How the composition of trade has affected economic growth in the Netherlands through a time series analysis

(1)

How the Composition of Trade has Affected

Economic Growth in The Netherlands through a

Time Series Analysis

James Dance

Student number: 11124539

Supervisor/First reader: Professor Aslan Zorlu Second reader: Professor Sako Musterd

University of Amsterdam

Master’s programme in Human Geography: Economic Track

Email address: james.dance@student.uva.nl Email address 2: james.al.dance@gmail.com

(2)

Abstract

As global economies have become increasingly integrated the importance of international trade has evolved over the years due to freer and more accessible flows of trade and capital. Imports and exports of goods have emerged as a meaningful source for economic growth for any country: especially for more developed countries who can no longer depend on their stagnating domestic markets. The Netherlands is a prime example of a country that has established itself as a key distribution node for the ‘Western World’ and Europe. This paper explores the significance of trade in relation to Dutch economic growth:

investigating the causality of exports and imports in reference to economic growth from 1946-2012. This thesis will analyse the impact of industry-specific and continent-specific trade over time through the use of econometric

(3)

Chapter 1: Introduction

The Dutch have a long history of mercantilism and trade. From the 17th_{and 18}th

centuries Dutch ships transported spices and other primary commodities from the likes of India, Asia and the Caribbean to Europe to be manufactured. After processing these raw materials The Netherlands would then sell these products back to the markets from where they originated and also globally. The

Netherlands was the world’s major trading nation until the late 1700s until the British overtook them (Nationsencyclopedia, 2016). The Netherlands was responsible for initiating seaborne trade with both China and Japan; their

success was due to the design of their ships and their approach to trade – hosting large cargo holds with small crews. The Dutch manner of trade reduced

transport costs and therefore increased profit margins. Today The Netherlands’ trading heritage continues; as the nation remains focused on generating foreign trade to contribute towards economic growth. The Netherlands is currently the sixth largest economy in the Euro Zone (Theodora, 2016) and is well known for its vital role as a European distribution and transportation hub. It is for this reason that I have chosen to focus on how and which Dutch trade characteristics have contributed to economic growth over time: aiming to distinguish what factors have played a pivotal role in shaping The Netherlands into the developed economy it is today.

The link between economic growth and trade has been an important and an attractive area of research interest for many years; especially over the past fifty years (for example Michaely, 1977; Balassa, 1978; Feder 1983; Awokuse, 2008). Initial global trade theory stemmed from David Ricardo’s comparative advantage theory that he devised in 1817: claiming that countries benefit from specializing in industries in which they can produce goods more efficiently than their

competitors.

(5)

status of becoming a ‘developed’ economy by capitalizing on exogenous demand for surplus goods produced in an economy.

As the world we live in continually evolves and becomes increasingly digitalized, so does global trade. Especially in the more developed western world. In the developed world we have witnessed services leading the forefront of economies and becoming increasingly prominent in regards to revenue generation and therefore economic development. Thomas Friedman refers to the digitalization of the world as Globalization 3.0; where the world is becoming increasingly integrated and interconnected due to technological advancements. The purpose of this essay is to investigate whether goods are becoming progressively

irrelevant or if they are playing an increasing role in initiating economic development; in the form of Real Gross Domestic Product (Real GDP).

This thesis focuses on how the Dutch economy has evolved into the developed economy it is today – reviewing the dynamic roles that different trade

characteristics have played over time in reference to Dutch economic growth. For the purpose of this assignment this thesis will be using quantitative methods via the application of statistical models in an attempt to successfully answer the research questions found in Chapter 4. This assignment is looking to discover and investigate the potential univariate relationships between Real GDP (the dependent variable) and different trade characteristics (independent variables); establishing how trade in goods has contributed to Dutch economic growth in the 20th_{and 21}st_{century. This will be achieved through using a time-series and}

decomposition analysis via the computer program STATA. This thesis will visualize the data and results through a combination of histograms, tables and forecasting line graphs. This data visualization will be produced through a using both STATA and Microsoft Excel computer programs.

The selected trade characteristics cover continent-specific destinations (in the case of Dutch exports) and origins (in the case of Dutch imports) and industry specific trade statistics (imports and exports). Before this assignment addresses the statistical significance of specific trade characteristics a comprehensive theoretical framework in Chapter 2 has been outlined. This literature review

(6)

covers key theoretical concepts behind global trade and how trade contributes towards a country’s economic growth. The key causal theories discussed are import-led growth and export-led growth: as suggested in the name these theories focus on how trade can directly and indirectly contribute to economic growth. Chapter 2 goes on to analyze how and why countries trade and why it benefits a country to specialize in an industry in order to trade. Alongside trade specialisation and the reasoning behind it, this thesis focuses on the

characteristics of trade (both geographically and industry-specific) and how these characteristics can determine and affect a country’s economic growth over time. Chapter 3 contains a detailed description of the data, and how this data has been manipulated and adapted for a fair and comprehensive analysis allowing us to investigate the statistical causality of trade over time in relation to economic growth. Chapter 4 explains this paper’s research questions and hypotheses after looking at our Data’s descriptive statistics. Chapter 5 looks at the Methodology and Operationalization of this investigation; giving the reader the theoretical backgrounds to different econometric models and how this investigation has attempted to employ these models in order to produce significant and reliable results that can effectively answer the research questions found in Chapter 4. In Chapter 6 the paper’s empirical results are documented and discussed. In this chapter the mathematical coefficients and economic forecasts of trade scenarios are interpreted and analyzed to determine the statistical contribution of trade characteristics in reference to Dutch economic growth: giving the reader an insight into which trade characteristics play pivotal roles in promoting economic growth over different time periods. Finally Chapter 7 of this paper focuses on the conclusions and recommendations of this study. Chapter 7 critically reflects on the paper’s interpretations on the results witnessed in Chapter 6. Alongside these interpretations recommendations on how this study could be furthered and improved by future investigations are highlighted. These recommendations allow future investigations to further investigate the influence of global trade in relation to Dutch Real GDP growth.

(7)

Chapter 2: Theoretical Framework

The bulk of econometric literature has focused on the widely renowned export-led growth (ELG) hypothesis (see Santos, Ribeiro and Carvalho, 2013; Giles and Williams 2000; Kónya 2004; Palley 2011). Accompanying the ELG hypothesis, academics have also analysed the import-led growth (ILG) hypothesis (see Amiri and Gerdtham, 2012; Gkagka and Zarotiadis 2011). These studies have

concentrated on the statistical relationship between country-specific year-on-year changes in economic growth/contraction in relation to year-on-year-on-year-on-year changes in trade characteristics. These econometric concepts have been tested empirically to a wealth of nation-states with contentious and conflicting results. The debates between ILG and ELG studies fall down to scholars using different models and variables to produce statistical findings. These concepts are well documented and have been thoroughly investigated and due to the existence of this extensive literature this paper is going to focus on a less coveted subject; how export characteristics have contributed towards economic growth over time in a developed economy.

Rather than looking at the volume or value of trade, development economists such as Prebisch (1950, 1959) and Singer (1950) have evaluated the significance of trade characteristics; contesting Richardo’s theory by emphasizing the

existence of a link between export concentration (especially in primary

products) and degradation. Prebisch and Singer argue that countries should have a diversified export portfolio, rather than a concentrated and narrow one and countries should aim for an export portfolio consisting of more manufactured goods than primary goods. More recent studies from Sachs and Warner (1999, 2000 and 2001) and Sachs and Rodriguez (1999) have also supported the Prebisch-Singer proposition of diversifying exports and opposed the

effectiveness of David Richardo’s comparative advantage theory in the global market we live in today. Effectively you can apply a metaphor of a country’s export portfolio as being similar to a stock broker’s trading portfolio – any good stock broker will tell you that risks must be taken but also minimized/mitigated. One way of reducing risk is by diversifying your investments and not “putting all

(8)

of your eggs in one basket”. This paper will be analyzing the statistical contribution of trade characteristics on Dutch economic growth – looking at which industries and geographies have played the most significant role in

promoting The Netherlands’ economic growth over time. This paper investigates if the relative market capitalization of industries/continental trade destinations (in regards to other Dutch exports or imports) correlates with a characteristics’ influence on Dutch Real GDP.

The formation of trade can be determined by multiple factors, these include: the need of major trading partners (typically there is trading history between the two nation-states) and the need for proximity to sizeable markets (limiting the amount of space and time required to transport goods). Geography has always played a large part in trade due to transport costs; although this factor is being slowly eroded over time due to improved transport technologies (Rodrigue, Comtois and Slack 2006). Despite falling transport costs and the continual

reduction of travel time, physical distance is a variable that human beings cannot manipulate or change. Therefore the distance needed for goods to travel will always play an important role in determining a country’s trade. Furthermore Baliamoune-Lutz (2011) argues that the destination of a country’s export could determine the development and long-term growth of a country’s economy; even channelling export-related inward foreign direct investment (FDI) for distinct export sectors – for example The Netherlands has witnessed extensive FDI in Rotterdam in the form of international logistics firms having a significant presence: with global and renowned commodity broking firms such as Cargil, Glencore and Vitol. Additionally the port of Rotterdam promotes innovative programs to encourage business developments in both existing and new markets associated with trade (Portofrotterdam, 2016). This is known as economic

clustering and Rotterdam is an example of government policy looking to stimulate organic economic growth through localized innovations.

(9)

associated with socio-cultural links with former colonies, the demand for more trading partners or the need to establish a presence in new and large external markets (Baliamoune-Lutz, 2011). Amador and Opromolla (2008) go as far to say that the choice to penetrate a new market is as critical as deciding to start up a new firm. In literature concerning destination diversification arguments

conflict and contest each other. Literature highlights the optimal characteristics trading partners should target. A key foundation argument supporting export diversification comes under the premise of: the more markets you target, the higher the potential technology and knowledge spillovers you expose yourself to. This is down to producing different sets of knowledge (some tacit) through production and research experiences. De Loecker (2007) states that the more trading partners you have, the greater chance you have of absorbing positive externalities by witnessing ranging production techniques and ideas. Coe and Helpman (1995) and Grossman and Helpman (1991) further this argument by stating that the adoption of foreign technologies helps improve productivity and therefore facilitates the potential for increased economic growth. Kali et al. (2007) develop this idea further by declaring that increasing competition in our contemporary global market requires firms to continually innovate and a

constantly pursue increasing productivity gains; increasing the scope for economic growth. Lederman and Maloney (2003) argue that increasing your exposure to new knowledge and foreign technologies, reaching out to new markets gives MNCs an increased number of customers, new consumer tastes, external government regulations and alternative business climates. Santos et al. (2013) add an extra dimension to this argument by stating that potential

knowledge gains can vary from country to country – with some markets offering more than others. Keller (2004) furthers Santos et al.’s (2013) claim by declaring that knowledge is concentrated in a few countries; specifically the most

developed countries.

As well as increasing your exposure to knowledge and new customers,

Baliamoune-Lutz (2011) argues that if you diversify your exports geographically you intrinsically reduce the risk of export-dependency and lessen your exposure to idiosyncratic shocks. However due to the increasing integration of regional

(10)

and national markets these shocks are difficult to avoid in our contemporary and global economy: for example the 2008 global financial crisis significantly affected the world’s economies (although some more than others; Europe’s economies suffered greater losses than Asian economies with exception to Japan).

Historically one would expect that the geographical diversification of exports would have a more significant affect when considering global exogenous shocks as the world is becoming ever more integrated. As with most econometric theories there is of course a counterargument favouring export concentration. Kali et al. (2007) and Frankel et al. (1995) both declare that concentration can reduce costs that are associated with inadequate commercial infrastructure; in the form of seaports, airports, diplomatic locations etc. Kali et al. (2007) and Frankel et al. (1995) both highlight that a major cause for trading blocks: when trade is dependent on infrastructure it is imperative that trade can be effectively facilitated by a strong transport network which can help reduce costs. The renowned quality of The Netherlands’ infrastructure (most notably Rotterdam’s seaport and Schiphol airport) is a key factor, alongside its geographic location that has contributed to The Netherlands being one of Europe’s largest trade distribution hubs.

Aside from the total number of trading partners, the characteristics of the country that goods are exported to play an important role in regards to promoting economic growth. The most distinct factor is the rapid growth of exogenous markets. Arora and Vamvakidis (2005) put this in Lehman’s terms by explaining that the higher the overall growth rate of trading partners, the higher their demand is for Dutch exports. This however cannot be taken as a given: as there may be higher quality alternative goods (substitute goods) available elsewhere; a higher purchasing power parity of the growing importer may cause demand to shift to alternative markets. An example of this could be seen with the demand for cars; Chinese cars may be cheaper, however, as an economy grows it may switch its demand for higher quality cars that are produced in Germany: which is renowned for its high-quality automotive industry. This demand is

(11)

of higher quality – which fundamentally result in industries targeting these economies to continually innovate to improve performance. The selection of customers is intended to retain current customers and attract new ones by “pushing the world technological frontier forward” (Santos et al., 2013). This implies that a trade network with more developed countries in an export

portfolio gives the exporting economy better access to superior knowledge flows due to buyer expertise and access to a highly educated market (Damijan et al., 2004 and Clerides et al., 1998). Therefore exporting to less developed economies can result in lowering productivity due to reduced demand for the

quality/delivery of goods: which may result in decreased efficiency (Vacek, 2010). A counter argument to this is that businesses may have a wide range of destination countries for their goods: from the most developed nations to less developed economies. Due to the standardisation of goods, the destination country of exports I expect has become less important; firms now seek to target specific socio-economic groups rather than countries – especially with highly manufactured products such as cars. This is partly due to the inelastic nature of demand for high-end goods combined with the improvement of transportation technologies that can now overcome geographical barriers and allows industries to specifically target groupings of customers rather than whole countries or continents. Creating different standards of the same model would be an

inefficient process of production and therefore a standardized model is produced that is made for the global market, rather than specific markets. An exception to this is in the soft drinks market where it is common knowledge that drink companies such as Coca-Cola have to limit their sugar content of their products in European economies, however these regulations do not exist in other

continental markets such as Africa. Under the assumption of standardisation exporting goods globally to a range of markets (rich and poor) typically have to meet the minimum requirements of the toughest government regulations: these goods would by proxy be able to enter less regulated markets. A critique of this can be applied by stating regional production centres may produce goods for their immediate and localised market rather than for global production. Although for the purpose of this essay we will not be focusing on regional production hubs.

(12)

Further to the geographic distribution of goods, the structure of a country’s trade portfolio can influence economic growth as mentioned by Sachs and Warner (2001), Baliamoune-Lutz and Ndikumana (2007): supporting the hypothesis that the structure of trade can be critical to long-term economic growth. With studies investigating the effect of the manufacturing industry on the economic growth of developing countries; this is referred to as ‘the fallacy of composition’ by

academics such as Cline (1982, 1984, & 2008), Ranis (1985) and Martin (1993). Whilst the manufacturing industry is widely associated with economic returns due to the inherent nature of adding value to primary products, researchers such as Hausmann, Hwant and Rodrik (2007) have investigated the influence of specialization patterns of countries and how this can result in differences in economic development. Mazumdar (1996) states that a country’s trade portfolio is a major principle of the ‘engine of growth’. Lewer (2002) and Lewer and Van den Berg (2003) discovered evidence supporting that if countries export consumer goods and import capital goods they will grow faster than their

counterparts that export capital goods. Baliamoune-Lutz and Ndikumana (2007) support this idea by stating:

“capital-intensive sectors such as oil [are] not likely to generate growth that is sustainable, especially because of the low gains in employment creation and limited spillover effects on non-oil sectors.”

Hesse (2008) proved that there is a non-linear association between export diversification and income per capita in developing countries. He states that whilst developing countries benefit from a diversified export portfolio, most developed countries in fact benefit from having a narrow portfolio by

specializing in specific industries. Baliamoune-Lutz and Ndikumana (2007) further Hesse’s claim by declaring that export portfolio heterogeneity enhances the growth of economies in Africa:

(13)

Increased support for this argument is found by Carrère et al. (2009) who identify an ‘inverted-U relationship’ between economic growth and export diversification; believing that a turning point is found around a purchasing power parity of $24,000 per capita – signifying the discovery that above this threshold the phenomenon of re-concentrating exports occurs in more

developed countries. Agosin (2007) discovers that a large export portfolio that is heterogeneous has a highly significant positive impact in South American and Asian countries in comparison to their global competitors. A critique of this argument comes from Hausmann et al. (2007) who have developed a theoretical framework suggesting that local cost discovery produces knowledge spillovers: demonstrating that trade specialization becomes partly indeterminate. This argument by Hausmann et al. (2007) deduces that the combination of goods that an economy produces has key ramifications for economic growth. To support their argument Hausmann et al. (2007) have created an index from empirical evidence that shows the ‘income level of a country’s exports’ – which can anticipate economic growth based in export demand. Hausmann et al. (2007) highlight that countries can become trapped with lower income goods due to a lack of entrepreneurship. To overcome this ‘trap’ countries can manipulate levels of entrepreneurship through constructing policies that encourage

entrepreneurship in new industries and out of primary resource markets (as seen in the example mentioned earlier for the case of Rotterdam) - enabling economies to enter and subsequently increase trade revenues by expanding their scope of trade into new markets whilst inherently spreading export

portfolio ‘risks’. Lederman and Maloney (2003) contradict this by stating that the relationship between resource abundance and economic growth is positive, this however is contentious and is down to a variety of factors; one of their

aforementioned factors, which is a key consideration, is the requirement for strong institutions that promote the redistribution of economic gains. The United States, Norway and Finland have demonstrated that economic growth can be sustained through natural resource based production and therefore they reject the ‘resource curse’ hypothesis (Wright and Czelusta, 2002; Blomoström and Kokko, 2003). The Netherlands however, doe not own a substantial amount of

(14)

natural resources, therefore this natural resource theory cannot be investigated in this case.

In conclusion to this literature review Amador and Opromolla (2008) established that export diversification and the geographical destination of products work in conjunction to contribute towards economic growth. Their study relied on Portugal-based MNC micro-data: analysing export composition from 1996-2005. They concluded that economic growth is related to the versatility of firms’ exports. More specifically firms that exported over four products in over four different markets were responsible for over two thirds of total exports. This is an article that is extremely useful and insightful as it focuses on major interests in trade characteristics within a developed country.

Accompanying this discovery in Amador and Opromolla’s work, the production of new goods was also positively correlated with export growth and therefore economic growth. Arora and Vamvakidis (2005) support Amador and

Opromolla’s claim with their study that used panel data for over 100 countries showing that trading partners’ economic expansion has a strong effect on domestic growth: with relative income growth of trading partners having a positive relationship with an exporter’s economic growth. This proves that the wealthier your trading partner, the more likely you are going to experience economic growth. In contradiction to the previous findings sited above Baliamoune-Lutz (2011) makes an interesting discovery in her analysis by concluding that African nations benefit by exporting a singular commodity in bulk to China, rather than diversifying their portfolio both geographically and structurally. This however could be due to the developing nature of African economies, as well as the importance of proximity from Africa to China compared to other nations such as Portugal.

(15)

Chapter 3: Data

3.1 Dependent Variable

The dependent variable in this investigation is the annual observation of Dutch Real Gross Domestic Product (Real GDP) from 1946-2012. This variable has been chosen to represent the economic output of The Netherlands as it replicated the aggregate demand of The Netherlands. Aggregate demand is calculated through the combination of:

1. Government Spending 2. Investment

3. Consumption

4. Balance of Trade (Imports and Exports)

Alternatively in its equation form:

GDP = G + I + C + (X-M)

To understand the relative increase of GDP this investigation calculates the nominal value of Real Gross Domestic Product (Rgdp) – this has been achieved through removing year-on-year inflation that is represented through The Netherlands’ Consumer Price Index (CPI). Obtaining this real value allows us to investigate the real increase or decrease in reference to demand for ‘real’ imports and exports in The Netherlands. Rgdp has been chosen as the dependent variable (in STATA) as it represents the total value of goods and services produced by an economy in a year. Rgdp gives us an insight into the health of an economy. Positive Rgdp growth is desired by every economy. Below is the equation that was used to calculate the real values of GDP. This equation was applied to other variables to by simply substituting the

dependent variable (Rgdp) with the other explanatory variables (i.e. Total Imports, Asian Exports or Chemical Products Imports). The CPI is divided by 100 in order to obtain an index value that can be applied to the GDP value to create a ‘real’ statistic.

(16)

Rgdp = GDP

(CPI/100)

The rationale behind choosing Rgdp as the dependent variable is down to the paper’s objective of analysing the economic contribution of trade in relation to the Dutch economy over time rather than focusing on the more holistic topic of a society’s development (some of which cannot be fully and efficiently

quantified). Due to the quantitative nature of this study Rgdp was chosen, as it is an objective index that measures the changing throughput of an economy and does not have subjective elements that may result in data inconsistencies and inaccuracies such as quantifying ‘happiness’ or ‘life satisfaction’. This paper does not claim that economic growth is always synonymous with improved ‘well-being’ but this paper is investigating the changing significance of trade and how it contributes towards domestic economic growth in The Netherlands, not development as a whole.

3.2 Independent Variables

To undertake this time-series analysis this paper uses historical import and export data that was extracted from the Centraal Bureau voor de Statisiek (CBS) online database from 1921-2012. This is The Netherlands’ official national statistics database. As with the dependent variable (Rgdp) this paper has calculated the ‘real’ values of imports and exports through removing annual inflation values. This was achieved by substituting the CPI into the annual values logged by CBS using the same method we used to generate Rgdp. The import and export data that was chosen gives us the accumulative values that are designated to 6 different industry goods baskets:

1. Food, drinks and tobacco

2. Animal and vegetable oils, fats and wax 3. Mineral fuels

(17)

6. Machinery and transport equipment

On top of these industry specific categories this paper uses historical regional-specific import and export values. These characteristics allow us to investigate the significance of the geographical distribution of Dutch trade. This paper has managed to extract 11 different countries’ data regarding respective import and export values over time. CBS has provided ‘other countries’ categories within some regions due to smaller trading partners with The Netherlands having small values such as Spain or Portugal within Europe or Canada within North America. The identification of these countries by CBS is not dependent upon their wealth or status, but the importance of a trading partner is based upon the value of trade done between The Netherlands and country X. Due to countries being aggregated into ‘other’ groups this paper is going to solely focus on continent-specific trade rather than country-specific trade in order to keep the analysis consistent throughout this empirical study. Oddly it seems that The Netherlands has minimal trade with South American countries – with CBS deeming South America not even significant enough to be categorized in the data set: consequently we have obtained the other five continents to analyse:

1. Africa 2. America 3. Europe 4. Asia

5. Australia and Oceania

3.3 Preparing the data

The data was stored in STATA and declared as time-series data through inputting the command “tsset Year” – with “Year” being our time variable. Initially the time series analysis was set from 1921 to 2012. However, due to data inconsistencies and missing data values before WW II this paper will

analyze historical Dutch trade data from 1946 onwards. If we had used data with missing values our econometric models would have been inaccurate. This

(18)

development made little difference as global trade began to grow exponentially after the WW II as nations and Multinational Corporations (MNCs) began to target exogenous markets. In figure 1 we can see that Rgdp is very low before growing at an exponential rate almost instantly after WW II.

Figure 1 – Dutch trade growth over time

On top of using categorical trade independent variables, “dummy” variables were created for significant time periods in order to investigate the connotations of exogenous global economic shocks and how commodities’ influences change in different global economic conditions. Major global economic crises such as “The First Global Oil Crisis” in 1973 and 1974, the “Second Global oil Crisis” in 1979 and the 2008 “Global Financial Crisis” were identified. These dummy variables allow us to identify which geographies and industries are the most

significant/important in reference to Dutch economic growth within periods of economic recession and global uncertainty. These shocks were selected through a combination of online research and overlooking our raw Real GDP data; determining periods of negative Real GDP growth. Interestingly periods of negative Dutch Rgdp growth coincided with global recessions – furthering

0 5000 10000 15000 20000 25000 30000 35000 1946 1952 1958 1964 1970 1976 1982 1988 1994 2000 2006 2012 Value in Millions of Euros Year

Annual Dutch Real Global Exports vs

Imports

Total Dutch Imports Total Dutch Exports

(19)

With results that cover different time periods of recession and crises we can pinpoint changes of trading patterns and analyze how geographical demand for Dutch exports and imports has changed over time. Using different time periods shows us the relative strength of industries within and outside of The

Netherlands, as well as identifying the strength of continental trading

partnerships (i.e. The Netherland’s relationship with Asia vs. North America). The specific time periods I have chosen are:

1. Post EU Trade Agreement (1960-2012) 2. The First Oil Crisis (1973-1974)

3. The Second Oil Crisis (1978-1979) 4. Globalization (2000-2012)

5. Financial Crisis (2008-2012)

Our first dummy variable is when The Netherlands joined the EU trade

agreement in 1960 with other European countries such as Germany and France to create a ‘single market’. This dummy variable was selected to see the effects this trade agreement and to investigate this agreement’s effects on trade characteristics. Within this trade agreement’s timeframe other factors such as advanced telecommunications (the internet and mobile phone devices) and improved transport technologies have also changed the global landscape but this is still an key time period to analyze in comparison with the whole time period of 1946-2012. Our second and third dummy variables are associated with the global oil crises in the 1970’s. These two periods were filled with consumer and business uncertainty as the price of oil dramatically crashed; resulting in two periods of global economic recession. The increased accessibility of

telecommunications is such a large factor that has impacted global trade that the dummy variable ‘Globalization’ was constructed. This variable stemmed from the previously mentioned Globalization 3.0 theory of Thomas Friedman. This dummy variable runs from 2000-2012 and will give us some great insight into how increased telecommunications and improved transport technologies have affected Dutch trading patterns. Finally the effects of the 2008 ‘Financial Crisis’ was chosen to be investigated as the most recent time period due to its

(20)

renowned effects on the global economy, and its effects are prominent to this day in 2016. Although this recession has been dubbed ‘The Financial Crisis’ in

reference to the global financial services defaulting: this recession had an affect on all industries in the world – especially in the Western world of Europe and North America. Investigating this period’s affect on Dutch GDP and trade in goods is interesting as it further explores Friedman’s hypothesis of

interconnectivity.

Figure 2 – Real GDP growth over time

As you can see from looking at Figures 1 and 2 our real trade and Real GDP data both carry upward trends and are therefore non-stationary. However, a cyclical pattern (seasonality) is absent. This makes our regression analysis easier – as we do not need to “smooth out” the data for analysis. For time-series analysis the application of decomposition techniques are required. Prior to modeling the data we need make the data stationary. Stationary data is data that does not show a trend, it is data that is consistent throughout a time series. Data is made

0 5000 10000 15000 20000 25000 30000 35000 40000 45000 50000 Ye ar 1949 1953 1957 1961 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 2005 2009 Millions of Euros (€)

Real GDP Over Time

(21)

for a number of reasons: to obtain residuals that are evenly distributed around 0 and to transform our data from “heteroscedastic” to “homoscedastic” (removing the systematic change in the data’s spread). This decomposition linearizes the relationship between the dependent and independent variables. Data can be made stationary through various techniques. The two most commonly applied methods are by either differencing your data or by applying natural logarithms to your data. Our data was made stationary through obtaining the natural

logarithms of our data and adding a first order difference to our data. This paper chose to naturally log the data for interpretation purposes. A natural logarithm function is calculated in reference to “base”, which is a positive number. This number is a boundless number “e” whose decimal enlargement is 1.73452.., the natural log function and the exponential function (ex_{) are inversed versions of}

each other. Using a logarithm can simplify the number and complexity of “interaction” terms by giving the reader an explanatory variable’s percentage influence on Rgdp. Adding a first order difference calculates the change from one period to the next. If Yt denotes the value of an observation in time series Y at

period t, then the first order difference of Y at period t is equal to Yt – Yt-1.

The naturally logged data was not fully stationary through the Augmented Dicky-Fuller (ADF) test and the Phillipe Perron (PP) test: tests that identify if our data is stationary or not. In order to make our data fully stationary we had to

difference our data. A major advantage of the PP test is that it is non-parametric: meaning that it does not require the user to select the level of serial correlation as in ADF, however, in Stata this is computed automatically and therefore we neither had to specify this. The PP test takes the same estimation as the ADF test, as well as correcting the statistics that test for autocorrelations and HAC type corrections (known as heteroscedasticity). The PP and the ADF both also share disadvantages of sensitivity to structural breaks and small sample power that typically results in unit root conclusions with smaller samples. This is not a problem for us as we have a large time-series data set to work with. By using both of these tests we can guarantee that our data is 100% stationary if we can reject the null-hypothesis of both tests. The null hypotheses of theses tests state that if the p value is above 0.05, then the data is not stationary. Figure 3 shows all

(22)

of our variables and their ADF and PP test scores. As we can see in figure 3 prior to differencing and by just having naturally logged variables this did not

guarantee significant stationarity, therefore we had to difference our naturally logged variables to fulfill the key assumption of stationarity in order to

effectively model and analyze the data.

Figure 3 – pperron and dfuller tests: naturally logged and differenced and logged variables

Variable pperron dfuller pperron after

differencing dfuller after differencing Ln Real GDP 0 0 0 0 Ln Total Exports 0 0 0 0 Ln Total Imports 0.01 0.02 0 0

Ln Total African Imports 0.02 0.02 0 0

Ln Total North American Imports 0.6 0.56 0 0

Ln Total European Imports 0 0 0 0

Ln Total Asian Imports 0 0 0 0

Ln Total Australasian and Oceania Imports 0.13 0.14 0 0

Ln Total African Exports 0 0 0 0

Ln Total North American Exports 0.01 0 0 0

Ln Total European Exports 0 0 0 0

Ln Total Asian Exports 0.02 0.01 0 0

Ln Total Australasian and Oceania Exports 0 0 0 0

Ln Food, drinks and tobacco Imports 0 0.01 0 0

Ln Animal and vegetable oils, fats and wax

Imports 0 0 0 0

Ln Mineral fuels Imports 0.25 0.25 0 0

Ln Chemical products Imports 0.05 0.24 0 0

Ln Manufactured goods Imports 0 0.01 0 0

Ln Machinery and transport equipment Imports 0.01 0.01 0 0

Ln Food, drinks and tobacco Exports 0 0 0 0

Ln Animal and vegetable oils, fats and wax

Exports 0 0.01 0 0

Ln Mineral fuels Exports 1 1 0 0

Ln Chemical products Exports 0 0 0 0

Ln Manufactured goods Exports 0 0 0 0

(23)

3.4 Descriptive Statistics

Prior to interpreting the statistical modeling/regression results we will look at what the raw data sets tell us. Looking at our data over time prior to significant data manipulation is key to get a good understanding of what characteristics the data holds. All of our raw data has been converted into ‘real’ terms as with our data used in Stata – this was carried out in order to get a fair and

non-inflationary understanding of our data. As stated previously in figure 2 we have already identified that we can see there is an upward trend in Rgdp –

demonstrating year-on-year real economic growth from 1946 to 2012 with an exception to a few years of recession. We have also seen that in conjunction with this growth, the real value of trade growing over the years (see figure 1). In figure 4 the real absolute value of Dutch trade (the total value of Dutch imports plus the total value of Dutch exports) in relation to Dutch Rgdp is visualized in a line graph.

Figure 4 – Real Absolute Trade vs Real GDP over time

In figure 4 the real absolute value of trade surpassed the total value of Rgdp in 2000; interestingly at the beginning of Thomas Friedman’s Globalization 3.0 theory. This tells us that global trade in goods is becoming increasingly significant and that global trade has been growing at a much faster rate than domestic GDP over our chosen time period. In fact Rgdp has seen a 1059.1% increase from 1946 to 2012; whilst the absolute value of Dutch trade has seen a

0 10000 20000 30000 40000 50000 60000 70000 1946 1951 1956 1961 1966 1971 1976 1981 1986 1991 1996 2001 2006 2011 Millions of Euros (€) Year

Real GDP Vs. Real Absolute Trade Over

Time

Real Absolute Value of Trade in Goods Real GDP

(24)

huge 5498.5% increase. Meaning that Rgdp has increased by over ten fold whilst absolute trade has increased by over fifty four times since 1946. Alongside The Netherlands’ increasing trade in goods growth we can also see in figure 4 that trade is far more volatile than GDP – due to its inherent exposure to global markets. This is especially apparent between 2008-2009 where we see trade dropping 19% in one year due to the global financial crisis.

3.5 Geographical Distribution of Trade

The importance of global markets further highlighted by figures 5 and 6 – showing the geographical distribution of Dutch imports/exports over time. Interestingly we can see in figures 5 and 6 that North America initially played a vital role in the Dutch import market from 1946-1947, straight after WWII. However, North America’s significance greatly fell to a 20% share before tailing off and stagnating to just above 10% from the 1970 onwards. Most importantly we can see that Europe has dominated the geographical compositions of Dutch imports and exports from 1946 to 2012 with an average market share of 65.7% and 78% respectively throughout the data set. Interestingly we can see that Dutch Exports are not as globally competitive as initially expected, and Europe, in the case of Dutch exports, is an extremely significant trading partner for The Netherlands.

In contrast we can see that Asian exports and imports from the 1990s onwards have become increasingly significant with high economic growth rates

correlating with increasing Asian productivity and demand. These findings contradict Tobler’s first law due to Asia surpassing both North American and African demand for Dutch exports, whilst also achieving even higher figures in comparison to all other continents for Dutch imports. Instead, these

observations support Thomas Friedman’s argument of an increasingly shrinking world, as price and quality competitive goods can overcome geographical

(25)

“Everything is related to everything else, but near things are more related than distant things’’ (Tobler 1970).

Janelle (1969) contradicts this statement by highlighting that the past two centuries have seen space-time convergence; as transportation and

communication technologies have “shrunk” the world. In the world we live in today geographical locations are much closer to each other in respect to the time requires for travel.

Figure 5 – Geographical distribution of Dutch imports over time

Figure 6 – Geographical distribution of Dutch exports over time

0 10 20 30 40 50 60 70 80 1946 1951 1956 1961 1966 1971 1976 1981 1986 1991 1996 2001 2006 2011 Percentage of Trade Year

Distribution of Dutch Imports from

Continents Over Time

Africa

North America Europe Asia

Australia and Oceania

0 10 20 30 40 50 60 70 80 90 100 1946 1951 1956 1961 1966 1971 1976 1981 1986 1991 1996 2001 2006 2011 Percentage of Trade Year

Distribution of Dutch Exports to

Continents Over Time

Africa

North America Europe Asia

(26)

3.6 Industry-Specific Trade

Dutch trading patterns (both imports and exports) can give us some insight into the relative competitiveness of Dutch goods. If we witness a Dutch industry exporting vast quantities globally we can assume that the industry in question is globally competitive. For instance Siemens (a Dutch technology manufacturer) will be one of the largest Dutch exporters due to its renowned world-class and globally competitive goods; this company produces far more goods than those solely needed in The Netherlands. Due to their extremely competitive goods Siemens can export its goods globally – as consumers are willing to pay for these goods to be transported to their respective markets. In figures 7 and 8 we can see that the Machinery and Transport Equipment is the largest industry in respect to both imports and exports. The Mineral fuels industry is the second largest trade industry; overtaking Manufactured goods in both trade categories in 2011 with an average 17.5% market share gain since 1946. Manufactured goods have lost the largest share of The Netherlands’ Import portfolio with a 17.1% loss from 1946 to 2012. Animal and vegetable oils, fats and wax and Food, drinks and tobacco imports have also lost their market share by 12.4% and 7.7%

respectively from 1946-2012. Similarly to the import data, both the Machinery and Transport Equipment and Mineral fuels exports gained the most, making these the two largest Dutch trading industries with 25.6% and 19.3% market share respectively in 2012.

Looking figures 7 and 8 we can clearly see that Dutch imports and exports are positively related with each other, as every industry has either both positive or negative market share gains from 1946 to 2012 over the same time periods. For example when Chemical Product imports fall, so do Chemical Product exports. This is surprising to me, as one would expect that if you import specific industry related goods, then you would not export many goods in this industry and visa versa. We would expect imports and exports to act as substitutes for each other, however in these descriptive statistics it does not seem to be the case.

(27)

Figure 7 – Distribution of Industry-specific Dutch imports over time

Figure 8 – Distribution of Industry-specific Dutch exports over time

The mathematical significance of these geographies/industries and their impact on Dutch economic growth through trade will be further explained in the

statistical modeling results section. 0 5 10 15 20 25 30 35 40 45 1946 1952 1958 1964 1970 1976 1982 1988 1994 2000 2006 2012 Percentage of Import Portfolio Year

The Netherlands' Import Portfolio:

Distribution of Industries Over Time

Food, drinks and tobacco Imports

Animal and vegetable oils, fats and wax Imports Mineral fuels Imports Chemical products Imports

Manufactured goods Imports

Machinery and transport equipment Imports 0 5 10 15 20 25 30 35 40 45 1946 1952 1958 1964 1970 1976 1982 1988 1994 2000 2006 2012 Percentage of Export Portfolio Year

The Netherlands' Export Portfolio:

Distribution of Industries Over Time

Food, drinks and tobacco Exports

Animal and vegetable oils, fats and wax Exports Mineral fuels Exports Chemical products Exports

Manufactured goods Exports

Machinery and transport equipment Exports

(28)

Chapter 4: Research Questions and Hypotheses

After Looking at our data and descriptive statistics we are now able to apply some hypotheses to our data. The questions asked are based upon the trade-specific literature and are trade-specifically tailored to our data. These questions have been created to investigate the possibilities of Import and Export Led growth and what characteristics have promoted/hindered Dutch economic growth from and within 1946-2012. The questions are as follows:

1. Which continents have had the greatest impact on economic growth in The Netherlands over time?

a. Does the relative distance of these continents from The Netherlands have an effect on the demand for traded goods?

2. What industries have had the greatest impact on economic growth in The Netherlands over time?

a. Does the market capitalization of these industries in The Netherlands’ import and export portfolios correlate with their contributions to Dutch Real GDP growth?

3. How important is international trade in relation to Dutch economic growth?

a. Have imports and exports increased their contribution to economic growth over time?

(29)

4.1 Thesis Statement and Hypotheses

Dutch exports’ and imports’ influence will become more dispersed

geographically (i.e. Dutch trade has a further reach and has become more reliant on more distant markets) due to Freidman’s hypothesis of the world shrinking over time and becoming increasingly interconnected;

“This is Globalization 3.0. In Globalization 1.0, which began around 1492, the world went from size large to size medium. In Globalization 2.0, the era that introduced us to multinational companies, it went from size medium to size small. And then around 2000 came Globalization 3.0, in which the world went from being small to tiny.”

(Thomas Friedman)

In respect to the importance of geographical economies Europe, North America and Asia will be the most significant trading partners due to their inherently large importing and exporting market capitalization sizes. Just from

overlooking The Netherlands’ trade data we can see that the rest of Europe is The Netherlands’ largest importing and exporting market (see figures 5 and 6 in the Descriptive Statistics section in Chapter 3), followed by North America and Asia. The vast contribution of Europe will contribute the most value to Dutch economic growth, however, newer and rapidly expanding markets in Asia will also play a significant role in economic growth; showing increasing contributions, especially in the last 15 years (i.e during the Globalization tie period). Australasia and Oceania will have the least impact on Dutch Real GDP growth due to the sheer distance and lack of trading history (Common Wealth ties exist between Australasia and Oceania and Britain instead). North America will have a significant impact due to its renowned global status as an economic super power, whilst Africa will be a less significant trading partner alongside Australasia and Oceania.

After looking at the descriptive statistics one would expect mineral fuels, chemical products and machinery and transport equipment trade to be the most influential industries on Dutch economic growth due to The Netherlands’

(30)

globally renowned fuel and chemical refining reputation. Whilst the smaller industries such as Food, drinks and tobacco and Animal and vegetable oils, fats and wax will have a lesser impact on Dutch Real GDP growth.

We would expect both imports and exports to play a significant role in

promoting Dutch economic growth – becoming increasingly important as time goes on; as the world becomes increasingly interconnected through

telecommunications and technologies, it is these technological advancements that encourage greater integrations between regional and global markets. Additionally all trade will display long-term (both future and past values) positive causal effects on Rgdp growth in the Netherlands. One would exports to have a greater effect on Dutch Rgdp due to their inherent nature of instantly creating revenue for the Dutch economy. Dutch exports also have a greater market value than Dutch imports (see figure 1 in the Descriptive Statistics section in Chapter 3): this paper expects imports’ and exports’ influence to correspond with their respective market capitalizations.

(31)

Chapter 5: Method and Operationalisation

5.1 Multiple Linear Regression Model

Initially this paper attempted to analyze our data with the most parsimonious model first; a multiple linear regression model. This model analyzes the relationship between a scalar dependent variable (dlnRgdp) and multiple

explanatory/independent variables (i.e. chemical products exports, manufactured goods exports and machinery and transport equipment exports). This model uses multiple linear dependencies to establish the statistical relationships between our data. Every value of the independent variable x is associated with a value of the dependent variable dlnRgdp. For example if we look at how Asian exports, African exports and North American exports affect Rgdp from 1946-2012; the model will then produce coefficients that give us an indication into how these continental exports have affected dlnRgdp on average within this time period in reference to an assumed linear dependency. In figure 11 we can see how

industry-specific imports from 1946-2012 have affected dlnRgdp in the form of coefficients. Multiple linear regression in its mathematical form is expressed as:

𝑌 = 𝑏0+ 𝑏1𝑥1+ 𝑏2𝑥2+ ⋯ 𝑏𝑝𝑥𝑝

Where Y is the predicted or expected value of the dependent variable (dlnRgdp), X1 through Xp are p distinct independent variables, b0 is the value of Y when all of

the independent variables (X1 through Xp) are equal to zero, and b1 through

bp are the estimated regression coefficients. Each regression coefficient

represents the change in Y relative to a one unit change in the corresponding independent variable. In the multiple regression situation, b1, for example, is the

change in Y relative to a one unit change in X1, when the other independent

(32)

5.1.1 Post Estimation

After producing results for the effects of continental exports on dlnRgdp

(displayed in figure 12) the results were tested for autocorrelation. The results were tested for autocoreelation prior to adding any paramters (i.e. lags or leads): this is essential as it is always necessary to double check the reliability and quality of a statistical model before producing significant results; reaffirming that the model used is one of best fit. Autocorrelation, or “serial correlation” occurs when a data point is strongly affected by its previous values: this implies there is a correlation between an observation’s past values and its current value (there is a correlation between yt and yt-k). We do not want this because

autocorrelation gives us scewed and innacurate results. The Durbin-Watson Test (DW) is used to test for for first order serial correlation. This test assumes variables are stationary as well as normally distributed with an average of zero. We know that all our variables are stationary from our previous ADF and PP stionarity test results (found in Chapter 3: Data, figure 3). Since the DW’s creation in 1951, the DW test has been extremely popular in statistics and especially for time series analysis.

Figure 9 – Durbin Watson Test Results

(Moksony, 2016)

The DW has two critical values, not just one, such as the ADF test. In the DW test there is a lower critical value (dL) and an upper critical value (dU). The

magnitudes of these two threshold values depend on four factors: sample size, number of independent variables, the selected significance level (usually 5%), and the type of the test selected (one-sided or two-sided). In our case this is a two-sided test; looking for evidence of both positive and negative

(33)

autocorrelation scores are below two, and negative autocoreelations scores are above two). In figure 10 we observe a score below one out of four (0.985) – not allowing us to reject the test’s null hypothesis and leading us to believe that we have first order serial autocorrelation in our data set. To confirm this result a statisics table was consulted (that provides confidence intervals to establish if the model that was have chosen and its specifications) to reaffirm if our data is serially autocorrelated or not (Stanford University’s statisicts website, 2016). The 0.985 figure did not fall within the required confidence intervals –

concluding that the data was, as expected, serially autocorrelated. Even though the DW test is widely used in statistics it does carry a few

shortcomings. First, the model’s form assumed by the test is measured by the dimension p and its explanatory variables. In reality this is seldom the case; instead a data based procedure must be implemented to effectively identify the model in question. Secondly, the test is sometime inconclusive if your data falls within specific confidence intervals. Lastly, the model does not account for in fact there are Moving Average (MA) errors rather than Autoregressive (AR) errors, or even a random walk; the DW test can sometime suffer from its own

presumptions and fail to detect alternative issues to serial autocorrelation as mentioned. We do not need to worry about MA and AR errors in this model as these parameters cannot be applied in a multiple linear regression model. MA and AR parameters will be explained later on in this chapter under our

Autoregressive Integrated Moving Average (ARIMA) model section (5.3).

Even though the DW test suggests that we have serially autocorrelated data, as with everything in statistics we want to proof this: this is done by looking at the Auto Correlation Function (ACF) or Partial Autocorrelation Function (PACF) plots of our model’s residuals to second the test with our own

opinion/interpretations. If we observe any of these models’ residuals outside the 95% confidence interval (the grey area on the chart), we then have

autocorrelated data. As we can see in Figure 11 (an ACF plot of the model’s residuals from figure 12) we have autocorrelated data, as one residual is clearly located outside the 95% confidence interval. This discovery forces us to abandon

(34)

this linear regression model due to our data not fulfilling a key statistical

assumption of not having autocorrelated residuals that derive from our results.

Figure 10 – Durbin-Watson Test statistic telling us that our data is serially autocorrelated

Figure 11 – ACF plot reaffirming that autocorrelation is present in our model displayed in figure 11

Figure 12 - Multiple Linear Regression : Industry-specific exports and their effects on Real

GDP

Durbin-Watson d-statistic (7, 66) 0.9854256

Number of Observations 66

Adj R-squared 0.3731

Independent Variables Coefficient Standard Error

Food, drinks and tobacco Exports -0.02 0.06

Animal and vegetable oils, fats and wax Exports 0.07 0.06

Chemical products Exports 0.34 0.05

Manufactured goods Exports 0.07* 0.04

(35)

5.2 Vector Autoregression Model

After attempting a linear regression model and discovering that the data had issues of serial autocorrelation we now turn to a Vector Autoregression Model (VAR). This model is an econometric model that is used to analyze the linear interdependencies in a multivariate time series. Structurally this model defines a variable as a linear function of its own past values and other variables – it

addresses issues of serial autocorrelation. An issue we clearly need to address after the DW test results for our linear regression model’s residuals as found with our post estimation results in 5.1.

In the case of continental exports in relation to dlnRgdp: contemporary dlnRgdp values would be calculated through the past values of dlnRgdp and the past values of continental exports. All variables are considered to be symmetrical in a structural sense (albeit the response coefficients differ). Each of the selected variables has its own lags, as do the other variables plugged into the model: each contributing towards a variable’s own evolution. A major positive of using a VAR model is that it does not need as much knowledge about forces affecting

variables as structural models with simultaneous equations. The only preceding knowledge is the requirement of a hypothesis for each independent variable being related with its previous values. The mathematical equation for this model is as follows:

𝑌𝑡= 𝑐 + 𝛽1𝑌(𝑡 − 1) + 𝛽2𝑌(𝑡 − 2) + ⋯ 𝛽𝑝𝑌(𝑡 − 𝑝) + ℰ𝑡

Where Yt represents dlnRgdp, c represents a vector of constants and B1Y1

represents a matrix of trade characteristics (i.e. continental exports) in relation to dlnRgdp in year one and so on. Finally ℰ𝑡 represents the error term in the

equation. Figure 16 shows us this equation applied to continental exports in a VAR(1) model.

In the process of applying this VAR model it was necessary to establish the number of lags and leads to effectively analyze our data. Lags are applied to see if

(36)

past values of an independent variable have an effect on the contemporary dependent variable; for instance Chemical Exports a year ago may still be

influencing dlnRgdp today due to long-term ‘trickle down’ effects. Leads are the opposite of lags; this condition is applied to a variable to account for future changes in demand affecting contemporary dlnRgdp. For example future

Machinery imports demand may have an effect on contemporary Rgdp as reduced future imports can imply low confidence in the Dutch economy. Reduced future import demand reflects less construction occurring in The Netherlands;

therefore showing a curtailing and stagnating Dutch economy.

In order to locate the optimal lags needed for our data we apply the STATA command “varsoc”. This command produces a number of information criterion that help our analysis by suggesting the number of optimal lags or leads for the dependent variable in question. The information criteria include the AIC, FPE, HQIC and SBIC pre-estimation test results, as well as the progression of

likelihood ratio (LR) tests. For the lag ‘p’, the LR compares a VAR with p lags with one lag. The measurement FPE is not strictly an information criterion but it is included in our discussion because we naturally select the lag length with the lowest corresponding score. FPE furthers our lag selection creation by

minimizing the prediction error through matching other predicting information criterions. The AIC score quantifies the inconsistency between the given model and the true model – we want to minimalize inconsistencies and therefore we select the AIC lag criteria with the lowest score. The SBIC and the HQIC

information criterions can be interpreted similarly to the AIC. Arguably the SBIC and the HQIC have a theoretical advantage over the AIC and the FPE. Lutkepohl (2005) states that the SBIC and the HQIC provide consistent estimates of the true lag order. Lutkephol also argues that by minimizing the AIC or the FPE one can overestimate the true lag order with positive probability, even with an infinite sample size. The null hypothesis of this varsoc test is that all the coefficients on the pth lags of endogenous variables are nil. You can see in Table figure 13 an asterix (*) appears next to the optimal lag. Figure 13 tells us that the optimal

(37)

Figure 13 – Varsoc command illustrating the optimal number of lags to be used in a VAR model with dlnRgdp as our dependent variable

5.2.1 Post Estimation

After establishing our optimal lags via the aid of the mentioned information criterions in figure 13 combined with a structural trial an error approach (as these tests do not always guarantee the optimal recipe to create the most significant model) we must test our models. Positively we can see in figures 14 and 15 that our PACFs and ACFs show that issues of autocorrelation have now been addressed with an AR(1) function with our VAR model. For now we can see in figures 14 and 15 that the majority of the data points fall within the 95% confidence interval (grey area) and ones that fall outside these intervals are very close to the 95% confidence interval or are beyond or equal to 5 lagged years (making them insignificant for our model), so under this paper’s interpretation we will overlook these residual plots. This is what we require to prove we do not have autocorrelated data. The theories and full explanations behind these ACF and PACF plots will be explained in the 5.3 of this chapter.

Sample 1951 - 2012 Number of observations 62

Lag LL LR df p FPE AIC HQIC SBIC

0 122.13 .001176 -3.90743 -3.89396 -3.87313

1 130.76 17.25* 1 0 0.000923* -4.15053* -4.12377* -4.0825*

2 131 0.49014 1 0.484 .000943 -4.12896 -4.08855 -4.02604

3 131.04 .08167 1 0.775 .000972 -4.09802 -4.04414 -3.96079

(38)

Figure 14 – ACF displaying the residuals produced from our VAR (1) model

(39)

After investigating the relative causality of our trade characteristics through the our VAR models’ tabular results, we can see that the VAR models produced highly insignificant results and therefore the best results produced would be with a complicated model with many lags and leads, whilst still having highly insignificant results.

Interestingly, the only significant value in our VAR(1) model was that past values of dlnRgdp affect contemporary values of dlnRgdp. Past values of dlnRgdp should in theory affect future values (especially in consecutive years) due to its inherent characteristics of investing in public services and long-term projects such as education and infrastructure. Due to these insignificant results we now move onto a model that used non-linear characteristics and a dynamic approach to statistical analysis to help determine the causality of trade in reference to Dutch GDP; the ARIMA model has these required characteristics.

(40)

Figure 16: VAR(1) Continental exports and their effects on Real GDP with one lag

dlnRgdp

(t)

= ∑𝜷lnE𝑐Rgdp

(𝐭 − 𝟏)

+ 𝒅lnE𝑐𝑅𝑔𝑑𝑝 + ℰ𝑡

Number of observations 65

R-squared 0.3

Log likelihood 438.45

Independent Variables Coefficient

Rgdp 0.57***

African Exports -0.04

North American Exports 0.01

European Exports -0.01

Asian Exports -0.01

Australasian and Oceania Exports 0.01

𝜷 =

Autoregressive correlation coefficient

ln

= Natural logarithm

d =

First order difference

Rgdp

(t) = Real GDP

Rgdp

(t-1) = Real GDP from the previous year

How the composition of trade has affected economic growth in the Netherlands through a time series analysis