tests, while chapter five is going to conclude my thesis.
2. Literature Review
2.1 Four principal predictions regarding the relationship between
inflation and growth
There are four different theories regarding the relation of inflation and economic growth. Tobin (1965) suggests positive relationship between inflation and long-run growth since in his model money and capital are substitutes of each other. Sidrauski (1967) creates a model where inflation has no effect on growth because money is neutral or super-neutral in optimal circumstances. The former means that changes in the stock of money will affect only nominal variables2, while super-neutrality of money is an even stronger concept since it alleges that neither the level of money supply nor the rate of money supply growth affect real variables3. Stockman (1981) develops a framework where money is a substitute to capital. In this model inflation
negatively affects growth in the long-run. In reaction to his results, newer studies also suggest similar relationship between the two variables but only above a certain threshold level.
The early theories about inflation and economic growth were built upon cyclical observations where periods of inflation were followed by periods of deflation. Because of its cyclical movements (or because it reacted later), inflation was called „lazy dog”. Haslag (1997) argues that before the World War II, theorists expected positive association between inflation and growth.
Until the 1970s, when stagflation came into the middle of interest, the aggregate-supply aggregate-demand (AS-AD) framework also predicted positive relationship between these two variables, but during the years of the oil crises the validity of this association was questioned (Gokal and Hanif, 2004). The AS-AD model contains two curves, where aggregate supply is the total supply of goods and services that firms of an economy aim to sell in one period, while aggregate demand refers to the total demand for final products and services in one specific period.
In the classical growth theories, connection between inflation and the level of output was not highlighted but it can be expected as a negative association since it will reduce the profit levels of firms in the form of high wage costs. The traditional keynesian model encompasses the AS-AD curves which are able to illustrate this relationship. According to this theory AS curve is upward sloping and in this case both prices and output are affected if changes happen in the AD curve (Dornbusch et al., 1996). If AS is vertical demand side changes will affect prices. According to this theory many factors can affect inflation and economic growth in the short run. The AS-AD model eventuates that we can curb inflation if we fill the gap between the two curves.
increased4. In this case output remains on the same level but inflation will increase. Furthermore, they mention the periods of stagflation when output falls while inflation increases.
2.2 Other theories related to inflation and economic growth
The theory of monetarism also mentions the effects of inflation. Milton Friedman – who created the term „monetarism” – argues that inflation occurs simply because the increase in the rate of money supply is higher than the growth rate of the economy (Friedman, 1956). It means that growth rate of money affects price levels in the long run. Furthermore, he claims that individuals can forecast the future rate of inflation and adjust their behaviour to it, thus it will not have any effects on employment or output5. It postulates the harmlessness of inflation, however, it has adverse effects on another macroeconomical variables (e.g. investment, capital accumulation or export) in the real life.
One of the early neo-classical models, which were elaborated by Solow (1956) and Swan (1956), stipulates technological change instead of investment, as the primary factor of growth. Todaro (2000) finds that it was determined exogeneously (independently from inflation) according to these growth theorists. Mundell (1963) elaborates this idea and argues that higher inflation or the expected level of it will decrease the wealth of people since the rate of return on the real balances of individuals falls. In this case people will save more money to accumulate the preferred wealth (they will buy more assets). Since the price of assets will increase the interest rates go down. This leads to greater capital accumulation and faster output growth. This theory works with diminishing returns to capital which suggests that countries with lower initial GDP relative to the steady-state position, will develop faster (Briault, 1995). If every country has the same steady-state output poorer countries will converge to the industrialised ones. With the concept of steady state, Solow and Swan refer to a long-term outcome of an economy since every country will converge to its relatively stable position (and to the output level of this point) in the future if it is away from its steady state.
Stockman (1981) created a model where greater inflation results in lower steady state output and reduction in the welfare of individuals. He argues that purchasing power is decreased by inflation, thus people will reduce their consumption of capital and cash goods which deteriorates the output level. Greenwood and Huffman (1987) explain this effect in the context of labour-leisure choice. They find that return to labour will fall if the rate of inflation increases. It implicates the substitution of working hours (and therefore the level of consumption as well) with more leisure time which ultimately leads to lower output level. The Neo-Keynesian theory improved the original Keynesian model and its theorists argue that inflation rate is determined endogeneously. In this framework the output and the rate of employment specify the level of inflation.
The growth rate has only one independent factor in endogeneous growth theory which is the rate of return on capital. Economic growth can be seen as a result of factors within the process, according to the endogeneous growth theory. Inflation has negative effect on growth in these models since it reduces the rate of return both on physical capital and on human capital (López-Villavicencio and Mignon; 2011). It has a negative knock-on effect on capital accumulation and in the end it results in lower growth rate (Lucas, 1980; McCallum and Goodfriend, 1987). In both cases inflation can be seen as a tax on capital (physical and human as well). In the latter case it will lead to substitution between goods and leisure, affecting output growth. Haslag (1995) carried out a research within the framework of this theory and finds that if economies had decreased the inflation rate from 10 percent to zero their growth rate could have increased by 0.2 percent during the examined period.
2.3 Empirical results about the effects of inflation
As it can be seen the connection between inflation and economic growth is not a new question in the literature, however, the number of empirical studies started to increase significantly in the last two decades in this field. For example, Johnson (1967) argues that there cannot be found indisputable evidence for either negative or positive effects. In the 1960s many IMF Staff Papers found little adverse effect of inflation (Bhatia, 1960; Dorrance, 1963) but experience was ambiguous in the 1970s in South America (Galbis, 1979).
The first researchers, who tried to shift the earlier belief of positive association, were Kormendi and Meguire (1985). After conducting cross-sectional analyses, they find that inflation affects economic growth negatively and significantly mainly through the investment channel. Contrary to their findings, Levine and Renelt (1992) and Fischer (1991, 1993) find that inflation affects growth not only through the investment but also through the efficiency channel. Grier and Tullock (1989) conclude that a single model is not appropriate for different countries and that is the reason why they interpret different results for different country groups.
After those investigations, in 1993 Fischer finds negative association between inflation and economic growth for a high number of countries by using panel data method. He argues that a 10 percent increase in inflation will result in a 0.4 percent decrease in the annual growth rate of GDP. These findings were confirmed by De Gregorio (1993) and by Barro (1995). Barro also finds that this relationship may not be linear in his research with 100 countries. He concludes that if the inflation rate increases by 10 percent the growth rate will decrease by 0.2-0.3 percent and investment share of GDP by 0.4-0.6 percent. Moreover, he argues that the causality runs from inflation to growth.
Many authors mention the rate and the volatility of inflation as well. Barro (1995) also confronts these two concepts in his study and concludes that higher volatility coexists with higher average inflation rate, which confirms the earlier findings of Okun (1971). Furthermore, Barro includes the two variables into one regression and he finds that the variability of inflation is not significantly correlated with economic growth.
2.4 Breaking points in the level of inflation
Another important feature of inflation is nonlinearity, which presumes a structural break where the effects of inflation start to be significantly worse or in others’ view they start to be detrimantal for economic progress. In the latter view low level of inflation is beneficial (or at least neutral) for growth and can be seen as “grease” for economic processes. Fischer (1993) was the first who tried to model this idea. He creates different ranges of inflation6 and finds negative relationship between the dependent and independent variables but the effect of inflation was found to be diminishing. It means that if inflation grows from 10 to 20 percent it will be more destructive as if it grew from 40 to 50 percent. Furthermore, he concludes that the statistical significance decreases as the interval of inflation gets wider.
Sarel (1996) estimates a breaking point at 8 percent and finds that below this rate inflation has no (or slightly positive) effect on economic growth. However, above this 8 percent level its effects are very powerful, significant and robust. Sarel’s results can explain why researchers could not figure out the negative effect of inflation during the period of 1950-1970 since there were not so many years when inflation rose above this breaking point. According to his results, if the average level inflation had doubled the average worldwide growth rate would have decreased by 1.7 percent which would have meant global stagnation since the average per capita growth rate was 1.7 percent during the 1980s. However, it is important to mention that Sarel did not distinguish between industrialized and developing countries.
Ghosh and Phillips (1998) agree with Sarel’s findings about non-linearities and in their own analysis they find that very low (2 percent) level of inflation can be seen as lubricator for the economy but higher rates have significant, negative and robust effect on economic growth. Furthermore, they take into account a possibility of fast
disinflation and find that it may have serious effects in the short run, since rapid disinflation can cause fall in the GDP growth. Using cross-sectional frameworks, Bruno and Easterly (1998) find that the effects of inflation start to be detrimental above 40 percent level for the growth rate of developing countries. The reason why their estimated threshold is much higher than the others’ ones is because of their methodology. They used cross-sectional analysis with period averages of a long interval.
This idea was improved by Khan and Senhadji (2001) who used unbalanced panel for 140 countries to find this structural break. In their study these rates, below which inflation does not have negative effect on economic progress, are around 1-3 percent for developed and 7-11 percent for non-industrialised countries. Drukker et al. (2005) use the same method on 138 countries and confirm the earlier results. They find the two thresholds at 2.7 percent and 17 percent for the two groups of states7. Both articles confirm the idea about the thresholds in the effect of inflation. The small difference between their breaking points can be the results of the slightly distinct time period they used8.
More recently researchers suggest similar thresholds to the ones mentioned in the previous paragraph. Blanchard et al. (2010) argue that inflation rate should be below 4 percent for rich countries because it gives the possibility for monetary expansion in case of an exogenous shock. Kremer et al. (2011) claim that developed countries want to decrease their average inflation rate to 2 percent, while the calculated structural break is at 2.5 percent according to their analysis and we can say that it is in line with many central banks’ inflation targets9. Furthermore, they calculate this threshold also for developing countries and they estimate it at 16 percent level. They argue that there is no significant difference in the effects of inflation on growth between rich and poor countries, only the structural breaks are different. It means that the detrimental effects of inflation is independent from the level of GDP. The results of their panel analyses show that the value of this structural break is significantly lower than that of Bruno
7 However, they calculated a second threshold for developed countries at 12.6 percent where the effects of
inflation are started to be even more adverse.
8 Khan and Senhadji (2001) carried out their research on countries between 1960-1998, while the time period of
Drukker et al. (2005) is 1950-2000.
9 The 95 percent confidence interval is ([1.94 2.76]) which contains the 2 percent but not 4 percent level which
and Easterly (1998) who estimated this point at 40 percent level. Table 1 summarises the different thresholds found by these articles.
Author (year) Threshold for rich states Threshold for poor states Sarel (1996) 8 percent without distinguishing between the two groups
Bruno and Easterly (1998) - 40 percent
Ghosh and Phillips (1998) 2 percent -
Khan and Senhadji (2001) 1-3 percent 7-11 percent
Drukker et al. (2005) 2.7 percent 17 percent
Kremer et al. (2011) 2.5 percent 16 percent
Table 1 shows which thresholds were found by different authors.
2.5 Different thresholds for rich and poor countries
What may be the sources of higher inflation in case of developing countries? This question is not easy to answer since there are too many differences among countries.
The most common way we distinguish between the two groups is the significant gap between the per capita GDP rates of emerging and developed countries. Since poor countries are specialised in labour-intensive industries, return on capital is much higher in those economies. Loungani and Swagel (2001) argue that developing countries’ inflation is many times a mark of their overheated economy and influenced by output gap.
Montiel (1989) claims that inflation is linked to fiscal imbalances in those countries. The role of the state in emerging countries is also unavoidable in many cases since governments apply price control on different articles which influences the rate of inflation (Milne and Ryan, 1994). Moreover, state intervention is also observable in the running of central banks which eventuates higher frequency of governor turnover. It can be the result of lack of independence of the monetary authority or the general instability of the state. Cukierman et al. (1992) find that central bank governor turnover is positively correlated with inflation rate in developing countries.
wages, the financial system and public debt. It can facilitate economic arrangements and in case of inflation shocks, indexation helps the system of relative prices to survive. It has the advantage to help to reduce the adverse effects of high inflation to a limited extent, increase the credibility and simplify future planning since it decreases the frequency of renegotiations (of wages or other contracts), thus eventually reduces transaction costs.
Khan and Senhadji (2001) argue that the reasons for the higher average rates of inflation in developing countries can be associated with the convergence (or catch-up) effect and the Balassa-Samuelson effect. The former refers to the idea that relatively poorer countries grow faster than relatively rich countries, which means that they will converge to the industrialised states. This can be explained by diminishing returns (to capital) since its effect is not so strong in developing countries as in rich ones. The Balassa-Samuelson effect claims that because of the productivity growth in the (internationally) traded sector is higher than in the non-traded (or sheltered) sector the nominal wages will increase which spills over in the non-traded sector. This wage increase will eventually increase the inflation rate of developed countries. Industrialised countries also face with this effect but in their case this effect results in higher average price levels both in the internationally traded and non-traded sectors.
Today there is consensus about the presence of these structural breaks above which inflation has negative effect on economic growth and below which this association is positive or neutral (Singh and Kalirajan, 2003). According to López-Villavicencio and Mignon (2011) the reason of the positive association between low level of inflation and economic growth can be found in the endogenous growth theory because low inflation results in lower unemployment rate and it will lead to better capacity utilization and lower output gap (Palley, 2003).
Based on the above mentioned studies, analyses and theories, I choose the 2 percent and 15 percent level of inflation as breaking points and formulate five testable hypotheses:
Two hypotheses related to industrialised countries:
2. Inflation has negative effect on the level of economic growth of advanced countries if its rate is above 2 percent.
Two hypotheses for developing countries:
3. In case of developing countries the association between inflation and growth are positive (or neutral) if its rate is below 15 percent.
4. The correlation between inflation and economic progress is negative in case of developing countries if the rate of inflation is above 15 percent. Furthermore, I formulate one more hypothesis for the case if the threshold effects of inflation will not seem to be relevant or measurable.
5. The threshold effect of inflation is not measurable, which means, including the breaking points into the effects of inflation is pointless. In line with earlier analyses and studies I will use the five independent variables in my thesis. They are the annual rate of inflation, the investment share of gross domestic product (GDP), annual rate of population growth, secondary school enrollment rate, and openness to international trade. My dependent variable will be the annual growth rate of GDP. These variables are commonly used in the literature related to the relationship between inflation and economic progress.
3. Data and Methods
This part of my thesis will summarize the methodology I will use during my research. It will include the data sources, components of the panel analysis, which method I would like to use in my research, and the regression model with the necessary variables and their explanation.
3.1 Data Sources
One of the most important tasks for all researchers is to find appropriate datasets to test their hypotheses. Today many international organisations and institutions10 provide rich data files for macroeconomic analyses. I will collect my dataset from
three important sources. They are the Penn World Table11 (Heston and Summers, 2011), the data collection of the World Bank12 and the database of Barro and Lee13. The latter one contains observations only on education over the last five decades, while the first two sources comprise a considerable amount of factors of numerous countries and long period of time which make them appropriate for a panel data analysis.
3.2 Panel analysis
Panel analysis is one type of regression where data contains time series observations of a number of units. It means that the dataset contains two dimensions: a cross-section (e.g. countries, firms or individuals etc.) one and a time-series (e.g. years, months or averages of years, etc.) dimension. Because of this feature a panel analysis has many advantages in comparison with a simple cross-section or time-series study. Hsiao (2006) lists these benefits: panel analyses are more accurate, contain more degrees of freedom and they are more informative. Panel data can capture more aspects of an observation since this method is more complex and it may uncover dynamic relationships in the model. Moreover, panel studies are very useful to reveal long-term or cumulative effects which cannot be modeled in a “one-shot” study.
The association between inflation and economic growth was analyzed in many researches using numerous methods. Most of the earlier researchers conducted either time-series analyses (see: Grimes, 1991; Stanners, 1993; Fisher and Seater, 1993; Weber, 1994; etc.) or cross-section inspections (see: Kormendi and Meguire, 1985; Grier and Tullock, 1989; Levine and Renelt, 1990; De Gregorio, 1993; etc.). The majority of these studies finds negative association between inflation and growth but both methods are constrained from at least one point of view since they inspect either one country or only one year (possibly average of years) which cannot discover the dynamics or the long-time averages may obscure the changes in the inflation-growth nexus.
11 The latest version (7.0) was published in March 2011 and availabe here
http://pwt.econ.upenn.edu/php_site/pwt_index.php [Accessed 18.04.2011].
12 This database available at http://data.worldbank.org/ [Accessed 28.04.2011].
13 The most recent version of their database are available on the following website:
Thus, if we would like to examine this topic in the most diversified way, a panel data analysis seems the most useful method. Drukker et al. (2005) argue that panel data methods are more flexible and decrease the chance of misspecifications. A large number of studies utilizes the advantages of panel data method (see: Fischer, 1993; Hansen, 1999; Drukker et al., 2005; Bick, 2010; Omay and Kan, 2010 or López-Villavicencio and Mignon, 2011 etc.). In most of these papers authors build a big database with many countries and for a period of 30-50 years in order to model the inflation-growth relationship. In my thesis I would like to follow this way and examine the relationship between inflation and economic progress in a longer period with high number of countries.
The panel data approach embraces many distinct types of analyses. Two of them are the most common: the fixed effects and the random effects models. In advance it is impossible to decide which method suits my datasets better. In order to find the answer for the question, I will conduct a Hausman test (Hill et al., 2008) that compares the estimates of fixed and random effects regressions and provides an answer for this question.
3.3 The regression model
In this paper the following growth equation will be applied to test the hypotheses mentioned at the end of the previous chapter.
ΔYit = α0 + α1lnINFit-1 + α2INVit + α3EDUit-4 + α4POPit-10 (15) + α6Openit + D0 + εit
Where i = 1,2,3,…,N denotes cross-section units, t = 1,2,3,…,T are the time periods and α0 is the constant.
(i) ΔYit is the growth rate measured in real per capita GDP.
(ii) lnINFit is measured as the logarithm of the value of the annual percentage change in consumer price index with the base year of 1974.
(iii) INVit controls for the physical capital and denotes the investment share of GDP. (iv) EDUit measures the human capital as the rate of net secondary school
enrollment rate during the period.
(vi) Openit is the sum of exports and imports of goods and services measured as a share of GDP.
(vii) D denotes dummy variables which are necessary to distinguish between industrialised and developing countries. Moreover, they will help to create the threshold levels of inflation.
(viii) εit is the error term of the equation.
This growth equation consists of five explanatory variables which are slightly less than the average number of regressors14 in this type of researches found by Sala-i-Martin (1997). Many studies conducted researches with similar variables (see for example: Khan and Senhadji, 2001; Hineline, 2007 or Omay and Kan, 2010, etc.). Hineline calculates the chances with investment share of GDP, openness to international trade, and inflation will influence growth. He finds that the percentage values are 100, 92 and 89 percent, respectively, which justify their presence in the model.
3.3.1 Variables
This section is going to introduce the variables included in the regression model. The dependent variable is the annual growth rate of gross domestic product (ΔY) which is expected to be influenced by the independent variables. Data of the dependent variable are taken from the online database of the World Bank.
Some observations on GDP growth will be excluded because there are some data points with more than plus or minus 30 percent GDP growth rate per year (for example in case of Liberia or Rwanda in some years), however, this step will eventuate an unbalanced panel dataset. Ghosh and Phillips (1998) argue that this step is reasonable because those few observations can occur only under exceptional circumstances (such as civil wars); therefore we can say that they may be suspicious for policy making.
The most important independent variable is the annual rate of inflation (lnINF) which is calculated from consumer price indices. Barro (1995) and Sarel (1996) argue that CPI data is better to use than GDP deflators since consumer price index is
calculated independently from output rates. Since the effects of inflation on growth are not immediate I will apply one-year lag to deal with this issue. Data on yearly inflation rate are gathered from the datasets of the World Bank. In general inflation is expected to have a negative effect on economic growth. In my regressions, this assumption is expected to be valid only above a certain limit which is called structural break. Below this threshold the effect is not unambiguous like above it.
Many studies uses average rates of inflation. The most common way is to use five-year averages (see: Grier and Tullock, 1989; Drukker et al., 2005; Bick, 2010 or Kremer et al., 2011) since it may help to avoid short-term fluctuations but this method has a serious problem: the explanatory power of the coefficients will decrease. Bruno and Easterly (1998) argue that results will get stronger if data frequency is higher. They claim that using, for example, 20- or 30-year averages is nonsense15. Furthermore, Hineline (2007) finds that in case of cross-country studies inflation is not robust but in panel analyses with fixed effects it becomes more robust.
The last problem with five-year averages was revealed by Drukker et al. (2005) who find that there are only few periods when average inflation rate of industrial countries is below the threshold level (2 percent). Because of these reasons I will use yearly data (in line with the researches of e.g. Temple, 2000 or Burdekin et al., 2004; etc.) to estimate the relationship between inflation and economic growth. It will result in a large, rich and unbalanced dataset but it may help to find stronger evidence for the threshold effect of inflation.
Sarel (1996) argues that the logarithm of inflation results in a much more balanced distribution. This assumption is very easy to test with histograms. Figure 1 on page 21 shows the distribution of inflation and the logarithm of inflation. The second histogram is very similar to a bell-shape, whereas the first one is not at all. Using log transformation of inflation leads more reasonable results since it means that multiplicative inflation shocks have the same effects16. Moreover, the logged values may help to avoid the distortion of extreme observations (Ghosh and Phillips, 1998), however, this method provides the opportunity to retain (extreme) high values of
15 This can be seen as a serious critique of cross-country studies which try to capture the effects of inflation on
economic growth.
16 Using simple inflation rates would mean that additive inflation shocks have the same effects on economic
inflation as well, since Barro (1995) finds that the significance of the results will decrease if he excludes countries with more than 40 percent rate of inflation from the regression.
Figure 1 shows the distribution of inflation and the distribution of the logarithm of inflation. In the first histogram we can see that most of the observations clusters close to zero but there are some very large inflation observations as well (e.g. hyperinflation in Congo, Brazil or Argentina) while the distribution of the second histogram is more similar to the bell-shape.
and it is practical since they cannot be logged. I will follow the fourth way – in line with Gosh and Phillips (1998) – and exclude the negative observations.
These regression analyses include other explanatory variables as well. The following paragraphs are going to introduce the important control variables.
Neoclassical theory emphasizes the importance of capital accumulation as a prerequisite of growth. Moreover, newer models highlight the importance of human capital as well. Mankiw et al. (1992) argue that the two types of capital – both human and physical – have significant impact on growth. The latter is measured by the investment share of GDP (INV) and taken from the Penn World Tables (Heston and Summers, 2011). The measure for investment does not contain public investments since they may be positively correlated with inflation because for example after an exogenous shock a government may start pursuing expansionary monetary policy to alleviate the negative effects.
According to Ghosh and Phillips (1998), investment share of GDP also can be influenced by inflation and it can have a further effect on growth. They argue that inclusion of investment share as a control variable may decrease the explanatory power of the coefficient of inflation. In order to deal with this problem and test their findings I will run regressions where investment share of GDP is included and where it is excluded.
Levine and Renelt (1992) emphasize that investment is one of those few variables that robustly impact growth. In their research they affirm the findings of Barro (1991) who claims that investment has a robust coefficient in growth regressions. De Gregorio (1993) argues that including investment share of GDP – as an explanatory variable – into the regression will not change the coefficient significantly. A possible explanation of this finding is that inflation works through productivity of investment rather than its level. This variable is expected to have a positive sign since we may expect that the higher the amount of (foreign) investments in a country the greater is the level of economic progress.
school enrollment rate (EDU) and the observations come from the most recent dataset of Barro and Lee (Barro and Lee, 2011).
Unfortunately, this data collection contains observations only on five-year base. In order to transform the data into a suitable form (yearly base) I have to apply a linear interpolation method which will lead to annual observations. It is important to remark that the value of school enrollment changes slowly which fact makes linear interpolation applicable for the dataset.
Glaeser et al. (2004) claim that human capital is an important control variable in growth equations, since it has impact on growth. Moreover, Levine and Renelt (1992) argue that countries tend to grow faster with higher rate of secondary school enrollment. They also find significant, positive and robust coefficient for secondary school enrollment rate.
The effect of this variable is not immediate either. Since I use enrollment rates it will mean that those students, who have just started to attend the secondary school, need approximately four years to finish it and enter the labour market. Because of this reason I will apply a four-year lag for the rate of secondary school enrollment during my research.
Openness to international trade (OPEN) is calculated by the sum of export and import divided by GDP17. Ghosh and Phillips (1998) emphasize the importance of openness since trade is not only a determinant of growth but it is a channel of technological transfer. Drukker et al. (2005) include this variable in their model and find positive association between openness to international trade and economic growth. Their findings are in line with the endogenous growth theory and the neo-classical growth models. Hineline (2007) justifies the importance of openness to international trade and calculates that it influences economic growth with 92 percent probability.
The last explanatory variable is the annual growth rate of population (POP). Baumol et al. (1989) find that it has positive effect on growth since the higher number of people can increase the level of production and consumption. Drukker et al. (2005) argues that population growth is a necessary control variable because of the findings of endogenous growth theory. Levine and Renelt (1991) find that out of the 41
articles, which they examined about the determinants of economic progress, 29 used population growth as a control variable. In their later study (Levine and Renelt, 1992) they also included population growth – among others – but they find negative and significant (or with different specifications insignificant) correlation with economic progress.
An important problem will arise if I include population growth into my analysis. Infants and children cannot work immediately so they will appear on the job market only more than a decade later. This is the reason why I have to use fifteen-year lags for this control variable. With this specification, the association between population growth and economic progress is expected to be positive since the growing number of workers can increase the rate of total consumption and production.
Andrés and Hernando (1999) find that if they add additional variables to their bivariate regression, such as investment share of GDP, population growth or school enrollment rates, the negative correlation between inflation and economic growth will survive. Gylfason and Herbertsson (2001) justify their results and conclude that inflation seems to be a robust variable.
3.3.2 Dummy Variables
Dummys are binary variables which can have the values of 1 and 0. They are easy to use and they can help to divide the dataset into smaller groups if it is necessary. In my analysis they will have very important roles since they can help to distinguish between rich and developing countries and determine the threshold values of inflation.
A dummy variable will define whether a country is an industrialised or a developing one. I will use the classification of the International Monetary Fund’s World Economic Outlook18 which defines 34 developed and 150 emerging economies. Since there are no data for many of these countries during the period some of them have to be be excluded. Eventually my dataset will consist of 106 countries for the period 1975-2005 which divide into two groups: 27 industrialized countries and 79 non-industrialised states. It is important to remark that there are some advanced economies (for example Cyprus and Malta) which will be considered as
18 This list is available at: http://www.imf.org/external/pubs/ft/weo/2010/01/weodata/groups.htm#oem [Accessed
developing countries because their labelling was changed only after 2005. In case of Germany between 1975-1991, data of the Federal Republic will be used. The states, which are included in the research, are summarized in table 2 on page 25.
Threshold values of inflation will be also defined by dummys for both types of countries. Dummy variables are appropriate for this role since they can get value 1 above a certain threshold and 0 otherwise. In my growth regressions in general 2 dummy variables will be included in the same time: one for the labelling the country (advanced or emerging) and one for adequate thresholds of the group (2 percent or 15 percent, respectively).
Advanced Economies Developing Countries
Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong SAR, Iceland, Ireland, Israel, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Singapore, South Korea, Spain, Sweden, Switzerland, United Kingdom, United States of America
Algeria, Argentina, Bahamas, Bahrain, Barbados, Bolivia, Botswana, Brazil, Bulgaria, Burkina Faso, Burundi, Cameroon, Colombia, Congo, Democratic Republic of, Costa Rica, Côte d'Ivoire, Cyprus, Dominica, Dominican Republic, Ecuador, Egypt, El Salvador, Fiji, Gabon, Gambia, Ghana, Grenada, Guatemala, Honduras, Hungary, India, Indonesia, Iran, Jamaica, Jordan, Kenya, Lesotho, Liberia, Madagascar, Malaysia, Malta, Mauritius, Mexico, Morocco, Nepal, Niger, Nigeria, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Qatar, Rwanda, Samoa, Saudi Arabia, Senegal, Seychelles, Solomon Islands, South Africa, Sri Lanka, St. Lucia, St. Vincent and the Grenadines, Sudan, Suriname, Swaziland, Syrian Arab Republic, Tanzania, Thailand, Togo, Tonga, Trinidad and Tobago, Tunisia, Turkey, Uruguay, Vanuatu, Venezuela, Zimbabwe Table 2. It contains those countries (27 rich and 79 developing) which were included into the research.
3.4 Other Important Issues
3.4.1 Heteroskedasticity
Heteroskedasticity happens when the standard deviation of a variable is non-constant. This problem can arise during my thesis because I will include developed and developing countries in my regressions as well and these states have different average level of inflation or population growth rate. In order to test for this possible problem I will conduct modified Wald tests for groupwise heteroskedasticity (Hill et al., 2008) for all the four groups which I will run a regression on. After running these tests, if I have to reject the null hypothesis (which says my data is homoskedastic19) I will apply White’s robust standard errors (White, 1980) to alleviate the problem of heteroskedasticity.
3.4.2 Endogeneity
The problem of endogeneity may arise in my analyses because for example a potential shock can generate higher inflation and cause inverse relation between inflation and economic progress. However, it is also possible that an economic shock is followed by expansionary monetary policy that causes higher inflation (Barro, 1995).
Khan and Senhadji (2001) argue that the seriousness of this issue depends on the causality and if it runs from inflation to growth endogeneity will not be a crucial problem. Fischer (1993) finds that the direction of causality runs more likely from inflation to economic growth.
If endogeneity arises in the context of inflation and economic growth, we can ease this problem by using instrumental variables. Barro (1995) applies lagged values of inflation20 and Hineline (2007) claims that lagged inflation is the most common instrument in the literature of inflation-growth relationship. Since I will include the (1-year) lag of inflation in the model I will try to ease this problem from the beginning. The lagged value of inflation is a reasonable instrument since the GDP growth of the second year cannot influence the rate of inflation of the previous year.
19 H
0: sigma(i)^2 = sigma^2 for all i.
20 He claims: „Earlier values of a country’s inflation rate have substantial explanatory power for inflation. Lagged
In case of one group of regressions (developing states with high rate of inflation), where I will not use lags, I will run a Hausman test (Hill et al., 2008) for endogeneity and if it is a problem I will try to alleviate it by using lag variables.
3.4.3 Robustness
In the results part of my paper I am going to examine the robustness of the relationship between inflation and growth as well. According to Hineline (2007) inflation has a quite robust coefficient in this relation if he carries out a panel data analysis.
Robustness will be tested by adding and omitting right hand side variables of the equation. Andrés and Hernando (1999) find robust evidence for inflation-growth relation in their research. They argue that the negative sign of the correlation survives the inclusion of other explanatory variables, such as population growth or investment share of GDP. I will be able to say that my results are robust if the coefficients of the interested variable (inflation) can keep its significance after including other explanatory variables.
An interesting feature of this study that I will be able to test whether or not inflation has a robust coefficient above and/or below the structural breaks or not robust at all.
3.5 The structure of the research
This part is going to summarise briefly the next main chapter of my thesis, which is the results part, and present the structure of the regressions and tests.
After creating a table for description I will run regressions in order to use its results for a Hausman test. This test will show me whether I have to use fixed or random effect model. In the next step I will test the normality of the variables. I will conduct a skewness and kurtosis test for normality21 to test the distribution of the dependent variable. I am also going to create a correlation matrix for testing the possible problem of multicollinearity in the model.
After this step I will carry out the main regressions. First of all with all countries to check the general effects of inflation on growth then I will decompose my dataset into four parts according to the level of development (rich or poor countries) and the
level of inflation (above or below the structural breaks) and run four distinct groups of regressions with using dummies to distinguish between units. Furthermore, I will run regressions where I will apply the low threshold (2 percent) for developing countries and the high structural break (15 percent) for advanced states to check whether those regressions can result in any significant outcome.
To test for the possible problem of heteroskedasticity I will carry out modified Wald tests for groupwise heteroskedasticity22 in each case and use White’s robust standard errors (White, 1980) if they are necessary.
Up to date with conducting regressions I will perform the robustness check as well which means that I will include and exclude explanatory variables in order to figure out whether or not inflation is a robust variable.
Table 1 It summarizes the thresholds of earlier researches.
Table 2 It contains the countries which are included into the analyses. Table 3 Current table.
Table 4 Table for the descriptive statistics. Table 5 It shows the result of the Hausman test.
Table 6 It comprises the result of the normlity test of the dependent variable. Table 7 This table includes the correlation matrix.
Table 8 Table with the outcome Wald heteroskedasticity test for regression 4. Table 9 Results of regressions with every country without applying thresholds. Table 10 It includes developed countries with lower than 2 percent inflation rate. Table 11 It comprises developed countries with higher than 2 percent inflation rate. Table 12 It contains developing countries lower than 15 percent inflation rate. Table 13 It shows developing countries with higher than 15 percent inflation rate. Table 14 It shows the results of regressions of table 13 with lagged inflation. Table 15 This table is for regressions with switched breaking points.
Table 16 It lists the justified and refuted hypotheses. Table 3 lists all the tables of my thesis.
The following parts of my research will contain many tables. To make it easier to overview this thesis, I will include a table (see table 3 on page 28) which contains all the 16 tables of my paper.
4. Results
This part of my thesis is going to introduce and explain the results of my analyses which were introduced in the previous sub-chapter.
4.1 Preliminary tests
Table 4 on page 29 contains the descriptive statistics for the whole dataset. First of all I would like to mention the reason of the difference in the number of observations (N). In my sample there are 106 countries (n) for the period of 1975-2005 but because of the lags of inflation, population growth and school enrollment rates, the number of observations are higher for those control variables.
Variable Mean Std. Dev. Min Max Observations
ΔY overall 3.393349 4.633061 -26.66846 29.7 N = 3196 between 1.55448 -1.215139 8.47982 n = 106 within 4.364922 -24.4779 31.89055 T-bar=30.1509 INF overall 37.49359 503.2319 -100 23773.13 N = 3300 between 130.3567 1.850987 1110.694 n = 106 within 486.1696 -1069.206 22699.93 T-bar=31.1321 lnINF overall 2.030655 1.288185 -4.190631 10.07631 N = 3176 between .8193531 .4711641 4.669449 n = 106 within 1.003139 -3.713372 8.569415 T-bar=29.9623 INV overall 24.01481 10.26411 -15.84472 86.34332 N = 3275 between 8.408038 6.34098 50.31005 n = 106 within 5.948913 -9.14524 79.03196 T-bar=30.8962 EDU overall 32.39051 15.32178 .9 80.1 N = 3198 between 4.398115 21.51636 42.98282 n = 106 within 14.68264 -2.955603 82.07204 T-bar=30.1698 POP overall 1.851526 1.255036 -8.271369 11.18066 N = 4863 between .8551056 .2018579 3.750798 n = 106 within .9223877 -7.96311 9.808881 T-bar=45.8774 OPEN overall 76.62996 45.83803 6.320343 428.4591 N = 3166 between 51.37433 18.99886 397.1306 n = 106 within 16.32621 -4.576163 220.0612 T-bar=29.8679 Table 4 with the descriptive statistics.
(-100 or 23773 percent) but the log form helps to decrease these enormous rates (in this case the minimum rate is -4.19 while the maximum is 10.07).
The values of the other variables seem reasonable. The average GDP growth rate (ΔY) of the 106 countries was 3.39 percent during the period. The average of population growth (POP) was also positive (1.85 percent) between 1975-2005 but there were many countries with negative rates or in some cases with extreme high negative rates (minus 6-8 percent) which were the consequence of serious internal conflicts or wars (e.g. in Rwanda in 1992-1993). Furthermore, openness to international trade (OPEN) is showing very high differences in its rate (6.32 – 428.45 percent). It means that there are closed economies (e.g. periods of autarky in Argentina or India) and opened ones (e.g. Singapore) in my sample as well.
In the next step I conduct a Hausman test (Hill et al., 2008). First of all I run two regressions with the same control variables, but one with fixed effect and with random effect. The Hausman test compares the residuals of the two regressions and formulates a null hypothesis (H0) which claims that the coefficient estimates of the regressions are equal one another. The result of this test can be seen in table 5 on this page. The p-value is 0.000 which means I have to reject H0 and conclude that fixed effect estimator is better to use in my analysis.
H0: Difference in coefficients not systematic
chi2(5) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 37.81
Prob>chi2 = 0.0000
Table 5 which contains the result of the Hausman test.
A very important test of every regression analysis is the normality test since it can prove whether or not a variable can be described by normal-distribution. The easiest way for testing normaility is to create histograms and then decide if they are bell-shaped. The other option is to run statistical tests for normality which check the values of skewness and kurtosis. The former one describes the symmetry of distribution, while the latter measures the “peakedness” of the normal-distribution.
According to the normality tests I have to say that the observations of my dependent variable (GDP growth) are not normally distributed since the p-value is strikingly low which means I have to reject the null hypothesis of normality.
A possible solution of this problem is the application of log transformation of the variable. In my case this method will not work since I would lose too many observations. However, I will use robust standard errors in the regressions which may help to ease the problem. Furthermore, I included the histogram of GDP growth (see: figure 2 on page 31) which shows that the distribution of this variable is very similar to the bell-shape23.
Skewness/Kurtosis tests for Normality
--- joint ---
Variable Pr(Skewness) Pr(Kurtosis) adj chi2(2) Prob>chi2
ΔY 0.000 0.000 . 0.0000
Table 6 includes the results of the normality test of the dependent variable.
Figure 2 shows the histogram of the distribution of GDP growth.
The last important preliminary test is a correlation matrix which can be seen in table 7. The correlation matrix can show whether or not multicollinearity is a serious problem in the resarch. Multicollinearity means that two or more explanatory
23 A possible explanation of the shape of figure 2 is the central limit theorem (Rice, 1995) which
variables of the regression are highly correlated. Perfect multicollinearity happens when the correlation value is either –1 or 1 between two variables.
In table 7 we can see that –0.3 and 0.24 are the greatest variables which means that multicollinearity is not a serious problem in the model. These rates are very far from the rates of perfect multicollinearity and it means that I do not have to deal with this problem in my analysis.
ΔY lnINF lagged INV EDU lagged POP lagged OPEN ΔY 1.0000 lnINF lagged -0.1001 1.0000 INV 0.2123 -0.1580 1.0000 EDU lagged 0.0132 -0.1232 0.0151 1.0000 POP lagged 0.0619 0.2405 -0.0374 -0.1672 1.0000 OPEN 0.1226 -0.3002 0.2902 0.0490 -0.1068 1.0000 Table 7 shows the correlation table.
The problem of heteroskedasticity is also an important issue in my models. I have to deal with it because I run modified Wald test for groupwise heteroskedasticity and it results in very low p-value which means that I have to accept the alternative hypothesis and reject the null one about homoskedasticity. To deal with heteroskedasticity I will use White robust standard errors (White, 1980) which slightly increase the p-values and standard errors of the coefficients. I will not include all the Wald tests in this part because the results are very similar with almost zero p-values but part 1 of the Appendix contains the outcomes of the 29 distinct regressions. In table 8 we can see one of these tests which shows the presence of heteroskedasticity. In Appendix it can be seen that heteroskedasticity is a serious problem in every regression. It means that I will use White robust standard errors in every regression of this thesis.
Modified Wald test for groupwise heteroskedasticity in fixed effect regression model
H0: sigma(i)^2 = sigma^2 for all i
chi2 (106) = 11800.32 Prob>chi2 = 0.0000
4.2 Main regressions
This part contains the main regressions which were introduced in the methodology part.
Table 9 (page 34) contains the results of the first group of regressions. In this case I included developed and developing countries as well. The structure of the table is the following24: the first column comprises a bivariate regression where only inflation rate (lagged lnINF) and the growth rate of gross domestic product (ΔY) is included. Moving to the right the number of explanatory variables increases. In the second column I add investment share of GDP (INV), and openness to international trade (OPEN) in the third one. This is in line with Hineline’s (2007) findings who calculated that these variables have the strongest impact on economic growth. Column number four includes all the independent variables (controls of column 3 plus the lagged value of secondary school enrollment rate [EDU] and the lagged rate of population growth [POP]). In the last column I excluded the investment share of GDP (INV) to test whether or not this control variable decreases the explanatory power of inflation (Ghosh and Phillips, 1998).
As it can be seen in all the five columns inflation has negative effect on economic growth between the period of 1975-2005 and these results are significant at 1 percent level. The coefficient tells us that if inflation ceteris paribus increases by one percent it will lead to a -0.002425 percent fall in the growth rate of the next year in case of bivariate regression (1) and a -0.0031 percent change if all the control variables are included into the regression (4). The explanatory power of the main control variable (lnINF) increases as I include other control variables. Moreover, when I exclude investment share the explanatory power of inflation decreases which is in contrast with the idea of Ghosh and Phillips (1998). This table shows that inflation is a robust variable because if I include new controls lnINF does not turn insignificant.
The other variables also have the expected sign: higher investment share of GDP and openness to international trade ceteris paribus result in higher growth rate and these variables are significant at 1 percent level in all cases. If the former increases by 1 percent growth will change between 0.1299 (4) and 0.1344 (2) percent while the
24 All the main result tables will follow this structure.
25 Because of the logarithm value I have to use the following formula to estimate the effect of lnINF on ΔY:
