The importance of vertical linkages in business cycle comovements: Evidence from NAFTA.

(1)

Faculty of Economics and Business

MSc in International Economics and Business (IE&B)

Master Thesis

The importance of vertical linkages in

business cycle comovements: Evidence

from NAFTA.

Author:

Elke Oude Weernink

Student number:

S2170302

E-mail address:

e.c.m.oude.weernink@student.rug.nl

Supervisor:

T.M. Tarek Harchaoui, PHD

Co-assessor:

dr. A. (Anna) Minasyan

(2)

2

Abstract

This paper examines the importance of vertical linkages in explaining business cycle

comovements using data from the free-trading zone NAFTA as a quasi-natural experiment. Using industry-level panel dataset analysis over the 1996-2009 period, the results identify both bilateral trade and vertical trade linkages as significant explanatory vehicles in the synchronization of business cycles. Furthermore, it seems that vertical linkages have evolved as an explanatory mechanism, however the results are not clear cut as doubts are placed on the robustness of the data. The results help to identify the channels through which demand- and supply shocks breed between trading industries. Once identified, they could help an industry detect forthcoming output declines.

Keywords: business cycle comovements, NAFTA, vertical linkages.

(3)

3

1. INTRODUCTION

On the 1st of January 1994, the United States, Canada and Mexico established the North American Free Trade Agreement (NAFTA), becoming among the largest free trade area in the world. NAFTA’s tariffs reductions had considerable impacts on its member’s economies, in particular for Mexico (Caliendo & Parro, 2015). These impacts range from expansion of scale of exporting firms, exit of non-competitive firms, jobs creation/destruction, expansion of the spectrum of products available, and decline of prices. These effects are widely known and constitute the direct manifestation of an increased economic integration between the three trading partners. Whether this development enhanced the synchronization of business cycles between these economies has not been an important source of inquiry in the literature and public policy debate. Yet, it has an important implications on the propagation of shocks. When industries are heavily involved in trade, demand- and/or supply shocks in either of those industries could influence output in the other sector a great deal. Economic recession starting in one country or industry within the NAFTA could therefore startle a possible domino effect.

While the landmark contribution by Frankel and Rose (1998) has established the fundamental result that countries that trade more experience enhanced business cycle synchronization and, and their contribution has been refined recently by di Giovanni and Levchenko (2010) and Ng (2010) for a large sample of economies, little is known about NAFTA. This paper uses this free-trading zone as a quasi-natural experiment to estimate the extent to which improved trade integration influences business cycle synchronization within the NAFTA. Trade integration is measured on the basis of both bilateral trade linkages and vertical integration between

industry-pairs between Canada, the United States and Mexico are calculated, using data from the World Input-Output Database1 (WIOD). The detailed data of the WIOD on input-output linkages provides the opportunity for the realistic measure of vertical integration, where past literature had to rely on assumptions for this. With this new way of measuring vertical integration, it complements existing literature, which did not have the chance to do so.

The results show that vertical linkages are a significant explanatory mechanisms in explaining business cycle comovement, where the magnitude of vertical linkages are about three times

1_{The World Input-Output Database (WIOD) 2013 release consists of a series of databases and covers 27 EU}

countries and 13 other major countries in the world for the period from 1995 to 2011. Timmer, M. P., Dietzenbacher, E., Los, B., Stehrer, R. and de Vries, G. J. (2015),

"An Illustrated User Guide to the World Input–Output Database: the Case of Global Automotive Production",

(5)

5

the size of the importance of bilateral trade linkages. Furthermore, the results show the growing importance of vertical linkages over time. The erratic behavior of the coefficients in both split periods however, raise concerns about the robustness of the data in these cases.

The rest of the paper will be as follows. Chapter 2 describes and analyzes existing literature on business cycle comovement. Chapter 3 outlines the empirical strategy and data used for the research. Chapter 4 will present the main results and chapter 5 will conclude with the

discussion of results, limitations and suggestions for further research.

2. LITERATURE REVIEW

The purpose of this section is to review the existing literature on business cycle comovement, with the specific focus on the research that devote special attention towards the role of vertical linkages as an important explanatory variable.

2.1 Theory on business cycle comovement

Past and current literature on business cycle synchronization or so-called business cycle comovement, predominantly make their first reference to the research of Frankel and Rose (1997, 1998). Frankel and Rose argue (in both the 1997 and 1998 papers) that from a theoretical viewpoint, closer international trade could result in either tighter or looser correlation of national business cycles. International trade between a set of countries could loosen business cycles between them as they could become more specialized in the goods in which they have comparative advantage, making them more sensitive to industry-specific shocks. However, when demand shocks predominate, or there are important shocks which are common across countries, or when intra-industry trade accounts for most trade, business cycles may become more synchronized across countries when they trade more. Using a panel of bilateral trade and business cycle data for twenty industrialized countries over 30 years, empirical tests confirm their expectation that closer international trade links results in more closely correlated business cycles across countries.

(6)

6

at the best possible explanation but typically did not control for country pair factor and common global shocks that could have influenced the business cycle comovement relationship. But more importantly, the studies lack the inclusion of the role of vertical linkages, largely reflecting a paucity in the data.

2.2 The recognition of vertical specialization in business cycle models

Vertical specialization - the use of intermediate inputs in production – has and still is changing the nature of international trade (Hummels et al., 2001). There is more “back-and-forth” trade as trade is increasingly characterized by countries specializing in particular stages of a goods-producing sequence, rather than producing the entire good. As result, goods

crosses borders multiple times. Together with this changing nature, the channels through which business cycle synchronization are influenced, are changing too.

Kose and Yi (2001) introduce the idea that an increasing amount of trade is vertical or fragmented and the allowance of this “back-and-forth” trade of complementary goods where at each stage the industries add value, might lead to greater business-cycle-correlation

responsiveness to stronger trade ties. They extend a basic international-business-cycle model2 with the inclusion for vertical specialization. Household consumption consists of goods that embody both domestic and foreign value added. Therefore, increased demand for the foreign consumption good will generate additional imports of both the foreign goods and domestic intermediate value added which is needed to produce the foreign consumption good. However, both their basic and their extended model with the inclusion of vertical

specialization fall short of quantitatively replicating the empirical finding that higher trade intensities induce higher business cycle correlations.

Examining the mechanisms underlying the relationship between trade and business cycle comovement, di Giovanni and Levchenko (2010) recognize the importance of vertical linkages next to bilateral trade ties. Using sector-level data, they have the chance to investigate whether vertical production linkages across industries help to explain business cycle comovement between those pairs. In order to measure the extent of vertical linkages, the authors use input-output matrices to capture the intensities of which individual industries use each other as intermediate inputs for their production output. As they do not possess the exact intensity ratios for each sector-pair across the country-pairs, they multiply the (known)

2_{Kose and Yi (2001) focus on the model by David Backus et al. (1994), which is the workhorse two-good}

(7)

7

bilateral trade data with fixed ratios for the use of intermediate inputs coming from an U.S. 1997 benchmark I-O table. With this they focus in their research on a particular identifiable channel: the use of intermediate inputs in production. Their main results show that the Frankel-Rose effect is evidently present at the sector level: sector-pairs that experience more bilateral trade exhibit stronger comovement. Furthermore, their empirical evidence shows that a given increase in bilateral trade lead to higher comovement in sector pairs that use each other heavily as intermediate inputs.

Although their research provides an worthy insight in the role of vertical linkages in business cycle comovement, it has some essential weaknesses. Using data for the period 1970-1999 to measure the intensity of both bilateral trade linkages and vertical integration, the data does not account for the second phase of globalization. This second phase is characterized by the rise of international production fragmentation, also known as global value chains. These chains, which are various tasks in production are driven mainly by declining communication and information costs fed by the rise of the internet. Therefore, the time period where vertical integration left its most important mark, is unfortunately not included in the sample of the research.

Furthermore, the sample of di Giovanni and Levchenko (2010) includes data for 55 developed and developing countries but did not incorporate China into the sample. Since opening up to free-trade, China has become among the world’s fastest-growing economies but moreover, led China to become one of the largest economies in the world. Holding such a global economic influence, the exclusion of China in the sample causes some concern on external validity of the results of the research. Lastly, the intensity-ratios with which individual sectors use each other as intermediate inputs in production are based on an single year benchmark I-O table and multiplied with bilateral trade data between the pairs. With this approach, the authors assume the industrial structures of both the developed and the developing countries in the sample being similar to the one of the United States which is something worth placing doubt on.

(8)

8

examines the effect of bilateral production fragmentation on synchronization of GDP growth of country-pairs.

For the construction of the bilateral production fragmentation variable, OECD Input-Output tables are used, which contain the aggregate imported input coefficient matrix for each

country. To construct the bilateral imported input coefficient matrix, the author uses an import proportionality assumption based on the bilateral import share of each sector in total imports. So essentially, the authors assume that when 30% of U.S. total imports of automobile parts come from Germany, for any sectors in the U.S. that use automobile parts as intermediate inputs for their, 30% of the imported inputs come from Germany as well.

Essential in the research is the theoretical discussion on linkages between trade and business-cycle comovement where the author digs deeper into the role of trade in substitutes and complements. The substitutability of goods or complementarity of goods being traded can potentially have opposite effects on business-cycle synchronization. As countries specialize in producing the good where they have a comparative advantage, they involve in trade in

substitutes with each other. As these products are varieties, a demand or supply shock for a particular good only applies to the one country producing it, resulting in a less correlated business cycle. When countries are heavily involved in the trade of goods that are

complementary to each other, resulting in a “back-and-forth” trade in intermediates, it can potentially generate positive demand- and supply side spillovers, resulting in more correlated business cycles. Since trade in substitutes and trade in complements can thus have opposite effects, the overall effect of trade on business cycle synchronization depends on the one being most dominant force.

The empirical results of Ng (2010) suggest that, once vertical linkages are taken into account, the findings from past literature on bilateral trade intensity and intra-industry trade being positively related to business cycle comovement are not robust. Furthermore, the results suggest the positive effects of vertical trade linkages (trade in complements) dominates the negative effect of bilateral trade linkages (characterized by trade in substitutes). Therefore, its conclusion is that bilateral vertical trade linkages between countries play a key role in the business cycle synchronization between them.

(9)

two-9

country Dynamic General Equilibrium model which allows for the inclusion of trade in differentiated capital goods and the technology included in these goods. From the evidence, it is argued that global synchronization of business cycles become somewhat less important while it appears that regional synchronization of business cycles appears to be increasing. These regional clusters are for example North America, Western Europe and emerging Asia but are certainly not limited to geographical proximity. Groups of countries which are at the same stage of development or have close ties regarding trade and/or capital flows could be viewed as regional clusters as well.

The paper argues that the structure of trade rather than the volume of trade within a group of trading partners is key in determining the scope of their business cycle synchronization. The research models the fragmentation of production as an exchange of capital goods among the countries concerned. In examining business cycle synchronization, the authors uses data on seasonally adjusted real GDP growth pairwise correlations between the countries in the sample to measure comovement. The conclusion of the paper is that the increase in business cycle synchronization in the sample can be mainly attributed to the growing fragmentation of production linkages.

2.3 Lessons learned from past literature and the opportunities for improvement

(10)

10

Moreover, research of Kose and Yi (2001) and Takeuchi (2011) have built models that simulate output structures between country-pairs instead of using real data on input-output structures. The introduction of the World Input-Output Database (WIOD) addresses these issues with the contribution of time-series of national input-output tables combined into a cross-country input-output matrix on intermediate usage on sector level. The availability of this data gives the opportunity to explore the channels through which trade influences

business cycle synchronization within the NAFTA, and more importantly investigate the role of vertical linkages.

2.4 The research questions

To give structure to the paper, a series of research questions are formulated. The leading goal of the paper is to determine the specific mechanisms of trade driving business cycle

comovement within the NAFTA which brings us to the first main research question, which is formulated more broadly:

1. Which types of trade were driving business cycle comovements within the NAFTA between 1996-2009?

To arrive at a comprehensive answer for this main research questions, a set of sub-questions are formulated. These set of questions focus on how to measure and evaluate business cycle comovements on the sector-level:

2. How are the cross-country bilateral trade linkages on the sector-level within the NAFTA measured, and how are they related to business cycle comovements?

3. How are the cross-country vertical linkages on the sector-level within the NAFTA measured, and how are they related to business cycle comovements?

Lastly, two questions are formulated based on the findings of past literature on the (growing) importance of vertical linkages in explaining business cycle synchronization:

4. Were vertical linkages a stronger mechanism than bilateral trade ties in explaining business cycle synchronization within the NAFTA in the period 1996-2009?

(11)

11

The next chapter will outline the modelling strategy and source data in which the answers to the sub-questions two and three are extensively discussed.

3. MODELLING STRATEGY AND SOURCE DATA

In this section, the modelling strategy, the source data and measurement of the variables will be outlined. Furthermore, it contains the descriptive analyses of the data on trade, vertical linkages and the business cycle comovement within the NAFTA.

3.1 Modelling strategy

We follow di Giovanni and Levchenko (2010) by modeling in the following ways the articulation between trade and comovements take form. Business cycle comovement is estimated using the impact of sector-level trade on the correlation between individual sectors in the country-pair. The following specification is estimated, using comovement and trade data for each sector pair:

(1) 𝜌

_𝑖𝑗𝑐𝑑

_{= ∝ + 𝛽}

1

𝑇𝑟𝑎𝑑𝑒

𝑖𝑗 𝑐𝑑

+ 𝒖 + 𝜖

𝑖𝑗𝑐𝑑

𝑇𝑟𝑎𝑑𝑒

_𝑖𝑗𝑐𝑑 embodies one of the four possible trade intensity measures (more on that below). Using sector-level data within the sample of the three countries gives me two advantages over the use of cross-country data alone: first, it allows for the inclusion for three sets of fixed effects since the dataset is four dimensional3. Secondly, using the sector-level of the WIOD allows for a better control of the effects of aggregate common shocks between countries that influence the interpretation of results of cross-country data

The specification includes four configurations of fixed effects u. The trade measures are recorded at the exporter x sector x importer x sector level allowing for the four configurations.

First, the configuration includes the control for importer, exporter and sector effects. Country effects capture the average effects of country characteristics on comovement across trading partners and sectors. These include macro policies, country-level aggregate volatility, country size and population and the level of income. Sector effects capture any possible characteristics

3_{The dataset is indexed by country 1 (exporter) and country 2 (importer), sector 1 (sector using the input for}

(12)

12

of sectors for example capital, R&D intensity, liquidity needs, tradability, overall volatility, reliance on external finance.

The second configuration of fixed effects is with exporter x sector and importer x sector effects. This configuration controls for the average comovement properties of each sector within each country across trading partners, for instance tariffs and nontariff barriers. As we are dealing with a sample with three countries operating within a free trade area, attention in the evaluation of the results will go towards this set of fixed effects.

The third and last configuration of fixed effects are the country-pair and sector –pair effects. The use of country-pair effects erases impacts of common shocks that can occur at the country-pair level, such as similarity in industrial structure, aggregate demand and currency unions or other types of monetary policy coordination. The inclusion of sector-pair effects take in the average synchronization for a particular pair of sectors in the sample.

To observe intra-industry trade and isolate the impact, a variant of equation (2) is estimated allowing the coefficient on 𝑇𝑟𝑎𝑑𝑒_𝑖𝑗𝑐𝑑_{to be different when it i = j. Specification (2) resembles}

this with the inclusion of the indicator function

𝟏[·].

The indicator function takes the value of 1 for observations when i = j and zero otherwise, functioning as a dummy variable,

identifying intra-industry trade when present.

(2) 𝜌

_𝑖𝑗𝑐𝑑

_{= ∝ + 𝛽}

1

𝑇𝑟𝑎𝑑𝑒

𝑖𝑗 𝑐𝑑

+ 𝛽

2

𝟏[𝑖 = 𝑗]𝑇𝑟𝑎𝑑𝑒

𝑖𝑗 𝑐𝑑

+ 𝒖 + 𝜖

𝑖𝑗𝑐𝑑

The investigation on the transmission of shocks takes a step further with the inclusion of vertical production linkages. As trade in intermediates has grown to be a main actor in today’s world trade, it mostly likely alters the nature of transmission of shocks and therefore, has to be taken into account.

Business cycle comovement is estimated with variable 𝐼𝑂_𝑖𝑗𝑐𝑑 which captures vertical

production linkages between each possible sector-pair and country-pair. It captures the value of intermediate inputs from sector i required to produce $1 of final output of good j plus the value of intermediate inputs from sector j required to produce $1 of finale output of good i. The assumption is made that dependency works both ways as vertical linkages are

characterized by “back-and-forth” trade (Kose and Yi, 2001).

(13)

13

(3) 𝜌

_𝑖𝑗𝑐𝑑

_{= ∝ + 𝛽}

1

𝑇𝑟𝑎𝑑𝑒

𝑖𝑗 𝑐𝑑

+ 𝛾

1

(

𝐼𝑂𝑖𝑗𝑐𝑑

) + 𝒖 + 𝜖

𝑖𝑗𝑐𝑑

Specification 4 is the full specification with focus on intra-industry for both bilateral trade flows and input-output linkages between the sector-pairs.

(4) 𝜌

_𝑖𝑗𝑐𝑑

_{= ∝ +𝛽}

1

𝑇𝑟𝑎𝑑𝑒

𝑖𝑗 𝑐𝑑

+ 𝛽

2

𝟏[𝑖 = 𝑗]𝑇𝑟𝑎𝑑𝑒

𝑖𝑗 𝑐𝑑

+ 𝛾

1

(

𝐼𝑂𝑖𝑗𝑐𝑑

) +

𝛾

₂

𝟏[𝑖 = 𝑗]

𝐼𝑂_𝑖𝑗𝑐𝑑

+ 𝒖 + 𝜖

_𝑖𝑗𝑐𝑑

3.2 Source data and measurement

As discussed in the literature review, past research on business cycle comovement use several types of approaches to measure vertical linkages between industry-pairs and/or country-pairs. However, none of them use data which contains exact numbers on input-output linkages between industries. Here, the significant contribution of the WIOD comes in. The tables of the WIOD, containing realistic input-output structures, are constructed in a clear conceptual framework. They are built on the basis of officially published input-output tables in

collaboration with national accounts and international statistics. The WIOD contains data on both sectoral output growth for the industries of the countries in the sample as well as data on vertical linkages between those industries. Therefore, data sectoral output growth for the construction of both the dependent and explanatory variables come from this database. Data for the 30 sectors used for analysis are classified according to the International Standard Industrial Classification revision 3 (ISIC Rev. 3). Although the tables of the WIOD contain data on 35 sectors, this research excludes the 5 industries of non-market services4 as they are non-trade intensive. The tables are adhere to the 1993 version of the SNA.

The remaining 30 sectors of the WIOD used in the sample can be divided up into 3 broad types of sectors: other goods producing sectors and market services. Appendix A contains information on the specific industries per type of sector.

Business cycle comovement is based on sector-level data and measured through pairwise (Pearson’s) correlations where

∆𝑌

_𝑖𝑐 is real output growth based on previous year prices of sector i in country c.

4_{Non-market services of the WIOD are: (1) Public Admin and Defence; Compulsory Social Security, (2)}

(14)

14

𝜌

_𝑖𝑗𝑐𝑑

= 𝑝𝑤𝑐𝑜𝑟𝑟 (∆𝑌

_𝑖𝑐

, ∆𝑌

_𝑗𝑑

)

The research includes four measures of bilateral trade intensity as introduced by di Giovanni and Levchenko. The trade measures differ from each other in the denominator as the

nominator stays the same in all four measures where

𝑋

_𝑖,𝑡𝑐𝑑 represents the value of exports in sector i from country c to country d. Subsequently, 𝑋_𝑗,𝑡𝑑𝑐 represents the value of export in sector j from country d to country c. Having 30 sectors means, bilateral trade volumes for 30 times 30 sectors can be computed.

Trade volumes for each sector-pair are divided differently for each trade measure. In the first two measures, bilateral sectoral trade is measured with output either at the aggregate or sector level. 𝑌_𝑡𝑐_{is the GDP}5_{of country c in year t and 𝑌}

𝑖,𝑡𝑐 is the output of sector

i in country c in year t.

𝑇𝑟𝑎𝑑𝑒

_𝑖𝑗𝑐𝑑

= log (

1 𝑇

∑

𝑋_𝑖,𝑡𝑐𝑑+ 𝑋_𝑗,𝑡𝑑𝑐 (_1000000)1 )(𝑌_𝑡𝑐+ 𝑌_𝑡𝑑) 𝑡

)

(Measure I)

𝑇𝑟𝑎𝑑𝑒

_𝑖𝑗𝑐𝑑

= log (

_𝑇1

∑

𝑋𝑖,𝑡𝑐𝑑+ 𝑋𝑗,𝑡𝑑𝑐 𝑌_𝑖,𝑡𝑐+ 𝑌_𝑖,𝑡𝑑 𝑡

)

(Measure II)

The third and fourth measure of trade normalize bilateral sector-level trade volumes by the overall trade in the two countries:

𝑇𝑟𝑎𝑑𝑒

_𝑖𝑗𝑐𝑑

= log (

_𝑇1

∑

𝑋𝑖,𝑡𝑐𝑑+ 𝑋𝑗,𝑡𝑑𝑐 (𝑋_𝑡𝑐+ 𝑀_𝑡𝑐)+(𝑋_𝑡𝑑+ 𝑀_𝑡𝑑) 𝑡

)

(Measure III)

𝑇𝑟𝑎𝑑𝑒

_𝑖𝑗𝑐𝑑

= log (

_𝑇1

∑

𝑋𝑖,𝑡𝑐𝑑+ 𝑋𝑗,𝑡𝑑𝑐 (𝑋_𝑖,𝑡𝑐 + 𝑀_𝑖,𝑡𝑐 )+(𝑋_𝑖,𝑡𝑑+ 𝑀_𝑖,𝑡𝑑) 𝑡

)

(Measure IV)

Where 𝑋_𝑖,𝑡𝑐 (𝑀_𝑖,𝑡𝑐 ) is the total exports (imports) of sector i of country c and 𝑋_𝑡𝑐_{is the total}

industry exports of country C.

The four trade measures provide an overview of trade intensities between sector-pairs, however, for the analysis of results later in the paper, attention will go towards the third measure of trade. In this measure of trade, bilateral trade linkages between industries are divided by the total industry exports of a country. This measure, normalizes trade volume and

5_{The data in the WIOD is in millions of US dollars, therefore the GDP of the country is divided by one million to}

(15)

15

furthermore, gives an indications of the ratio of an industry’s trade over the total trade in the country. This gives an indication of the importance of an industry when accounting for trade intensity within the country.

In all estimations, natural logs are used in estimation because the trade ratios in levels are extremely skewed, which would result to a tiny share of the top values of the trade ratios affecting the estimated coefficient heavily.

To dig deeper into the transmission of shocks through trade, vertical production linkages are impossible to ignore in today’s world trade. Fed by decreasing coordination and transports costs, production processes increasingly fragment across borders (Timmer et al. 2015). Through these vertical production channels, demand or supply shocks to a sector in one country increases or decreases that sector’s demand for intermediate goods, which in turns stimulates output of intermediates in the partner country.

The availability of the realistic input-output structure of the WIOD allows me to construct a different approach for the measurement of intermediate input usage between the industries in the country pairs in the sample. Where 𝐼𝑂_𝑖,𝑡𝑑𝑐_{is exports of intermediates of sector j from}

country d to sector i in country c. Dividing this value by the output of sector i in country c, results in the ratio representing the extent to which sector i in country c uses sector j of country d as an intermediate input used to produce $1 of final good i.

The second parts, 𝐼𝑂_𝑖𝑗𝑐𝑑_{represents the exports of intermediates of sector i from country c to}

sector j in country d divided by the value of the output of sector j in country d. This gives the ratio of the extent to which sector j in country d uses sector i of country c as an intermediate input used to produce $1 of final good i.

𝐼𝑂

_𝑖𝑗𝑐𝑑

= 𝑙𝑜𝑔 (

_𝑇1

∑

𝐼𝑂_𝑦𝑗𝑖,𝑡𝑑𝑐

𝑖,𝑡𝑐

+

𝐼𝑂_{𝑗𝑖,𝑡}𝑐𝑑 𝑦_𝑗,𝑡𝑑

𝑡

)

(Vertical specialization measure)

When two industries in different countries trade intensively in intermediates, they depend on each other influencing their business cycle comovement. When industry i in country c is a supplier in intermediate goods to industry j in country d, its output and thus output growth is dependent on the demand of industry j in country d.

(16)

business-16

cycle-correlation responsiveness to stronger trade ties

If industry j in country d demands intermediate inputs for its output from sector i in country c, changes in the supply quantities or price-levels from sector i in country c can influence output in industry j in country d. The two sectors in the country-pair are interdependent, influencing each other from both sides.

Next section will contain the descriptive analysis on the variables introduced. Furthermore, it will provide an overview of the extent of vertical integration within the NAFTA.

3.3 Descriptive analyses

The research includes three datasets, one for the full period of analysis (1996-2009), one for the first and one for the second split periods (1996-2001 and 2002-2009 respectively).

The 3 datasets of the research include 5400 observations of business cycle comovement of industry output growth. The dependent variable is measured through Pearson’s correlation coefficient. Because these coefficients are bounded at 1 and -1, the error terms in a

regressions model of the determinants of business cycle synchronization are unlikely to be normally distributed which raise concerns about reliable inferences.

To check for normality in the dependent variables, the variables are graphed in histograms (figure 1 and 2).

(17)

17

From the graph representing the dependent variable in the full period, normality can be assumed and therefore interpretation and inference of the results are therefore reliable.

From the graphs of both split periods however, the dependent variables are somewhat non-normal distributed. Therefore, in the regressions for the two split period, I use Fisher’s z-transformation of the correlation coefficients as dependent variable. Used by Inklaar et al (2008), they introduce this transformation for the dependent variable to ensure the dependent variable does not suffer from non-normality and the non-normal distribution of errors. The transformed correlation coefficient are calculated as 𝜌_𝑖𝑗𝑐𝑑transformed = 1/2ln((1+ 𝜌_𝑖𝑗𝑐𝑑)/(1-𝜌_𝑖𝑗𝑐𝑑)).

After transformation, the histograms of the dependent variables follow a normal distribution. 6

3.3.1 Graphing trade measures, vertical integration and business cycle comovement

Next, to observe the relationship between the four trade measures of trade, vertical integration and the correlation of industry output growth, scatterplots are shown in figure 3. The graphs are for the full period.

The scatterplot indicates a positive relationship between the four measures of trade and the dependent variable in the form of industry output growth correlation. Next, a scatterplot is shown to observe the relationship between the vertical integration and correlation of real output growth between industries (figure 4). The scatterplot points towards a strong positive association between the degree of vertical integration and comovement. More importantly, both figures gives a first indication relative importance of the types of trade on comovement where vertical integration is clearly stronger positively related than bilateral trade.

6_{Histograms of both the transformed variables for both split periods and the untransformed variable split}

(18)

18

Figure 3. Correlations of real industry output growth versus trade ratios

Note: the y-axis variable for all four figures is the correlation of real industry output growth between industry-pairs. The x-axis is based on a log scale and the variables are the four different kind of trade measures. Panel A: industry bilateral trade/GDP; panel B: industry bilateral trade/industry total output; Panel C: industry bilateral trade/total trade; panel D: industry bilateral trade/total trade within an industry.

Figure 4. Correlation of real industry output growth and vertical linkage ratios

(19)

19

To observe, whether the influence of vertical integration differs for the two split periods, scatterplots for both periods are depicted in figure 5 and 6. From the figures, first evidence of the rising importance of the explanation of vertical linkages in business cycle comovement is shown.

To complement figures 5 and 6 in the exploration of vertical linkages in the NAFTA, contour plots offer great insights.

Contour plots graph the intensity of vertical linkages between industries on both axes. It tells how much of the industries on the x-axis, use input from the industries on the y-axis for its output, as a ratio. It is calculated by dividing total intermediate usage of industries by the total output of those industries and can be built for both domestic input-output linkages and cross-country industry linkages. The plots used in this report are built from the data from the WIOD, including all 35 sectors in an economy.

The interpretation from the plots are as follows: the numbers of both axes indicate which type of industry we are dealing with. The industries on the x-axis are the industry which use intermediate inputs for their output from the industries on the y-axis. The color of the area indicates the ratio of which the output for the given industry on the x-axis consists of inputs from the industry on the y-axis. When an industry shares an area of a certain color with a corresponding industry on the y-axis (draw a line up from the industry, up to the colored area and take the line to the left. You will arrive at the corresponding industry), that particular industry on the x-axis is using inputs up to a ratio of a certain extent for its output from that

Figure 5. Correlation of vertical trade linkages and real industry output

growth. First split period (1996-2001)

Figure 6. Correlation of vertical trade linkages and real industry output

(20)

20

industry on the y-axis. The color bar on the right, indicates the ratio intensities. To enhance the understanding of the contour plots, an numerical example is given: Suppose, that total output of the 15th_{industry (Transport Equipment) in Canada was 10}

million of US$. This Canadian industry, used intermediate inputs from the industry Rubber and Plastics (industry 10) in the United States for its output. This input usage amounted to 1 million of US$. The ratio of which the 15th Canadian industry used the 10th U.S. industry as an intermediate inputs for its output is therefore 100/10 = 0.1

When viewing the plots, a reader will probably notice the horizontal trend in the figures. This horizontal nature of the figures represent the intermediate trade between the same type of industries, and are an indicating of the vertical nature of trade.

Below, three figures are presented containing the visualization of cross-country intermediate input use by industries in the NAFTA.

(21)

21

Figure 8. The use of inputs by Mexican industries for both 1995 and 2011. Mexican industries using Canadian industry-input (above) and Mexican industries using U.S. industry-input (below)

(22)

22

The plots confirm the existence of vertical linkages between the countries in the sample, but at the same time, indicate that the intensities across the country-pairs are not of the same extent. This inequality of vertical intensity are important when interpreting the results of the variable vertical linkages. When interpreting the results for the influence of vertical linkages on business cycle comovement within the NAFTA, one must keep in mind that this influence can be different for the different country-pairs.

Predominantly, the contour plots, surprisingly, do not show an increase of intermediate input use within the NAFTA as expected. The use of Canadian intermediates by U.S. industries even declined between the year 1995 and 2011. Although these contour plots are snapshots of the behavior of the industries between the country-pairs for two years, it does say something about the development between those country-pairs regarding vertical linkages.

The country-pairs which did experience tighter links regarding intermediate usage is Mexico-United States. In special, Mexican industries used inputs from the U.S. to a greater extent than before and moreover, the range of industries which used inputs from the U.S. grew.

Next to the fact that industries import intermediates, they also have the possibility to search for these intermediates domestically. Buying domestic intermediates inputs can have advantages over foreign intermediate inputs like for example lower transport costs, no language barriers, connecting production processes and possible subsidies from governments to stimulate domestic production.

To observe the behavior of the industries regarding the use of domestic intermediate input usage, contour plots are presented in figures 10, 11 and 12.

(23)

23

The plots containing the Canadian intermediate input usage by Canadian industries reveal a surprising trend: more industries in Canada used Canadian intermediates for their output, furthermore, it seems that the intensity these use ratios also increased. Moreover, the plots for the U.S. domestic input usage indicate the same trend, but with even a more drastic extent. The plots for Mexico stayed more or less the same, indicating that industries within the country did not use domestic inputs to a great or lesser extent for their output.

Where literature repeatedly have stressed the rise of global value chains and increased vertical specialization in the world, the contour plots for domestic industry intermediate input usage in Canada and the United States offer some contractions.

The scatterplots of figure 5 and 6 have provided some first evidence of the growing importance of vertical trade integration on business cycle comovement.

On the other side, the contour plots, indicate that with the regard to vertical trade ties in the

Figure 11. Domestic intermediate input usage in Mexico. 1995 and 2011.

(24)

24

NAFTA, one could speak of marginal gains for vertical integration. Even more, the

strengthening of domestic vertical linkages could indicate slow vertical disintegration within the NAFTA. Although these figures gives some valuable information on the trends on vertical integration, they cannot provide scientific answers on the questions raised in the paper,

therefore we turn our heads towards the quantitative side of the research.

(25)

25

4. RESULTS

This chapter presents the results for estimations of all three datasets. First, results are

presented for the total time period (1995-2011) before making a distinction between the two time split periods (1996-2001 and 2002-2009).

4.1 Business cycle comovement in the NAFTA between 1996 and 2009.

Table 1 presents the results of estimating equation (1). There are four columns, one for each trade measure as described in chapter 3. Column (1) estimates the impacts of trade on comovement without any fixed effects on a pooled OLS regression, with column (2) adding country and sector effects. Column (3) includes country times sector effects and column (4) and final model is estimated using country-pair and sector-pair effects.

Since we are dealing with a linear-log model, some consideration has to be taken into account when interpreting the coefficients. A one percentage change in the independent variables represent an unit change in the dependent variable. Here, this means that an one percent increase in the bilateral trade ratio for example, increases the ratio of which two trading industries are correlated, with the coefficient of the bilateral trade variable.

Panel A. Trade/GDP Panel B. Trade/Output

(1) (2) (3) (4) (1) (2) (3) (4)

Trade 0.0133*** 0.00585*** 0.00840*** 0.0101*** Trade 0.0102*** 0.00185 0.00559*** 0.00310

(0.00148) (0.00226) (0.00173) (0.00266) (0.00138) (0.00206) (0.00163) (0.00217)

Observations 5,392 5,392 5,392 5,392 Observations 5,392 5,392 5,392 5,392

R-squared 0.015 0.243 0.318 0.350 R-squared 0.011 0.242 0.317 0.348

Panel C. Trade/Total trade Panel D. Trade/Sectot total trade

Trade 0.0157*** 0.00593*** 0.0101*** 0.0102*** Trade 0.0128*** 0.00311 0.00845*** 0.00454* (0.00135) (0.00226) (0.00157) (0.00267) (0.00147) (0.00232) (0.00167) (0.00234) Observations 5,392 5,392 5,392 5,392 Observations 5,392 5,392 5,392 5,392 R-squared 0.024 0.243 0.321 0.350 R-squared 0.014 0.242 0.319 0.349 N Y N N N Y N N N N Y N N N Y N N N N Y N N N Y 𝑐1 + 𝑐 + 𝑖 + 𝑗 𝑐1 x 𝑖 + 𝑐 𝑗 𝑐1 x 𝑐 + 𝑖 𝑗 𝑐1 + 𝑐 + 𝑖 + 𝑗 𝑐1 x 𝑖 + 𝑐 𝑗 𝑐1 x 𝑐 + 𝑖 𝑗 Table 1 Impact of trade on comovement at the sector level: pooled estimates.

Note: Robust standard errors in parentheses. The sample period is 1996-2009. The dependent variable is the correlation of real output growth between sector i and j of the country pair. µc1 and µc2 are country 1 and country 2 fixed effects, respectively. µi and µj are sector i and j fixed effects, respectively.

(26)

26

As explained in chapter 3, our attention goes towards the results in column 3. This is done not only because these are identified to be the most finest way to evaluate the impact of trade on business cycle comovement but also because the bulk of results are just too extensive to evaluate in its complete.

Table 2. Impact of trade on sector-level comovement: within- and cross-sector estimates.

The coefficients for the variable “same sector” in column 4 are omitted because this set of fi ed effects multiplies the sector dummies, whereas the variable only has effect when sector i = j and the fixed effect would again control for this. The coefficients for this combination turned out to be very unusal.

*** significant at the 1% level ** significant at the 5% level * significant at the 1% level

(1) (2) (3) (4) (1) (2) (3) (4)

Trade 0.0121*** 0.00659*** 0.00810*** 0.00987*** Trade 0.00926*** 0.00240 0.00538*** 0.00276 (0.00152) (0.00227) (0.00176) (0.00272) (0.00141) (0.00208) (0.00165) (0.00219) Trade x same sector 0.0347*** 0.0311*** 0.0311*** 0.00459 Trade x same sector 0.0318*** 0.0280*** 0.0286*** 0.0115*

(0.00585) (0.00485) (0.00506) (0.00764) (0.00547) (0.00445) (0.00475) (0.00694) same sector 0.490*** 0.448*** 0.452*** - same sector 0.342*** 0.310*** 0.320***

-(0.0664) (0.0555) (0.0597) - (0.0434) (0.0378) (0.0407) -Observations 5,392 5,392 5,392 5,392 Observations 5,392 5,392 5,392 5,392 R-squared 0.025 0.252 0.327 0.350 R-squared 0.020 0.250 0.325 0.349

Trade 0.0147*** 0.00668*** 0.00972*** 0.00994*** Trade 0.0121*** 0.00397* 0.00840*** 0.00433* (0.00139) (0.00228) (0.00160) (0.00271) (0.00150) (0.00233) (0.00170) (0.00236) Trade x same sector 0.0313*** 0.0288*** 0.0288*** 0.00604 Trade x same sector 0.0302*** 0.0260*** 0.0264*** 0.00827

(0.00538) (0.00454) (0.00475) (0.00654) (0.00655) (0.00526) (0.00560) (0.00680) same sector 0.412*** 0.384*** 0.388*** - same sector 0.280*** 0.252*** 0.260***

(27)

27

From both table 1 and 2, it is evident that there is a positive relationship between the strength of bilateral sectoral trade links and the corresponding sector-level comovement.

Table 2 provides an additional insight in the effect of intra-industry trade on business cycle synchronization. The table includes equation (2), in which the coefficient on the trade variable is allowed to be different for observations with i = j. The structure in the table is the same as in table 2, with the same order of the configurations of fixed effects. From the table, it becomes clear that the within-sector transmission of shocks through trade play an important role in explaning business cycle comovement.

Table 3. Impact of trade on sector-level comovement: vertical linkage estimates

The coefficients for the variable “Input-Output” in column 4 are omitted because this set of fi ed effects multiplies the sector dummies yielding very unusual results.

Trade -0.0264*** -0.0289*** -0.0262*** -0.0393*** Trade -0.0251*** -0.0254*** -0.0222*** -0.0271*** (0.00411) (0.00640) (0.00445) (0.00708) (0.00360) (0.00480) (0.00391) (0.00468) Trade X IO -0.000958*** -0.00144*** -0.000827*** -0.00232*** Trade X IO -0.000900*** -0.00116*** -0.000539** -0.00164*** (0.000273) (0.000339) (0.000276) (0.000370) (0.000253) (0.000273) (0.000255) (0.000287) Input-Output 0.0200*** 0.00173 0.0174*** - Input-Output 0.0238*** 0.00920*** 0.0223*** -(0.00316) (0.00401) (0.00311) - (0.00221) (0.00265) (0.00220) -Observations 5,392 5,392 5,392 5,392 Observations 5,392 5,392 5,392 5,392 R-squared 0.099 0.263 0.364 0.361 R-squared 0.102 0.264 0.365 0.361

(28)

28

The results for table 3 estimates equation (3) which includes the role of vertical linkages. What notable about the results, are the negative signs for the coefficient of both bilateral trade and the interaction effect of bilateral trade and vertical linkages when the effect of vertical linkages enters into the model. This could point towards a confirmation of the theory which assumes that the goods embodied bilateral trade are substitutions to each other, resulting in lesser correlated business cycles when supply- and demand industry shocks are flowing through the economies. However, to really observe whether bilateral trade is influencing business cycle correlation negatively, we need to take the full model into consideration. Therefore, we turn our attention to table 4.

The results for the estimation of equation (4) are presented in table 4. With the focus on Panel C, column 3, we observe significance for all variables at at least the 10% level. While bilateral trade solely has a negative effect on business cycle comovement, the interaction variables trade times same sector and trade x same sector x IO yield positive, significant effects on business cycle comovement. Therefore, it is not safe to say that bilateral trade in total has a negative effect on business cycle synchronization.

The effect of vertical linkages has remained positive and significant in all specifications and configurations of fixed effects. There is one interaction variable however (Trade x IO) that alters the effect of vertical linkages in a negative way in the model.

(29)

29

(1) (2) (3) (4) (1) (2) (3) (4)

Trade -0.0280*** -0.0181*** -0.0274*** -0.0407*** Trade -0.0258*** -0.0162*** -0.0231*** -0.0286***

(0.00432) (0.00580) (0.00463) (0.00722) (0.00374) (0.00451) (0.00402) (0.00479)

Trade x same sector 0.0361*** 0.0344*** 0.0345*** 0.0245 Trade x same sector 0.0199 0.0233* 0.0255* 0.0349*

(0.0137) (0.0114) (0.0119) (0.0292) (0.0148) (0.0131) (0.0132) (0.0187)

Trade x IO -0.00107*** -0.00101*** -0.000997*** -0.00243*** Trade x IO -0.000968*** -0.000828*** -0.000693*** -0.00174***

(0.000289) (0.000325) (0.000290) (0.000380) (0.000266) (0.000275) (0.000266) (0.000296)

Trade x same 0.000139 0.00151** 0.00166** 0.00285** Trade x same 0.000172 0.00131* 0.00173** 0.00219**

sector x IO (0.000932) (0.000762) (0.000750) (0.00130) sector x IO (0.000856) (0.000748) (0.000721) (0.00103)

Same sector x IO -0.00861 0.0210* 0.0224** - Same sector x IO -0.000657 0.0183** 0.0221***

-(0.0156) (0.0119) (0.0111) - (0.0112) (0.00922) (0.00843)

-Same sector 0.351*** 0.455*** 0.440*** - Same sector 0.170** 0.284*** 0.290***

-(0.125) (0.105) (0.114) - (0.0802) (0.0734) (0.0768)

-Input-Output 0.0182*** 0.0111*** 0.0141*** - Input-Output 0.0225*** 0.0163*** 0.0196***

-(0.00332) (0.00364) (0.00326) - (0.00231) (0.00246) (0.00231)

-Observations 5,392 5,392 5,392 5,392 Observations 5,392 5,392 5,392 5,392

(1) (2) (3) (4) (1) (2) (3) (4)

Trade -0.0201*** -0.00340 -0.0204*** -0.0357*** Trade -0.0198*** -0.00599 -0.0182*** -0.0265***

(0.00403) (0.00495) (0.00415) (0.00666) (0.00416) (0.00507) (0.00432) (0.00530)

Trade x same sector 0.0337** 0.0325*** 0.0307*** 0.0291 Trade x same sector -0.00430 0.00560 0.00613 0.0163

(0.0131) (0.0108) (0.0110) (0.0219) (0.0169) (0.0141) (0.0142) (0.0208)

Trade x IO -0.000779*** -0.000447 -0.000670** -0.00212*** Trade x IO -0.000765*** -0.000516* -0.000613** -0.00171***

(0.000267) (0.000284) (0.000266) (0.000341) (0.000290) (0.000304) (0.000289) (0.000324)

Trade x same -3.81e-05 0.00140* 0.00150** 0.00261** Trade x same -0.000653 0.000925 0.00122 0.00180

sector x IO (0.000884) (0.000730) (0.000712) (0.00107) sector x IO (0.00102) (0.000875) (0.000844) (0.00115)

Same sector x IO -0.0120 0.0181* 0.0192* - Same sector x IO 0.00163 0.0208*** 0.0225***

-(0.0144) (0.0110) (0.0102) - (0.00985) (0.00797) (0.00725)

-Same sector 0.275*** 0.399*** 0.364*** - Same sector 0.0739 0.215*** 0.204***

-(0.102) (0.0862) (0.0932) - (0.0749) (0.0671) (0.0701) -Input-Output 0.0209*** 0.0154*** 0.0181*** - Input-Output 0.0230*** 0.0175*** 0.0197*** -(0.00284) (0.00299) (0.00282) - (0.00199) (0.00215) (0.00202) -Observations 5,392 5,392 5,392 5,392 Observations 5,392 5,392 5,392 5,392 R-squared 0.097 0.246 0.365 0.361 R-squared 0.097 0.245 0.364 0.360 N Y N N N Y N N N N Y N N N Y N N N N Y N N N Y 𝑐1 + 𝑐 + 𝑖 + 𝑗 𝑐1 x 𝑖 + 𝑐 𝑗 𝑐1 x 𝑐 + 𝑖 𝑗 𝑐1 + 𝑐 + 𝑖 + 𝑗 𝑐1 x 𝑖 + 𝑐 𝑗 𝑐1 x 𝑐 + 𝑖 𝑗

Table 4. Impact of trade on sector-level comovement: vertical linkages, within- and cross-sector estimates

Note: Robust standard errors in parentheses. The sample period is 1996-2009. The dependent variable is the correlation of real output growth between sector

i and j of the country pair. µc1 and µc2 are country 1 and country 2 fixed effects, respectively. µi and µj are sector i and j fixed effects, respectively.

The coefficients for the variables “Same sector IO”, “Same sector”, and “Input-Output” in column 4 are omitted because this set of fi ed effects multiplies the sector dummies yielding very unusual results.

(30)

30

To assess the total effects of both bilateral trade and vertical linkages on busines cycle comovement, the effect is estimated by taking the coefficients of the designated variables sum them, while multiplying the interaction variables with the sample mean of vertical linkages, trade and/or with the dummy variable same sector.78910

The results are presented in table 5.

Panel A Panel B Panel C Panel D

Trade

_0.0071

_0.0024

_0.0103

_-0.0121

Vertical Linkages

_0.0365

_0.0417

_0.0373

_0.0422

Table 5. The estimated effect of trade and vertical linkages on business cycle comovement at the

sector level within the NAFTA

For consistency, we turn our attention towards the results for Panel C. There, evidence of the relative importance of vertical linkages in explaining business cycle comovement is found. But moreover, the results also indicate that the effect of bilateral trade on business cycle comovement is positive when taking all variables and interactions into consideration.

At this moment, there is prove for the (relative) importance of vertical linkages in explaining business cycle comovement in the NAFTA over the full sample period (1996-2009).

However, to see whether vertical linkages have become a more important trend in explaining business cycle synchronization over the years, we split the period into two subperiods,

simultaneously exploring the manifestation of an increased economic integration between the three trading partners. The first split period consists of data for business cycle comovement at the sector level over the period 1996 – 2001 whereas the second split period estimates the same specifications over the period 2002-2009.

7_{Coefficients are taken from column (3), insignificant coefficients are included.} 8 𝜕𝑐𝑜𝑚𝑜𝑣𝑒𝑚𝑒𝑛𝑡

𝜕𝑏𝑖𝑙𝑎𝑡𝑒𝑟𝑎𝑙 𝑡𝑟𝑎𝑑𝑒 = β1 Trade + β2 (Trade * same sector) + β3 (Trade * µIO) + β4 (Trade * same sector * µIO) + 𝜖_𝑖𝑗𝑐𝑑

9 𝜕𝑐𝑜𝑚𝑜𝑣𝑒𝑚𝑒𝑛𝑡

𝜕𝑣𝑒𝑟𝑡𝑖𝑐𝑎𝑙 𝑙𝑖𝑛𝑘𝑎𝑔𝑒𝑠 = β1 IO + β2 (IO * same sector) + β3 (µTrade * same sector * IO) + β4 (µTrade * IO) + 𝜖𝑖𝑗

𝑐𝑑

(31)

31

4.2 Estimating business cycle comovement: splitting the periods

This section presents the results for the estimation the effect of trade and vertical linkages on business cycle comovement for both split periods. To avoid an unnecessary long list of tables, only the results for the full specification (specification 4) are presented below.

Table 6. Impact of trade on comovement at the sector-level, split period 1 (1996-2001): vertical linkages, within- and cross-sector

estimates.

Note: Robust standard errors in parentheses. The sample period is 1996-2001. The dependent variable is the correlation of real output growth between sector i and j of the country pair. µc1 and µc2 are country 1 and country 2 fixed effects, respectively. µi and µj are sector i and j fixed effects,

respectively. The coefficients for the variables “Same sector IO”, “Same sector”, and “Input-Output” in column 4 are omitted because this set of fixed effects multiplies the sector dummies yielding very unusual results.

(1) (2) (3) (4) (1) (2) (3) (4)

Trade -0.0194* -0.0763*** 0.0347*** 0.0656*** Trade -0.0147 -0.0399*** 0.0169* 0.0544***

(0.0103) (0.0139) (0.0115) (0.0162) (0.00918) (0.0115) (0.00987) (0.0114)

Trade x same sector 0.0544* 0.0367 0.0709** 0.0701 Trade x same sector 0.0297 0.0233 0.0604** 0.0277

(0.0319) (0.0296) (0.0312) (0.0618) (0.0319) (0.0293) (0.0290) (0.0397)

Trade x IO -0.000577 -0.00200** 0.00217*** 0.00301*** Trade x IO -0.000145 -0.000226 0.00167** 0.00270***

(0.000693) (0.000795) (0.000731) (0.000843) (0.000659) (0.000704) (0.000658) (0.000698)

Trade x same sector -0.00192 -0.00154 0.000327 0.00381 Trade x same sector -0.00229 -0.00241 0.000414 0.000580

x IO (0.00200) (0.00208) (0.00191) (0.00279) x IO (0.00187) (0.00188) (0.00177) (0.00216)

Same sector x IO -0.0236 -0.0187 0.00104 - Same sector x IO -0.0102 -0.0193 0.00476

-(0.0336) (0.0321) (0.0311) - (0.0250) (0.0235) (0.0235)

-Same sector 0.848*** 0.625** 1.033*** - Same sector 0.550*** 0.420** 0.721***

-(0.260) (0.256) (0.282) - (0.164) (0.167) (0.176)

-Input-Output -0.0172** -0.0279*** -3.75e-05 - Input-Output -0.0114* -0.0111* -0.00640

-(0.00841) (0.00916) (0.00859) - (0.00615) (0.00638) (0.00611)

Panel C. Trade/Total trade Panel D. Trade/Sectot total trade

(1) (2) (3) (4) (1) (2) (3) (4)

Trade -0.0393*** -0.0839*** 0.00652 0.0729*** Trade -0.0484*** -0.0778*** -0.00194 0.0501***

(0.00954) (0.0116) (0.0102) (0.0151) (0.0100) (0.0121) (0.0105) (0.0124)

Trade x same sector 0.0466 0.0307 0.0654** 0.0520 Trade x same sector 0.00493 -0.00321 0.0397 0.0582

(0.0305) (0.0282) (0.0292) (0.0465) (0.0323) (0.0288) (0.0288) (0.0433)

Trade x IO -0.000864 -0.00174** 0.00151** 0.00343*** Trade x IO -0.00120* -0.00149** 0.00116 0.00286***

(0.000645) (0.000694) (0.000672) (0.000760) (0.000707) (0.000750) (0.000719) (0.000745)

Trade x same sector -0.00116 -0.00141 0.000498 0.00270 Trade x same sector -0.00166 -0.00252 0.000398 0.00208

x IO (0.00194) (0.00198) (0.00184) (0.00219) x IO (0.00227) (0.00216) (0.00197) (0.00233)

Same sector x IO -0.00721 -0.0145 0.00243 - Same sector x IO 0.0169 -0.000944 0.0178

-(0.0311) (0.0293) (0.0291) - (0.0223) (0.0207) (0.0207)

-Same sector 0.708*** 0.493** 0.850*** - Same sector 0.471*** 0.309** 0.592***

-(0.209) (0.208) (0.227) - (0.147) (0.142) (0.150)

-Input-Output -0.00890 -0.0129* 3.14e-05 - Input-Output -0.00779 -0.00745 -0.00695

-(0.00729) (0.00757) (0.00746) - (0.00535) (0.00557) (0.00544)

(32)

32

Table 6 presents the results for the estimations of trade on comovement for the first split period. Table 7 presents the results for the estimations of trade on comovement for the second split period.

There are again four columns, one for each trade measure as described in chapter 3. Column (1) estimates the impacts of trade on comovement without any fixed effects on a simple OLS regression, with column (2) adding country and sector effects. Column (3) includes country times sector effects and column (4) and final model is estimated using country-pair and sector-pair effects.

Attention will towards the results in panel C, column 3 in both tables to the sake of consistency in evaluating the impact of trade on business cycle comovement on the sector-level within the NAFTA.

At first sight, the tables evidence of the rising importance of vertical linkages in explaining of business cycle comovements within the NAFTA. But, the erratic behavior of the coefficients in the different configurations of fixed effects and measures of trade give some concerns about the robustness of the data when the period is split. The fact that both sector output growth correlations as well as trade and vertical integration ratios in the first and second split periods are measured over respectively 6 and 8 years, could be an important factor in this behavior. The observations are the weighted averages, which could be more heavily

influenced by outliers when the period of analysis gets shorter. The results for especially the first split period show that, the short time span of analysis turned out to be a limitation for the explanatory power of the variables.

(33)

33

(1) (2) (3) (4) (1) (2) (3) (4)

Trade -0.0168*** 0.0126*** -0.0196*** 0.0210*** Trade 0.00229 0.0287*** 0.00157 0.0275***

(0.00291) (0.00362) (0.00302) (0.00449) (0.00560) (0.00677) (0.00519) (0.00696)

Trade x same sector 0.0146 0.0220 0.00667 0.0204 Trade x same sector -0.0376 -0.0114 -0.0279 0.00593

(0.0251) (0.0195) (0.0203) (0.0475) (0.0271) (0.0210) (0.0210) (0.0326)

Trade x IO -0.00336* -0.000278 -0.00197 0.00129 Trade x IO 0.00161*** 0.00156*** 0.00144*** 0.00162***

(0.00175) (0.00135) (0.00134) (0.00313) (0.000411) (0.000405) (0.000347) (0.000457)

Trade x same sector - - - - Trade x same sector -0.00577*** -0.00201* -0.00320*** -0.00278

x IO x IO (0.00153) (0.00122) (0.00121) (0.00221)

Same sector x IO -0.0827*** -0.0142 -0.0450** - Same sector x IO -0.0766*** -0.0194 -0.0365**

-(0.0289) (0.0208) (0.0211) - (0.0197) (0.0148) (0.0147)

-Same sector -0.189 0.192 -0.113 - Same sector -0.506*** -0.0715 -0.291**

-(0.238) (0.188) (0.194) - (0.161) (0.127) (0.131)

-(0.00217) (0.00226) (0.00214) - (0.00383) (0.00417) (0.00339)

Panel C. Trade/Total trade Panel D. Trade/Sectot total trade

(1) (2) (3) (4) (1) (2) (3) (4)

Trade 0.0249*** 0.0650*** 0.0133** 0.0807*** Trade 0.0184*** 0.0520*** 0.0110* 0.0409***

(0.00570) (0.00805) (0.00538) (0.00918) (0.00621) (0.00804) (0.00573) (0.00768)

Trade x same sector -0.0104 0.00175 -0.00809 -0.00166 Trade x same sector -0.0774** -0.0295 -0.0515* -0.00577

(0.0247) (0.0188) (0.0192) (0.0342) (0.0362) (0.0271) (0.0271) (0.0340)

Trade x IO 0.00278*** 0.00329*** 0.00218*** 0.00385*** Trade x IO 0.00239*** 0.00264*** 0.00190*** 0.00218***

(0.000406) (0.000446) (0.000351) (0.000511) (0.000457) (0.000470) (0.000388) (0.000498)

Trade x same sector -0.00550*** -0.00154 -0.00302** -0.000905 Trade x same sector -0.00726*** -0.00232 -0.00408** -0.00251

x IO (0.00177) (0.00133) (0.00134) (0.00270) x IO (0.00204) (0.00159) (0.00163) (0.00245)

Same sector x IO -0.101*** -0.0235 -0.0519** - Same sector x IO -0.0633*** -0.00996 -0.0295**

-(0.0269) (0.0197) (0.0202) - (0.0167) (0.0127) (0.0132)

-Same sector -0.434** -0.00306 -0.246 - Same sector -0.557*** -0.0835 -0.316**

-(0.209) (0.160) (0.164) - (0.168) (0.132) (0.136)

-(0.00466) (0.00521) (0.00404) - (0.00339) (0.00380) (0.00303)

Table 7. Impact of trade on comovement at the sector-level, split period 2 (2002-2009): vertical linkages, within- and

cross-sector estimates.

The coefficients for the variables “Same sector IO”, “Same sector”, and “Input-Output” in column 4 are omitted because this set of fi ed effects multiplies the sector dummies yielding very unusual results.

The results for the interaction variable Trade x same sector x IO in Panel A are omitted by STATA because of multicollinearity. *** significant at the 1% level

(34)

34

When viewing the variable Input-Output solely, the two tables indicate a rising importance of vertical linkages in explaining business cycle comovement. However, again we are dealing with the full model from which we already have learned that the interaction variables have the power to alter the overall impact of both bilateral trade and vertical linkages on business cycle comovement. Therefore, again we turn towards a summations of results11. The results are presented in table 8.

Attention in table 8 will go towards the results in panel C. The results show the declining importance of bilateral trade in explaining business cycle comovement at the sector level whereas vertical linkages doubled in strength as an explanatory mechanisms through which business cycle comovement is influenced. The table also shows however, that the both the trade as well as the vertical linkage variable are very sensitive to the measurement of trade used. As both periods are measured over data on sectoral output and bilateral trade over the course of 6 and 8 years for the first and second split period, respectively, the influence of outliers in for example GDP levels can have a significant effect in the observations, resulting in the different outcomes for different measures of trade. In special, the variable “Same sector x IO” influence the total impact of vertical linkages a great deal. Because this variable yields a

11_{The overall results are measured with the same approach as for the full period. For details, see footnote 6,7,}

and 8.

(35)

35

negative number for the coefficient in panel C, column 3 and this coefficient is relatively large, the total estimated impact of vertical linkages is therefore heavily influenced by this specific variable. It leads to the fact that the total results don’t add up intuitively: it would make sense namely, that the total estimated impact of vertical linkages of the total period (0,3037) lied somewhere in the range of the estimated impacts of the same variable between the two split periods (0,025 and 0,0048). Now, this is not the case, pointing out the flaws in the research.

The next section will provide a discussion of the main results of the paper, together with its limitations and suggestions for further research.

5. DISCUSSION OF RESULTS, LIMITATIONS AND SUGGESTIONS

FOR FURTHER RESEARCH

The aim of this paper was twofold: first, it attempted to confirm the findings of past literature on the importance of vertical linkages on business cycle comovement at the sector-level. Second, it tried to find evidence for the increase of the relative importance of vertical linkages strengthening trade integration within the NAFTA.

It can be confirmed that the strength of vertical linkages on business cycle comovement within the NAFTA was about three times stronger than the effect of bilateral trade linkages in the full sample period (1996-2009). Moreover, the results also pointed out that next to vertical linkages, bilateral trade is also positively related to business cycle comovement. This finding contradicts the findings by Ng (2010) who suggests that, once vertical linkages are taken into account, the findings from past literature on bilateral trade intensity and intra-industry trade being positively related to business cycle comovement are not robust. As for the trade

linkages in the NAFTA between Canada, United States and Mexico, this turned out to be not the case. In the research, no assumptions were made about the elasticity of substitution of the goods being traded in both types of trade. Bilateral trade was simply taken by the final

(36)

36

Whether importance vertical linkages on business cycle comovement were stronger once the manifestation of the trade integration of the NAFTA settled in, remains a question for debate. The results from the regressions for both split periods indicate the effect of vertical integration in the second split period to be twice the size of the effect of vertical integration in the first split period. These outcomes could confirm the expectations of the rising importance of vertical trade linkages on business cycle comovement, but they have to be taken into account with some careful considerations as the results raised some questions about the robustness of the data. Although because most of the variable coefficients for regressions turned out to be insignificant for the first, they were taken into account in the summation. The short time span for the analysis for both split periods turned out to be a limitation for the research, as

outcomes were both erratic as well as often insignificant.

As for suggestions for further research, I would recommend the next step in analyzing

business cycle comovement to add a trade measure which interchange gross outputs for value added to capture bilateral trade linkages between sector-pairs. Introduced by Duval et al. (2015), the reason why using value-added trade in assessing the importance of bilateral trade linkages on business cycle comovement is quite simple. It is the value added exported by an specific industry that contributes to overall value added and therefore is a pure measure of growth within the sector. In today’s world, industries and countries increasingly specialize in adding value at a particular stage of production, with goods typically crossing borders

multiple times. Gross outputs therefore, do not capture how much growth an industry is experiencing because goods are typically double counted in this measurement. As the time for this research did allow for such an analysis, this remains a fruitful avenue for further research.

The importance of vertical linkages in business cycle comovements: Evidence from NAFTA.

Faculty of Economics and Business

MSc in International Economics and Business (IE&B)

Master Thesis

The importance of vertical linkages in

business cycle comovements: Evidence

from NAFTA.

Author:

Elke Oude Weernink

Student number:

S2170302

E-mail address:

e.c.m.oude.weernink@student.rug.nl

Supervisor:

T.M. Tarek Harchaoui, PHD

Co-assessor:

dr. A. (Anna) Minasyan

Table of contents

1. INTRODUCTION

2. LITERATURE REVIEW

3. MODELLING STRATEGY AND SOURCE DATA

(1) 𝜌

= ∝ + 𝛽

𝑇𝑟𝑎𝑑𝑒

+ 𝒖 + 𝜖

𝑇𝑟𝑎𝑑𝑒

𝟏[·].

(2) 𝜌

= ∝ + 𝛽

𝑇𝑟𝑎𝑑𝑒

+ 𝛽

𝟏[𝑖 = 𝑗]𝑇𝑟𝑎𝑑𝑒

+ 𝒖 + 𝜖

(3) 𝜌

= ∝ + 𝛽

𝑇𝑟𝑎𝑑𝑒

+ 𝛾

(

) + 𝒖 + 𝜖

(4) 𝜌

= ∝ +𝛽

𝑇𝑟𝑎𝑑𝑒

+ 𝛽

𝟏[𝑖 = 𝑗]𝑇𝑟𝑎𝑑𝑒

+ 𝛾

(

) +

𝛾

𝟏[𝑖 = 𝑗]

+ 𝒖 + 𝜖

∆𝑌

𝜌

= 𝑝𝑤𝑐𝑜𝑟𝑟 (∆𝑌

, ∆𝑌

)

𝑋

𝑇𝑟𝑎𝑑𝑒

= log (

∑

)

(Measure I)

𝑇𝑟𝑎𝑑𝑒

= log (

∑

)

(Measure II)

𝑇𝑟𝑎𝑑𝑒

= log (

∑

)

(Measure III)

𝑇𝑟𝑎𝑑𝑒

= log (

∑

)

(Measure IV)

𝐼𝑂

= 𝑙𝑜𝑔 (

∑

+

_{= ∝ + 𝛽}

_{= ∝ + 𝛽}

_{= ∝ + 𝛽}

_{= ∝ +𝛽}

_0.0071

_0.0024

_0.0103

_-0.0121

_0.0365

_0.0417

_0.0373

_0.0422