Economic Growth in Developing Countries

Economic Growth in Developing Countries

Master thesis

International Economics & Business University of Groningen

Laurens de Nooijer

1258052 Faculty of Economics University of Groningen Groningen, The Netherlands

Supervised by

Many developing countries are highly dependent on primary commodities for their export income.

Commodity prices can therefore be expected to impact long-run growth rates in these countries. Using

panel data analysis, the effect of price trend and volatility on growth is tested for 93 developing

countries for the period 1965-2005. Country-specific price indices are constructed using 65 commodity

price series, which are then separated into a long-run trend component and a component capturing

the variability around this trend. Price volatility is found to significantly impact growth rates, especially

for non-oil exporting countries. This might indicate that oil exporters, despite high oil price volatility,

have become resilient to price shocks. A higher degree of openness to trade does not significantly

mitigate the negative effect of volatile prices, nor is this effect found to be stronger in small island

states and landlocked countries. The results are robust to changes in the estimation method.

Economic Growth in Developing Countries

Preface

The idea of resource dependence as the subject for my thesis first occurred to me during my

internship on Curaçao. While researching the economic performance of the Netherlands Antilles

compared to other countries in the Caribbean, I came across the resource curse hypothesis. This

immediately intrigued me. Is there really a relationship between resource abundance and economic

performance? The Caribbean basin itself shows a mixed picture: Trinidad & Tobago, a major oil

exporter, has enjoyed a healthy economic growth, while the Guyana’s have not seen much progress

despite vast natural riches. During the past months, I delved deeper into the large body of literature

surrounding this topic, and this paper is the outcome of my research. I would like to thank everyone

who took out time to review and comment on my work. I owe much gratitude to Dr. Van Leeuwen for

his time and insightful comments, and to Dr. Koetter for his advice especially on the methodology of

the paper.

Contents

Contents ... - 4 -

1. Introduction... - 6 -

1.1 Research objectives... - 7 -

1.2 Paper outline ... - 8 -

2. Resource dependence and price movements ... - 9 -

2.1 Commodity prices and terms of trade ... - 10 -

2.2 Terms of Trade trends... - 11 -

2.2.1 The Prebisch-Singer hypothesis... - 11 -

2.2.2 Empirical evidence... - 12 -

2.3 Terms of Trade volatility... - 13 -

2.3.1 Origins of volatility in commodity prices... - 14 -

2.3.2 Empirical evidence... - 15 -

3. Price movements and economic growth ... - 16 -

3.1 Price trend and growth ... - 16 -

3.2 Price volatility and growth ... - 17 -

3.2.1 Effects on producers... - 17 -

3.2.2 Effects on governments ... - 18 -

3.2.3 Empirical evidence... - 19 -

3.3 Economic growth models ... - 20 -

3.3.1 Openness ... - 21 -

3.3.2 Small islands and landlocked countries ... - 22 -

4. Methodology ... - 23 -

4.1 Commodity price trend and volatility ... - 24 -

4.1.1 Seasonal adjustment ... - 25 -

4.1.2 Stationarity... - 26 -

4.2 Other independent variables ... - 27 -

4.3 Dummy variables ... - 27 -

4.4 Data... - 27 -

4.4.1 Data concerns... - 28 -

4.5 Panel data analysis specifications ... - 29 -

4.5.1 Normality, outliers and descriptive statistics... - 30 -

4.5.2 Multicollinearity ... - 31 -

5. Empirical analysis and results ... - 32 -

5.1 Price movements and growth... - 33 -

5.1.1 OLS and within-effects estimation ... - 33 -

5.1.2 System GMM model specification ... - 34 -

5.1.3 System GMM estimation ... - 35 -

5.2 Regional effects ... - 37 -

5.3 Type of exporter effects ... - 39 -

5.4 The effect of openness... - 40 -

5.5 Small islands and landlocked countries ... - 42 -

5.6 Growth regression robustness tests ... - 42 -

6. Conclusions ... - 45 -

Bibliography... - 48 -

Appendices

1. Introduction

“We are in part to blame, but this is the curse of being born with a copper spoon in our mouths”

This quote by Kenneth Kaunda, the former president of Zambia, touches the origin of one of the most persistent problems for developing countries in international trade. Zambia, a large landlocked country in southern Africa, has always been dependent on the exports of copper for most of its income. Once considered to be one of the most developed countries on the continent, it rapidly plunged into poverty when a severe depression hit the copper market in the mid-1970s. Up to this day, the country’s economy still did not fully recover.

The example of Zambia is no exception: resource dependence is a widely known characteristic of many developing country economies. Especially African, Caribbean and Latin American nations derive an important part of their national income from the exports of primary goods, as can be seen in appendix 1. Whereas developed countries on average only depend on primary goods for 6 percent of their export income, in the least developed countries (LDCs) the average is 46 percent. For some countries, for instance Zambia, the ratio is higher than 80 percent.

Such dependence can create several serious problems, chief among them declining trends and high volatility in commodity prices. Primary products are usually traded on global commodity markets, where they are bought and sold at world market prices. These prices are influenced by uncontrollable and unpredictable factors such as the weather, foreign economic crises, the development of synthetic product alternatives, and other supply and demand shocks. Most of these factors are outside the sphere of influence of the individual exporting countries.

It has been empirically shown that, in general, commodity prices have been in a significant downward trend over the last few decades. The UNCTAD combined non-fuel commodity price index, depicted in figure 1, stood at 50 percent of its 1980 value in 2002. According to UNCTAD, this amounted to a loss of billions of dollars in potential trade revenues for the world’s poorest countries.

Besides adverse trends, commodity price volatility can have significant impact by creating instability of export earnings and causing significant uncertainty on both micro- and macroeconomic levels.

Governments, especially in resource-dependent developing countries, are highly dependent on exports for their income, and instability in export revenues can complicate budgetary planning and the ability to meet short-term debt obligations. On a microeconomic level, farmers and rural households face unpredictable income, forcing them to make suboptimal decisions in order to reduce risk.

1Commodity dependence is usually measured as the share of the three leading commodities in the total exports of a country.

This definition is also used in appendix 1.

Figure 1: World market prices for commodity sub-groups, 1960-2002 (index, 1980=100)

Source: UNCTAD Least Developed Countries Report 2002

1.1 Research objectives

This paper aims to investigate the macroeconomic effects of commodity price trends and volatility in developing countries. Appendix 1 lists the countries included in this research, and the major products these countries export. As can be seen, a large number of countries are highly undiversified within the resources trade, obtaining most income from the marketing of only one or two commodities. Because export earnings are the most important source of national income for many (if not most) developing countries, the exposure of the economy to price changes in primary goods can be substantial, and this constitutes large risks on both macro- and microeconomic levels.

On the basis of these facts, it could be expected that commodity prices significantly impact economic growth, but empirical research into the relationship has failed to universally support this proposition.

First, employing new analytical techniques and country-specific data, I attempt to find an answer to the question What is the impact of commodity price movements on economic growth in developing

the effects of volatility and price trends? Are there differences between regions or types of exporters?

Second, the role of trade openness of the economy is considered. Here, it is examined whether open

countries are more adversely affected by price volatility and whether this conditioning effect is

significant or not. This issue is still an open question in the literature.

Third and last, it is tested whether small island states and landlocked countries are at a structural disadvantage when it comes to resource dependence. Several authors as well as international organisations have stressed the important detrimental effect of remoteness and vulnerability to economic and ecological shocks on these countries’ economies. Are commodity price movements having a structurally higher impact on these countries?

These relationships are tested using advanced econometric methods and country-specific price indices for 93 resource-dependent countries. Price movements are separated into a long-run trend and short-term movements around the trend to enable their separate analysis.

1.2 Paper outline

In this first chapter, the paper’s research problem is specified and the importance of the topic is discussed. The next chapter deals with the theoretical background on resource dependence in general, and the empirical research into commodity price movements in specific. In chapter three, a more in-depth analysis of the links between commodity prices and economic growth is presented.

Based on existing literature, a number of hypotheses that will be empirically tested in this research are

formulated. The methodology and data used to do so are discussed in chapter four. Chapter five then

presents the empirical analysis and the results, and the final chapter discusses the findings of this

research.

2. Resource dependence and price movements

There is a wide body of literature on the effects of resource dependence on economic growth. Several recent studies have reported that since the 1960s resource-abundant countries have experienced considerably slower growth than resource-poor countries (Sachs and Warner, 2001; Gylfason, 2001).

Sachs and Warner, who published an influential series of papers on this issue, summarize the link in the scatter plot depicted in Figure 2. Despite the empirical evidence in these papers, however, economic theory provides no satisfactory explanation as to why resource abundance should be inherently detrimental.

Figure 2: The relationship between resource dependence and economic growth, 1970-1989

Source: Sachs and Warner (2001)

Recent contributions to the discussion have focused on a range of explanatory factors that are rooted

in political circumstances. Ross (1999) divides explanations of the political aspects of the ‘resource

curse’ into three categories: cognitive theories, which argue that failures to profit from resource wealth

are to be blamed on the short-sightedness of policy makers; societal theories, which contend that

resource exports tend to politically empower certain groups that favour growth-impeding policies; and

state-centered theories, which claim that when states obtain most of their revenues from external

sources, they become less accountable to their populace, making them ardently guard the status quo

and hence less focused on improving economic policies. Examples of these three categories can be

found in practice: cognitive behaviour in Caribbean sugar-exporting countries, societal behaviour in the 1970s Latin American countries, and state-centered behaviour in Middle Eastern oil exporters (Ross, 1999).

Such theories that find their origins in the field of political science have been used by economic scholars to empirically test the existence of a resource curse. Papers in this category include Sachs and Warner (1995) on poor governance, Collier and Hoeffler (2001) on the greater incidence of civil war in resource-rich countries and Auty (2001) on the role of export diversification. A number of these authors also consider the role of price movements as a possible explanation for slow growth. The issue of declining relative commodity prices was first raised in the 1950s, and has been investigated ever since, employing more extensive data sets and increasingly sophisticated statistical tools. It is to this discussion that I turn now.

2.1 Commodity prices and terms of trade

The literature on terms of trade (TOT) behaviour in developing countries is discussed in this chapter.

Throughout the chapter, terms of trade trend and volatility and commodity price trend and volatility are used interchangeably, for the following reason. Across developed and developing nations, manufactured goods invariably constitute about 65 percent of the import basket. These products are mainly manufactured in developed countries, who import raw materials to process them. Hence, the direction of trade flows between developed and developing countries can roughly be presented as in .

Figure 3: International trade flows

Given figure 3, the TOT for developing countries can be thought of as the relative price of the exported primary commodities to the imported manufactures. The latter being quite stable, movements in the TOT of resource-exporting countries are mainly caused by movements of export prices (Baxter and Kouparitsas, 2000). The prices of primary resource exports can thus be considered a proxy for the TOT

, hence trend and volatility in commodity prices are an approximation of the trend and volatility in the TOT.

2For this reason, the terms of trade of resource-exporting countries are often referred to as ‘Commodity Terms of Trade’ or ‘Net Barter Terms of Trade’.

Primary commodities

Manufactured goods

Developed Countries Developing

Countries

2.2 Terms of Trade trends

Classical international trade theory dictates that it is beneficial for countries to specialize in the production of goods in which they have a comparative advantage. If these goods are then freely traded in the world market for products which the country itself can produce less efficiently, both countries derive economic benefit. This theory, first systematically explained by David Ricardo in 1817, was extended in the 1930s by Eli Heckscher and Bertil Ohlin.

Free trade theory gained much support from the 1970s onwards, as it was generally believed to be advantageous to all countries. However, the theory was undermined by real-world observations about the lagging prosperity of many countries, especially former colonies in Africa, Latin America and Asia.

According to two influential economists, Hans Singer and Raúl Prebisch, this was due to the high dependence of these economies on primary commodities, and the adverse effects of relative prices for these products. Their work set the stage for a large body of research on relative prices.

2.2.1 The Prebisch-Singer hypothesis

Independently of each other, Prebisch (1950) and Singer (1950) noted that over the long run, the prices of primary commodities tend to decline relative to the prices of manufactured goods. Their belief was based on the decades leading up to 1950. According to the authors, the observed downward trend is caused by three main factors:

I. Commodities markets are purely competitive, because demand for primary commodities, such as basic food stuffs, is relatively inelastic (Engel’s Law), and there is little to no diversification among producers.

II. The market structure for manufactured goods is less competitive, which leads to monopoly pricing.

III. There exists an unequal distribution of the benefits of technological progress in primary goods production versus manufacturing production.

The reasoning of the Prebisch-Singer hypothesis (PSH) runs as follows. As technology progresses,

the productivity in manufactured goods production increases. According to the PSH, in the developed

world the associated cost savings are transferred to labour in the form of higher wages, mainly

because of well-organised labour (e.g. due to labour unions). In addition, markets for manufactured

goods are oligopolistic, enabling producers to determine prices. This combination leads to only modest

price decreases for manufactured goods. In developing countries, where labour is unorganised and

abundant, the gains of technology do not fall into the hands of the workers. Instead, producers of

primary goods have to compete over buyers for their products, but since global demand for primary

goods is relatively inelastic they are forced to lower their prices in a race to the bottom. The result is

that all benefits from technological progress and international trade eventually accrue to the developed

countries. The resource exporters, meanwhile, face deterioration in the terms of trade.

2.2.2 Empirical evidence

Probably because of its implications for the global distribution of wealth, the Prebisch-Singer hypothesis is one of the most debated issues in international trade research. The detection of negative commodity price trends is supported by empirical research by several authors (Spraos, 1980; Basu and McLeod, 1992). Most prominent among the papers supporting the PSH, however, is an important study by Grilli and Yang (1988) on 20

century commodity prices. The authors construct new price series for 24 commodities and build seven indices using these price series. They detect a negative trend of 0.6 percent per annum, which is less steep than the results posited by Prebisch.

Grilli and Yang (GY) also investigate for structural breaks in the series. These can be thought of as

‘once-and-for-all’ changes in commodity prices, meaning sharp shifts that become permanent features of the price series because there is no subsequent reversal to the long-run trend. Such breaks can lead to incorrect perceptions about trends: a single sharp price drop can create the impression that a certain price series is downward trending over a long period of time, even when it is in fact not displaying any trend or an upward one.

GY found no structural breaks in their data, only an acceleration in the downward trend from 1921 onwards.

Among the first to question this finding were Cuddington and Urzúa (1989), who found no downward trend but detected a large structural break in the GY series after 1920, the same moment where GY noticed trend acceleration. The question thus arose whether the observed decrease in prices over the long run was caused by a single downward break or by a shift in trend.

While this discussion on the cause of long-run downward price movements seems feeble (Ocampo and Parra, 2002), it is important from an econometric analysis point of view. When price series contain structural breaks, they have to be considered non-stationary, requiring different methods of analysis.

Scholars are now more or less in agreement on the issue, with the majority of them agreeing that TOT are indeed non-stationary with breaks. However, they disagree on the number and timing of the breaks. Powell (1991), for instance, concludes that there exists no downward trend, but that commodity prices are depressed by three breaks, in 1921, 1938 and 1975, whereas Bleaney and Greenaway (1993) only find evidence for one or two breaks.

There are several reasons why the search for long-term trends has led to such different findings. First, one of the problems with TOT analysis lies in the fact that commodity prices do not display co- movement, contrary to what some authors claim

. TOT experiences differ from country to country depending on their predominant type of export products. For instance, in any given year, producers of coffee can enjoy a price increase, while coal exporters are experiencing a price slump. This is empirically made clear by Cuddington (1992), who tests price movements of 26 commodities over the period 1900-1983. His conclusion is that 16 of these prices are trendless, 5 of them trend upward and the remaining 5 display a downward trend. This leads him to reject the PSH as a universal truth.

3Commodity price developments during World War I were excluded from the GY series.

4An example of this is Pindyck and Rotemberg (1990). These authors state that even the prices of completely unrelated commodities displayed ‘excess co movement’. Cashin, McDermott and Scott (1999) and other authors prove that commodity prices do in fact not move in unison to the extent previously thought.

Second, the outcome of any price trend analysis depends greatly on the period taken into account.

Spraos (1980), for instance, finds a downward TOT trend in the 70 years before World War II (the period studied by Prebisch and Singer), but notes that when data up to the 1970s are included in the analysis, the trend largely disappears. Diakosavvas and Scandizzo (1991) discover that the tendency of prices to decline, although present, is quite small and reverses itself given a sufficiently long time period.

Last, the improvement of statistical and econometrical time-series techniques has led to new methods to analyse price behaviour. The ability to detect and model structural breaks, for instance, has led to improved insights on long-term secular price trends.

Two major problems with the PSH have been pointed out by Davis and Tilton (2005), among others.

First, there is the issue of quality improvements. These are quite common for manufactured goods, but less so for primary resources. Hence, even if there is a downward trend in the prices of commodities relative to those of manufactures, this may simply reflect quality enhancements of the manufactured goods, in which case negative relative price trends are likely to be overstated. Besides, quality enhancements in import goods are not disadvantageous to developing countries; therefore, declining relative prices overstate the negative effects for these countries. Removing this factor from prices is extremely difficult.

Second, there is a trade-off between the price one receives for a product and the cost of producing that product. Hence, even if commodity prices have declined, the costs of growing or mining them might have declined even more, for instance due to technological advances. This means benefits from the resource sector to the economy may actually have increased. Davis and Tilton (2005) mention the example of Chile, where copper mining operations have increased despite a decline in real copper prices over the past three decades.

In summary, the proposition that commodity prices are in a long-term declining trend relative to prices for manufactured goods has been difficult to support empirically. Observed downward trends are possibly due to a small number of structural breaks in the price series, in which case the PSH does not hold. Furthermore, the detection of trends greatly depends on the commodities and time period investigated and the research tools selected for this. Even when declining trends are found, these might be due to quality improvements in manufactured goods, and they might be offset by declining production costs for primary resources. These issues constitute weaknesses in the PSH, and they should be kept in mind when interpreting the results of this research.

2.3 Terms of Trade volatility

More recently, the debate has turned from trends to the issue of TOT volatility. Several authors have empirically shown that the effects of instability in the TOT are much more significant to developing countries than a declining trend, if there is any trend at all.

As mentioned earlier, TOT instability is mainly brought about by volatility of export prices. Commodity

prices have historically been much more volatile than manufactured goods prices; the origins for this

fact can be found in their distinct characteristics. These origins of price volatility will now be briefly discussed.

2.3.1 Origins of volatility in commodity prices

The main channels through which commodity prices are determined are the distinct supply and demand shocks. Supply shocks are often of a natural kind, for instance earthquakes, droughts, floods, or locust plagues, and primarily impact agricultural production. Demand shocks, on the other hand, are usually of an external kind (e.g. foreign economic crises, interest rate instability, business cycle effects), and mainly affect commodities that are still to undergo some form of industrial processing, for instance base metals such as copper and tin.

In addition, demand and supply for primary commodities are relatively inelastic, which implies higher price volatility. On the demand side, this is because many commodity inputs cannot easily be replaced by substitutes and because the commodity’s price may not be a large part of the overall value of the final product (e.g. the price of cocoa in the value of chocolate). On the supply side, inelasticity is due to the slow responsiveness of production to prices. In both agriculture and mining, decisions about production levels have to be made before new prices are known to the producers. The time lag between the production decision and the sale can be several months.

These time lags tend to aggravate the pro-cyclical behaviour of prices. When demand rises, prices go up and inventories are depleted. By the time supply can finally react to the higher profits, demand can be in a slump again and there is greater downward pressure on the price. Hence, any supply shock triggers far greater price variability for primary commodities than would be the case for manufactured goods.

In addition to this, the production decision is often highly irreversible. In mining, substantial capital investments need to be made before minerals can be extracted, but production can be reduced or expanded relatively easily, and so can supply to the market due to storage possibilities. In agriculture, however, irreversibility of investment is problematic, especially in case of so-called perennial crops such as coffee or cocoa. In the case of coffee, after the trees have been planted it can take up to 5 years before they bear fruit; they will then continue to do so for about 60 years. Replacing coffee plants with another crop would be extremely costly for most producers, and it would make very little economic sense to do so on the basis of a single price slump. Thus, even when prices are very low it is not likely that producers will decide not to harvest their crop.

Unfortunately, storage of most agricultural products is very costly. The cost of holding inventories is affected by the prevailing interest rate, hence prices are significantly affected by fluctuations in the interest rate. Furthermore, storage of perishable goods is only possible for a limited time. For these reasons, producers will be forced to sell their produce, even if market conditions are unfavourable.

5Note that demand can also be affected by the weather/seasonality, e.g. the demand for oil and natural gas.

2.3.2 Empirical evidence

There is a substantial body of work on the occurrence and effects of commodity price instability. The impact of volatility on growth will be discussed in the next chapter. This section gives a short overview of the literature on its occurrence and the detected changes over time.

One example of a paper that raises price volatility as an important issue is the work of Cashin and McDermott (2002). The authors find a downward price trend of 1.3 percent annually in price series from 1862 to 1999, but conclude that price volatility is so severe that it completely overshadows this negative trend in importance. In extreme cases, they find short-term price changes in certain commodities of up to 50 percent. Likewise, Deaton and Laroque (1992) study 13 commodity prices over the period 1900-1987 and report no detectable trends, but “extreme volatility”.

Does volatility increase or decrease over time? Cashin and McDermott (2002) empirically show that the standard deviation of price movements has increased after 1971. According to Cuddington and Liang (2003), this is due to the collapse of the Bretton Woods system and the return to flexible exchange rates. The authors find strong evidence supporting the presumption that commodity price volatility is higher under flexible exchange rates, which is probably caused by the higher systematic risk to world commodity markets that countries face during flexible exchange rate periods. The causation of this link is not exactly clear, but may perhaps be considered a part of the greater puzzle of exchange rate determinants. Ocampo and Parra (2002) note that in recent decades price volatility has also been intensified by the move towards international trade liberalisation, which has led to the dismantling of commodity agreements and stabilisation boards.

Before proceeding to discuss the effects of commodity price movements, I will quickly recap this

chapter. As the analysis above shows, commodity prices are a good proxy for the terms of trade for

resource-dependent developing countries. In the 1950s, Prebisch and Singer instigated the interest in

commodity price movements with their well-known thesis that commodity prices are in a structural

decline, a decline that deteriorates the terms of trade and the economic growth rates of developing

countries. It has been difficult to find conclusive evidence on this proposition, however; as a matter of

fact, many authors have found that price instability is the greater threat to economic growth. This

volatility in commodity prices is mainly caused by supply and demand shocks, but originates in the

inelasticity of supply and demand of primary resources, in slow responsiveness of production to price

changes, and in the irreversibility of investment decisions. It is important to make the distinction

between agricultural and mining products here: prices for both types of commodities are influenced by

different factors. Over the past decades, commodity prices have been highly volatile at times, and

volatility seems to have increased since the early 1970s. Possible explanations for this are the

increasing number of countries that have currencies with flexible exchange rates, and the global move

towards more open trade.

3. Price movements and economic grow th

The previous chapter discussed the reasons why commodity prices behave as they do, and outlined the empirical proof found in the literature. This chapter proceeds with a survey of the effects of price movements on growth in resource-dependent countries, and the channels through which this works.

First, the impact of commodity price trend and volatility on growth is discussed. I will then proceed to describe some of the findings from economic growth literature, which provides a background for the control variables that will be included in the regression models. This section also discusses the expected role of openness and geography on the link between price movements and growth.

3.1 Price trend and growth

The predicted relationship between price trends and growth depends on how one views the productivity of the resource sector. As was discussed in section 2.1, supporters of the resource curse view, such as Sachs and Warner (2001) argue that the natural resource sector is inherently unproductive because of rent-seeking and failure by governments to invest earnings in successful long term economic projects. Under this view, even upward price trends will not improve economic growth in the long run.

Another group of authors assumes that natural resource activities generate the same economic effects as the manufacturing sector, thus predicting a positive relationship between price trends and growth.

An example of this is Mendoza (1997), who claims that commodity price increases raise the expected rate of return on investment in that sector, leading to capital accumulation and hence economic growth. His findings are corroborated by Bleaney and Greenaway (2001) who find that terms of trade improvements led to higher growth and investment in Sub-Saharan Africa between 1980 and 1995.

Deaton and Miller (1995) investigate a set of African economies and conclude that a 10 percent increase in commodity prices brings about a 6 percent increase in output.

In line with these authors, a positive relationship between price trend and economic growth is expected here. Therefore, the first hypothesis is:

1. Secular price trends are positively related to economic growth

3.2 Price volatility and growth

Commodity price volatility gives rise to several problems. In terms of output (GDP), there can be two effects:

I. volatility of export prices leads to volatility in GDP II. volatility of export prices depresses the level of GDP

As regards the first issue, the link runs through the trade component in the growth model (net exports, or X-M). Fluctuations in export prices can lead to fluctuations in net exports, and this directly affects output. This link explains volatility in GDP, but not how GDP is actually reduced by price volatility. For this, we need to examine the behaviour of economic agents when facing price volatility in the export sector.

Price volatility leads to uncertainty regarding future income, both for governments and individual households. Price movements have a direct impact on commodity producers (farmers, mining companies), and through them an indirect impact on other sectors (suppliers, labour-providing households). It also impacts on the macroeconomic level, through government expenditure, and this in turn affects the whole economy. The outcomes for primary goods producers and governments are described in this section.

3.2.1 Effects on producers

When discussing the effects of volatility on the economy, it is important to distinguish between ex ante effects and ex post outcomes. The ex ante effects are based on uncertainty regarding future prices, which are in turn based on volatility in past price realizations. This will lead risk-averse households to alter their allocations away from risky activities: it is a form of risk management. The ex post effects manifest themselves as households adjust their expectations of future earnings in response to current income, or as income shortfalls force them to adjust current expenditure; hence, it is a form of risk coping (Alderman and Paxson, 1992).

First consider risk management. For local producers of (especially) agricultural products, such as

coffee growers, ex ante price uncertainty means income uncertainty, and often they have little

opportunity to insure themselves against income loss. As a result, households tend to spread the risk

over multiple income-generating activities, for example by growing several different crops or by

entering into non-agricultural activities such as wage employment. This diversification creates all sorts

of adverse effects, such as the sacrifice of gains from crop specialisation and reduced scope for

learning-by-doing. Also, as Dercon (2002) shows, the entry costs into other crops or non-agriculture

activities may be high due to lack of skills, working capital and other requirements. Furthermore, the

perceived need to reduce risk can mean that households engage in activities that have low returns,

just because these activities are low-risk or have risks that are perceived to be uncorrelated with their

other activities (Collier and Gunning, 1999).

Risk coping actions occur after the shock has happened. Households have two options: risk coping using assets, such as self-insurance, or using credit markets. In many developing countries, the latter solution is rarely available to rural households, firstly because credit markets are typically not well developed in these countries, and secondly because households usually have no useful collateral:

land is abundant, and livestock is vulnerable to sickness and theft. Self-insurance using assets is difficult in many circumstances, because the accumulation of assets is unpractical or risky: the most commonly available assets, livestock and food items, are susceptible to spoilage, pests and falling cattle prices. Accumulating liquid assets can be constrained by the ability to defend them against theft (Collier and Gunning, 1999). Even in cases where assets can be successfully accumulated, in periods of crisis all households tend to sell them at the same time, resulting in lower prices.

Although arguably these responses by risk-averse households do not necessarily lead to bad outcomes for them, Collier and Gunning (1999) note that the ‘purchase’ of lower risk is likely to reduce economic growth at the national level. Risk management, they claim, lowers mean income and thereby savings; risk coping requires holding assets, which lowers consumption and investment in future production capacity or productivity enhancements. For these reasons, commodity price volatility could lead to lower growth.

3.2.2 Effects on governments

Besides private investment, the macroeconomic effects of volatility are considered to work through government expenditure. For governments, the linkage between commodity prices and revenues can either be direct, through export taxes and royalties obtained from mining firms, or indirect, when export revenues impact the economy as a whole and this in turn affects government revenues.

Why are government revenues important? First of all, government spending invariably is a major component of GDP in developing countries. Not only does expenditure have a direct effect on output growth, it also affects growth in the long run: government investments in infrastructure and human capital are necessary building blocks for economic development. Furthermore, government officials set trade policy and, especially in developing countries, are in a position to subtract substantial rent from export income. Clearly, government revenue is a main mechanism through which export price volatility affects economic growth. Two important ways in which this can happen are discussed here.

The first is related to indebtedness, the second to institutional quality.

Developing countries typically have a high level of external debt. When facing an unsure stream of

income, governments can unexpectedly run into trouble servicing this debt. For debt to be sustainable,

the rate of growth of export proceeds must exceed the interest rate on outstanding debt, but due to

declining and instable export prices, this condition is often difficult to fulfil. Both debt default and new

debt issues increase servicing costs, which in turn lowers the amount of funds available for public

investment. Also, given revenue uncertainty caused by export price volatility governments may under-

invest in public goods, depressing current GDP and leading to lower human and physical capital in the

long run. This notion was empirically proven by Ramey and Ramey (1995) in a seminal paper to which

I return later.

As is well known institutional quality is generally considered one of the factors inhibiting economic growth in developing countries. These countries are usually small markets: low income in combination of a small population means that these countries are less able to develop efficient market instruments to cope with price volatility, leaving governments incapable of foreseeing price crashes and unable to benefit from price booms. Collier (2003) summarizes the problem with relation to African governments as follows: “not only is Africa atypically exposed to these shocks, it is atypically bad at coping with them.”

3.2.3 Empirical evidence

The discussion on terms of trade instability fits in the broader literature on macroeconomic volatility, a topic that was brought to the attention of economic scholars by Ramey and Ramey (1995). These authors find a strong negative link between volatility and long-run growth, a relationship that is robust for both developing and developed countries. Subsequent research has consistently found proof for this link, which appears to be stronger in developing countries (e.g. Kose et al., 2005; Aizenman and Pinto, 2005).

Likewise, the idea that trade instability harms economic growth has a strong intuitive appeal, and it was accepted as fact without much evidence from research until the 1960s. In an important work in 1966, Alasdair MacBean challenged the assumption, and in his empirical investigation he failed to find clear support for the hypothesis that export instability harms economic growth. A similar conclusion was reached earlier by Coppock (1960) using country-specific export instability indices. Employing the same methodology, Glezakos (1973) refutes the results of both authors, and finds that export instability is detrimental to economic growth in developing countries, but not in developed countries.

Subsequent contributions reached mixed conclusions (Basu and McLeod, 1992; Lutz, 1994;

Guillaumont and Chauvet, 2001).

Some of the research on trade instability has looked into the role of investment levels as a channel

through which trade instability affects growth. In general, investment is believed to be positively linked

with economic growth (see Levine and Renelt, 1991), hence unstable export earnings might deter

private and public investment and affect economic growth rates in this way. Aizenman and Marion

(1996) find that volatility in the terms of trade does indeed have a negative effect on private

investment, which is in line with the findings by Ramey and Ramey (1995) who started the debate on

volatility and growth. Servén (1998), who focuses on uncertainty instead of volatility, reaches the same

conclusions. Looking more specifically at commodity price movements, Deaton and Miller (1995) find

that these strongly affect output in countries in Sub-Saharan Africa, and that this link runs largely

through investment. The causation appears almost immediately and is reinforced in subsequent

periods, with the multiplier-accelerator links transmitting price changes into output growth and

investment. The authors provide no explanation for this observation, but their discussion of the

behaviour by African governments in the face of price booms hints that the problem might be

economic mismanagement. However, in contrast to the findings by Deaton and Miller, Dehn (2000a)

finds that negative commodity price shocks do indeed exert a substantial negative impact on growth, but that this seems to work independently of investment.

Mendoza (1997) models a small commodity-exporting nation and focuses specifically on uncertainty and risk in foreign trade. He concludes that there is a large adverse effect of TOT variability on consumption growth, which his model interprets as the result of the role of TOT as an indicator of country risk. However, as Bleaney and Greenaway (2001) comment, the predictions of the Mendoza model apply to consumption growth, and do not necessarily apply equally well to output growth. In fact, the sign of the relationship between terms of trade volatility and output growth depends on the country’s relative risk aversion; since this cannot be known with certainty, the link between price movements and growth is ambiguous.

Even so, both in Mendoza (1997) and in Bleaney and Greenaway (2001) a negative impact of terms of trade volatility on growth is found. Blattman et al. (2004) also find this negative link in developing countries prior to 1913. Based on these findings, the second hypothesis for this research is constructed, which is:

2. Commodity price volatility has a negative effect on economic growth

3.3 Economic growth models

There is a very extensive body of literature on economic growth and the many possible variables that can explain growth. Since this research employs a growth regression in which control variables are used besides the measures of price movements, a short summary on growth theory is given in this section. This serves to explain which control variables are used in this research.

The model employed here is an augmented neoclassical growth model similar to the model by Barro (1991). The model is based on the Solow growth model. Using the growth rate of real per capita GDP, Barro finds support for the observation by Solow (1956) that GDP growth is negatively related to the initial level of GDP, causing economic convergence between countries. Barro also finds that GDP growth is positively related to initial levels of human capital, and negatively related to the share of government consumption in GDP.

Subsequent work has attempted to determine which factors determine economic growth. Besides initial income and human capital, other often-employed variables are the investment rate, school enrolment ratios, and policy indicators. Sala-i-Martin (1997) finds 60 variables that have been found to be significantly correlated with economic growth in at least one regression. He tests these variables in a model with three ‘fixed’ control variables, i.e. variables that appear in every regression. These are selected because they have been found significant in the great majority of regressions in previous literature. Two of these are similar to Barro: initial GDP and initial human capital (proxied by school enrolment rates), the third is initial life expectancy as a measure of non-educational human capital.

Levine and Renelt (1991) find that besides initial level of GDP the population growth rate, school

enrolment rates and the share of investment in GDP are robust variables in their growth regressions.

Barro (1991), Sala-i-Martin (1997) and Easterly and Levine (1995) also include dummy variables for African and Latin American countries in their regressions, and these are found to be significantly negative. Trade openness was found to have a significantly positive correlation with growth by a number of authors, including Sachs and Warner (1995) and Edwards (1998). According to Dollar and Kraay (2002), the notion that greater openness to trade accelerates economic development is one of the most widely-supported views in economics.

The variables included in the regressions in this research are based on these results from the growth literature.

3.3.1 Openness

Trade liberalisation for developing countries remains a hotly debated topic. It would be interesting to investigate the effect of openness specifically on resource price volatility: does openness diminish the effect of volatility on growth, or does it make the country in question more vulnerable to such external instability?

The link seems unambiguous at first. Intuitively, openness enlarges the exposure of a country to external trade shocks, such as commodity price shocks, and this would lead to a negative impulse to economic growth. However, empirical testing on the role of openness has resulted in conflicting outcomes, sometimes by the same authors: Kose et al. (2003) find that economies that are more open to trade are more vulnerable to external macroeconomic shocks, while they find the opposite effect in a paper published two years later (Kose et al., 2005). In contrast, Hnatkovska and Loayza (2005) find that openness does not play a role in moderating or exacerbating the effect of macroeconomic volatility on growth. These authors use trade as a percentage of GDP as their openness variable.

Looking more specifically at volatility in the terms of trade of developing countries, Combes and Guillaumont (2002) divide their measure for openness, the ratio of exports to GDP, into a structural component and a policy component. The structural component accounts for the stylized facts that larger countries are less dependent on trade (hence, they are less open) and richer countries have stronger and more differentiated demand (hence, they are more open). The authors find that while structural openness increases exposure to terms of trade shocks, openness policies increase resilience to such shocks.

Blattman et al. (2004) investigate the effect of openness on the link between commodity prices and growth in primary resource exporting countries, and find that more openness to trade helps to mitigate negative effects from price movements. Following this finding, the fourth hypothesis under investigation is formulated as follows:

3. Trade openness dampens the effect of commodity price volatility on growth

6There are some dissenting views, most notably by Rodríguez and Rodrik (1999) who find no significant proof that openness policy is significantly associated with growth.

3.3.2 Small islands and landlocked countries

The thesis that small islands (and small nations in general) are at a disadvantage in the global economy is a view that is widely subscribed to by scholars, and it has led the United Nations to devote special attention and resources to small island development. Due to small size, remoteness from world markets, vulnerability to natural disasters and other factors, these nations are believed to face more obstacles to economic development. Other authors, however, have refuted this theory, with finding that small states do not have different growth rates from other countries (Easterly and Kraay, 2000). Landlocked countries are also frequently claimed to be facing structural disadvantages: for instance, not having a coastline significantly increases transport costs (Sachs and Warner, 1997;

Limao and Venables, 1999).

It would be interesting to test whether the geographical features of a country also play a role for resource-dependent countries. Easterly and Kraay (2000) find that small states face higher instability in the terms of trade; this might harm growth through the effects outlined above. Therefore, the last hypothesis is:

4. Small islands and landlocked countries are more adversely affected by commodity price movements than other countries

7For opposing views on the importance of ‘insularity’, see Briguglio (1995) and Easterly and Kraay (2000). For a more elaborate discussion on the importance of geography for growth, see Gallup, Sachs and Mellinger (1999).

4. Methodology

This research applies panel data analysis methods to investigate the impact of commodity price movements on economic growth. A panel data approach combines cross-section data with time series to allow for a regression analysis that studies a set of cross-sectional subjects over time. In this research, the cross-sectional aspect refers to the 93 countries, whose commodity price indices are observed over a period of 41 years, from 1965 to 2005.

The benefits of using panel data over the cross-section approach and aggregate time series methods are substantial. Cross-section methods cannot estimate dynamic relationships because there are only observations taken at one point in time. Aggregate time series data suffers from aggregation biases, which can obscure the underlying country-specific dynamics (Bond, 2002). Furthermore, the inclusion into a regression model of distinctly different countries may give rise to concerns about the validity of the outcomes (Levine and Zervos, 1993), because regression analysis presupposes that countries are drawn from a distinct population. Panel data enables the researcher to identify an individual country’s behaviour across time periods, making it easier to recognise the parameters of economic interest.

There are many possible estimation methods that can be applied to this data set. However, most of them suffer from serious biases that affect the validity of the results. Simple ordinary least squares (OLS) estimation suffers from omitted variable bias, which arises because variables that are not included in the model are correlated with the included regressors. There are two possible ways to correct this. To correct for omitted variables that either vary over time or between countries, a random effects model is recommended. When omitted variables differ between countries but are constant over time, a fixed effects model is preferred.

Unfortunately, both methods are not satisfactory for use with this data set, because they suffer from dynamic panel bias. This bias occurs because the lagged dependent variable and the error term are correlated (Bond, 2002). Another endogeneity problem arises because this method assumes the regressors to be strictly exogenous, which in many cases does not hold.

To avoid these issues, the method used here is the first-differenced Generalized Method of Moments (GMM) introduced by Arellano and Bond (1991) and further developed by Blundell and Bond (1998).

This method is suitable for dynamic panel data with a relatively small time period and a large set of countries, country-fixed effects, and regressors that are possibly endogenous. Authors that have used GMM methods to estimate empirical growth models are Caselli et al. (1996), who test convergence in a Solow model; Greenaway et al. (2002), who analyse the impact of trade liberalisation on growth in developing countries; and Hoeffler (2002) for testing the Solow model on African countries. Dehn (2000a) uses GMM, among other methods, to assess the impact of commodity price shocks and volatility on growth. A technical summary of the GMM estimator is given in appendix 9.

8However, running an OLS and a fixed- or random effects model is still useful, as will be explained in the next chapter.

The choice for the GMM estimator restricts the sample somewhat, because the method requires a minimum of 5 years of consecutive observations per country (Dehn, 2000c). Hence, it drops any country from the sample for which less than 5 years of observations are available, or in case these observations are not consecutive. This means four countries are dropped from the sample.

I will now proceed to describe the variables that are used in the regression, starting with the main variables of interest: price trend and volatility.

4.1 Commodity price trend and volatility

Our main object of study is commodity price fluctuations. In order to study this, a suitable measure of price movements is needed, of which several have been created in past studies. Most of these are price indices of individual commodities, aggregate commodity price indices, or terms of trade indices.

However, as Dehn (2000b) notes, none of these methods are satisfactory, for the following reasons.

A single commodity price series does not fully capture the effects of resource dependence, since only a few countries (mostly oil exporters) are undiversified to the extent that they export only one commodity. Aggregate commodity price indices, on the other hand, do not give an accurate picture.

Since commodity prices do not display the degree of co-movement as was previously thought (Cashin, McDermott and Scott, 1999), such indices would be unrepresentative of the volatility a resource exporter faces. Terms of trade indices, although frequently employed by other authors, also possess undesirable features: they capture not only the price movements of raw materials but also those of manufactured goods and imports, making it impossible to determine with certainty if the observed volatility is caused by commodity prices or by these other factors.

Since we are interested in the effects of resource price volatility on individual countries’ economic performance, it is of interest to construct country-specific price indices. Such an index, akin to a Laspeyres index, would require a fixed ‘basket’ or goods, or exports in this case, against which the price series is geometrically weighted.

I follow Deaton and Miller (1995) in constructing country-specific price indices of commodity price movements for each of the countries in the sample. The index takes the following form:

[1]

=

i

 P for all i = 1, 2,…, n

where P

is the world commodity price (in real 1990 US dollars) for commodity i, and W

is the weighting item. W

is the country-specific value of commodity i in the total export value (in real 1990 US dollars) of all commodities, n, for the constant base period t*:

9These are Angola, Central African Republic, Libya and Myanmar.

[2]

W P Q

P Q

 

The year 1990 is chosen as the base year for determining the weights. The indices are constructed for 93 countries (k = 93) using 65 quarterly price series (n = 65) on 54 different commodities, including nearly all important primary resource export products from developing countries. The products can be divided into 5 distinct categories; an outline is given in appendix 3. A detailed description of the price series that are included in the research can be found in appendix 4, while appendix 5 gives a more detailed description of the construction of the country price indices.

The advantage of the country-specific indices is that they capture the commodity price movements faced by individual countries on the basis of the importance of the individual commodities within the country’s export bundle. This is practical, since we are interested in the effects of resource prices on individual exporters. Furthermore, since developing countries are considered price takers on the world market, the commodity prices can be assumed exogenous to the individual countries that export them.

Even in cases where a country does have a certain degree of market power in a commodity

, they have historically been unable to successfully manipulate prices (Deaton and Miller, 1995).

The main drawback to this approach is that the index does not capture changes in the structure of exports: the composition of the export basket is assumed not to have changed over the period of observation. This means that the index does not include resource discoveries and other production shifts after the base year. However, since the aim here is to measure price movements rather than quantity movements, keeping volumes constant is a desirable property of this approach. In this way, endogeneity problems are also avoided in case primary resource production volume responds to price changes. The effect is that the index will understate income effects of a given price change (Dehn, 2000b). Please note that beside effects on commodity prices, quantity changes can affect prices in other sectors, most notably the non-tradable sector, through the so-called ‘Dutch disease’ effect.

However, discussion of this topic is beyond the scope of this paper.

4.1.1 Seasonal adjustment

Besides a long-run trend and an irregular component (‘shocks’), a time series can contain a third component: a seasonal component. In case of primary products, seasonal effects could occur when certain price series exhibit intra-year movements caused by, for instance, the harvesting season.

Seasonal patterns can be assumed to be predictable to producers, and therefore their occurrence in the price series cannot be regarded as instability. Therefore, before identifying and separating trend and shocks, this seasonal component needs to be removed (Maravall and Del Río, 2001).

This removal is usually done by using a method such as the US Census Bureau’s X-12-ARIMA or by regression on seasonal dummies. In this paper, the X-12 method is used, because seasonal dummies

10For instance Ivory Coast (cocoa), Brazil (coffee), DR Congo (diamonds) or Zambia (copper).

11For a discussion on the Dutch Disease thesis, see Van Wijnbergen (1984).

such as those applied by Dehn are unsatisfactory in that they assume the seasonal pattern to be identical in every single year (Abeysinghe, 1991). Since the seasonal adjustment method is applied to the country-specific price indices, where seasonal patterns might well be flexible, this assumption is too strong. Seasonally adjusted price series are only used when the X-12 output in Eviews indicates that identifiable seasonality is present, in other cases it is preferred to use non-deseasonalized price series in order to minimize distortion of the original data.

4.1.2 Stationarity

When running regressions on time series data, one should be aware of the possibility of non- stationarity. When time series are non-stationary, they contain a drift or a trend; in other words, their variance and mean change over time. The regression of two nonstationary variables on each other can lead to spurious (i.e. potentially misleading) results (Granger and Newbold, 1974). Many statistical methods assume stationarity; hence it is necessary to transform the data before starting analysis.

To test for stationarity, time series are usually checked for the presence unit roots. When no unit roots are found, the series are regarded as trend stationary (TS), meaning that temporary price shocks have only short-term effects, i.e. do not lead to permanent shifts. In this case, the series tend to revert to a long-term trend. In case of unit roots, the series are difference stationary (DS), in which case they follow a random walk, possibly with structural breaks, and do not return to a trend.

To test whether the country price indices are characterised by TS or DS processes, a number of econometric tests are available.

However, as Dehn (2000b) points out, such tests are not powerful when the data contains structural breaks and fewer than 250 observations. Since both facts are the case in the current data set, this paper follows Dehn by assuming the time series are DS, and thus need to be detrended. The most compelling argument for this is that the enduring debate on the Prebisch-Singer hypothesis has shown that commodity prices are likely to possess either no trend or a weak one. Hence, they are somewhere between TS and DS, in which case detrending the series has been proven to produce smaller forecast errors (Enders, 1995).

A frequently used detrending tool, the Hodrick-Prescott (HP) filter, is applied to detrend the price indices.

This method is employed by Blattman et al. (2004), among others. Besides taking care of non-stationarity, the HP filter conveniently separates the country price index series into trend and variance around the trend (volatility), and allows the researcher to investigate both separately. The resulting trend and volatility measures are entered into the regression model as independent variables. I will describe how this is done in the next chapter.

Graphs depicting the trend and volatility per country can be found in appendix 6. In the graphs, the blue line represents the commodity price index for the country in question (1990 = 100). The red line is the HP-filtered long-run price trend, whereas the green line represents the volatility around this trend line. As can be seen from these graphs, price movements differ widely between countries. Although a

12For instance the Augmented Dickey-Fuller test or the Phillips-Perron test.

13For the HP filter, the smoothing parameter is set to 1600. This implies a relatively slow-changing trend. The value of 1600 is suggested by Hodrick and Prescott for quarterly data and is common in the literature.

few country indices exhibit a secular downward trend (i.e. a decreasing trend over the entire period;

see for instance Cambodia or Zimbabwe), at face value most commodity price indices do not show evidence for the predictions of the Prebisch-Singer hypothesis.

4.2 Other independent variables

Besides price trend and volatility, the following control variables are included in the model, based on the analysis in section 3.3: lagged per capita GDP growth, population growth, human capital, and openness. To test for the impact of openness on the link between price volatility and growth, an interaction variable is included that quantifies this effect. This is similar to the approach taken by Blattman et al. (2004); however, these authors chose to interact openness with price trend instead of volatility.

As is common in the literature on Solow growth models, technological progress and the depreciation rate are assumed to be constant across countries. Following Mankiw, Romer and Weil (1992) and Hoeffler (2002), their joint value is assumed to equal 5 percent. Hence, the population growth rate variable is calculated as (p + g +

used as a proxy. Sachs and Warner (1997) use life expectancy as a measure of human capital because poor health impacts growth through reduced labour productivity. Similarly, shorter life expectancy affects the level of investments in schooling (Galor, 2005).

Openness is measured using the share of total trade in GDP, as is common in the literature (see, for instance, Rigobon and Rodrik, 2004). Lagged GDP per capita is used as a proxy of initial wealth.

4.3 Dummy variables

The effects of being an island or being landlocked on the link between commodity price movements and growth are tested for with dummy variables. For the selection of small island states, the criteria used by the United Nations are followed. These include not only island nations but also small coastal countries with similar characteristics (e.g. higher vulnerability to economic and ecological shocks).

Landlocked countries are all countries that have no direct access to the sea. There are 12 small island states and 20 landlocked countries in the sample.

4.4 Data

To construct country-specific price indices, two main data sets are used. Commodity price data was obtained from the International Monetary Fund’s IFS database, which can be accessed online. Data from this source was complemented by price data from UNCTAD’s Handbook of Statistics 2005.

Prices in these databases are quoted in nominal US dollars. Real prices are obtained by deflating the nominal series using the World Bank’s MUV index. This index of the unit value of manufactured

14The regressions also include time dummies, which are not reported in order to conserve space.