THE VALUE OF WATER
A quantitative analysis of the effect of water scarcity on commodity pricing
Master Thesis Alexander van Aken
August 1st 2014
Supervisor: prof. dr. E.J.S. Plug Second reader:
Department of Economics
Faculty of Economics and Business
2
Abstract
This thesis analyses whether water scarcity increases import commodity prices or quantities for various agricultural product categories in the Netherlands. This is done for the partner countries China, India, Spain, Turkey, South Africa and Mexico. For this research a panel dataset is used based on monthly data for the period 1996-2005. An ordinary least squares (OLS) regression model is used to test theory which predicts that water scarcity negatively affects quantity supplied, which in a market of supply and demand will result in a higher equilibrium price. A second OLS regression tests the effect of drought in one country on average prices or quantities of the five other countries. A fixed effects model has been adopted in all OLS regressions to control for country and time fixed effects. The estimation results show mixed results both in terms of sign and significance, but more significant results are found for the effect on prices than for the effect on quantity. Several possible explanations are presented in this thesis, yet not a single theory can explain all the results found.
3
Table of contents
Abstract 2
List of tables 4
1. Introduction 5
2. Theoretical framework and literature study 7
2.1. Water and water scarcity
2.2. Water scarcity, trade and pricing
3. Data description and summary statistics 12
3.1. Blue water scarcity 3.2. Precipitation
3.3. Commodity prices, quantities and selected products
4. Empirical analysis 17
4.1. Empirical strategy
4.2. Ordinary least squares regression: model setup and threats to validity
5. Results 24
5.1. Correlation between blue water scarcity and precipitation 5.2. The effect of drought on prices using OLS regression
5.3. The effect of drought on coffee prices and quantities using OLS and IV 5.4. The effect of drought in country i on prices and quantities of other countries 5.5. Limitations
6. Conclusion 35
References 36
4
List of tables
1. EUROSTAT product categories used in this research 15
2. Summary statistics 16
3. Test for multicollinearity of the independent variable drought 23 4. Correlation between blue water scarcity and average precipitation 25 5. The effect of drought on log prices of different product categories 26 6. The effect of drought on log quantities of different product categories 27 7. The effect of drought in country i on average prices or quantities of other countries 33 8. Average monthly blue water scarcity in percentages, 1996-2005 40
9. Average monthly precipitation in mm, 1996-2005 41
5
1. Introduction
Every year the World Economic Forum publishes its Global Risk Report. This report lists the largest systemic risks the world faces that require international collaboration to restrain. In their latest report, the Global Risk Report 2014, water crises are on the third place of the ten global risks of highest concern in 2014 (WEF 2014). Freshwater accounts for only 2.5% of the total global water stock (Hoekstra and Mekonnen, 2011a). Water security, both from a perspective of drought and of having too much water, requires more attention and cooperation from international leaders as well as more awareness with the population. The drought in Russia in 2010 showed the economic impact of water scarcity not just on the price of crops within the country itself, but for the international market in particular. The result was the largest jump in prices of wheat since 1973 as well as a temporary ban on the export of grain (Bloomberg, 2010). This in turn resulted in lower food security and higher prices of food in the Middle East, which has been identified as one of the causes of tensions that lead to the Arab Spring (WEF, 2014).
The growing problem of water scarcity demonstrates itself in the increased usage of groundwater resources which lead to exhaustion. Water scarcity is an indicator used mostly to measure water quantity, but water quality plays an important role in water scarcity as well: water which has too high a concentration of pollution cannot be used (UNDP, 2006). A better understanding of the usage of fresh water in production can help manage the water resources around the globe. Fresh water is a not an infinite resource and therefore it should not be treated as one (Hoekstra and Mekonnen, 2011b). There are three main fresh water users: the agricultural sector, industrial sector and the household water supply. Although large differences exist between countries, globally 92% of the estimated fresh water from ground resources is consumed by the agricultural sector (Mekonnen and Hoekstra, 2011).
Not all of the goods produced are for national consumption. International trade leads to the availability of products in for example the Netherlands, despite the fact that production in this country is not be possible or would be too expensive due to climate circumstances. The Netherlands imports both agricultural and industrial goods from countries around the globe. The Netherlands is one of the largest exporter of agricultural goods, though there are many kinds of agricultural goods that the Netherlands imports from other countries. With the import of these products the Netherlands is basically importing water in virtual form. That is; during production water is used to grow crops for example, but that water is not part of the final product that is imported by the Netherlands. Therefore by trading goods, countries are trading virtual water. Despite the serious water scarcity that some countries experience part of the year, different
water-6
intensive goods are produced in these countries for external use. In periods of water scarcity it is expected that for agricultural, water-intensive goods the price increases due to a lower supplied quantity. The expected relationship between water scarcity and its effect on prices and quantities is more elaborated in chapter 2. This research will test for the relation between water scarcity and prices using an ordinary least squares regression (OLS). It aims to contribute to existing literature on the interconnectivity of the water footprint. Using data on import prices, blue water scarcity and precipitation the following research question will be addressed: do periods of water scarcity
have a positive impact on commodity prices?
Next to this, the research aims to find if the effect of drought on prices runs through quantities supplied. Water scarcity or drought is expected to only affect supply, not demand for water-intensive goods. A shock in drought, for example a period with extreme drought compared to other periods, could reduce agricultural production at a given price. The supply curve would shift and this again will lead to an increase in equilibrium price (Stock and Watson, 2012). Furthermore, in thesis it will be investigated how the impact of water scarcity in one country will result in a change of quantity and prices in other countries. It is expected that water scarcity in country i affects prices in country i, which could result in a shift in demand away from country i and towards other trading partners. This is further elaborated in chapter 2.
This research uses the Netherlands as a case study. For the Netherlands 89% of its water footprint is external (Van Oel et al., 2009) and it is therefore highly dependent on the use of water resources around the world. Six partner countries have been selected for this research: China, India, Spain, Turkey, South Africa and Mexico. These countries have been selected based on the water scarcity indicators of Van Oel et al. (2009), which show that for the period 1996-2005 the Dutch external water footprint has the most significant impact on these countries.
The remainder of this thesis is structured as follows. Chapter 2 contains a theoretical framework and literature study that introduces the relevant concepts. The data is introduced in chapter 3. Chapter 4 contains the empirical analysis and discusses the validity of the OLS regression model. The estimation results and the implications and limitations of the results found are presented in chapter 5. Chapter 6 concludes and is followed by the appendix.
7
2. Theoretical Framework and literature study
2.1 Water and water scarcity
Fresh water is an input production factor in all different kinds of products. In this research, the distinction is made between agricultural products and industrial products. This distinction is made as agricultural products are known to require a large amount of fresh water and moreover this fresh water is the most important input (Postel, 2000). Not only is fresh water an finite good, the growing world population will require additional food production and therefore fresh water usage in the future. Next to this, the fast rate of economic development, demographic pressures and larger levels of pollution are all threats to finite availability of fresh water. Demand is rising, supply is diminishing. The pressure on fresh water resources is intensified by the monthly fluctuations in climate and weather circumstances. Countries around the globe experience variability in temperature, precipitation and other climate indicators. These seasonal fluctuations also exist in the countries studied in this research: China, India, Spain, Turkey, South Africa and Mexico. These countries have in common that they have certain months a year of serious low amounts of precipitation. The distribution however differs as is shown in chapter 3 of this thesis.
This research will test if there is a strong relationship between blue water scarcity and precipitation. More specifically, if there is a strong relationship between periods of low precipitation and high level of scarcity at groundwater resources. The fresh surface and groundwater resources are also known as blue water resources. The trade of commodities that require fresh water during production leads to the trade in virtual water. The concept of virtual water was firstly introduced in 1998 by John Allan and was defined as: ‘the water embedded in key water-intensive commodities such as wheat’ (Allan, 1997). A more recent definition comes from Chapagain: ‘The fresh water used in the production process of an agricultural or industrial product’ (Chapagain, 2003). Virtual water is therefore not part of the final product.
Not all agricultural or industrial products consumed by a country are produced internally. Many commodities are produced elsewhere in the world and imported for consumption or perhaps re-exported. With the international trade of commodities another form of trade arises: the trade of virtual water. The total amount of fresh water used for the consumption of agricultural- and industrial commodities by a country’s population leads to a nation’s total water footprint (Chapagain, 2003). With the consumption of imported commodities in for example the Netherlands, freshwater resources elsewhere in the world are affected. The total volume of fresh water used in other countries for the production of goods that are consumed in the Netherlands is
8
called the external water footprint of the Netherlands (van Oel et al., 2009). The negative externalities as a result of Dutch consumption are mostly impacting six countries: China, India, Spain, Turkey, South Africa and Mexico (Chapagain, 2008). These countries are not the largest contributors to the external water footprint of the Netherlands, but as these countries experience different levels of water scarcity each year, the negative externality of Dutch consumption is the largest (Hoekstra, 2011a). In their research, Hoekstra et al. developed an assessment of the global water scarcity. This paper on global water scarcity is the first that calculates the water scarcity based on the usage of ground- and surface water instead of the withdrawal of fresh water from these sources. Namely, large amounts of water withdrawn will return to local rivers and can therefore be used again (Perry, 2007). The water scarcity indicators used by Hoekstra are therefore with more accuracy than has done before (Hoekstra, 2012). As the implications of a shortage of fresh water now and in the future poses serious challenges for policy makers and businesses, an accurate assessment of the global water scarcity is of high importance.
Next to this, specifically for the Netherlands the water footprint has been calculated. Per capita the total water footprint of Dutch consumers is 2300 m3 per year for the period 1996-2005 (Hoekstra, 2008). Of this total footprint per capita, 67% arises from agricultural goods. The Netherlands does not produce all its agricultural goods that it consumes internally. More specifically, the Netherlands only fulfils the water needs of 5% of its own agricultural goods consumption. For the other 95% of the fresh water needs for the production of agricultural commodities consumed in the Netherlands, they are depended on sources outside the Netherlands (Hoekstra, 2007).
The extraction of fresh water from groundwater resources for the usage in production is an important example of man-induced processes influencing water scarcity (Pereira et al., 2009). Next to this there are natural processes affecting water scarcity of which a low average precipitation or high variability precipitation is the most important one. This theory predicts the strong relationship between precipitation and blue water scarcity. This research will test its correlation.
2.2 Water scarcity, trade and pricing
The export of water-intensive commodities by water scarce countries is counter intuitive and not in line with the Ricardian or Heckscher-Ohlin (H-O) model of trade theory. Classical Ricardian trade theory states that countries should export products in which they have a comparative advantage in production (Nielsen et al., 1994). Countries that have better technical solutions than others enables them to get ‘more crop per drop’ (FAO, 212). They have a comparative advantage
9
and should therefore produce and trade water-intensive commodities. The Heckscher-Ohlin (H-O) model takes a different perspective in trade theory. The model takes states that countries have differences in availability of factors of production, which is also the case with the availability of water resources. A country should, according to the H-O model, export products that intensively uses the factor of production in which the country is abundant and import products that intensively uses the factor of production in which the country is scarce (Nielsen et al., 1994). Based on the H-O principles of international economics, one would expect that countries which experience periods of drought adjust their production portfolio and only produce commodities that require low or none fresh water in production. Another option would be to shift the production towards higher value agricultural goods (Arcadis, 2012).
The Ricardian model states that trade arises due to differences between countries in terms of labour productivity, whereas the H-O model explains international trade due to differences in factor endowments (Nielsen et al., 1994, Krugman and Obstfeld, 2009). New International trade theory argues that trade need not be the result of comparative advantage. Both models of comparative advantage were based on the assumption of constant returns to scale. Instead new trade theorists base international trade on increasing returns or economies of scale. This gives countries an incentive to specialize and trade with countries in their resources or technology. Even in the absence of differences between countries (Krugmann and Obstfeld, 2009).
There is not one single trade theory that can explain the international trade patterns we see for agricultural commodities. For a full understanding of why trade occurs as we can see in the data, all players, government, producers and consumers, have to be taken into account. This however will not be the focus of this research.
Since some water scarce countries produce and trade commodities that are water-intensive during production, there has to be an economic incentive to do so. India for example is one of the largest net virtual water exporters but also characterized by periods of serious water scarcity. Water scarcity is a natural hazard that proves to be a great threat for agricultural production (Liverman, 1990). Fresh water is an important factor of production for agricultural goods. When this factor of production becomes scarce, it negatively affects the quantity produced within the agricultural sector (FAO, 2012). This type of shock affect all individual producers in a similar way and therefore market supply is affected.
Economic theory predicts that in an environment of supply and demand, the effect of a negative shock on quantity supplied will lead to an increase in the equilibrium price (O’Sullivan et al., 2007). The expected effect of water scarcity on supply is displayed in figure 1. This figure shows the expected effect of water scarcity in country i on prices and quantity of country i. The quantity demanded is expected to be affected by fluctuations in price, but not directly as a result
10
of drought. More specifically, it is expected that drought in country i does not directly affect the demand for agricultural products of a country that imports these commodities from country i, in this case being the Netherlands.
Figure 1
Water scarcity could have a second effect on the international market. The Netherlands imports the same type of products from different trade partners. The assumption is made that these products from different trade partners are substitute goods. They have a positive cross elasticity of demand (Snyder and Nicholson, 2008). When the price of a product in country i rises, this has a positive effect on the demand for the same product in country j. If water scarcity in country i increases prices in country i than it is possible that the Netherlands will shift its demand away from country i and towards other trading partners due to the effect of water scarcity on prices and quantity supplied in country i. This shift in demand towards other trading partners, for example country j is displayed in figure 2. The effect of this higher demand in country j shifts prices upwards. Therefore, it is expected that water scarcity in one country increases prices of that same country, as well as others.
11
Figure 2
The above theory shows that a exogenous shock which negatively affects quantity supplied results in a higher equilibrium price level. The data used in this research is data on prices and quantities traded. It is expected that the effect on producer prices and quantity supplied are also part of the quantity traded, as well as the trade price. It is expected that the effects are passed on to the customers.
12
3. Data description and summary statistics
This chapter describes the data and methodology used in this thesis. It will elaborate on the description of the data on blue water scarcity, precipitation and import prices and quantities. The result and data of the research of Hoekstra is the starting point of this research. The data of Hoekstra contains estimations for the monthly blue water scarcity of the world’s major river basins, a total of 405, for the period 1996-2005. Van Oel et al. (2009) show that the impact of the negative externalities which are a result of Dutch consumption is the largest in countries China, Spain, India, Turkey, South Africa, Mexico, Pakistan and Sudan. Unfortunately, due to incomplete data on precipitation in Pakistan and Sudan for the period 1996-2005, these two countries will not be part of this research.
3.1 Blue water scarcity
For this research data on blue water scarcity is used as calculated by Hoekstra and Mekonnen (2011a). In this report monthly data on blue water availability is presented as well as the natural runoff and the blue water footprint for 405 of the largest river basins in the world. With this, the monthly blue water scarcity is calculated for the period of 1996-2005. The data shows that for the major river basins of the six countries involved, severe drought is experienced certain months a year. For the largest river basins around the world the blue water scarcity is calculated based on the blue water footprint and blue water availability. These results per river basin are used in order to calculate the blue water scarcity per country. Hoekstra et al. (2011a) calculated the blue water scarcity as follows:
Blue water scarcity= Blue water footprint Blue water availability
Of which:
Blue water availability= Natural runoff
Environmental flows requirements
The blue water footprint is a measurement of the consumption of blue water resources, with ‘blue’ meaning surface and groundwater (Mekonnen and Hoekstra, 2011). This calculation method gives a more precise result than previous water scarcity indicators, like Falkenmark’s indicator for water scarcity which is the most common indicator used in previous studies (Hoekstra and
13
Mekonnen, 2011a). The Falkenmark indicator uses water withdrawal as measurement for water demand. Perry (2007) however shows that large amounts of the water withdrawal, for example in agriculture, will return to local river basins and can therefore be withdrawn again. Therefore Hoekstra and Mekonnen (2011a) use blue water consumption, also called blue water footprint, instead of blue water withdrawal to get a more accurate indicator for blue water scarcity.
The blue water scarcity is reported in a percentage that indicates if the blue water footprint is lower than or exceeds a certain percentage of the blue water availability. More information on the methodology and explanation of the different categories of blue water scarcity can be found in appendix 2.
As this dataset only contains an average per month for this 10-year period, it is not sufficient for testing the hypothesis that drought significantly affects prices. The data for blue water scarcity is used to test its correlation with precipitation in order to use the data on precipitation to analyse the effect on import prices and quantities by the Netherlands. This is discussed in chapter 3.2.
3.2 Precipitation
Data on precipitation is collected from the National Climatic Data Center (NCDC) of the United States National Oceanic and Atmospheric Administration (NOAA). This data includes total monthly precipitation for the different weather stations located worldwide. China has 187, Spain 82, India 77, Turkey 6, South Africa 447 and Mexico 128 weather stations that report precipitation data for the period 1996-2005. All six countries have more weather stations, but these are not included as they partly or fully do not report data on precipitation. The total precipitation is reported per weather station. This station level data is used to compute a country average per month, per year for the period 1996-2005. For example, for Spain the month January 1996 has 82 precipitation observations, which are summed to one total amount of precipitation. This gives a total of 12 (months) * 10 (years) observations of precipitation for Spain. These observations are divided by the amount of weather stations in Spain, 82. The result is the country average total precipitation for a specific month. The same is done for the other countries. The data on total precipitation is reported in tenths of mm. Therefore, all data on precipitation is divided by ten in order to get the amount of precipitation in mm.
The data on precipitation is used in all parts of this research. First it is used to correlate precipitation with blue water scarcity. Secondly it is used to estimate the relationship between drought and trade prices and quantities. For this latter estimation, a binary variable drought was created which indicates whether a certain month was characterized by low precipitation. The third
14
and last part of the empirical analysis the effect of drought in country i on average prices and quantities in other countries is estimated. More information on the chosen threshold of the binary variable drought can be found in Appendix 3.
3.3 Commodity prices, quantities and selected products
In this research a correlation test will indicate whether precipitation is a strong indicator for blue water scarcity and if so, data on precipitation will be used to estimate the effect of a period of drought has a significant positive effect on the prices and quantities of different commodities. In this regression the dependent variable is either logarithmic prices or quantities.
The commodity prices and quantities are derived from the EU28 trade since 1988 by HS2-HS4 database of Eurostat. This database reports monthly import quantities in hundred kilogram and value in euros for different product categories. All the different physical products that are traded with the Netherlands and other European countries are categorized within a product classification according to the Harmonized Commodity Description and Coding System or the Harmonized System (HS). More than two hundred countries use this system inter alia for the collection of international trade statistics. The data on value in euros is divided by the quantity traded in order to get the value or price per unit of traded good. Therefore, the price of one unit of a commodity refers to the price of hundred kilogram of that commodity.
For this research, aggregate data on prices of different commodity categories is used. For example, coffee, tea, maté and spices are combined into one category. The choice to use data for product categories is made as data on values and quantities traded for the subcategories are only available sporadically. The availability of the data depends on what products are imported by the Netherlands from the different partner countries. For this research, a selection of commodity categories has been made based on the dependency of fresh water during production. Commodities for which fresh water is one of the most important factors of production are expected to show price fluctuations when the availability of this factor of production fluctuates as well. This is for example the case with crops and other agricultural products like fruits and vegetables. These products are water-intensive in production. When the availability of this factor of production is limited, the quantity produced could be affected. Next to this, there are commodities which are less dependent on fresh water during production as is the case with industrial commodities like footwear, metals and plastics. These industrials commodities may require some fresh water during production, however for these commodities there are other inputs which are far more expensive than fresh water. These more expensive inputs as well as costs of labor and capital are the key factors behind the price fluctuations in these industrial commodities
15
(Hoekstra and Mekonnen, 2011a). The following commodity categories are selected for this research:
Table 1. EUROSTAT product categories used in this research 1. Live trees & other plants
2. Edible fruits & nuts 3. Edible vegetables
4. Coffee, tea, maté & spices 5. Cereals
6. Oil seeds & oleaginous fruits 7. Sugars
8. Cotton, including yars & woven fabrics
9. Optical, photographic & cinematographic instruments
For commodity categories 1-8 fresh water is of major importance in the production process (Mekonnen and Hoekstra, 2011). It is expected that one or more periods of drought, low precipitation and high blue water scarcity, will lead to an increase in the price of that commodity category in the same month or in months after that. This thesis aims to find empirical support for the effect of drought on commodity prices.
Commodity category 9 contains only industrial products which is included in this research as a control group. The expectation is that drought will have a positive effect on price levels for water-intensive commodities. The industrial commodity category is not water-intensive and therefore water scarcity should not have a significant effect on its price level.
The results of the correlation test for blue water scarcity and precipitation will determine whether there is a lag in the effect of precipitation on ground- and surface water resources. Depending on the results, the estimation of the effect of drought on prices and quantities traded will also be performed with one or more lags. More details about the methodology and the empirical setup can be found in chapter 4. Table 2 shows the summary statistics of all variables used in this research.
16
Table 2. Summary statistics
Variable Obs. Mean Std. Dev. Min. Max.
Water variables
Blue water scarcity 72 184.333 167.245 3 669
Total average monthly precipitation* 72 56.806 44.254 9 206
Average precipitation* 720 56.799 49.692 0 301 Drought 720 0.474 0.500 0 1 Drought China 120 0.383 0.488 0 1 Drought India 120 0.467 0.501 0 1 Drought Spain 120 0.425 0.496 0 1 Drought Turkey 120 0.425 0.496 0 1
Drought South Africa 120 0.550 0.500 0 1
Drought Mexico 120 0.592 0.494 0 1
Product category prices (log)
Prices of live trees and other plants 720 5.834 0.511 0.693 6.934
Prices of edible fruits & nuts 720 4.910 0.647 3.850 6.750
Prices of edible vegetables 711 4.543 0.541 2.708 7.971
Prices of coffee, tea mate & spices 720 5.343 0.414 4.143 7.349
Prices of cereals 509 4.004 0.699 2.079 8.364
Prices of oil seeds & miscellaneous grains 719 4.880 0.725 2.833 9.673
Prices of sugars 595 5.354 0.142 0 12.324
Prices of cotton 555 6.306 0.735 4.407 9.908
Prices of optical & photographic instruments 720 8.565 0.121 2.079 12.107 Coffee prices (log)
Coffee prices China 120 5.068 0.343 4.357 6.087
Coffee prices India 120 5.290 0.340 4.663 6.420
Coffee prices Spain 120 5.530 0.179 4.984 6.250
Coffee prices Turkey 120 5.184 0.414 4.143 6.232
Coffee prices South Africa 120 5.278 0.457 4.369 7.349
Coffee prices Mexico 120 5.709 0.334 4.905 6.819
Coffee quantity (log)
Coffee quantity China 120 8.261 0.640 7.024 9.670
Coffee quantity India 120 8.833 0.466 7.718 9.869
Coffee quantity Spain 120 7.186 0.481 5.394 8.336
Coffee quantity Turkey 120 7.018 0.802 4.043 8.831
Coffee quantity South Africa 120 6.737 0.976 3.466 9.012
Coffee quantity Mexico 120 8.050 0.693 5.966 9.446
Notes: * For explanation on the calculation of both averages see appendix 2. Quantity is measured in units
of hundred kilogram. Prices are therefore prices per hundred kilogram of a certain product.
Source: All prices and quantities are import prices or quantities reported by EUROSTAT (2014). Data on
17
4. Empirical analysis
This chapter contains the strategy and model used in the three different parts of the empirical analysis. Chapter 4.1 describes the empirical strategy that is used. The setup of the OLS regression model and possible threats to validity of the OLS model are discussed in chapter 4.2.
4.1 Empirical strategy
The empirical analysis consists of three parts. First a correlation test will investigate whether or not precipitation is a strong indicator for blue water scarcity. For this monthly averages are used for the period 1996-2005 as is elaborated in more detail in chapter 3. Secondly the data on precipitation will be used to test if fresh water availability, or drought, has a significant effect on commodity prices and quantities of that same country. The third part of the empirical analysis is to test, using OLS, if drought in country i has a significant effect on the commodity prices or quantities in country j.
With the use of the data from Hoekstra and Mekonnen (2011a) and NOAA the country average blue water scarcity and the country average precipitation per month is calculated for the period 1996-2005 as can be found in table 8 and 9 of the appendix. First the relationship between precipitation in month t and the blue water scarcity in month t will be tested. Next to this, the relationship between precipitation in month t and the blue water scarcity in month t+1 – t+4 will also be tested. There are different reasons why drought could have a lagged effect on prices. For example, the data on precipitation is reported on the first day of each month, which is the summation of the daily precipitation measured at the weather stations. It is unknown at what date each month the blue water availability, footprint and therefore scarcity is measured. Therefore, it is possible that large amounts of precipitation will ‘fall’ after the moment that a month reports a certain level of blue water scarcity.
As mentioned earlier, the data for blue water scarcity contains only monthly averages for the period 1996-2005. For each country the data on the local river basins is averaged which gives a total of 12 observations per country. With this data it will be tested if precipitation is a strong indicator for blue water scarcity. Note that for this correlation test average precipitation is used instead of the binary regressor droughti,t. This latter will be used in the second and third part of
the empirical analysis to test the relationship between commodity prices, quantities and precipitation which is explained in more detail below. The correlation test between blue water scarcity and average rainfall will show not only whether low precipitation co-moves with high
18
blue water scarcity (as is expected), but also whether high precipitation co-moves with low blue water scarcity. The results of this test will indicate whether or not precipitation is an strong indicator for blue water scarcity.
If indeed the relationship is strong, it allows for the usage of precipitation as an indicator for blue water scarcity in the further stages of this research, to analyse the relationship between blue water scarcity (low precipitation) and prices and quantities of certain good categories.
Commodity prices, quantities and precipitation
Prices of imported goods by the Netherlands are calculated by dividing the total trade value in euros reported by Eurostat (2014) with the trade quantity in hundred kilogram. The prices used in this research are therefore the trade value per hundred kilogram of a commodity. Natural logarithms of these trade prices and quantities are used in the regression. The logarithms convert changes in prices and quantities into percentage changes which facilitates the interpretation of the results. A panel dataset is constructed as for different entities, or countries, the prices and quantities are observed for 120 time periods (months). The advantages of -and conditions for- using panel data are discussed in the next chapter.
The test for the relationship of drought and prices involves multiple least squares regressions. In each regression, logarithmic prices of a certain product category in period t will be the dependent variable, and the binary variable which indicates one for drought (<40 mm) in period t will be the independent variable. Next to the regression between prices and the same periods’ precipitation, another regression will be testing the relation between prices in period t and lagged values of the binary variable droughti,t. This enables to test if drought, or water
scarcity, in period t-1 or t-2 results in an increase in import prices in period t.
Droughti,t = 1 if precipitationi,t < 40 mm. Droughti,t = 0 if precipitationi,t ≥ 40 mm.
4.2 Ordinary least squares regression: model setup and threats to validity
The first part of the empirical analysis contains a correlation test. No further explanation for this test is covered in this chapter. The second and third part of the empirical analysis use OLS regression.
The OLS analysis uses panel data in the regression. For each of the six countries 120 different time periods are observed. The advantage of panel data is that it allows to include country fixed effects and control for variables that are not observed and differ across countries, but are constant over time. At the same time it enables to include time fixed effects which controls for
19
variables that are not observed and are constant across countries, but vary over time (Stock & Watson, 2012).
Ordinary least squares model
The model used for the second part of the empirical analysis is an OLS regression model. It contains the natural logarithm of prices or quantities as the dependent variable, lnPe
i,t or lnQei,t,
and the binary variable droughti,t as independent variable. Super- and subscript e, i and t denote
the commodity category, country and time period (month) respectively. For reasons explained in more detail below a fixed effects regression is employed and country fixed effects ci and time
fixed effects tt are included in the model. µe,i,t indicates the error term.
𝑙𝑛𝑃
𝑖,𝑡𝑒= α
i
+ β
1DROUGHT
i,t-1+ c
i+t
t+ μ
e,i,t𝑙𝑛𝑄
𝑖,𝑡𝑒= α
i
+ β
1DROUGHT
i,t-1+ c
i+t
t+ μ
e,i,tThe third and final part of the empirical analysis will focus on the effect of drought in country i on average prices or quantities of the five other countries. It is expected that drought in country i influences the quantity supplied in country i and this again results in higher prices in country i. Furthermore, the higher prices in country i will cause a shift in demand away from country i towards other countries that supply the same commodity. This increase in demand in other countries will bring along a rise in prices in the other countries. Summarized, the expected relation is as follows:
Droughti,t → lnQs,i,te → lnPi,te → lnQd,i,te → lnQd,j,te → lnPj,te
The OLS model used to estimate the effect of drought in country i on average prices or quantities of the other countries is as follows:
𝑙𝑛𝑃
𝑘,𝑡𝑒= α
i+ β
1DROUGHT
i,t-1+ c
i+t
t+ μ
e,i,t𝑙𝑛𝑄
𝑘,𝑡𝑒= α
i
+ β
1DROUGHT
i,t-1+ c
i+t
t+ μ
e,i,tSubscript k does not denote one particular country, but denotes the average of the five other countries than country i. The effect of drought in country i on average prices or quantities of the other countries is expected to be especially strong for countries that are have a large market share for a particular product. Therefore for this third part of the analysis a selection is made of large
20
players in the market for each of the nine product categories. For each of these large players it will be estimated how its drought affects other countries prices and quantities.
Fixed effects regression model
In this regression panel data is used which deals with two-dimensional observations: time and entity (country). It is possible that there are other variables that affect commodity import prices but which are not included in this regression. To control for such variables a fixed effects regression model is used. The variation in the country fixed effects arises from omitted variables which vary across countries, but not over time (Stock Watson, 2012). Next to country specific fixed effects, it is also possible that there are time specific fixed effects which affect all countries in a similar way, but change over time. Implementing time fixed effects in the regression enables one to control for these unobserved variables (Stock and Watson, 2012). For example, one could think of local wages to be such a variable. Unfortunately, monthly data on wages in the countries observed are not available and therefore this variable cannot be included in the regression. The combined use of country fixed effects and time fixed effects eliminates any omitted variables bias arising from both kind of unobserved variables.
Threats to model (OLS)
Before using the model it is necessary to verify that the model complies with its conditions in order to get a valid, unbiased and consistent coefficient estimator. Therefore the possible threats to the internal validity of the regression must be checked. Not all of these threats apply to this research. Therefore only the ones that are important for this research will be discussed. In general, the different threats are (Stock & Watson, 2012):
1. Omitted variable bias
2. Misspecification of the functional form
3. Measurement error and errors-in-variables bias 4. Sample selection
5. Simultaneous causality
6. Inconsistency of the Ordinary Least Squares standard errors 7. Multicollinearity
Omitted variable bias arises when a regression leaves out a variable which determines the dependent variable and is correlated with one of the independent variables. In this research the only regressor is the binary variable droughti,t. It is expected that drought has a significant effect
21
this assumption it is expected that the independent variable droughti,t is uncorrelated with any
omitted variables that are part of the error term. Therefore, omitted variable bias is not a threat to the OLS regression model used in this research. The assumption that precipitation is an exogenous variable rules out the problem of simultaneous causality as prices or quantities traded will not have a causal effect on drought.
Errors in variables bias can be the result when an independent variable is incorrectly measured (Stock and Watson, 2012). Measurement error can result in correlation between the regressor and the error term which causes an inconsistent OLS estimator that is biased towards zero. It is possible that the independent variable drought is measured imprecisely. It is a country average based on different local observations from weather stations. Large variations of drought within a country could exist, especially for large countries like China or India. However, observations of import prices and quantities on a local scale are not available. Therefore, the data reported by NOAA, used to calculate the binary variable droughti,t , is the best dataset available.
It is measured in an uniform manner for all the countries observed. When there is data missing which is related to the dependent variable, then this selection can introduce correlation between the error term and the regressor (Stock and Watson, 2012). In this research the data on the independent variable droughti,t is complete, however there is data missing on some of the prices
of product categories, as can be seen in table 2. An example would be cereals: the Netherlands does not import cereals from Mexico and only sporadically from South Africa. This missing data does not introduce bias of the estimator, but it does result in a less precise estimated coefficient which could be reflected by large standard errors (Stock and Watson, 2012). The results in chapter 5 will show whether or not this is the case.
The problem of simultaneous causality arises when there is not only a causal effect of the regressor on the dependent variable, but also the other way around. If the causality runs also from the dependent variable to the independent variable the OLS estimator is biased and inconsistent (Stock and Watson, 2012). In this research the independent variable is precipitation which is assumed to be beyond human control. Therefore the independent variable is exogenous and simultaneous causality is not present in this research.
Another possible threat is that of inconsistency of the OLS standard errors. Inconsistent standard errors are a threat to internal validity. It is caused when the error terms are correlated over time, also known as autocorrelation, or when homoscedasticity-only standard errors are used when actually the regression error is heteroskedastic. Both can lead to invalid standard errors. The solution for this is to use heteroskedasticity- and autocorrelation-consistent (HAC) standard errors. The HAC standard errors, or clustered standard errors implemented in the regression model
22
used in this research. It allows the standard errors to correlate within a country, but not across countries or clusters (Stock & Watson, 2012).
The third and last part of the empirical analysis will be to test whether drought in country
i affects average prices or quantity of the other countries observed. As explained in chapter 2, it
is expected that drought in country i influences the prices in country i which would lead to a shift in demand towards other countries. This shift in demand would lead to a similar change in price in both countries. However, it is important to test if there is a relationship between drought in country i and country j. If so, the coefficient could be estimated imprecisely (Stock & Watson, 2012). One of the conditions for a valid regressor is its exogeneity. If there is indeed a relationship between drought of different countries, this condition does not hold.
The exogeneity of drought in country i ensures that the prices of country j are only affected by the change in prices of country i and not directly by the drought in county i. For this test an OLS regression has been used instead of a correlation test, which allows controlling for time effects. The results are presented in table 3.
Table 3. Test for multicollinearity of the independent variable drought Independent variable
Dependent
variable China India Spain Turkey South Africa Mexico
China 1 -0.012 (0.016) -0.006 (0.049) 0.107* (0.057) -0.006 (0.051) -0.034 (0.060) India -0.030 (0.037) 1 0.035 (0.060) -0.063 (0.072) 0.029 (0.081) 0.022 (0.127) Spain -0.029 (0.244) 0.071 (0.122) 1 0.070 (0.103) -0.059 (0.116) Turkey 0.500*** (0.190) -0.119 (0.132) 0.064 (0.095) 1 -0.059 (0.096) -0.090 (0.163) South Africa -0.030 (0.258) 0.030 (0.081) 0.035 (0.060) -0.063 (0.072) 1 0.022 (0.127) Mexico -0.088 (0.079) 0.024 (0.135) -0.070 (0.063) -0.050 (0.092) 0.024 (0.104) 1
Notes: Dependent variable is the first lag of drought of country i. Independent variable is the first lag of
drought of country j. Robust standard errors and a dummy for time (months) are included. Asterisks indicate the statistical significance of the estimated coefficients at the *.10 level, **.05 level, ***.01 level.
23
The results show that with two exceptions, there is no relationship between drought in country i and country j. These two exceptions are considering the relationship between drought in Turkey and China. There is however, besides the above results, no reason to believe that drought in Turkey affects drought in China and the other way around. Therefore also for Turkey and China exogeneity of drought is assumed. In the third part of this empirical analysis the regressor drought of country i will be used to test for its effect on price levels and quantities in other countries.
24
5. Results
This chapter will show the results for the three parts of the empirical research. First the results for the correlation test will be presented in chapter 5.1, followed by the results for the OLS regression of drought on prices and quantities in chapter 5.2. The last part of the empirical research that investigates the effect of drought in country i on average prices and quantities of the other countries in this research can be found in chapter 5.3.
5.1 Correlation between blue water scarcity and precipitation
Table 4 shows the result for the correlation test of blue water scarcity and precipitation. The results show that there is a strong negative relationship between blue water scarcity in period t and precipitation in period t. The result of China with no lag is the only exception within the results of this correlation test. Not only does it show a weak correlation but also a positive one in contradiction to the rest of the results. However, when looking at the first lag of China, the correlation already shows a strong negative relationship which is also the case for the second and third lag. The results for the first and second lag are even stronger for the other five countries than the correlation of China. The negative correlation results should be interpreted as a high correlation between low precipitation and high ground and surface water scarcity.
25
Table 4. Correlation between blue water scarcity & average precipitation
Country Lag 0 1 2 3
All six countries -0.282 (72) -0.517 (66) -0.550 (60) -0.497 (54) China 0.127 (12) -0.386 (11) -0.667 (10) -0.872 (9) India -0.532 (12) -0.830 (11) -0.795 (10) -0.612 (9) Spain -0.509 (12) -0.761 (11) -0.660 (10) -0.134 (9) Turkey -0.734 (12) -0.936 (11) -0.764 (10) -0.369 (9) South Africa -0.375 (12) -0.701 (11) -0.874 (10) -0.840 (9) Mexico -0.253 (12) -0.743 (11) -0.810 (10) -0.718 (9)
Notes: For reasons explained in chapter 2 the correlation is tested for
different lags of average precipitation. Number of observations between brackets beneath the estimated correlation.
These results confirm the expectations that precipitation in period t is a strong indicator for blue water scarcity in period t+1 and t+2. Therefore the dataset of precipitation can be used for the OLS regression models. As the result for China, zero lags, is odd compared to the other countries, the OLS regression models will not investigate the relationship between drought in period t (independent variable) and prices or quantities in period t. Summarized, the regression models will test the following relations:
Droughti,t & Pricesei,t-1
Droughti,t & Pricesei,t-2
Droughti,t & Pricesei,t-1 & Priceset-1
That is, the OLS models will test the effect of drought last month, or two months ago, on prices or quantities this month., Next to this it will also test for the effect of two consecutive months of drought on prices of quantities traded this month.
26
5.2 The effect of drought on prices or quantities traded using OLS regression
The results for the OLS regression which tests the relation between a period of drought and commodity prices or quantities traded can be found in tables 5 and 6 respectively. The OLS regression tested for the effect of drought on prices or quantities of nine different product categories. For each product category a total of thirty-two regressions have been executed to test for the different lags of drought. Next to this, for each lag four different regressions show what the estimated coefficient and its significance is when country fixed effects, time fixed effects and heteroskedasticity- and autocorrelation-consistent (clustered) standard errors are included. The dependent variable is reported in natural logarithms. Therefore a reported coefficient of for example 0.053 should be interpreted as an increase in prices or quantities of 5.3%.
27
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10 (11) (12) (13) (14) (15) (16) Dependent variable
Prices of live trees and other plants 0.120*** (0.026) 0.119*** (0.026) 0.119 (0.075) 0.112 (0.078) 0.092*** (0.026) 0.090*** (0.026) 0.090 (0.063) 0.093 (0.065) 0.102*** (0.030) 0.101*** (0.030) 0.101 (0.065) 0.091 (0.069) 0.039 (0.030) 0.038 (0.030) 0.038 (0.042) 0.046 (0.047) Prices of edible fruits & nuts -0.121***
(0.024) -0.121*** (0.024) -0.121 (0.096) -0.134 (0.109) -0.062** (0.025) -0.062** (0.025) -0.062 (0.074) -0.061 (0.079) -0.126*** (0.029) -0.126*** (0.028) -0.126 (0.087) -0.145 (0.102) 0.003 (0.029) 0.003 (0.028) 0.003 (0.036) 0.013 (0.032) Prices of edible vegetables -0.074*
(0.038) -0.077** (0.038) -0.077 (0.114) -0.038 (0.091) -0.067 (0.038) -0.070* (0.038) -0.070 (0.139) -0.025 (0.109) -0.057 (0.045) -0.064 (0.044) -0.064 (0.064) -0.043 (0.055) -0.032 (0.045) -0.036 (0.044) -0.036 (0.108) -0.003 (0.088) Prices of coffee, tea mate & spices -0.053**
(0.027) -0.053** (0.027) -0.053** (0.016) -0.063* (0.028) -0.033 (0.027) -0.033 (0.027) -0.033 (0.023) -0.033 (0.031) -0.050 (0.031) -0.051 (0.031) -0.051*** (0.007) -0.067*** (0.015) -0.007 (0.031) -0.007 (0.031) -0.007 (0.023) 0.002 (0.026) Prices of cereals 0.036 (0.051) 0.033 (0.051) 0.033 (0.076) 0.030 (0.090) -0.030 (0.052) -0.036 (0.052) -0.036 (0.036) -0.029 (0.040) 0.073 (0.060) 0.073 (0.060) 0.073 (0.082) 0.066 (0.093) -0.068 (0.060) -0.074 (0.036) -0.074* (0.031) -0.062** (0.019) Prices of oil seeds & miscellaneous grains 0.137***
(0.048) 0.137*** (0.048) 0.137 (0.114) 0.151 (0.081) 0.034 (0.048) 0.034 (0.048) 0.034 (0.095) 0.051 (0.084) 0.165*** (0.056) 0.165*** (0.056) 0.165 (0.115) 0.172 (0.094) -0.051 (0.056) -0.051 (0.056) -0.051 (0.098) -0.038 (0.093) Prices of sugars -0.127 (0.105) -0.112 (0.104) -0.112 (0.104) -0.082 (0.057) -0.262** (0.105) -0.245** (0.104) -0.245 (0.165) -0.214 (0.132) -0.009 (0.120) 0.007 (0.119) 0.007 (0.078) 0.024 (0.074) -0.264** (0.120) -0.249** (0.119) -0.249 (0.171) -0.225 (0.156) Prices of cotton 0.118** (0.049) 0.119** (0.049) 0.119 (0.082) 0.138 (0.083) 0.050 (0.050) 0.052 (0.050) 0.052 (0.070) 0.070 (0.061) 0.132** (0.060) 0.131** (0.057) 0.131 (0.082) 0.148 (0.086) -0.015 (0.057) -0.013 (0.057) -0.013 (0.059) -0.004 (0.053) Prices of optical & photographic instruments 0.009
(0.072) 0.007 (0.072) 0.007 (0.097) -0.003 (0.088) -0.016 (0.072) -0.019 (0.072) -0.019 (0.080) -0.025 (0.086) 0.040 (0.084) -0.037 (0.084) 0.038 (0.087) 0.031 (0.077) 0.038 (0.084) -0.039 (0.084) -0.039 (0.049) -0.041 (0.064)
Country fixed effects - Yes Yes Yes - Yes Yes Yes - Yes Yes Yes - Yes Yes Yes
Time fixed effects - - - Yes - - - Yes - - - Yes - - - Yes
Clustered standard errors - - Yes Yes - - Yes Yes - - Yes Yes - - Yes Yes
Notes: Dependent variables are converted to natural logarithms. Standard errors are given in parentheses under the coefficients. Regression (9)-(12) and (13)-(16) show the results for the first and second lag respectively.
Asterisks indicate the statistical significance of the estimated coefficients at the *.10 level, **.05 level, ***.01 level.
First and second lag
Independent variable: one or multiple lags of drought (binary variable) Table 5. The effect of drought on log prices of different product categories
28
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10 (11) (12) (13) (14) (15) (16) Dependent variable
Quantities of live trees and other plants 0.131*** (0.050) 0.132*** (0.049) 0.131 (0.065) 0.144* (0.067) 0.144*** (0.049) 0.145*** (0.049) 0.145 (0.089) 0.149 (0.113) 0.072 (0.057) 0.073 (0.057) 0.073 (0.065) 0.084** (0.024) 0.107* (0.057) 0.107* (0.057) 0.107 (0.095) 0.106 (0.107) Quantities of edible fruits & nuts 0.056
(0.070) 0.056 (0.070) 0.056 (0.310) 0.074 (0.286) -0.019 (0.070) -0.019 (0.070) -0.019 (0.221) -0.017 (0.225) 0.092 (0.081) 0.092 (0.081) 0.092 (0.289) 0.112 (0.271) -0.066 (0.081) -0.066 (0.081) -0.066 (0.128) -0.075 (0.165) Quantities of edible vegetables -0.004
(0.085) -0.001 (0.084) -0.001 (0.350) -0.114 (0.259) 0.028 (0.085) 0.030 (0.085) 0.030 (0.352) -0.080 (0.264) -0.012 (0.099) -0.010 (0.099) -0.010 (0.240) -0.090 (0.185) 0.035 (0.099) 0.036 (0.099) 0.036 (0.239) -0.034 (0.189) Quantities of coffee, tea mate & spices -0.036
(0.053) -0.036 (0.053) -0.036 (0.057) -0.030 (0.071) -0.003 (0.053) -0.003 (0.053) -0.003 (0.103) -0.043 (0.090) -0.037 (0.062) -0.037 (0.062) -0.037 (0.016) -0.003 (0.066) 0.016 (0.062) 0.016 (0.062) 0.016 (0.126) -0.041 (0.089) Quantities of cereals 0.019 (0.142) 0.021 (0.142) 0.021 (0.121) -0.048 (0.106) 0.026 (0.143) 0.036 (0.142) 0.036 (0.126) 0.013 (0.114) 0.004 (0.167) 0.003 (0.166) 0.003 (0.151) -0.090 (0.153) 0.020 (0.167) 0.034 (0.166) 0.034 (0.148) 0.056 (0.164) Quantities of oil seeds & miscellaneous grains 0.020
(0.062) 0.020 (0.062) 0.020 (0.088) 0.033 (0.092) 0.161*** (0.062) 0.161*** (0.062) 0.162 (0.104) 0.176 (0.119) -0.104 (0.072) -0.104 (0.072) -0.103 (0.095) -0.098 (0.101) 0.215*** (0.072) 0.215*** (0.072) 0.215 (0.112) 0.227 (0.129) Quantities of sugars -0.306 (0.225) -0.304 (0.225) -0.304 (0.251) -0.296 (0.239) 0.201 (0.226) 0.206 (0.226) 0.206 (0.283) 0.190 (0.306) -0.525** (0.257) -0.523** (0.258) -0.523 (0.326) -0.496 (0.319) 0.454 (0.257) 0.458* (0.257) 0.460 (0.356) 0.425 (0.388) Quantities of cotton -0.300*** (0.083) -0.300*** (0.082) -0.298 (0.148) -0.317 (0.160) -0.177** (0.084) -0.170** (0.083) -0.170 (0.108) -0.193 (0.104) -0.302*** (0.096) -0.302*** (0.095) -0.302 (0.151) -0.318 (0.177) -0.025 (0.096) -0.020 (0.095) -0.020 (0.095) -0.033 (0.104) Quantities of optical & photographic instruments 0.147
(0.090) 0.147 (0.090) 0.147 (0.090) 0.176 (0.104) 0.211** (0.090) 0.211** (0.090) 0.212 (0.067) 0.229 (0.082) 0.054 (0.104) 0.054 (0.104) 0.055 (0.184) 0.077 (0.104) 0.183* (0.104) 0.183* (0.104) 0.184** (0.052) 0.189** (0.058)
Country fixed effects - Yes Yes Yes - Yes Yes Yes - Yes Yes Yes - Yes Yes Yes
Time fixed effects - - - Yes - - - Yes - - - Yes - - - Yes
Clustered standard errors - - Yes Yes - - Yes Yes - - Yes Yes - - Yes Yes
Notes: Dependent variables are converted to natural logarithms. Standard errors are given in parentheses under the coefficients. Regression (9)-(12) and (13)-(16) show the results for the first and second lag respectively.
Asterisks indicate the statistical significance of the estimated coefficients at the *.10 level, **.05 level, ***.01 level.
Table 6. The effect of drought on log quantities of different product categories
Independent variable: one or multiple lags of drought (binary variable)
29 Sign
The effect of drought on quantity traded is expected to be negative, due to the expected negative effect of drought on quantity produced. For all prices of product categories with the exception of prices of optical & photographic instruments a positive sign is expected: a period of drought leads to a lower production quantity which shifts the supply curve and leads to a higher price in equilibrium for water-intensive commodities. For the product category optical and photographic instruments, neither a positive or negative sign is expected because it is expected that drought has no effect on commodity prices of industrial products. The results for the product category of optical and photographic instruments show that drought does positively affect quantity traded, but this does not statistically significantly affect trade prices. It is unclear why drought positively affects quantity traded of this industrial product category, but the effect is quite large. The expectation that drought has no effect on prices of optical & photographic instruments is confirmed by the results.
The results for the effect of drought on prices of water-intensive goods are mixed. Four out of eight product categories show the expected sign. The other four product categories show a negative sign which states that one or two periods of drought will result in an increase in prices. The mixed results show that the prices of the eight different agricultural product categories investigated all respond differently on a period of water scarcity or drought.
The effect of drought on quantity traded is mixed as well. The product categories of fruits & nuts, vegetables and coffee, tea, mate & spices show coefficients that are close to zero. Therefore, the significant effect found for prices of these three product categories cannot be explained by the expected effect via quantity traded. Several reasons could explain why drought affects prices but not quantity traded, which are elaborated below.
Significance
Immediately it is clear that for the effect of drought on prices of five of the product categories the estimated results show high level of significance for the random effects model and for the fixed effects model when only country fixed effects are included. However, as soon as the clustered standard errors are included the significance disappears. These clustered standard errors capture the correlation of omitted factors, which are part of the error term, over time for a given country. When one does not correct for autocorrelation, the estimated standard errors are often too low (Stock and Watson, 2012). This can be seen in the estimated results of for example prices of edible fruits & nuts. Regression (2) and (3) have the same estimated coefficient, but including the clustered standard errors causes an increase in the standard errors of four times the value of regression (2).
The results of regression (1)-(4) compared to (5)-(8) show that more coefficients contain significance when the first lag of drought is used as the regressor. The exception of this is prices of sugar where the first lag shows no significance, but the second lag does. Therefore it is expected that also the effect of drought on quantity of sugar traded will be stronger for the second lag than for the first lag of
30
drought. This however is not the case. The results for the effect of drought on quantity of sugar traded is very large and in some cases significant.
An interesting result is that of the prices of cereals. For regressions using the first lag of drought as regressor (1)-(4) no significant coefficient is estimated. The same holds for regression (5)-(8) where only the second lag of drought is included as a regressor. However, including both as regressor leads to significant results for the coefficient of regression (15) and (16). That is, when both lags are included in the regression, the second lag of drought has a significant negative impact on prices of cereals. However, the effect of drought on quantity of cereals traded show no significance, and is not consistently positive or negative. The estimated coefficients for prices of coffee, regression (1)-(4), show significance even when country fixed effects, time fixed effects and clustered standard errors are included. The results of regression (11) and (12) are highly significant for coffee. Including both lags of drought as a regressor causes the regressors to explain more of the variation in coffee prices than using just one, which is reflected in lower standard errors. Again, the significant negative effect of drought on prices cannot be explained via quantity traded. If the results show that drought negatively affects import prices, it is expected, according to economic theory, that drought positively affects quantity produced. This can be explained by figure 1, when all the effects and thus arrows are opposite.
The effect of drought on prices and quantity of cotton is the only result that confirms the expectations. Drought has a statistically significant negative effect on quantity traded and a statistically significant positive effect on prices. Although its significance disappears when including time fixed effects and clustered standard errors, the estimated coefficient remains very large.
Interpretation
The mixed results for the effect of drought on trade quantity or prices makes interpretation difficult. To explain why quantity traded or prices are positively or negatively affected by a period of water scarcity one has to look at the production process of each product in more detail. For example, coffee beans are the inside of a fruit that grows on a coffee plant. For the plant and fruit to grow fresh water is needed. The same holds for tea: tea leaves are part of a tea plant that requires fresh water to grow. A period of water scarcity or drought could therefore result in a lower quantity produced and this again could positively affect the import price in a competitive market in which supply and demand interact. The results however show that water scarcity or drought has a negative impact on import prices of coffee, tea, mate and spices. Details on the production process of coffee and tea shows that this could possibly be explained by a later stage of the production process. Coffee beans and tea leaves are dried after harvesting which is commonly done on large fields in the open air. In case of precipitation, the producers have to take additional measures to complete this drying process and prevent precipitation affecting the product. This could raise the production costs without affecting the total quantity produced. Next to this, it could be the case that the customers in the Netherlands and suppliers in the countries investigated have
31
agreements on a fixed amount traded, but that these higher production costs are passed on to the customer.
Another reason why drought could affect prices but not quantity traded could be because of variation in quality of the agricultural commodities. Quality is not measured but it is possible that drought negatively, or perhaps positively, affects the qualtity of for example vegetables, and that this fluctuation in quality results in a different price of vegetables traded. Next to this, all the different agricultural products investigated, with the expection of vegetables and fruits, are easily stored in stock. This allows suppliers to produce additionaly in periods when there is no (extreme) water scarcity, and use this stock to compensate for the lower production in times when water scarcity affects the quantity produced.
Despite the several reasons why water scarcity or drought could affect import price but not import quantity, the mixed results for all of the agricultural product categories are difficult to interpret as a whole. Only a few significant results are found for the effect of drought on quantity traded, yet more significant results are found for the effect of drought on import prices. For explanation of the results found, as it turns out, it is necessary to dig deeper into the production process of each product. It is unfortunate that for this research only trade prices and quantities are available. For the exact effect of water scarcity on agricultural commodities, producer prices and data on quantity produced are necessary.
Another possible explanation for the mixed results could be the variability in production time of all different products. For all product categories, it is tested with one or two lags, months, if drought affects prices or quantities traded. However, the time needed to produce the commodities is likely to vary between the different products. This however, is one of the limitations of this study.
5.3 The effect of drought in country i on prices and quantities of other countries
The third and last part empirically tests if drought in country i affects average prices or quantities of the other countries observed. It is expected that when droughti,t-1 positively affects pricesi,t this causes a shift
in demand from the Netherlands away from country i and towards other countries. This increase in demand in other countries will result in a higher price in other countries according to the economic theory on interaction of supply and demand.
For each product category the country with the largest market share is selected based on quantity traded with the Netherlands. It is expected that when drought affects the quantity supplied of a large player, that the shift in demand away from that country is large as well. Therefore, other smaller players could face a significant increase in quantity traded with the Netherlands. This effect is demonstrated in figure 2.
The estimation results can be found in table 7. The first coefficient reported for Spain is -0.040. This should be interpreted as follows: drought in Spain, the largest producer of live trees & other plants,
