Are Chinese domestic shocks propagated through trade?
A VAR analysis of the SARS epidemic
Matthijs van Bolhuis s2514265
Msc. Economic Development & Globalization University of Groningen
16-06-2020
1
Abstract:
Using a VAR model, this paper quantifies the direct and linkage effects on the economic performance of European economies in terms of output, unemployment and consumer prices resulting from the trade effects ascribed to the SARS epidemic. A supply shock in Chinese trade, such as the SARS epidemic, has long lasting effects on European economies. In particular, a 1% drop in trade with China leads to a slowdown in growth between 0.04 and 1.52 percentage point slowdown in economic growth and between a 0.06 and 0.19 percentage point increase in unemployment. Prices barely seemed to respond at all.
2
1. Introduction
During the start of January 2020 news appeared of a new type of respiratory virus in China. This new virus, currently known as Covid-19, made many headlines. As an aspiring economist, the headline that struck me as most interesting was an article on the website of the NOS, a major Dutch news-organization, that said: “Possibly no Christmas decorations in store this year due to new Chinese epidemic” (Schutijser, 2020). This raised the question how integrated Chinese trade has become with European economies and how Chinese domestic shocks affected European economies. What are the extent- and duration of shocks? What are possible indirect effects? We consider the SARS pandemic as a natural experiment to answer these questions and to extrapolate some of these answers in an effort to speculate on some of the potential effects of Covid-19.
The Covid-19 epidemic has become a global pandemic and Europe has just experienced a pretty severe lockdown in which stores were closed and social contact was restricted (RIVM, 2020). While it is difficult to ascertain all the effects of Covid-19, it may be useful to use the SARS experiment to advance our knowledge in better understanding the ramifications of other forms of shocks arising from the Chinese economy.
In particular, this paper aims to map the direct- and indirect effects of the 2002-2003 SARS epidemic that arose in China on the economic performance of the European Union (EU), including some of its constituent major economies—Italy, France, Germany and the Netherlands. These shocks are assumed to be transmitted to the European economy through bilateral trade (Yamamoto, 2013; Jääskäla & Smith, 2013). This paper sets out to measure what the effects are of such transmitted shocks on the economic performance of several EU member states and the EU as a whole. Ultimately these results are applied to see, as crude as this currently may be, what the possible effects are of Covid-19 on the EU its trade with China and economic performance.
Untangling these effects is difficult, since GDP changes and trade changes are always dependent on a multitude of variables. Using a vector auto regression model, or VAR for short, we untangle these effects and see which effects are due to the Chinese domestic shocks and which effects are ‘regular’ economic fluctuations. A VAR model offers the possibility to not only distinguish between direct- and linkage effects but also offers a possibility to see which of these effects were still persisting after a long time (≥2 years). A separate VAR is ran for each of the individual locations.
3
Relevant literature is discussed in chapter 2, in which we look at how domestic shocks can be propagated and what its possible effects are. We also look at the economic figures of the SARS epidemic. Chapter 3 introduces the variables used for measurements. Chapter 4 discusses the framework and its restrictions. Chapter 5 discusses the relevant tests on stationarity, normality and co-integration. As well as discussing trends in the variables.
In chapter 6 we discuss the results based on Impulse Response Functions (IRFs) and Forecast Error Variance Decompositions (FEVDs): we find that there was a significant direct impact of the SARS epidemic on the economic performance of the European Union as a whole. These direct effects were also present in the German- and French economies. A standard deviation shock in Chinese trade leads to an astonishing 2% change in output Y of the EU. These results are similar for France and Germany, both experienced quite heavy negative effects from the SARS shocks. Surprisingly, the direct effects on the Italian economy were positive on the short term. Even more surprising is that the effects of the Dutch economy were found to be very minor. The indirect effects of the shocks were found to be much smaller than the direct effects. In all of the locations the indirect effects had the same direction of effects as the direct effects, except in Italy in which it had the opposite effect from the direct effects. Chapter 7 tries to extend these results to the Covid-19 pandemic, using relevant trade data and policy recommendations. Chapters 8&9 conclude the paper and discuss relevant caveats.
2. Literature review
Surprisingly, there has been ample research on the international economic effects of the SARS epidemic. Papers from Beutals et al. (2009) and Siu & Wong (2004) found that there was a substantial negative effect on the Chinese consumption and travel industries. The negative effects of the SARS epidemic on the tourism industry are further supported by the research of Hai et al. (2004), who found that the effects of the SARS epidemic were most prevalent in the Chinese tourism industry. However, substantial research to what the effects of the SARS epidemic was on Western economies is lacking, leaving a crying gap as particularly given that a large span of the literature suggests that domestic economic performance and trade flows are correlated (Burstein, Kurz & Tesar, 2008). This would suggest that, if trade flows were negatively affected, Western economies suffered both direct- and ‘linkage’ economic damages. In recent years, the question has arisen for European policy makers whether they have become too dependent on Chinese exports and to what extent future calamities in China will be felt in European economies (Fox & Godement, 2019).
The goal of this paper is to measure the effects of internationally transmitted Chinese domestic shocks on the real European economy. This is because taking both the monetary and the real economy into account overloads the VAR model with variables, making the measured shock its individual effects hard to measure on the endogenous variables (Gottschalk, 2001; Blanchard, 1989). For this reason this paper ignores the propagation of shocks through the financial integration channel and solely focusses on the propagation of shocks trough trade linkage of industries.
4
transmitted through trade value (Yamamoto, 2013; Jääskäla & Smith, 2013) and that a shock taking place in China affects the value of trade between the U.S. and the EU through China’s trade with the U.S. (Kireyev & Leonidov, 2018). This concept is referred to in this paper as triangular trade linkage, since this paper only measures trade between three different locations at a time. The second set of assumptions has to do with distinguishing between direct- and indirect effects and is explained in subsections 2.2 and 2.3.
2.1 International propagation of shocks
According to Lee (2019), domestic shocks are transmitted through two major channels: trade- and vertical linkage of industries and the financial integration of global markets. Trade- and vertical linkages are the industries’ dependencies on imports and the use of Global Value Chains (GVCs). Financial integration of global markets is the extent to which corporations and industries are connected through accounts at the same banks/shared financial instruments. This idea was preceded by a paper of Carare & Mody (2012), who found that domestic shocks are faster transmitted currently due to the global fragmentation of production. In other words, they find that domestic shocks are transmitted faster through to trade integration of economies. For the scope of this paper this means that a domestic shock in China, specifically the SARS epidemic of 2002-2003, is propagated to the EU through a loss of trade value with the EU. Eickmeier & Ng (2016) measured the propagation of international credit shocks and found bilateral trade linkage of countries to be an important determinant of the international propagation of the shocks.
2.2 Direct effects of trade shocks
The main aim of this paper is to measure what the effects the SARS epidemic on the European economies. For this sake the direct effects of the SARS epidemic are the effects of the SARS epidemic propagated through a shock in bilateral trade with China. This means that direct effects are measured as a loss of economic performance, most notably unemployment and output, of the European Union and its member states due to a trade shock with China. (Keogh Brown & Smith, 2008).
Shocks are propagated internationally through trade in a number of ways: a shock in demand, a shock in supply and price shocks. The direct effects of such shocks differ depending on the type of shock (Peersman & van Robays, 2009). Since China has manifested itself as the manufacturing/export hub of the world (Li, 2018) it can be expected that the type of shocks which are internationally propagated are supply shocks. This idea is supported by Hansen & Robertson (2010), who find that the domestic shocks propagated through Chinese trade with its western trading partners are likely to be supply shocks. Supply shocks have strong effects on unemployment, economic output and national price levels (David, Dorn & Hanson, 2013). These effects differ across nations, since import competition with the country in which the shocks occur is a major determinant of whether the effects are positive or negative (Autor, Dorn & Hanson, 2016).
5
2.3 Indirect effects of trade shocks
Indirect effects are difficult to tackle, since it is difficult to classify which effects are related to the shock. Each downstream occurrence related to the initial shock can be classified as an indirect effect (Alfaro, García-Santana & Moral-Benito, 2019). The most important determinant of indirect effects in this paper is that indirect effects are effects of the initial shock which are propagated through downstream channels. This idea of downstream effects is preceded by Autor, Dorn & Hanson (2016), who find that downstream damages occur due to the interconnection of Global Value Chains (GVCs) and international trade. This means that the effects of the shock are propagated through triangular trade linkage.
This definition of indirect effects along the lines of Alfaro,García-Santana & Moral-Benito (2019) and Autor, Dorn & Hanson (2016) is applied in this paper. Applying this results in the following definition: Indirect effects are the downstream effects of shocks which are measured as a loss of trade value with the EU and other trading partners. This definition is backed by a paper of Kireyev & Leonidov (2018), which comes to a similar conclusion: The indirect ‘spillover’ effects of internationally propagated shocks occur through network linkage of countries and industries.
To map the indirect effects of the domestic Chinese shock the European economic performance propagate through trade with other trading partners there are several interesting options: Mapping trade with the rest of the world, intra-European trade and trade with the United States (OECD, 2020). Trade with rest of the world can paint a complete picture on the ‘rippling effects’ of Chinese trade shocks. Trade values with the ROW is not homogenous however, which means that is hard to specify in which these downstream effects occurred and what specific implication has for policy makers. (Benedictis & Tajoli, 2011). Choosing intra-European trade as a measure of downstream effects is another option, the problem with choosing intra-European trade is that intra-European trade is almost fully endogenous with economic performance of the Eurozone (Guerrieri & Esposito, 2012). For these reasons the downstream channel used in this paper is trade with the U.S., as this is Europe’s largest trading partner in absolute value (OECD trade statistics, 2020). Supporting this ‘rippling’ effect of the triangular trade between the U.S. the EU and China is a recent book on the U.S. – China trade war. Syed & Yilmaz Genç (2020, pp. 138) find that bilateral trade value and -shocks between the U.S. and China has a substantial negative effect on the European economic performance.
2.4 Effects of the SARS epidemic on trade
6
propagated through trade it makes sense to include these as main endogenous variables in the analysis.
Judging from theory it is likely that SARS had a negative impact on Europe’s output and a magnifying effect on prices and unemployment (in the sense that these went up due to a trade loss).
Figure 1: Bilateral trade between the EU and China (in billion €)
Notes: Data is taken from the OECD database on trade (2020). Bilateral trade is the added exports and imports between both locations. The y-axis represents the bilateral trade in trillion €, the x-axis represents the period. Bilateral trade is measured in real prices.
Figure 1 shows there was a decline in bilateral trade between the EU and China during the SARS epidemic (Q4 2002 – Q2 2003), this decline in trade seems to have been minor, being 560€ million from Q4 2002 – Q1 2003, but it took almost a full three quarters to recover. More notably is the decline in bilateral trade of around 5.8€ billion during Q2 2004 – Q3 2004. This drop in trade bounced back one quarter after its occurrence. There does not seem to be any other specific event during that period except the after effects of the SARS epidemic (Lee & McKibbin, 2004).
A recent paper by Fernandes & Alexander (2020) supports the finding that the 2004 dip in trade is due to the SARS epidemic. In their paper they find that Chinese firms across the board experienced a loss in imports and exports. The paper also found that products that were relatively capital- or skill-intensive suffered a lower export decline and a stronger recovery (Fernandes & Alexander, 2020). The European Union’s main imports from China consist of consumer- and industrial goods, machinery and clothing & apparel (European Commission trade data, 2020). These goods, except machinery & industrial goods, are relatively quite low-skill and low-capital intensive goods (Ma, Tang & Zhang, 2014). This means that it can be expected that European trade was hit harder than the trade of trading partners who imported relatively higher skill- and capital intensive goods.
7
Figure 2: Output European Union (in billion €)
Notes: Data is taken from the OECD database on trade (2020). Output is measured as real GDP. The y-axis represents the output of the EU in trillion €, the x-axis represents the period.
Figure 2 supports this idea, as it can be clearly seen that European output suffered a quite significant dip of 13€ billion during the SARS epidemic (Q4 2002 – Q2 2003). European output seems to go up again after the second quarter of 2003, albeit quite slowly. Looking at Figure 2 it takes almost a full year for European output to recover to the levels it had before the SARS epidemic. This hints at pretty strong economic damages due to the SARS epidemic.
This is supported by the previously mentioned paper of Fernandes & Tang (2020), stating that the SARS epidemic lead to a wide array of small exporters exiting the market, slowing down trade recovery quite substantially. They found that firms affected by SARS had a significantly lower trade growth for about two years after the epidemic, slacking roughly 12% behind the Chinese firms that were not affected by SARS (Fernandes & Tang, 2020).
2.5 Possible Covid-19 effects on trade
Extending this paper’s topic to Covid-19 is tricky, since the economic damages of the current widespread lockdown are likely to far exceed the damages caused by a Covid-19 related trade shock (Coibion, Gorodnichenko & Weber, 2020). However, the papers by Fernandes & Tang (2020) and Lee (2019) showed that a widespread trade shock as a result of exogenous shocks hinders the speed of economic recovery. This means that either returning to a more autarkic economic model, or a loss in international consumption of traded goods due to the astounding loss of consumer’s trust can extend the duration of the crisis. Because trade value and the number of exporters positively affects economic recovery (Fernandes & Tang, 2020; Lee, 2019). The estimated World Uncertainty Index (WUI) is expected to go up to a staggering 300% because of the Covid-19 crisis. For reference: during the financial crisis of 2008 the WUI was 200% and during the SARS epidemic the WUI was 70% (Kilic & Darin, 2020; Ahir, Bloom & Furcery, 2018). Not only is the expected WUI much greater than during the SARS epidemic, or even the financial crisis of 2008, China’s share of World GDP and exports is currently also
8
almost triple of what it was during the SARS epidemic (The Economist, 2020). China’s share of global GDP went up from 4% in 2003 to 16% currently. China’s exports’ share of total exports grew from 5.6% in 2003 to 12.6% currently (The Economist, 2020; OECD database, 2020). This means that even if Covid-19 had only remained as a domestic trade shock, its damages would far exceed the damages from SARS. To later extend the results we use the estimation of the World Trade Organization, who estimate a global trade decline of a staggering 32% in 2020 (WTO, 2020), as well as the OECD data on Bilateral trade from January to March 2020.
2.6 Is there any difference in impact on the individual EU member states?
The period up until 2008 was called the period of hyper-globalization, in which Global Value Chains (GVCs) accounted for 60% of world trade growth (Timmer et al., 2016). Around 2011, the usage of Global Value Chains peaked and remained roughly the same amount until this day. The dependency on GVCs for a countries production and exports is still a good measure of a country’s susceptibility to shocks transmitted through trade (Vandenbussche, Garcia & Simons, 2019; Autor, Dorn & Hanson, 2016). This means that the more shared GVCs a country has with China, the more susceptible it is to suffering from effects of Chinese domestic shocks.
To estimate the degree of competitiveness of the European countries and Chinese imports we assess the most important industries of each location. In Germany, France and Italy the industry with the most reshoring activity is the chemical industry. For Germany the industry most dependent on imports is the manufacturing industry. Italy’s industry that is most dependent on imports is the electronic industry, albeit very small, for France this is the automotive industry and for the Netherlands this is electronics (WIOD, 2016). The biggest exports of China to Europe are consumer goods, machinery and equipment, footwear and clothing and manufacturing products (European Commission trade data, 2020).
Judging from these import numbers we can assess that the German and French industries have the strongest dependency on Chinese imports, which means that they will probably suffer stronger direct effects from trade shocks with China than the Netherlands and Italy. This is because Italy competes with a lot of Chinese exported products and is relatively independent on China for its GVCs. From this we expect that Italy has either enjoys a positive economic effect from trade shocks with China or that its direct effects are fairly minimal (WTO 2020; WIOD, 2016).
2.7 Hypotheses
Two hypotheses concerning the trade effects of SARS and the possible trade effects of Covid-19 are drawn up.
9
suffered trade growth damages till up to two years after the SARS epidemic. (Fernandes & Alexander, 2020; Fernandes & Tang, 2020; Kose & Riezman, 2013; Autor, Born & Hanson, 2016).
Hypothesis 2: Economic performance of the European Union suffers additional linkage effects through trade with the U.S. and its effects took around 2 years to fully recover.
The second hypothesis has to do with the indirect effects of trade shocks, it can be expected that domestic shocks in China lead to an overall decline of trade around the globe due to trade linkages between nations (Kireyev & Leonidov, 2018). Specifically for this paper, that means that European GDP and unemployment suffers due to decrease trade between the U.S. and the EU as a result of Chinese domestic shocks. Based on the paper of Autor, Dorn & Hanson (2016) these effects are expected to be much smaller and are estimated to be roughly 20% of the initial direct effects of the shock (Autor, Dorn & Hanson, 2016).
Ouput and unemployment of European member states are correlated with each other (Blanchard, 2006; McGann & Ortega-Argilés, 2015). For this reason the two hypotheses above are assumed for each individual location.
3. Variable introduction
In order to map the effects of Chinese domestic shocks on the European Union (EU), several assumptions have to be made: Firstly, a domestic shock is transmitted through trade (Yamamoto, 2013; Jääskäla & Smith, 2013). Secondly, trade between locations in which the domestic shock did not take place is affected through their linkage to the country in which the domestic shock occurred (Kireyev & Leonidov, 2018). To map these shocks transmitted through Chinese trade to the EU a VAR analysis is applied, which allows for measuring to what extent domestic Chinese shocks channeled through shocks in trade have an effect on the economic performance of the EU and several individual member states.
The VAR also measures the effect of said shock on trade of the EU with other major trading partners. In this paper, the other trading partner is the U.S., since it is the largest trading partner of the EU in terms of value in euros. The model also illustrates the duration of the shocks. The VAR is ran in two separate ways: (log) levels and first differences. As running the VAR in these two ways allows me to test the log-levels models robustness, seeing whether there are any significant differences depending on the type of measurement. The log-levels provide the most important results, with the first differences model serving as a robustness check.
After measuring the effects on the European Union, the individual effects on several major member states will be measured individually. These major states being: Italy, France, Germany and the Netherlands. Following Blanchard (1989) and Blanchard & Perotti (2002), economic performance is defined as: economic output Yt, unemployment level Ut and price level,
measured as the Consumer Price Index (CPI), Pt. The trade channels are encapsulated in the
variables: bilateral trade with China Tct , which measures all the trade between Europe and
China. This is an important variable as it is the main channel through which Chinese domestic shocks are transferred to Europe (Lee, 2019). The other trade variable focusses on the other most important trading partner of the EU in terms of value in euros, the United States, and is measured as bilateral trade with the U.S. Tust. The notation of the models is done along the
10
the GEO Federal Reserve Economic Data Centre (GEOFRED). An interesting thing about VAR modelling is that it yields a strong forecasting power, making the results also applicable to current trade shocks (Korobilis, 2013).
The variables are further outlined in Table 1: natural logs are used following Blanchard (1989) and Breitung & Pesaran (2008). The period of measurement is from 2002 to 2019 and uses monthly data. This period starts at 2002 to encapsulate the period of in which SARS took place and ends in 2019 in order to make sure the shocks were not persistent until today. The data is taken from OECD and GEOFRED is all seasonally adjusted by the OECD and GEOFRED.
Table 1: Variable definitions and measurements Variable definitions and measurements
Variable Definition Source Measurement Period
Tc Bilateral trade with China in €
OECD database
Natural logs (ln) 2002-2019, monthly Tus Bilateral trade
with the U.S. in €
OECD database
Natural logs (ln) 2002-2019, monthly
Y Industrial
Production Index (IPI)
GEOFRED Natural logs (ln) 2002-2019, monthly
U Unemployment
rate
GEOFRED Percentages (%) 2002-2019, monthly
P Consumer price
index (CPI)
GEOFRED Natural logs (ln) 2002-2019, monthly
Notes: For Ythe Industrial Production Index is used because of its availability of monthly data. For Italy, France, Germany and the Netherlands the measurement methods, as well as the sources, are exactly the same.
4. Empirical framework
In order to measure the effects of transmitted shocks with China I follow a model along the lines of Peersman & van Robays (2008) and Gottschalk (2001). The model is quite standard, but is given economic meaning using restrictions via Cholesky’s Decompositions. The model follows a set of equations which let variables be defined by its own lags and other variables and their lags: ( 𝑇𝑐𝑡 𝑇𝑢𝑠𝑡 𝑌𝑡 𝑈𝑡 𝑃𝑡 ) = ( 𝑐𝑇𝑐 𝑐𝑇𝑢𝑠 𝑐𝑌 𝑐𝑈 𝑐𝑃 ) + ( 𝑎11 𝑎12 𝑎13 𝑎14 𝑎15 𝑎21 𝑎22 𝑎23 𝑎24 𝑎25 𝑎31 𝑎32 𝑎33 𝑎34 𝑎35 𝑎41 𝑎42 𝑎43 𝑎44 𝑎45 𝑎51 𝑎52 𝑎53 𝑎54 𝑎55) ∗ ( 𝑇𝑐𝑡,..𝑡−𝑝 𝑇𝑢𝑠𝑡,..𝑡−𝑝 𝑌𝑡,..𝑡−𝑝 𝑈𝑡,..𝑡−𝑝 𝑃𝑡,..𝑡−𝑝 ) + ( 𝑒𝑇𝑐 𝑒𝑇𝑢𝑠 𝑒𝑌 𝑒𝑈 𝑒𝑃 ) (1)
In which the first matrix denotes the variables at time t, are dependent on the pth lagged values of all variables and their Lag polynomial a. The c matrix represents a constant intercept and the e matrix represent the values of the error term for each variable. The extent of exogenous shocks on the variables are captured by these error terms. Equation (1) is a long drawn out equation, which could be shortened in the standard vector form:
11
In which Yt represents a vector of the variables P, Y, U, Tus and Tc at time t, B(L) represents
the lag polynomial matrix containing the a coefficients and the constants c from equation (1). The second Yt represents the vector with the lagged values of the variables P, Y, U, Tus and Tc
to the number of lags p and et is the vector of the error terms e.
The problem with equation (1) and (2) is that the effects measured do not yet have economic meaning. This is because each of the variables are assumed to be endogenous and have a direct causal relationship with each other. In reality this is not the case, which means that some restrictions have to be imposed on the standard model to give the results any meaning. The first set of restrictions is a usual set of orthogonality restrictions, which makes sure that the covariance of the variables has to be 0. In example: the σy,u must be 0 in order to assume orthogonality of the variables, with y ≠ u. The second set of restrictions are the Cholesky decompositions, which restrict the “causal chain” of shocks. I.e.: the first shock influences all variables at time t and the last shock only influences the last variable at time t. This makes the recursive ordering of the variables very important. Note that Cholesky’s matrix is an Identity of 𝛾 effect-coefficient matrix as denoted in equation (3):
𝐶𝑀 = ( 1 0 0 0 0 𝛾21 1 0 0 0 𝛾31 𝛾32 1 0 0 𝛾41 𝛾42 𝛾43 1 0 𝛾51 𝛾52 𝛾53 𝛾54 1) (3)
This means that the ordering of the endogenous variables matters, which will be explained in the chapter below, because all values above the diagonal are set to 0. The standard model is applied to each of the researched locations: the EU, Italy, France, Germany and the Netherlands. 4.1 Compositions, reasoning and causality
The most important function of Cholesky’s decompositions in this paper is to differentiate between indirect- and direct effects of trade shocks with China. This means that direct- and indirect effects of trade shocks with China are differentiated through their causal relationships. For the Levels models and the First differences model the decompositions are the same, the difference between the two models lies solely in measurement.
4.1.1 Direct effects
12
This means that direct international effects of domestic shocks in China are transferred via a loss in trade volume, particularly a loss in Chinese exports. According to Herrero & Xu (2017), China’s exports are contributing highly to European output and job creation. Following these the following ordering of variables is proposed:
𝐶𝐷(1): 𝑌 = |𝑒𝑇𝑐𝑡𝑒𝑌𝑡𝑒𝑈𝑡𝑒𝑇𝑢𝑠𝑡𝑒𝑃𝑡| 𝐶𝐷(1): 𝑌 = |∆𝑇𝑐𝑡∆𝑌𝑡𝑈𝑡∆𝑇𝑢𝑠𝑡∆𝑃𝑡| (4)
Notes: the left Cholesky’s decomposition (CD) is in (log) levels, the right decomposition is in first differences. Unemployment is not taken in first differences since it is assumed to be stationary and its effects will always return to zero in the long run (Papell, Murray & Ghiblawi, 2000; Khraief et al., 2017)
In which Y represents the the endogenous variables in equation 2. For the set of equations, the causal ordering of contemporaneous effects follows: еTct → еYt → eUt → eTust → ePt. Of which
adaptation of prices is fixed. This is because prices are sluggishly to adapt, which means that none of the variables influences prices in the same period (Blanchard 1989). The intuition is that a trade shock with China will influence European output and unemployment down the line. The direct effects are differentiated by putting eTust right before prices. The intuition of this is
that trade shocks with China influence the functioning of the European economy, encapsulated by еYt & eUt, which then influences the trade with the U.S. This means that the decomposition
in equation (4) measures the direct damages due to domestic shocks in China channeled through a shock in trade volume (Jääskäla & Smith, 2013). The relationship between unemployment Ut
and output Yt is tricky, since it has been a topic of debates amongst economics to the direction
of causality. For the sake of simplicity we chose to let unemployment Ut, be caused by economic
output Yt. Padalino and Vivarelli (1997) found a strong one way relationship between aggregate
economic activity and employment, finding that unemployment was a weak predictor of economic output.
4.1.2 Indirect effects
A paper by Hanousek, Kočenda & Maurel (2011) defines indirect economic effects as ‘spillover effects’ causing damages, or gains in the case of positive shocks, through triangular trade linkage. These spillover effects can cause economic damages due to unforeseen circumstances, such as loss of trade with other nations where the shock did not occur due to triangular trade linkages. In this paper I follow along these lines to determine that economic damages sustained from loss of trade volume with the U.S. resulting from Chinese trade shocks are classified as indirect effects. This results in the following decompositions:
𝐶𝐷(2): 𝑌 = |𝑒𝑇𝑐𝑡𝑒𝑇𝑢𝑠𝑡𝑒𝑌𝑡𝑒𝑈𝑡𝑒𝑃𝑡| 𝐶𝐷(2): 𝑌 = |∆𝑇𝑐𝑡∆𝑇𝑢𝑠𝑡∆𝑌𝑡𝑈𝑡∆𝑃𝑡| (5)
Notes: the left Cholesky’s decomposition (CD) is in (log) levels, the right decomposition is in first differences. Unemployment is not taken in first differences since it is assumed to be stationary and its effects will always return to zero in the long run (Papell, Murray & Ghiblawi, 2000; Khraief et al., 2017)
Equation (5) has a causal chain of contemporaneous effects that follows: еTct → eTust → eYt
→ еUt → ePt. The intuition of equation (5) is that a domestic shock in China is internationally
transmitted through a shock in еTct. Rather than in equation (4), in this situation that does not
directly lead to a loss in economic performance, denoted by еYt & eUt , but a shock in еTct leads
to eTust. Possible economic damages sustained in this scenario are due to ‘spillover effects’,
since еTct & eTust → еYt & eUt (Fry, 2004). The same reasoning for the causal chain of
13
5. Data and Trends
This chapter shows the results of all preliminary tests for all of the VARs. The results for the tests concerning co-integration and normality for the individual countries are found in appendix A.
5.1 Trends- and variable behavior over the years
Figure 3: Bilateral trade with China, Bilateral trade with the U.S., Price level, Output and Unemployment over time in the EU
Sources: GEOFRED and the OECD databank. Notes: Both Bilateral trade variables Tc& Tus, as well as Price level Pand output Y , are taken in natural logs. Unemployment rate U is measured in percentages. The y-axis represents the endogenous variables, the x-axis represents the monthly period over 2002-2019.
14
bilateral trade with the U.S., prices and economic output follow a log linear trend. For prices the deviation from that trend is limited, because prices are generally slow to adapt (Sousa & Bradley, 2009). Most notably unemployment and output suffer from a break in the time series during the financial crisis of 2008. However, because of the estimated duration of the shocks and the time in which the SARS shock took place, this break in the time series of the variables is of minor concern. This is further supported by Table 3, which finds that Tc and P are stationary in I (1) which means that the variables revolve around their mean. If the structural break in the dataset would be of any concern that would mean that there would be a problem regarding stationarity of the variables, Table 3 shows that this is not the case. This is supported by the Dicky-Fuller t-test for structural breaks, which finds no structural breaks, even at the p<0.1 level. The results for the Dicky-Fuller t-tests can be found in Table A.1 of appendix A. This is further supported by the trend direction of the variables from after the recession, as it can be seen from Figure 3 that around 2014 each of the variables returns to its growth rate from pre-2008. There seems to be some downward fluctuations in output, trade with China and trade with the U.S. during the period of the SARS (2002 Q4 -2003 Q2) epidemic, but these were quick to recover. It is interesting to note that shortly after, in the second quarter of 2004 there was a quite substantial drop in both the trade variables, specifically in Tc. This could be due to late effects if the SARS epidemic, but these events could also be unrelated. The optimal amount of lags used for each of the separate locations is tested using a sequential Likelihood Ratio (LR) modifier. The results of the lag tests of each of the locations can be found in Table 2 below:
Table 2: lag selection and sample size
EU ITA FRA GER NL
Optimal No. of lags p
12 12 12 12 12
Sample size (N) 204 204 204 204 204
Notes: Lag length was tested following a sequential Likelihood Ratio (LR) modifier. The maximum amount was set to 12 lags (= 1 year).
5.2 Stationarity & co-integration
In order to check the variables for stationarity, a Dicky-Fuller GLS was used. For each of the locations the outcomes of the Dicky-Fuller GLS tests yield comparable results.
Table 3: Dicky-Fuller GLS test results for the EU
(log)levels First differences
trend intercept lag t-stat p-value trend intercept lag t-stat p-value
Tc yes yes 12 -1.3587 0.1759 no yes 11 -3.7804 0.0037
Tus yes yes 12 -2.0467 0.0421 no yes 12 -4.0733 0.0013
Y yes yes 3 -2.7435 0.0066 no yes 2 -5.3873 0.0000
U yes yes 2 -6.9776 0.0000 no yes 1 -3.8515 0.0001
P no yes 12 0.7061 0.4792 no yes 12 -2.9905 0.0811
15
Theory can be consulted in order to further develop the idea of stationarity of these variables. According to Rapach (2002), economic output Y is considered stationary if adjusted for seasonality. Only after adjusting for seasonality Y is stationary around a deterministic trend since it is highly influenced by the results of the previous time period. This assumption is the same for P and U. Papers by Papell, Murray & Ghiblawi (2000) and Khraief et al. (2017) both suggest that unemployment always circles around its natural rate and can thus be considered stationary. The results of the Dicky-Fuller GLS test on stationarity is shown in Table 3. Table 3 shows some issues regarding the stationarity of the variables Tc and P when measuring in levels. Trade variables are considered to have a trend and intercept, just as economic output Y, according to a study by Buluswar, Thompson & Upadhaya (1996), in which they state that trade value and aggregate economic activity tends to move in the same direction. There seems to be no issue regarding stationarity using a first-differences model, in which all variables are significantly stationary at the p<0.10 level. This implies that the variables P and Tc are I (1). This is the case for each of the individual locations, of which the tables for the Dicky-Fuller GLS test can be found in the appendices A2-A5. The VAR in both first differences and in log levels are stable, the AR Characteristics polynomial of all the locations were found to be stable and are found in the Appendices A1 to A5. This means that the VAR is stable, which implies that the effects of exogenous shocks are reduced to zero over time, making a VAR a suitable model. The AR Characteristics polynomial for each of the countries Italy, France, Germany and the Netherlands all imply a stable VAR as well.
In order to measure whether the variables suffer from co-integration, Johansen co-integration tests for time series were conducted. The results of the co-integration tests for the EU can be found in Table A1.3 in the appendix, results for Italy, France, Germany and the Netherlands can be found in Tables A2.3, A3.3, A4.3 and A5.3 of appendix A respectively. I performed three separate co-integration tests, because Tc has a stationary trend case three (linear deterministic trend, intercept & trend assumed) is preferred. The correlation matrices of the variables for each of the VARs are shown in Table A1.2 – A5.2 of the appendices A1-A5 of each of the locations, there seem to be no large correlations between the variables, except between Tc and Tus. This could be because of the fact that the U.S. and China have a lot of similar exportable products and the EU is an important importer of both countries (WTO, 2020). The summarized results of all the co-integration tests can be found in the table below:
Table 4: Summarized results of the Johansen co-integration tests Number of co-integrated equations
Location Trace Max EV
EU 0** 0**
ITA 1* 0*
FRA 1* 0*
GER 0** 0**
NL 0** 0**
Notes: full tables can be found in appendices A1-A5. ** = p<0.01, * = p<0.05.
16
the error terms of the models for Italy and France suffer from autocorrelation, making the effects of the shocks for Italy and France difficult to entangle (Gottschalk, 2001). We still run the results of the VARs for Italy and France, because under the max Eigenvalue statistic both locations have zero possible co-integrating equations at the p<0.05 level.
5.3 Normality tests
In order to test normality the Doornik-Hansen normality tests were performed, following along the lines set out by Doornik & Hansen (2008), of which the results can be found in Table 5 below:
Looking at Table 5, it can be seen that the only variable suffering from non-normality is German unemployment at p<0.01 and European prices if significance levels are stretched to p<0.1. However, due to the sample size (n = 204) this is of no concern.
Table 5: Summarized results of the Doornik-Hansen tests for multivariate normality
Χ2 p-value
EU ITA FRA GER NL EU ITA FRA GER NL
Tc 0.553 1.001 1.419 0.674 0.075 0.46 0.31 0.23 0.41 0.78
Tus 0.831 2.394 0.038 2.067 1.019 0.36 0.12 0.84 0.15 0.31
Y 1.776 0.256 1.441 0.742 0.591 0.18 0.61 0.23 0.39 0.44
U 0.012 0.594 0.208 27.86 0.000 0.90 0.44 0.65 0.00*** 0.99
P 3.013 0.023 0.087 0.262 0.568 0.08* 0.87 0.77 0.60 0.45
Notes: Normality is tested against the null-hypothesis of multivariate normality. * = p<0.1, ** = p<0.05, *** = p<0.01
6. Results
The results of the analyses are found in this chapter. First we estimate the Granger causality of each of the individual models. Granger causality is based on predictive causality, making it a useful tool for forecasting. This means that the Granger causality is useful in extending the effects measured in this paper to current and future events, such as the Covid-19 pandemic. Secondly I look at the Impulse response functions, which graphically depict how variables respond to a one standard deviation shock in Tc and Tus. This will tell the extent of the effects of shocks in eTc as well as estimating the duration for the effects of these shocks to return to 0. Lastly, the forecast error variance decompositions are estimated. The forecast error variance decompositions measure the extent of the variance in economic performance eY & eU caused by shocks in trade with China, eTc, (between t and t + s) and the extent that is caused by shocks in other endogenous variables (including eY & eU itself) (Gottschalk, 2001; Pfaff, 2008). 6.1 Results of the Granger causality tests
17
Table 6: Results of the Granger causality test for the EU. Granger Causality test
Tc Tus Y U P
IV. F Pr. IV. F Pr. IV. F Pr. IV. F Pr. IV. F Pr. Y 1.16 0.39 Y 1.13 0.32 U 3.94 0.02 Y 0.43 0.65 Y 0.11 0.90
U 3.87 0.00 U 1.94 0.15 P 0.47 0.63 P 0.46 0.63 U 1.08 0.34
P 8.52 0.00 P 10.2 0.00 Tc 1.92 0.15 Tc 0.36 0.03 Tc 1.93 0.15
Tus 1.59 0.20 Tc 7.18 0.00 Tus 5.9 0.00 Tus 9.8 0.00 Tus 3.3 0.04
Notes: 12 lags are included (n=204). Results for Italy, France, Germany and the Netherlands can be found in appendix B1.
The null hypothesis is X1 does not Granger-cause X2. The dependent variable is found in the top row of Table 6.
Note that Granger causality is not the same as direct causality, as it only measures if variables are useful for predicting variances in dependent variables. That being said, from Table 7 can be clearly seen that Tc is a useful predictor for Tus and P in four out of five models. The Granger causality test shows that the predictive causal relationship between Tc → U and Tc → Y is not present for each of the five locations, in only three locations Tc granger causes Y and in four locations Tc granger causes U.
Table 7: Summarized results of the Granger causality tests for all locations Number of models in which IV Granger causes DV
IV\DV Tc Tus Y U P Tc - 4*** 2**(3*) 2**(4*) 4*** Tus 2*** - 1*** 3*** 2***(3**) Y 3* 1* - 3*** 0 U 1*** 2** 1***(2**) - 2** P 5*** 5*** 2***(3*) 3*** -
Notes: The number represents the amount of models in which IV granger caused DV. Full results can be found in appendix B1. *** = p<0.01, ** = p<0.05, * = p<0.1. Possible number of Granger causalities is in brackets if significance levels are stretched.
The biggest issue is the significance of the relationships between Tc & Y and Tc & U. At p<0.05, there are only two locations in which Tc granger causes U and Y. This means that in only two locations bilateral trade with China Tc can be used as a reliable predictor for output Y and unemployment U. However, this is still sufficient to further test the hypotheses as Granger causality relies on predictive power rather than actual causality.
18
6.2 Impulse Response Function Graphs (IRFs)
The IRFs are the most important analyzing tool in this paper. IRFs introduce a hypothetical shock of one standard deviation and then measure the response of a moving average of the variables to that hypothetical one standard deviation shock in Tc and Tus, using the data over the period 2002-2019. The reaction of the variables to this shock can be used to estimate the damages caused by the SARS epidemic using the trade data from the OECD. The variables in which the shocks occur are Tc, which encapsulates the direct effects, and Tus through a shock in Tc, which encapsulate the indirect effects. Before reading the graphs below there are several things of note: the x-axis represents the number of periods and the y-axis represents the response. Since unemployment U is measured in percentages the response of U is in percentage points, whereas the response of the other variables is in coefficients between 0 and 1.
Because of the focus of this paper, the responses to shocks in U, Y and P are left out. This chapter is divided into two separate parts: direct effects in which we measure the effects of CD (1), written in equation (4), and indirect effects in which we measure the effects of CD (2), written in equation (5).
The magnitude of a shock of one standard deviation, as well as the actual shock of SARS is given in the following table:
Table 8: Shock values Hypothetical shock in value Hypothetical shock in % SARS-shock in value SARS shock in %
Tc Tus Tc Tus Tc Tus Tc Tus
EU 0.4904 0.2241 2.004% 0.944% 0.1232 0.1321 0.537% 0.529%
Italy 0.4107 0.2585 1.901% 1.180% 0.2181 0.1734 1.053% 0.729% France 0.4426 0.2005 2.043% 0.904% 0.1632 0.1932 0.786% 0.870% Germany 0.5010 0.2371 2.197% 1.008% 0.1621 0.1598 0.774% 0.697% Netherlands 0.6037 0.2430 2.732% 1.102% 0.1124 0.1521 0.534% 0.718%
Sources: OECD trade data (2020). Notes: The value of shocks is given in natural logarithms, The SARS shock is measured as the loss of trade during the period of Q4, 2002-Q2, 2003. The percentage of the shock is calculated setting the trade value of November 2002 as 100%.
Table 8 shows the hypothetical shock introduced in the model, as well as the actual shock in these variables during the SARS epidemic. One thing of note in Table 8 is that the actual shock due to the SARS epidemic was much smaller than the hypothetical introduced shock. This means that the measured effects from the IRFs are in general three to four times as large as the shock due to SARS. An interesting thing about this hypothetical shock is that it can later be extended to estimate the damages caused by the trade shocks of Covid-19.
6.2.1 Direct effects
To briefly recap chapter 2: Direct effects are measured the direct economic losses in the EU from trade shocks using CD (1) and follows the causal chain еTct → еYt → eUt → eTust → ePt.
This means that the response of the economic variables Ut and Yt to a shock in Tct are the main
19
6.2.1 (1) (Log) levels model
The response function for the (log) levels model can be found in Figure 4 below. The figure represents the responses of the endogenous variables measuring economic performance, Y, U and P, as well as the response of trade with the U.S. to a shock in trade with China eTc.
Figure 4: Impulse Response Functions EU from CD (1) in (log) levels
Response to Cholesky One S.D. (d.f. adjusted) Innovations +2 S.E.
Notes: The blue line indicates the IRF, the orange dotted lines represent the 95% confidence interval. The x-axis represents the number of periods (1 period = 1 month), the y-axis represents the response coefficients.
Looking at Figure 4, there is a lot to unpack. The results of the IRFs for the European Union are the four graphs in the first column in Figure 4. We see that the response of economic output YEU to TcEU is not very large at first, only after roughly 10 periods we see that the YEU start to respond to TcEU. After roughly 15 periods, the effect peaks at the 0.0194 (=1.94%) coefficient. After around 40 periods this effect hits the ‘steady state’ of the x-axis, which means that the effect of the shock has decreased to zero. This means that the effects of trade shocks with China on European output YEU are present for an incredibly long time and disappear fully after roughly 4 years. Unemployment UEU takes an even longer time to converge to zero. Looking at graph 2 of column 1, we see that the response of UEU to a shock in TcEU also peaks at around 15 periods at 0.29% after which it reverts back to the steady state after 75 periods. This means that for the EU, a 2.004% shock in Tc leads to a 1.94% change in economic output Y and a 0.29% change in unemployment U.
Periods;25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100
Periods;25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100 Periods;25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100
Periods;25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100
Response YEU to TcEU Response YITA to TcITA Response YFRA to TcFRA Response YGER to TcGER Response YNL to TcNL
Response TusEU to TcEU Response TusITA to TcITA Response TusFRA to TcFRA Response TusGER to TcGER Response TusNL to TcNL Response UEU to TcEU Response UITA to TcITA Response UFRA to TcFRA Response UGER to TcGER Response UNL to TcNL
20
For Italy, these effects appear to be much different than for the EU. The results for Italy can be found the second column of graphs in Figure 4. Starting with the response of economic output YITA to TcITA, the first thing of note here is that the sign of the coefficient changes after around 20 periods, going from a positive relationship to an inverse relationship. This means that after roughly 20 periods, a negative shock in Tc will lead to an increase for Italian output Y. Before the change of the direction of the effect of the shock on Y, it peaked at the coefficient 0.008 (=0.8%). After the change of the direction it peaked at 0.012(=1.2%). The effects seem to disappear completely after a staggering 55 periods, which means that after 55 months the effects of the shock disappear entirely. The inverse relationship between Italian economic performance and a shock in Tc is further supported by the response of Italian unemployment UITA to TcITA. We see that unemployment responds positively at first, peaking at 0.197%, but then the direction changes after 25 periods and unemployment peaks at the 0.247% mark after roughly 35 periods. After 60 periods, a staggering five years, the effects seem to disappear fully. This means that for Italy, a 1.901% shock in Tc leads to a small negative change in Y at first, but after two years this effect becomes positive, leading to a 1.2% increase in output Y. A similar effect occurs in unemployment U, in which the shock in Tc leads to a 0.247% decrease in unemployment U.
The responses of the French economy can be found in the third column of graphs in Figure 4. French output Y and -unemployment U seem to respond very similarly to a shock in TcFRA as the European output and unemployment did to a shock in TcEU. The response of French output YFRA peaks after 10 periods at the 0.0321 (=3.21%) coefficient, after which the effects quickly disappear. After two years the direction of the effect changes, leading to a slight increase in GDP until around 40 periods, but this effect can be due to the endogenous nature of economic output. The effects of a shock in Tc on French unemployment UFRA peak after roughly 15 periods at 0.389% and disappear fully after 40 periods. This means that for France, a 2.043% shock in Tc leads to a staggering 3.21% change in economic output Y and a 0.389% change in unemployment U. These numbers are surprisingly large.
The peaks of the responses of the German variables measuring economic performance seem to be much steeper than those of the other locations. The results for the German economy can be found in the fourth column of graphs in Figure 4. The response of German output YGER also takes roughly 10 periods to fully occur, after which it quickly peaks at 15 periods at the 0.0199 (=1.99%) coefficient. After 30 periods, the effect returns to zero, after which it has some minor fluctuations around the 0 line. German unemployment UGER responds immediately to the shock, peaking at 0.2% change in unemployment after 20 periods. This effect returns to zero, disappearing completely, after 48 periods. This means that a 2.197% shock in TcGER leads to a 1.99% change in German output Y and a 0.2% change in German unemployment U.
21
6.2.1 (2) First differences model
To test whether results differ substantially per measurement method, the VAR is ran in first differences assuming the same Cholesky’s decomposition CD (1). First differences assumes all variables to be I (1) for all variables except unemployment U. Recall from chapter 2 that U is assumed to be stationary and its effects will always return to zero in the long run (Papell, Murray & Ghiblawi 2000; Khraief et al. 2017). The first differences model serves as a robustness check, to find out whether the effects measured differ substantially depending on the measurement method used. This is why its results are discussed briefly, only substantial inconsistencies between the first differences- and the (log) levels model are discussed thoroughly.
Figure 5: Impulse Response Functions of EU from CD (1) in first differences
Response to Cholesky One S.D. (d.f. adjusted) Innovations +2 S.E.
Notes: The blue line indicates the IRF, the orange dotted lines represent the 95% confidence interval. The x-axis represents the number of periods (1 period = 1 month), the y-axis represents the response coefficients.
There are a two minor things of note in Figure 5, the first being that the response of Italian unemployment UITA to a shock in ΔTcITA has a different direction than in the (log) levels model. This could be because of the use of percentages for unemployment U, rather than first differences. This can also be seen in the duration of the effect of shocks in Tc on unemployment U, in all of the five locations the response takes an incredibly long time to converge to zero, after 50 periods there barely seems to be a dent in the duration of the shock. Apart from these effects, the general direction of the responses of each variable to a shock in Tc seem to not diverge much from the (log) levels model.
5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 4550 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 4550 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 4550 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 4550
Response ΔYEU to ΔTcEU Response ΔYITA to ΔTcITA Response ΔYFRA ΔTcFRA Response ΔYGER to ΔTcGER Response ΔYNL to ΔTcNL
Response UEU to ΔTcEU Response UITA to ΔTcITA Response UFRA ΔTcFRA Response UGER to ΔTcGER Response UNL to ΔTcNL
Response ΔTusEU to ΔTcEU Response ΔTusITA to ΔTcITA Response ΔTusFRA ΔTcFRA Response ΔTusGER to ΔTcGER Response ΔTusNL to ΔTcNL
22
6.2.2 Indirect effects
To briefly recap chapter 2 on measuring indirect effects: Indirect effects are defined as ‘spillover effects’ from trade shocks using CD (2), found in equation (5). CD (2) follows the causal chain of shocks: еTct → eTust → еUt → eYt → ePt. This means that, when analyzing the
indirect effects, the trade shock еTct is then responsible for eTust which then affects the
economic performance of the European Union as a whole and –of its individual member states. For this reason, the responses of Tus and Tc are left out of the analysis of the indirect effects. The (log) levels model and the first differences of both models are analyzed, with the first differences serving as a robustness check to see whether there are any substantial inconsistencies between the two methods of analysis.
6.2.2 (1) (Log) levels models
Figure 6: Impulse Response Functions of EU from CD (2) in (log) levels
Response to Cholesky One S.D. (d.f. adjusted) Innovations +2 S.E.
Notes: The blue line indicates the IRF, the orange dotted lines represent the 95% confidence interval. The x-axis represents the number of periods (1 period = 1 month), the y-axis represents the response coefficients.
The first thing of note in Figure 6 is the response the price variable P in each of the locations. In every one of the locations, except Germany, there seems to be little to no response in prices P to a shock in bilateral trade with the U.S. Tus. The problem with the shocks that occur in prices for Germany is that they take an incredibly long time to converge to zero, which could be expected considering the problems with its stationarity (Table 3).
Starting with the responses of Output Y and unemployment U of the European Union to a shock in Tus, which can be found in the first column of Figure 6. We see that the indirect effects, measured as responses to a shock in Tus, are much smaller than the direct effects of eTc. The response of European economic output YEU peaks at 0.067 (=0.67%) and disappears, after one fluctuation at 25 periods, completely after 40 periods. The disappearance of the effects on
Response YEU to TusEU Response YITA to TusITA Response YFRA to TusFRA Response YGER to TusGER Response YNL to TusNL
Response UEU to TusEU Response UITA to TusITA Response UFRA to TusFRA Response UGER to TusGER Response UNL to TusNL
Response PEU to TusEU Response PITA to TusITA Response PFRA to TusFRA Response UGER to TusGER Response PNL to TusNL
Periods;25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100 Periods;25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100
23
European unemployment UEU is much slower, with its effects peaking after 50 periods at 0.21%. This means that a shock in Tus, due to a shock in Tc, of 0.944% leads to a change in European output Y of 0.67%, which disappear completely after 40 periods, and a change in European unemployment U of 0.21%, which takes an incredibly long time to disappear completely (roughly 80 periods).
The indirect effects on the Italian economy are, unsurprisingly, also very minor. There does seem to be some issues regarding stationarity, since both the effects on unemployment U and output Y take an incredibly long time to converge to zero (>100 periods). This could also mean that the indirect effects on the Italian economy are very long-lasting however, since the Dicky-Fuller GLP tests regarding stationarity did not find any stationarity issues with Y and U. Another thing of note is that the direction of the responses of Y and U to the shock in Tus are in the opposite direction of the responses to a shock in Tc. The response of output YITA to a shock in Tus peaks at 0.0034 (=0.34%) and does not disappear within 100 periods, where-as the response of unemployment UITA seems to peak after 85 periods at 0.33%, since a small decline after 80 periods can be observed. This means that, if the Dicky-Fuller GLP tests regarding the stationarity tests were correct, that a 1.180% shock in Tus due to eTc leads an change of 0.34% in output Y and a 0.33% in unemployment U, both taking an incredible amount of time to converge to zero (>100 periods).
The French economy barely seems to respond at all. In the third column of graphs in Figure 6 we see a very minor dip of 0.0014 (=0.14%) in French economic output YFRA to a response in eTus after 5 periods. This response quickly inverses however, peaking at 0.0009 (=0.09%) at 20 periods after it gradually declines to zero at 45 periods. French Unemployment UFRA also responds very weakly, peaking at 0.12% at the 40th period, after which it gradually declines. The same issue appears to occur with French unemployment as with Italian Unemployment, as in that it takes an unbelievable amount of time to converge to zero, hinting at possible stationarity issues. No stationarity issues regarding unemployment and output were found in the Dicky-Fuller GLP tests, so the variables are assumed to be stationary. This means that a 0.904% shock in Tus leads to a 0.14% growth of output Y at five periods, after which it quickly shifts to a 0.09% decline in output Y, with its effects disappearing completely after 45 periods. Unemployment U changes by 0.12% to eTus and its effects take an incredible amount of time to converge to zero.
24
The Dutch economy seems to suffer stronger indirect effects than they did direct effects. Looking at the fifth column of Figure 6, we can see that Dutch output YNL reacts inversely to the direction of the shock in eTus, peaking at 0.004 (=0.4%) after 60 periods. It does appear that the effects linger on for a long time, since the Dicky-Fuller GLP tests show no stationarity issues at p<0.01, since they do not converge back to zero within 100 periods. The same issue regarding stationarity appears to occur for the effects on UNL, since these also do not converge back to zero within 100 periods. They do not even seem to peak within this period. Looking at the Dicky-Fuller GLP tests in appendix A5 we see that there appears to be some stationarity issues regarding Dutch unemployment U. This means that a 1.102% shock in Tus leads to a 0.4% change in output Y, which disappears incredibly slowly. No reliable measure of the indirect effects on Dutch unemployment U was found.
6.2.2 (2) First differences model
Similarly to the first differences model measuring the direct effects, the first differences model measuring the indirect effects is used as a robustness check whether any substantial inconsistencies are found between the two differing measurement methods.
The results of the IRFs from CD (2) measured in first differences is found in Figure 7 below: Figure 7: Impulse Response Functions of EU from CD (1) in first differences
Response to Cholesky One S.D. (d.f. adjusted) Innovations +2 S.E.
Notes: The blue line indicates the IRF, the orange dotted lines represent the 95% confidence interval. The x-axis represents the number of periods (1 period = 1 month), the y-axis represents the response coefficients.
There do not appear to be any substantial differences in the direction of the effects of eTus between the first differences and (log) levels. A thing of note is that the same issues regarding the stationarity of the effects of eTus on Dutch- and Italian unemployment arise, both not converging back to zero within 50 periods.
5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 4550 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 4550
5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 4550
Response ΔYEU to ΔTusEU Response ΔYITA to ΔTusITA Response ΔYFRA to ΔTusFRA Response ΔYGER to ΔTusGER Response ΔYNL to ΔTusNL
Response UEU to ΔTusEU Response UITA to ΔTusITA Response UFRA to ΔTusFRA Response UGER to ΔTusGER Response UNL to ΔTusNL
25
6.2.3 Impulse Response Functions (IRFs) results and SARS damages
The main takeaways from the IRFs is that in most locations the direction of the shock in Tc directly impacts the growth of economic performance Y and that eTc has an inverse relationship with the unemployment rate U. This is not true for Italy however, as it seems that Italy’s economic performance Y has an inverse relationship with the exogenous shock eTc, likely due to the competitiveness of Italian and Chinese export products (WTO, 2020). The effects of eTus are much smaller than the direct effects of a shock in Tc. This supports the idea that spillover effects tend to be much smaller than the direct effects. However, for unemployment U there appeared to be some stationarity concerns, as the effects of the shock in Tus took an incredible amount of time to converge back to zero.
If using the numbers of the SARS-shock from Table 8, the damages from direct- and indirect effects of the SARS epidemic on European economic performance are as follows:
Table 9: IRF takeaways for the SARS epidemic
Direct effects Indirect effects
EU The SARS shock was channeled
through a 0.537% loss in Tc, resulting in an output loss in Y of 0.52% and a rise in unemployment of 0.08%. The effects disappear after 75 periods.
The indirect effects channeled through a 0.529% loss in Tus resulted in a loss of output Y of 0.37% and an increase in unemployment U of 0.12%. These effects disappear after 80 periods.
Italy The SARS shock channeled through a 1.053% loss in Tc resulted in an overall increase of output Y of 0.66% and a decrease in unemployment U of 0.14%. The effects disappeared after 60 periods.
The indirect effects channeled through a 0.729% loss in Tus resulted in a decrease of output Y of 0.21% and an increase of unemployment U of 0.20%. The effects do not disappear after 100 periods.
France The SARS shock channeled through a 0.786% loss in Tc resulted in an overall decrease of output Y of 1.23% and an increase in unemployment U of 0.15%. These effects disappeared after 40 periods.
The indirect effects channeled through a 0.87% loss in Tus resulted in an initial increase in output Y of 0.13%, after which it decreased with 0.09%. Unemployment U increased with 0.12%. Effects lasted 80 periods. Germany The SARS shock channeled through a
0.774% loss in Tc resulted in a 0.7% decrease in output Y and a 0.07% increase in unemployment U. These effects disappeared after 48 periods.
The indirect effects channeled through a 0.697% loss in Tus resulted in a 0.56% decrease in output Y. Unemployment U measures had stationarity issues.
The Netherlands
The SARS shock channeled through a 0.534% loss in Tc only resulted in a loss of output Y of 0.06% and an increase in unemployment U of 0.02%. The effects disappeared completely after 50 periods.
The indirect effects channeled through a 0.718% loss in Tus resulted in a 0.26% loss of output Y. Unemployment U measures suffered from non-stationarity.
