Financial contagion on the international trade network.

Kali, Raja ; Reyes, Javier

I. INTRODUCTION

Does integration into the international trade network make a country more vulnerable to financial crises? A priori, the answer to this question is not obvious. On the one hand, if a country has many distinct trading partners, then there are many pathways through which a shock originating somewhere else can reach it. On the other hand, if a country is hit by an adverse shock, then having many trading partners provides more channels to diversify away the impact. Participation in international trade can thus amplify or cushion the impact of an adverse shock to a country's financial markets. In this article, we attempt to answer this question through a new approach that links the transmission of financial crises to the network of international trade relationships. We suggest that a network approach capable of incorporating the cascading and diffusion of interdependent ripples--that happen when a shock hits a specific part of the global trade network provides us with an improved explanation of financial contagion.

Understanding such mechanisms is important given the dramatic increase in international economic integration or "globalization" that has characterized the past decade together with the increase in the volatility of country-level performance, reflected in several recent episodes of economic and financial crises. (1) One striking characteristic of many of these crises is how an initial country-specific event was rapidly transmitted to markets around the globe. These events sparked the widespread use of a new meaning for the term "contagion" and a surge of interest in the determinants of a country's vulnerability to crises that originate elsewhere in the world. But despite this interest, there continues to be little agreement on why many of these crises that began in relatively small economies had such global repercussions and why shocks originating in one economy affected some markets, while markets in other countries were relatively unaffected.

Part of the problem may be the focus on bilateral trade and finance relationships as channels for transmission. Recent advances in the study of networks (Albert and Barabasi 2002; Newman 2003) have placed elegant and powerful tools at our disposal, enabling us to turn from a focus on bilateral relationships to a consideration of the pattern of linkages that tie countries around the world together as a whole. Our first step is thus to combine a network approach with data on international trade linkages in order to map the global trading system as an interdependent complex network, (2) with the countries as nodes and trade relationships as links between them.

Countries tied through trade are concurrently linked through credit arrangements as well as pure financial flows. Several influential articles (Glick and Rose 1999; Kaminsky and Reinhart 2000; Van Rijckeghem and Weder 2001) note that it is extremely difficult to disentangle these linkages, and estimates of the importance of trade linkages may actually capture the impact of financial linkages and vice versa. Our goal in this article is thus not to separate out trade flows from financial flows but rather to examine the extent to which the structure of the international trade network can explain the transmission of crises. Underlying this approach is the assumption that the structure of the international trade network is a proxy for meaningful economic linkages between countries. In this spirit, we take the network of international trade linkages to be the backbone that underpins and motivates trade and financial flows of various kinds between countries. We use trade flows to chart the structure of the network because these statistics are widely available and consistently measured across countries, as well as being highly correlated with other cross-country linkages. (3) In addition, using trade flow data allows us to construct a complete global network of linkages consisting of 182 countries. (4)

Since exports and imports are mirror images of each other, in principle, we could construct a trade network using either. However, for our purposes, we define links based on export shares since exporting countries are recipients of payments for their exports, while importing countries are sources of payments for their imports. In the context of financial contagion, this makes it possible to examine the vulnerability of exporting countries to shocks from importing countries. Thus, the links in our network are directed in terms of the flow of cash from importing to exporting countries. We explain our data and the construction of the network in detail in Section III.

Considering the pattern of international trade linkages explicitly as a network enables us to obtain indicators of country-level "integration" measuring how well connected a country is into the global trading system. The core of our empirical analysis is based on three distinct network measures of country connectedness into the global trading system. The first, known as node importance, is an index of network dependency in which nodes are defined as more important if other nodes depend more on them and if the other nodes depending on them are themselves important. For the international trade network, exporting countries depend on the importing ones for revenues. We therefore use export shares to construct a dependency matrix. A country that is more important according to this measure is likely to have a greater influence on the network if it is affected by an adverse shock such as a financial crisis. A second indicator is node centrality. Node centrality measures how central a given node is by measuring how "star-like" a node is relative to a perfect star, a node that is linked to every other. This can thus be considered an indicator of how well connected a country is into the international trade network. A third measure is maximum flow, which counts the number of paths, direct and indirect, that lead from a particular country (say country i) to another country (say country j). In the context of financial contagion, this provides a measure of the number of paths via which a shock from a crisis "epicenter" country can potentially travel through the network to impact a "target" country, directly and indirectly through intermediate countries. We describe these measures in more detail in Section II.

We examine how these network-based measures of country-level integration into the global trading system perform in explaining stock market returns during several recent crisis episodes. (5) Since our goal is to isolate the impact of network interconnections, our empirical strategy focuses on five nonoverlapping periods of financial crisis (6) in order to prevent, to the extent possible, our results being "contaminated" by continuous turmoil in financial markets. We find that the adverse impact of a crisis on the stock market of a country is amplified if the epicenter country for a crisis is better connected according to these network measures. On the other hand, the impact of a crisis on any target country is dampened if the country in question is better connected into the trade network. In other words, a better connected international trading network is a double-edged sword. Based on this line of reasoning, a network approach can help explain why the Tequila crisis, the Asian flu, and the Russian virus were highly contagious crises, while the financial crises that originated in Venezuela and Argentina did not have such a virulent effect. We elaborate on this in Section IV.

The rest of the article is structured as follows. Section II provides an overview of the recent debate on transmission channels for financial crises and explains the intuition behind the network approach and the network measures we use in our empirical strategy. Section III describes our empirical methodology and data. Section IV contains our principal empirical results. Section V provides robustness checks on our results. We conclude with Section VI.

II. A NETWORK APPROACH TO FINANCIAL CONTAGION

A poor understanding of the transmission of economic and financial crises has in the past few years prompted a surge of interest in international economic integration and its relationship to international financial contagion. The debate on the relative importance of trade linkages versus financial flows continues to be unresolved. A number of recent articles (Bae, Karolyi, and Stulz 2003; Kaminsky and Reinhart 2000, 2003) have emphasized the importance of financial sector links in the propagation of crises across countries, while others (Abeysinghe and Forbes 2002; Claessens and Forbes 2001; Forbes 2001; Forbes and Rigobon 2002) have stressed the importance of trade linkages. However, both strands of research note that it is difficult to separate the two because most countries that are linked in trade are also linked in finance.

The recent literature examining the role of trade in the transmission of crises has yet to reach a consensus. As a first pass at the issue, aggregate measures of bilateral trade do not seem to perform well in explaining crisis transmission (Kaminsky and Reinhart 2000, 2003). However, the use of such aggregate measures masks the fact that trade can be parsed into several different and possibly counteracting channels, such as a competitiveness effect (when changes in relative prices affect a country's ability to compete abroad), an income effect (when a crisis affects incomes and the demand for imports), and a cheap-import effect (when a crisis reduces import prices for a trading partner and acts as a positive supply shock), which are found to be significant in explaining variation in stock market returns during recent crises (Forbes 2001). In a similar vein, research using new and more detailed data that decompose stock market returns into various factors (global, sectoral, and cross-country) finds that direct trade linkages appear to be the strongest and most important determinant of cross-country linkages in both stock and bond markets (Forbes and Chinn 2004). The importance of trade linkages has been further emphasized in a recent investigation by Abeysinghe and Forbes (2002), who use a structural vector auto regression (VAR) model to find that even weak trade linkages can transmit shocks across countries quite powerfully through indirect multiplier effects on output and growth. (7)

Nevertheless, one consideration that is conspicuously absent from the current literature is the idea that since countries are interconnected through trade linkages, a shock originating in one country can be transmitted and amplified because of the pattern of interconnections in the network. Consider an example with three countries, A, B, and C as depicted in Figure 1 below. First, assume that A has trade links with B and C but B and C do not trade with each other. To fix ideas, suppose that the linkages embody trade credit relationships derived from trade in intermediate inputs. In other words, suppose firms in A have bought intermediate inputs from firms in B and C and have paid only a fraction of the price as a down payment on delivery. The remainder of the price will be paid on realization of sales of the final product. It could also be that firms in A (either the same firms or the different firms) have provided intermediate inputs to firms in B and C on similar terms. These kinds of trade credit relationships are conventional and ubiquitous in international trade. In this situation, if a negative shock hits country A, firms in A may be forced to default on their input suppliers in B and C. Thus, a shock originating in country A will have an impact on B and C, and any effects on B and C will "boomerang" back to A only through separate bilateral links. This would be the case if the linkages are as depicted in the left-hand panel in Figure 1. If, however, B and C are also trading partners, then any direct effect of the shock from A to B will ripple through to C and vice versa. These "second-order" effects will in turn affect A. This cascade of interdependent ripples may amplify the magnitude and duration of the initial shock to A. This would be the case if the linkages are as depicted in the right-hand panel in Figure 1.

The structure of interlinkages could also play a role in dampening the impact of a shock originating in some other country. Suppose again that A is the epicenter of an initial shock. If B is not linked to countries other than A, then B has to absorb the full brunt of the shock that is transmitted from A through its link with A. However, if B trades with other countries, then it may be able to spread the impact of its shock among its trading partners, thereby cushioning the blow. Again, in a trade credit context, if A defaults on an outstanding balance to B and B is an intermediate link in a credit network, B can in turn default on a proportional basis on its trade creditors. But if B is the final node in the chain, then B has to absorb the full effect of A's default by itself. It is a short step from here to argue that the denser the network of linkages, the greater the potential for such interdependent ripple effects that could potentially amplify or dampen a shock. The incorporation of such considerations may go toward clarifying some of the puzzles regarding transmission through trade linkages and asymmetric effects on different countries that we described earlier.

[FIGURE 1 OMITTED]

A. Network Measures

A network is a set of points, called nodes or vertices, with connections between them, called links or edges. In our context, each country is considered to be a node of the network. Since international trade is usually measured using the monetary value of exports and imports between countries, trading relationships are analogous to valued links in a network, and these vary from country to country. In order to chart the structure of the network for the present purposes, we wish to take into account the magnitude of these relationships but not specifically their exact value. Additionally, as previously discussed, we consider directed trade links based on the cash flows from importing to exporting countries.

We do this by considering a network link between two countries to be present if the value of the link between them is above a certain threshold. Specifically, we define a link between country i and country j to be present if the value of exports from country i to country j as a proportion of country i's total exports is greater than or equal to a given magnitude. Thus, if country i's exports to country j, out of the total exports of country i, are greater or equal to a given threshold, then the link j [right arrow] i is defined as present. Since trade levels vary considerably from country to country and there could be debate over what constitutes "meaningful" levels of trade, we construct the network and associated network measures (described below) for several different values of the trade-link threshold. The specific thresholds we use are 0%, 1%, and 2%. The 0% threshold indicates the mere existence of trade among two countries, and in this sense, it is the least restrictive threshold. It simply acknowledges the presence of positive trade. We choose the 1% and 2% thresholds because 88% of the trade shares in 1992 (89% in 2000) are between 0% and 1%, and this number increases to 92% when the range between 0% and 2% is considered for the 1992 data (and the same is observed in 2000). Therefore, we could say that these thresholds are close to embodying meaningful or representative trade.

By examining how the structure of the network changes as the trade threshold used to define the presence of links varies, we are able to understand the sensitivity of the country-level network connectedness measures to differing trade magnitudes. Constructing the network for different thresholds thus enables us to incorporate both magnitudes and network features and adds to the robustness of our analysis.

For a network with a small number of nodes, it is easy and appealing to use a graphical representation. However, as the number of vertices increases, graphical analysis of the network becomes more difficult, and its characteristics and patterns are hard to identify through the naked eye, and statistical analysis using the network represented as a matrix becomes indispensable. Recent advances in complex network analysis provide us with a number of measures that we can use to understand how well connected an individual country is into the network and assess the level of influence that it has on other countries (nodes) and on the network as a whole (Hanneman 2001). These indicators, which incorporate information on volumes of trade, also take into account the number of trading partners, the position in the network of the country in question, and the degree of influence that a country has on others. Our empirical strategy uses the following network measures. (8)

Node Degree Centrality. Just as nodes and links are the basic components of any network, node degree is the basic component of complex network analysis. The degree is the number of links connected to a given node. The number of in- and outbound links will ultimately determine the connectivity of an individual node, but there are different ways in which this connectivity can be measured. The simplest of these measures is node degree centrality. The degree centrality of an individual node can be simply represented by its degree, but a more standard way is to normalize the individual node centrality in the following fashion:

(1) [C.sub.D](i) = [d.sub.i]/g - 1,

where g denotes the number of nodes in the network and [d.sub.i] stands for the degree (number of links) of node i. This index shows which countries are at the core or close to the core of the network. If a country is at the core of the network, then its node degree centrality will be close to 1. For a periphery country, this number will be close to 0, given that the number of international trade linkages of such a country will be relatively small.

We calculate node centrality for all the countries in our data set (9) and present the top 30 countries ranked by this measure in Table 1. The indices for the whole sample of countries (182) for the years 1992 through 2000 are available in a data appendix from the authors. We find that countries such as Brazil, South Korea, Indonesia, Malaysia, the Russian Federation, Thailand, and Turkey are among the top 35 (some in the top 15) most central countries in the international trade network. This is especially noteworthy since these countries have been at the epicenter of several financial crises and contagion episodes of the 1990s. This is suggestive of the importance of international trade linkages for financial contagion.

In the table, we also provide country rankings according to the ratio of total trade (exports plus imports) to gross domestic product (GDP), the standard trade openness measure used in the literature, in order to compare with our network measures. It is noteworthy that country rankings according to the node centrality are quite different from those obtained when countries are ranked according to the ratio of total trade (exports plus imports) to GDP. This is the case with our other network measures (described below) as well, suggesting that a network approach embodies different information than standard measures of openness used thus far in the literature.

Node Influence or Importance. Node degree centrality provides a preliminary approach to the identification of influential nodes. It is based on the number of countries that can be reached through direct links by an individual country. But it misses important features of the international trade network. The number of trading partners is a relevant statistic, but the specific characteristics of these trading partners may amplify or dampen the influence that a specific country has on others and on the whole network. One could say that it is not only the quantity of your partners that matters for influence but also how influential they are in turn. If country A trades with country B and B trades with 50 other countries, then A exerts indirect influence on these 50 countries.

In a prominent article, Salancik (1986) argues that "accurate assessments of the structural power of several interdependent parties are hampered by the fact that parties depend on one another indirectly as well as directly and that any one's dependencies are not equally important for all parties." He goes on to propose an index for dependency networks in which nodes are defined as more important if other nodes depend more on them and if the other nodes depending on them are themselves important. Applying his index to our context, the importance of country i is a function of the dependence of other nodes on i and the importance of these other nodes.

(2) [imp.sub.i] = [[summation].sub.j][dep.sub.ij][imp.sub.j] + [int.sub.j] for all j [not equal to] i,

where [imp.sub.i] is the importance of country i, [dep.sub.ij] is the extent to which country i is depended on by country j, and [int.sub.i] denotes the intrinsic value of country i. Equation (2), which represents a system of n equations corresponding to the number of countries, determines that if a country is not depended on by other countries, then it will be unimportant. Also, if a country is depended on only by unimportant countries, then it would also be considered unimportant.

Equation (2) can be rewritten in matrix form As:

(3) [IMP.sub.i] = [[D].sub.ij][IMP.sub.i] + [INT.sub.i],

where [[D].sub.ij] denotes the matrix of dependencies of each country j on each country i. For the international trade network, exporting countries depend on importing ones. Therefore, we use as the elements of [[D].sub.ij] the share of exports of country j to country i out of the total exports of country j. We compute two different node importance indices. First, we use the world trade share of a country as a measure of its intrinsic importance, and second, we consider the case where the intrinsic importance is determined by the node degree centrality value for the 1% trade-link threshold (as defined above). (10) After calibrating the elements of [[D].sub.ij] and [INT.sub.i] in this way, we then solve the resulting system of linear equations for the n x 1 vector of importance. Note that since the matrix of dependencies uses the raw trade share data instead of a binary matrix based on trade-link thresholds, the node importance measure is free of threshold considerations.

Table 1 also shows the rankings for the top 35 countries according to the node importance using world trade share as the intrinsic value, but the indices for all 182 countries are available in a data appendix from the authors.

Maximum Flow. Another way to measure the influence that a given country has on others is by counting the number of direct and indirect ways in which a shock to the country in turn affects others. When a particular country is hit by a shock that forces a reduction of its domestic demand for foreign goods, this translates into a diverse number of direct and indirect hits on a number of countries in the international network. A reduced demand for foreign goods in country A translates into a reduced cash flow for country B, given that this country's exports to A have decreased. In turn, country B's domestic demand for foreign exports from country C may drop, and this implies an indirect effect of a shock to A on country C. Given the size and structure of the international trade network, shocks to one country translate into a cascade of direct and indirect effects on other countries. A meaningful indicator would be one that captures the number of times, directly and indirectly, that a shock from country A translates into effects on all the other members of the network. Such a measure has been used in complex network analysis and has been termed maximum flow (Hanneman 2001).

The maximum flow measures the number of pathways to a target through nodes in the neighborhood of a source. The neighborhood refers to the nodes to which a particular source node is directly linked. The maximum flow is thus a count of all pathways from a source node to a target node via nodes to which the source is directly connected. If a shock to country A can only be transmitted to country C through country B, then the connection of country A is weak (even if B affects C through many other countries). On the other hand, if a shock to A is transmitted to C via three different countries (each of which has one or more ways in which the shock is transmitted to C), then A's connection is stronger.

Table 1 shows the top 35 countries ranked according to this methodology as well. It is instructive to use maximum flow to look at specific cases, such as the countries involved in the contagion that started in 1997-1998 with the crisis of Thailand and ended with the collapse of Argentina in 2002. Thailand devalued its currency in July 1997. This event ignited the east Asian crisis in which Indonesia, Malaysia, and Korea were the main participants. The contagion effect of these events reached Russia and Brazil in 1998 when the ruble and the real were devalued and flexible exchange rate regimes were adopted in both countries. Thereafter, Ecuador, Turkey, and Argentina collapsed in 1999, 1999, and 2001, respectively.

From the maximum flow matrix, it is possible to identify the countries that would be more severely affected, directly or indirectly, by a shock to a given country. Table 2 lists the top 35 countries, and the number of paths leading to them, that would be affected by shocks to Mexico (1994), Venezuela (1995), Thailand (1997), Russia (1998), Argentina (2000), and Brazil (1998) (11) according to the maximum flow measure at the 1% trade-link threshold. These results should be interpreted as follows. For example, for the column entitled Mexico (1994), the numbers in this column represent the number of different paths that lead from Mexico to a given country, while those for Venezuela (1995) refer to the number of paths that lead from Venezuela to a given country. The fact that these numbers are different is not a result of the crisis; it is the result of the different node degree of each of the epicenter countries (Mexico in 1994 and Venezuela in 1995) and also of the node degree of the countries to which these countries are connected. The higher numbers observed for Mexico imply that Mexico affects these countries through many more channels than Venezuela. Notice also that the countries affected by each epicenter are different since they have different trading partners and therefore have different effects on different countries. The table shows that it is possible to identify Brazil, Indonesia, Korea, and Russia as candidate countries for severe aftershocks of the Thai currency devaluation. The maximum flow is thus valuable for a bird's-eye view of financial contagion.

However, while the maximum flow is an instructive measure, it suffers from the drawback that the different paths leading from the epicenter country to country i are not distance weighted, and no assumptions are made with respect to marginal propensities to import. In other words, a direct path is considered as important as one that went through n other countries. Hence, it is important to keep this caveat in mind when considering this measure (12) and the regressions with this measure in our empirical analysis.

We would also like to note that all the network indicators are hardly correlated at all to the ratio of total trade to GDP, buttressing the point that the information conveyed by the network indicators is substantially different from that implied by the standard measure of trade openness.

III. METHODOLOGY AND DATA

The data we use for the computation of the network and associated measures were extracted from the NBER-UN world trade database. (13) We use the U.S. dollar value of exports and imports of all commodities between 182 countries for the years 1992 through 2000. Our trading relationships are based on the flow of payments between countries. Exporting countries are thus recipients of payments for their exports, while importing countries are sources of payments for their imports. This approach allows us to capture the influence of importing countries on exporting ones as influential buyers. Countries are considered nodes of the network and a link between them represents a trading relationship. We use the share of exports of country i to country j out of the total exports of country i to define the presence of a trade link between i and j and construct binary matrix representations of the international trade network for different magnitudes of this trade ratio. We thus construct the network and associated network measures for three different values of the trade-link threshold (0%, 1%, and 2%) with 182 countries for the years 1992 through 2000.

In order to examine how these measures perform in explaining stock market performance during recent periods of crisis (referred to as crisis windows), we follow the approaches of Rigobon (2003), Forbes (2001), and Glick and Rose (1998) to specify the time periods for our study. Rigobon (2003) and Glick and Rose (1998) base their studies on nonoverlapping crises windows in order to prevent, to the extent possible, the analysis from being contaminated by continuous turmoil in financial markets. Rigobon (2003) specifies three such windows: the Mexican crisis in 1994-1995, the Asian crisis in 1997-1998, and the Russian crisis in 1998-1999. The selection criteria used to determine these windows are based on the empirical fact of observed low-volatility (tranquil) periods prior to these crises, followed by high-volatility periods after a specific date. Glick and Rose (1998) consider five different crises, pinpointed by the occurrence of currency crises--the breakdown of the Bretton Woods system in 1971, collapse of the Smithsonian Agreement in 1973, the EMS crisis in 1992-1993, the Mexican meltdown in 1994-1995, and the Asian flu of 1997-1998. Forbes (2003) follows a more conventional approach and builds an "exchange market pressure index" that is a function of the observed variations of the exchange rate, interest rate, and international reserves of a given country. (14) Based on this index, for the period July 1, 1994, to June 31, 1999, she determines the occurrences of crises. Specifically, if the index of country i at time t exceeds a given threshold, then a crisis is considered to have occurred in period t. Three different thresholds are considered to identify crisis periods of varying intensity: [mu] [+ or -] 5[sigma], [mu] [+ or -] 3[sigma], and [mu] [+ or -] 1.5[sigma], where [mu] and [sigma] denote the mean and the standard deviation of the exchange market pressure index. Applying this methodology to the data results in the identification of 16 (sometimes overlapping) crises windows, which are listed in Table 3. Thus, following the lead of Rigobon (2003) and Glick and Rose (1998) regarding nonoverlapping crises, and the results obtained by Forbes (2001) through the analysis of the EMP, we select five nonoverlapping crisis windows for our analysis. The crisis windows we use for our analysis are also listed in Table 3.

Since our interests lie in investigating network effects using international trade linkages as the backbone for the analysis, we specify not only crisis windows but also an epicenter country for each crisis. The selection of the epicenter is not random. Rigobon (2003) traces the beginning of each high-volatility window to a specific country. This can be seen in Table 3, which shows the crisis windows identified by Rigobon (2003). Venezuela is chosen as the epicenter country of our second crisis window since in Forbes' (2001) study, it is the first [mu] [+ or -] 5[sigma] country after a relatively tranquil period that followed the Mexican crisis in 1994. (15) We also include the Argentinean crisis of 2001, a crisis not considered in the studies described above. In June 2001, the Argentinean government performed a debt swap of over 30 billion U.S. dollars' worth of bonds. Even though the swap was initially considered a success, within weeks the Argentinean markets, and others within the region, were in distress, and this forced Argentina to turn again to the International Monetary Fund (IMF). This marked the beginning of the end for the currency board in Argentina, a system that was removed entirely in January 2002 when the government announced a formal default on its debt.

The base econometric model we use for the subsequent analysis is the following:

[SMR.sub.i,w] = [[beta].sub.1][NIC.sub.i,w] + [[beta].sub.2][NIEC.sub.w] [K.summation over (k=1)][[gamma].sub.k [Z.sub.k,i,w] +[[epsilon].sub.i,w], (4)

where [SMR.sub.i,w] denotes the stock market performance indicator for country i during crisis window w, and [NIC.sub.i,w] and [NIEC.sub.w] represent the network indicator for country i and the network indicator for the epicenter country during crisis window w, respectively. The [Z.sub.k,i,w]s denote control variables, most of which are the standard macroeconomic determinants of currency, banking, and financial crises. Following Forbes (2003), these variables are measured during the year prior to the starting date of the crisis window considered. In addition, we also include a control variable for trade between the crisis epicenter country and the country i in order to account for a direct bilateral transmission channel between the two countries. A direct bilateral link is arguably a potent transmission channel, and we include it in order not to confound network effects with the direct effect that is due to trade with the epicenter. In such a framework, network effects, if obtained, should be considered especially relevant.

We are interested in explaining how country i's vulnerability (assessed by stock market performance during a particular crisis window) is affected, positively or negatively, by the network characteristics of country i, as well as those specific to the crisis epicenter country. Based on the intuition from the previous section, our priors are that the better connected the epicenter country is to the whole network, the greater will be the impact on the network and on country i of a shock to the epicenter country. Also, following the intuition presented earlier, we expect that a better connected country i will be less affected by shocks to other countries since it can dissipate the negative impact among its many (links) trade partners.

The stock market performance indicator is computed by calculating the deviations of the weekly average stock market return of country i during crisis window w from the weekly average of the stock market return of country i for the period between January 1992 (the first observation of our data set) and December 1994 (last observation before the Mexican crisis). This indicator can therefore be considered a measure of abnormal returns. Weekly stock market data from January 1992 to December 2002, extracted from Bloomberg, were used to compute the weekly average abnormal stock market returns for each crisis window. The resulting data set includes data on 41 countries. The reason for limiting the time period for the baseline stock market average (the mean from which the deviations are measured) from January 1992 through December 1994 is because this period is characterized by relatively stable macroeconomic and financial conditions throughout the world. Forbes (2003) uses a similar abnormal returns indicator, but she computes the deviations of stock market returns from the mean of the stock market in the previous year. We believe that although this is a good measure of abnormal returns for the Mexican crisis since the period preceding this event was relatively tranquil, it is not for the other crisis windows, since from 1995 onward the world was characterized by a wide array of crises, including balance of payments, exchange rates, and financial and banking crises. In Section V, as a robustness check, we replicate the analysis using a different definition of the abnormal returns.

The macroeconomic control variables considered for the estimation of the base model, Equation (4), are bank reserves to assets ratio, yearly inflation rate (CPI), GDP growth rate, current account balance, government expenditures, overall government budget balance, private capital flows, and total trade as percentage of GDP. These variables, or the data necessary for their computation, were extracted from the World Development Indicators (WDI) database. The bilateral link variable is trade with epicenter, measured as the ratio of trade between the country i and the epicenter to the total trade of country i. The data appendix presents a detailed explanation of these variables and lists the specific WDI codes. Merging the stock market data with the data for the macroeconomic control variables and network indicators leads to a panel that contains between 34 and 40 countries, depending on the control variables included, for five different periods (crisis windows). The list of countries and tables presenting descriptive statistics are in the Table A1 at the end of the article. We should clarify, however, that network measures of connectedness for each country in this sample are computed from the complete international trade network consisting of 182 countries. We believe that this makes for a more complete estimate of the impact of the crisis epicenter country and the ability of a target country to dissipate the shock. We should note that the time period covered for the analysis is 1992-2002, but the trade data are only available up to 2000. We therefore use the network indicators computed using the 2000 data for the analysis of the Argentinean crisis window (June 2001 to April 2002).

IV. RESULTS

We present the results for the estimation of the base model using the original definition for the weekly abnormal stock market returns, the trade-based network indicators, and all the control variables. In Section V, the robustness section, we present and discuss the results obtained for the cases where the GDP-based indicators and the alternative definition for the abnormal stock market returns are considered. A close examination of Equation (4) shows that with the inclusion of the network indicator of the epicenter, [NIEC.sub.w], the regression resembles a fixed time period effects model; therefore, there is no need to control for this type of effect in our estimation. On the other hand, there could very well be fixed cross-section effects. (16) We explore this possibility, but the results for the F and chi-square tests (for the redundant fixed effects-likelihood ratio test) cannot reject the null hypothesis of redundant fixed effects for any of the cases considered. We estimate the base model using ordinary least squares (OLS) and also through generalized least squares (GLS) exploring the possibility of cross-section-specific heteroskedasticity.

Table 4 presents the results of the estimations of the base model using two different network indicators, node importance and node centrality, and all the control variables. The results for node importance are presented for two cases, one where the world trade share of country i (total trade of country i out of total world trade) is used for intrinsic value (17) and the other where the node centrality at the 1% trade-link threshold for country i is used as the intrinsic value. For the case of node centrality, the table presents results for three different trade-link thresholds defining the network: 0%, 1%, and 2%. Odd-numbered columns present the results estimated through OLS, while even-numbered columns show the results obtained via GLS.

A first glance at Table 4 shows the strong statistical significance of the network indicators. The coefficients for the epicenter indicators are statistically significant across the board, except in Regression 9, and are all negative, while those for the network indicators of country i are of the fight sign (+) and statistically significant in six out of the ten regressions presented. More importantly, the economic significance of these variables is considerable. For example, the coefficient for the epicenter importance in Regression 2 is -44.88 (significant at the 1% level), implying that a one-standard deviation increase in epicenter importance will reduce the abnormal return (meaning it affects stock market return negatively) by almost 0.19 of a standard deviation. (18) Regarding the importance indicator of country i, the coefficient of this variable in Regression 2 is 3.91, implying that a one-standard deviation increase in country i's importance translates into an increase in the abnormal stock market return (i.e., it positively affects stock market performance) of 0.08 of a standard deviation. The economic magnitude of the effects of these changes on the weekly abnormal stock market return is a fall of 0.24 percentage points for the epicenter country indicator and an increase of 0.11 percentage points for the country i indicator. For the centrality indicators, using the results for the 1% threshold (i.e., Regression 8), the effect of a one-standard deviation increase in epicenter centrality leads to a decrease in the weekly abnormal stock return of 0.07 of a standard deviation or 0.09 percentage points. The increase of a one standard deviation of country i centrality results in an increase in the weekly abnormal return by 0.043 of a standard deviation or 0.055 percentage points. Considering the fact that the average duration for the crisis windows exceeds 10 wk, the total cumulative effects are substantial. (19) Instead of discussing the specific magnitudes for each of the coefficients for the network indicators for each regression, we show the estimated effects for all of them in Table 5. The results are consistent across the board.

The Tequila crisis, the Asian flu, and the Russian virus were highly contagious crises, while the financial crises that occurred in Venezuela and Argentina did not have such a virulent effect. The results presented in Table 6 suggest that network effects may be able to explain some of these differences between crises. Table 6, using the estimated coefficients of Regression 2 in Table 4 and the network importance indicator for the epicenter countries, shows that the epicenter network effect on other countries was substantially higher for the Tequila, Asian flu, and Russian virus crises than for the Venezuelan and Argentinean crises. (20) Table 6 also shows how better connected countries like the United States, Canada, and Italy can dampen the negative effects of shocks originating in other countries, while less connected countries like Ecuador, India, Venezuela, and others cannot. For example, during the Asian flu, the magnitude of the effect from the epicenter country, based on the results for the importance indicators that use world trade shares as intrinsic values, was a decrease in the abnormal weekly stock market return of 0.51 percentage points for all countries. This result is computed by multiplying the specific network indicator for the epicenter country in question times the estimated coefficient for the epicenter network indicator reported in the second column of Table 4. But the United States, since its importance indicator is high, dampens the effect by 0.58 percentage points, computed by multiplying the specific network indicator of the country in question (target country) times the coefficient for the country i network indicator reported in the second column of Table 4. In contrast, countries like Greece and Venezuela, on account of a low value of importance in the network, can only dampen this negative effect by 0.01 percentage points. (21)

The third network indicator presented in Section IIA was maximum flow. This particular indicator measures the influence that a given country has on the network by counting the number of direct and indirect ways in which a shock to the country in turn affects others. More specifically, it is possible to measure the number of paths through which a shock originating in country i can be transmitted to country j (i.e., the maximum flow from country i to country j). We use this measure to estimate the following specification of the base model:

[SMR.sub.i,w] = [[beta].sub.1]Maxfloe [(e to i).sub.i,w] + + + [K.summation over (k=1)][[gamma].sub.k [Z.sub.k,i,w] + [[epsilon].sub.i,w], (5)

where Maxflow(e to i) denotes the maximum flow measure from the epicenter country to target country i. Once again, we estimate this model using OLS and also through GLS, exploring the possibility of cross-section-specific heteroskedasticity. We wish to emphasize caution in using the results obtained through this network measure because maximum flow has the shortcoming that different paths leading from the epicenter country to country i are not distance weighted. In other words, a direct path is considered as important as one that went through n other countries. It is important to keep in mind this important caveat with this measure. Nevertheless, we include the results, in Table 7, as they may be instructive. Comparing these results to those obtained for the other two network indicators, node importance and centrality, we conclude that the signs coincide but the magnitude of the effects is lower.

Even though the results for all the network indicators for the epicenter and country i match in sign, the magnitude differs across them to some extent. This suggests that even though the network indicators are related in principle (i.e., are strongly correlated), they provide different information regarding the role for each country in the trade network. The importance indicator is a better measure of how relevant a country is within the network since it takes into consideration the structure of indirect connectivity between countries, while node centrality is much more based on direct connections. From this perspective, it is not surprising that the node importance indicators for the epicenter and country i with world trade share as the intrinsic value give the strongest results.

It is also worth noting that results for two distinct node importance indicators are presented in Table 4. The results obtained using the centrality of country i at the i% threshold as the intrinsic value of the node (columns 3 and 4) are similar to those obtained when world trade share of country i is used as the intrinsic value (columns 1 and 2). Recall from our description of node importance in Section II that it is free of threshold considerations since it is based on raw trade share data. This gives additional confidence in the measures based on trade-link thresholds.

V. ROBUSTNESS AND SENSITIVITY ANALYSIS

To check the robustness of the previous results, we present results for different specifications of the base model, Equation (4), and we also consider alternative definitions for the dependent variable (abnormal stock market returns) and the network indicators. Previously, we defined abnormal stock market returns as deviations of the average stock market return of country i during crisis window w from the average of the stock market return of country i for the period January 1992 (first observation of our data set) to December 1994 (last observation before the Mexican crisis). The definition considered here computes these deviations from the average of the stock market return of country i for the whole sample period (January 1992 to December 2002). Also, instead of the original dependency measure described above, that is, share of exports of country i to country j out of the total exports of country i, we use instead the value of the exports of country i to country j out of country i's GDP. (22) The modified dependency statistic can be thought of as an economy-wide dependency measure, that is, how country i's economic activity depends on the sales to country j. The GDP data used for these calculations were extracted from the WDI database of the World Bank. (23) In addition, we perform sensitivity checks for our results by considering subsets of our sample period for some of the alternative specifications considered for the robustness analysis.

The regression results presented in Table 4 use all the macroeconomic control variables to estimate the base model, Equation (4). Columns 1 through 6 in Table 8 show the results for the estimation of the base model using different subsets of the macroeconomic control variables. Regressions 1 through 3 use macroeconomic control variables that are usually associated with the analysis of financial and currency crises: bank reserves to assets ratio, current account deficit, private capital flows, and government expenditures, while Regressions 4 through 6 use the rest of the control variables. Henceforth, we refer to these two groups of regressions as specifications A (1 through 3) and B (4 through 6). We only report the results for the GLS methodology since the estimated coefficients for the network indicators are lower than those obtained through OLS, and this therefore keeps our reported results on the conservative side. (24) The results for the network indicators, using the original definition for the dependent variable, are robust across specifications (A and B). The statistical significance of the network indicators is very strong, and their economic significance is similar to that reported for the original specification of the base model. For example, using the results from Regression 1, the estimated coefficient for node importance, -22.9, implies that a one-standard deviation increase in epicenter node importance leads to a decrease in the abnormal stock return by 0.10 of a standard deviation or about 0.12 percentage points. Regarding the effects of country i's importance, the estimated coefficient is 4.32, meaning that a one-standard deviation increase in country i's importance translates into an increase in the abnormal stock market return of 0.09 of a standard deviation or 0.12 percentage points.

Similar conclusions apply for the results shown in columns 10 through 18 of Table 8, where the regressions use the alternative definition of the dependent variable for the three different specifications considered previously, that is, all control variables in columns 10 through 12 and the two subsets (A and B) described above in columns 13 through 18. (25) As a final check, we reestimate the regressions used to obtain the results in Tables 4 and 8 using the alternative computation of the network indicators. That is, we compute the network indicators using exports of country i to country j out of country i's GDP to define trade links instead of the original dependency measure and exports of country i to country j out of the total exports of country i. The results for this robustness check exercise, presented in Table 9, show no meaningful differences from those discussed above. The estimated coefficients for the network indicators for the epicenter country and for country i are similar to those discussed previously. All the coefficients have the right sign, negative for the epicenter effects and positive for country i, and are statistically significant.

For reasons of space, we do not discuss in detail the economic significance for each of the estimated coefficients for the epicenter and country i's network indicators in Tables 8 and 9. But using the information in these tables and the descriptive statistics presented in the data appendix, it can be verified that there are no meaningful differences from those reported in Table 5.

As mentioned earlier, the trade data needed to compute the network indicators for 2001 were not available to us. Therefore, we used the network indicators for 2000 as proxies for 2001 values. Hence, as a final sensitivity check, we exclude the fifth crisis window (Argentina) from the analysis. The results for this sensitivity check are reported in columns 7 through 9 in Table 8, which follows the same specification as Table 4. The statistical significance and the magnitude of the effects are very similar to those obtained before.

Regarding macroeconomic control variables, we see from the results in Tables 4, 8, and 9 that, in general, more vulnerable countries (because they either have low growth rates, high current account deficits, and high rates of inflation or are susceptible to capital flow reversals) tend to have greater negative abnormal returns during the crises periods. Not surprisingly, higher levels of government expenditures and trade openness tend to dampen these negative effects as is the case of low inflation rates. Finally, one indicator for which statistical significance is rarely found in the regressions presented throughout the article is bank reserves to assets ratio. This may suggest that more detailed banking indicators are needed in order to capture the effects that could emerge from the banking/financial sector in times of crisis.

A. Fixed Effects versus Network Effects

As a final comment, we would like to mention that even though the estimated base model resembles a fixed time period effects model, there is a definite advantage to using the network approach. Table 10 shows the results obtained after reestimating the base model, but instead of including the network indicators for the epicenter country, a fixed time period effects specification is used. These regressions use the original definition for the weekly abnormal stock market returns as well as the original dependency measure for the network indicators, exports of country i to country j out of the total exports of country i. These results show that the network effects for country i are statistically significant when the network indicator considered is node importance with world trade share as intrinsic value, as well as node centrality at the 2% trade-link thresholds. But most importantly, Table 10 also shows the (recovered) fixed time period effect dummies for each crisis window. The direction of these effects is in line with those obtained when the network indicators for the epicenter country were used instead of using a fixed time period specification. But the magnitudes of these effects are similar across crisis periods.

We see that the fixed effects specification does not fully capture the differences observed between the different crisis windows considered. Moreover, if the focus of the study is to analyze the possible implications of future crises, then the fixed time period specification is no longer useful, while the network specification is still useful. The methodology using the network indicators can be used not only to describe and analyze previous crisis windows, but it can also be used as a predictive tool, albeit with caution, given that these are estimated effects. It is possible to forecast the magnitude of the effect that a hypothetical crisis in India, for example, would have on the average weekly abnormal stock market returns. For this purpose, the network indicators and the estimated coefficients presented in Table 4 can be used to forecast the epicenter country and country i's effects. For example, using the node importance indicator for 2000 (intrinsic value: world trade share) for India as proxy for the current one, it is possible to show that the network effects of the epicenter country (India) would be a decrease in the abnormal weekly stock market return by 0.36 percentage points for all countries, a value that is a bit lower than that observed for Thailand and Russia during the Asian and Russian crises, respectively, but considerably higher than those observed for Argentina and Venezuela during their crisis windows. (26)

VI. CONCLUSIONS

Given that countries around the world are tied together through a complex network of trade and financial linkages leading to a web of dependencies of various kinds between them, it seems reasonable to argue for a systemwide perspective on the transmission and diffusion of cross-country shocks. The absence of research that incorporates system-wide considerations may be one reason why the literature on financial contagion continues to puzzle over why many of the recent crises that began in relatively small economies had such global repercussions and why shocks originating in one economy spread to some markets, while markets in other countries were relatively unaffected. A network approach that is capable of incorporating the cascading of interdependent ripples that happen when a shock hits a specific part of the network may therefore provide us with a deeper understanding of economic and financial contagion.

Such an approach may also have useful policy implications. The actual structure of the interdependent global trade network can be used to model what would happen to the system if different parts of the network collapsed owing to economic crises or speculative financial attacks. In other words, "what-if" scenarios can be simulated with regard to crises originating in hitherto unexplored parts of the network. This analysis may provide predictive insight into the likelihood of contagion for hypothetical crises.

Moreover, since the actual structure of the international trade network evolves over time, and can in principle be monitored, it is conceivable that an understanding of the vulnerability of the network together with the topology of the network (27) could be a potentially useful policy tool in discussions of optimal financial architecture and intervention by organizations such as the IMF and the U.S. Department of the Treasury in the event of actual or anticipated crises.

doi: 10.1111/j.1465-7295.2009.00249.x

ABBREVIATIONS

GDP: Gross Domestic Product

WDI: World Development Indicators

GLS: Generalized Least Squares

APPENDIX

TABLE A1

Control Variables

 As Referred in the WDI
 As Referred in Database and/or
 the Article Computation WDI Code

Bank reserves to Bank liquid reserves to FD.RES.LIQU.AS.ZS
assets ratio bank assets ratio

Current account External balance on goods NERSB.GNFS.ZS
balance as and services (% GDP) (a)
percentage of GDP

Government General government final NE.CON.GOVT.ZS
expenditures as consumption expenditure (%
percentage of GDP GDP)

Private capital Gross private capital BG.KAC.FNEI.GD.ZS
flows as percentage flows (% GDP)
of GDP

Government budget Overall budget balance, GB.BAL.OVRL.GD.ZS
balance as including grants (% GDP)
percentage of GDP

GDP growth rate GDP growth (annual %) NY.GDP.MKTP.KD.ZG

Total trade as Trade (% GDP) NE.TRD.GNFS.ZS
percentage of GDP

Yearly inflation Calculated using consumer FP.CPI.TOTL
rate price index (1995 = 100)

Trade with epicenter Exports of country i to --
(calculated using epicenter plus imports of
the NBER-UN world country i from epicenter
trade database) over the total exports
 plus total imports of
 country i

(a) Used as proxy for current account balances since more data were
available for this variable.

Countries Included in the Analysis

Argentina
Australia
Austria
Belgium
Brazil
Canada
Chile
China
Czech Republic
Denmark
Finland
France
Germany
Greece
Hong Kong
Hungary
India
Indonesia
Israel
Italy
Japan
Korea
Luxemburg
Malaysia
Mexico
Netherlands
New Zealand
Norway
Peru
Philippines
Poland
Portugal
Russia
Singapore
Spain
Sweden
Switzerland
Thailand
United Kingdom
United States
Venezuela

Note: Because of data availability, italicized countries are not
present in all regressions.

Number of Cross-Sections Included in Regressions

Table 4 34
Table 7 34
Table 8 35 (1-6), 38 (7-9), 38 (10-12), 39 (13-15), 40 (16-18)
Table 9 34 (1-3), 35 (4-9), 38 (10-12), 39 (13-15), 40 (16-18)
Table 10 34

Data Descriptive Statistics

Network Indicator Statistics (Trade-Based Dependency: Exports of
Country i to Country j out of Total Exports of Country i)

 Epicenter Importance Country i Importance
 Index (IV: World Index (IV: World
 Trade Share) Trade Share)

M 0.00951 0.02059
Maximum 0.01691 0.16188
Minimum 0.00296 0.00075
SD 0.00533 0.02754

 Epicenter Importance Country i Importance
 Index (IV: Centrality Index (IV: Centrality
 1% Threshold) 1% Threshold)

M 0.01702 0.02874
Maximum 0.03048 0.09391
Minimum 0.00403 0.00000
SD 0.01112 0.02580

 Epicenter
 Centrality (0%
 Threshold)

M 64.82760
Maximum 83.90800
Minimum 45.97700
SD 12.99970

 Country i Epicenter Country i
 Centrality Centrality Centrality
 (0% Threshold) (1% Threshold) (I% Threshold)

M 73.59388 17.01160 28.710
Maximum 99.42500 30.46000 93.678
Minimum 0.00000 4.02300 0.000
SD 16.47745 11.11311 25.7661

 Epicenter Country i
 Centrality Centrality
 (2% Threshold) (2% Threshold)

M 10.920 19.576
Maximum 22.989 88.506
Minimum 2.299 0.000
SD 8.1622 22.1771

 Maximum Flow Maximum Flow
 Epicenter to Country i Epicenter to Country i
 (0% Threshold) (0.5% Threshold)

M 106.316 20.716
Maximum 146.000 40.000
Minimum 0.000 0.000
SD 26.0609 8.7785

 Maximum Flow Maximum Flow
 Epicenter to Country i Epicenter to Country i
 (1% Threshold) (2% Threshold)

M 12.836 4.747
Maximum 24.000 13.000
Minimum 0.000 0.000
SD 5.5689 3.6281

IV, intrinsic value.

Network Indicator Statistics (GDP-Based Dependency: Exports of
Country i to Country j out of Country i's GDP)

 Epicenter Importance Country i Importance
 Index (IV: World Index (IV: World
 Trade Share) Trade Share)

M 0.00950 0.02057
Maximum 0.01690 0.16176
Minimum 0.00296 0.00075
SD 0.00532 0.02752

 Epicenter Importance Country i Importance
 Index (IV: Centrality Index (IV: Centrality
 1 % Threshold) 1 % Threshold)

M 0.00931 0.01792
Maximum 0.02012 0.07878
Minimum 0.00287 0.00000
SD 0.00716 0.01925

 Epicenter
 Centrality (0%
 Threshold)

M 64.36800
Maximum 82.75900
Minimum 45.97700
SD 12.65624

 Country i Epicenter Country i
 Centrality Centrality Centrality
 (0% Threshold) (1% Threshold) (1% Threshold)

M 72.80972 9.31040 17.872
Maximum 97.12600 20.11500 78.736
Minimum 0.00000 2.87400 0.000
SD 15.95887 7.16075 19.235

 Epicenter Country i
 Centrality Centrality
 (2% Threshold) (2% Threshold)

M 5.747 11.101
Maximum 15.517 70.690
Minimum 1.149 0.000
SD 5.620 14.336

IV, intrinsic value.

Dependent Variable and Macroeconomic Control Variables

 Abnormal Stock
 Abnormal Stock Market Return
 Market Return (Deviation from
 (Deviation from Overall Average Total Trade
 Average before 1994) before 1992-2000) to GDP Ratio

M -0.42383 -0.28636 73.03769
Maximum 2.12612 2.34622 287.40550
Minimum -7.66728 -7.53382 15.92322
SD 1.27660 1.14775 52.00246

 Bank
 Yearly Reserves to
 Inflation Assets Ratio

M 13.47255 6.86191
Maximum 874.62190 34.13177
Minimum -5.20943 0.02018
SD 62.93572 7.69720

 Government
 Current Account to Consumption to Private Capital
 GDP Ratio GDP Ratio Inflows to GDP Ratio

M 0.6695 16.5772 21.6734
Maximum 22.1613 29.4407 179.1754
Minimum -16.1205 5.0128 2.5377
SD 5.6517 5.5678 22.7116

 Government Budget GDP Growth Trade with
 to GDP Ratio Rate Epicenter

M -1.7916 3.8807 0.0109
Maximum 16.0776 13.5000 0.1031
Minimum -14.9732 -12.5698 0.0000
SD 4.2842 3.2907 0.0178

REFERENCES

Abeysinghe, T., and K. Forbes. "Trade Linkages and Output-Multiplier Effects: A Structural VAR Approach with a Focus on Asia." Review of International Economics, 13(2), 2005, 356-75.

Albert, A., and A. L. Barabasi. "Statistical Mechanics of Complex Networks." Reviews of Modern Physics, 74, 2002, 47.

Bae, K.-H., G. A. Karolyi, and R. Stulz. "A New Approach to Measuring Financial Contagion." Review of Financial Studies, 16, 2003, 717-63.

Claessens, S., and K. Forbes. "International Financial Contagion: An Overview of Issues," in Introductory chapter of International Financial Contagion, edited by S. Claessens and K. Forbes. Norwell, MA: Kluwer Academic Publishers, 2001, 3-17.

Eichengreen, B., A. Rose, and C. Wyplosz. "Contagious Currency Crises: First Tests." NBER Working Paper No. 5681 1996.

Fagiolo, G., J. Reyes, and S. Schiavo. "On the Topological Properties of the World Trade Web: A Weighted Network Analysis." Physica A, 387, 2008, 3868-73.

Feenstra, R. C., R. E. Lipsey, H. Deng, A. C. Ma, and Hengyong Mo. "World Trade Flows, 1962-2000." NBER Working Paper No. 11040 2005.

Forbes, K. "Are Trade Linkages Important Determinants of Country Vulnerability to Crises," in Preventing Currency Crises in Emerging Markets, edited by S. Edwards and J. Frankel. Chicago: University of Chicago Press, 2001, 77-124.

Forbes, K., and M. Chinn. "A Decomposition of Global Linkages in Financial Markets over Time." Review of Economics and Statistics, 86, 2004, 705-22.

Forbes, K., and R. Rigobon. "No Contagion, Only Interdependence: Measuring Stock Market Comovements." Journal of Finance, LVII, 2002, 2223-61.

Glick, R., and A. K. Rose. "Contagion and Trade: Why Are Currency Crises Regional?" NBER Working Paper No. 6806 1998.

--. Contagion and Trade: Why Are Currency Crises Regional? Journal of International Money and Finance, 18(4), 1999, 603-17.

Hanneman, R. A. Introduction to Social Network Methods. Mimeo. Riverside: University of California, 2001.

Kali, R., and J. Reyes. "The Architecture of Globalization: A Network Approach to International Economic Integration." Journal of International Business Studies, 38, 2007, 595-620.

Kaminsky, G., and C. Reinhart. "On Crisis, Contagion and Confusion." Journal of International Economics, 51, 2000, 145-68.

--. "The Center and the Periphery: The Globalization of Financial Shocks." NBER Working Paper No. 9479, 2003.

Newman, M. E. J. "The Structure and Function of Complex Networks." S1AM Review, 45, 2003, 167-256.

Rigobon, R. "On the Measurement of the International Propagation of Shocks: Is It Stable?" Journal of International Economics, 61, 2003, 261-83.

Salancik, G. "An Index of Subgroup Influence in Dependency Networks." Administrative Science Quarterly, 31, 1986, 194-211.

Serrano, A., and M. Boguna. "Topology of the World Trade Web." Physical Review E, 68, 015101(R), 2003.

Van Rijckeghem, C., and B. Weder. "Sources of Contagion: Is It Finance or Trade?" Journal of International Economics, 54, 2001, 293-308.

(1.) Prominent examples and their colorful sobriquets are the Mexican "Tequila crisis" of 1994, the "Asian flu" of 1997, and the "Russian virus" of 1998.

(2.) Complex networks are large-scale graphs that are composed of so many nodes and links that they cannot be meaningfully visualized and analyzed using standard graph theory. Recent advances in network research now enable us to analyze such graphs in terms of their statistical properties. Albert and Barabasi (2002) and Newman (2003) are excellent surveys of these methods.

(3.) Van Rijekeghem and Weder (2001) find that trade and financial linkages were highly correlated ahead of the Mexican, Thai, and Russian crises. The correlation between their measure of trade competition and financial competition is 0.45, 0.70, and 0.33 for the Mexican, Thai, and Russian crises, respectively.

(4.) Reliable data on cross-country financial flows are severely limited. Using only financial flow data would provide us with a truncated and unrepresentative network of cross-country economic linkages.

(5.) Another option for the study of contagion across countries would be to use interest rate data. We decided to use stock market data because, in many cases, interest rates are affected (and/or controlled) directly or indirectly by policy makers. Therefore, interest rate movements would not reflect the complete story of financial contagion as perceived by market participants.

(6.) These are crisis periods in the 1990s with the epicenter countries being Mexico, Venezuela, Thailand, Russia, and Argentina. We describe our crisis definitions and crisis windows in detail in Section III.

(7.) There are a number of earlier studies that find the trade channel to be a significant explanation for contagion. Eichengreen, Rose, and Wyploz (1996) analyze contagion using data on 20 industrial economies from 1959 to 1993. They find that the probability of a crisis in a country increases the presence of a crisis elsewhere and that this increase is better explained by trade links among countries than by macroeconomic similarities. Glick and Rose (1999) use a much larger sample of countries and find that trade competition in third markets has high power in explaining contagion in five major crisis episodes between 1971 and 1997.

(8.) The node degree centrality and maximum flow indicators have been calculated using UCINET 6.7 software.

(9.) The data we use for the computation of the network and associated measures were extracted from the NBER-UN world trade database. We use the U.S. dollar value of exports and imports of all commodities between 182 countries for the years 1992 through 2000. We discuss our data set in detail in Section III.

(10.) The rationale for the intrinsic value in Equation (2) is to provide a weight for a country that is independent of its network position. For example, a country with a larger share of world trade receives a greater weight regardless of its network connectedness.

(11.) The data for Argentina in 2000 are used in our analysis as a proxy for the 2001 data since the data for international trade were only available through 2000.

(12.) Computing a distance-weighted version of maximum flow presents a daunting technical challenge. It involves calculating the network distance from each node to every other node in the network across the 182 nodes in our sample. We defer this task to future research.

(13.) This database is available online at www.nber.org/data (International Trade Data, NBERUN world trade data). A detailed description of the database can be found in Feenstra et al. (2005).

(14.) The argument in favor of such an index, widely used in the literature, is that it captures the main tools that central banks can use to fight off a speculative attack and/or adverse market behavior in response to new information. For further discussion, see Eichengreen, Rose, and Wyplosz (1996).

(15.) Given that both Mexico and Venezuela were emerging oil-exporting countries, we may suspect that the Venezuelan crisis is an aftershock of the Mexican one. Statistically, this is not the case since the Venezuelan crisis is the first [mu] [+ or -] 5[sigma] crisis after a relatively tranquil period. This leads us to conclude that it is a separate event in which Venezuela played the role of the epicenter country.

(16.) In the robustness section, we present and discuss the results obtained by estimating an alternative model. Instead of using the network indicators for the epicenter country, we use a dummy variable for each of the crisis windows (in essence, a time period fixed effects model). There, we show that the results of the fixed effects model match those presented here.

(17.) Total world trade shares were calculated with the same data used for the computation of the network indicators.

(18.) The Data Descriptive Statistics table, presented in the Table A1 at the end of the article, reports that the standard deviation of the abnormal stock return is equal to 1.27, while that for the importance of the epicenter country, using the world trade share as the intrinsic value, is 0.00533. Therefore, a one standard deviation of epicenter importance implies a change of -0.24 percentage points in the abnormal stock return (-44.88 x 0.00533), which is about 0.187 of a standard deviation of the dependent variable (0.24/1.27).

(19.) It should be noted that these cumulative effects are for the abnormal stock market returns (i.e., deviation from the average weekly return); therefore, their magnitude should be interpreted accordingly. The objective of the article is not to predict substantial daily declines, as is often observed in times of crises, but to provide the basis for relative comparisons across countries as it becomes clear in the discussion of Table 6.

(20.) Similar results hold if the other regressions are used for this analysis.

(21.) These calculations used the data of the network indicators for each country and the estimated coefficients of Regression 2 in Table 4.

(22.) The 1% and 2% thresholds are still appropriate when using this dependency measure since 97% of the exports to GDP ratios are between 0% and 1% in 1992 (96% in 2000).

(23.) Specific details are listed in the data appendix at the end of the article.

(24.) This can be verified in Table 4. It should also be noted that the statistical significance of the network indicators and the macroeconomic control variables of the OLS regressions for the specifications A and B very closely follows the results presented in Table 4.

(25.) The magnitude effects for the network coefficients are not presented or discussed here, but it is easy to verify that the results for these robustness checks match those presented in Table 5.

(26.) This can be verified in Table 6.

(27.) Several recent articles, such as Fagiolo, Reyes, and Schiavo (2008), Serrano and Boguna (2003), and Kali and Reyes (2007), examine the topology of the international trade network.

RAJA KALI and JAVIER REYES, We are grateful to Jon Johnson, Jungmin Lee, two anonymous referees, and the editor for comments that improved the article. We thank seminar participants at the Harvard Business School, the University of Arkansas, and the 2008 CIBIF conference in Groningen for their comments. Viktoria Riiman provided outstanding research assistance.

Kali: Associate Professor, Department of Economics, Sam M. Walton College of Business, University of Arkansas, Fayetteville, AR 72701. Phone 1-479-575-6219, E-mail [email protected]

Reyes: Associate Professor, Department of Economics, Sam M. Walton College of Business, University of Arkansas, Fayetteville, AR 72701. 1-479-575-6079, E-mail [email protected]

TABLE 1
Country Rankings for the Top 35 Countries according to the Network
Indicators

 Node Centrality Node Importance (IV:
 (1% Threshold) World Trade Share)

 1994 1997 2000 1994 1997 2000

United States 1 1 1 1 1 1
Germany 2 2 2 2 2 2
France, Monaco 3 3 3 3 3 3
United Kingdom 4 4 5 4 4 5
Italy 5 5 4 5 5 4
Japan 6 6 7 6 6 7
Spain 8 7 8 8 7 8
Belgium-Lux 9 8 9 9 8 8
Netherlands 7 9 6 7 8 6
Korea Republic 12 10 12 12 9 11
Canada 11 11 11 11 10 10
China 10 12 10 10 10 9
Russian Federation 168 14 15 46 11 14
Taiwan 13 13 13 13 11 12
China HK SAR 16 15 17 16 12 16
Thailand 15 16 14 15 13 13
Singapore 14 17 16 14 14 15
Turkey 23 18 18 22 15 17
Brazil 22 19 19 21 16 18
India 20 20 100 20 17 43
Portugal 17 21 21 17 18 19
Switz.Liecht 18 22 20 18 19 18
Austria 21 23 23 20 20 21
Malaysia 27 24 26 24 21 23
Australia 24 26 27 22 22 23
Poland 26 25 24 23 22 22
Indonesia 32 27 29 28 23 24
Sweden 19 28 25 19 24 22
Greece 30 29 28 27 25 23
Mexico 25 30 22 23 26 20
Norway 33 32 33 29 26 27
Philippines 39 31 30 33 26 25
Denmark 28 33 31 25 27 25
Finland 29 34 40 26 28 31
Iran 59 35 41 43 29 31
Czech Republic 37 37 34 32 30 28
South Africa 35 36 42 31 30 31
Hungary 34 38 35 30 31 28
Romania 41 40 38 35 32 30
Saudi Arabia 31 39 32 28 32 26
Israel 38 41 36 33 33 28
Chile 40 42 46 34 34 33
Pakistan 36 43 37 32 34 29
Bulgaria 69 44 49 44 35 35
Argentina 44 46 51 37 36 36
Venezuela 47 48 43 38 37 31

 Maximum Flow Total Trade (Exports
 (1% Threshold) plus Imports) to GDP

 1994 1997 2000 1994 1997 2000

United States 9 8 8 156 154 148
Germany 6 10 10 100 99 84
France, Monaco 3 6 4 113 114 105
United Kingdom 10 9 7 97 104 110
Italy 11 7 11 114 123 107
Japan 8 5 8 160 159 151
Spain 6 4 5 118 111 102
Belgium-Lux 4 4 3 18 13 14
Netherlands 4 3 3 31 26 27
Korea Republic 10 9 8 74 58 51
Canada 1 1 1 54 52 53
China 2 2 4 57 90 73
Russian Federation 52 11 15 29 33 NA
Taiwan 6 5 6 152 116 66
China HK SAR 7 5 9 10 8 8
Thailand 11 9 12 39 36 19
Singapore 9 6 6 5 4 4
Turkey 20 8 7 125 115 111
Brazil 16 17 19 158 162 150
India 14 15 56 155 155 165
Portugal 22 24 22 70 70 69
Switz.Liecht 12 18 11 58 54 55
Austria 5 6 14 68 61 63
Malaysia 15 16 17 8 7 6
Australia 13 12 16 104 96 96
Poland 18 14 13 130 140 124
Indonesia 21 21 23 86 89 57
Sweden 10 13 9 66 56 61
Greece 26 27 21 127 138 121
Mexico 17 20 2 115 68 67
Norway 28 31 26 49 35 23
Philippines 24 19 18 69 67 76
Denmark 19 22 20 80 77 80
Finland 30 32 30 65 59 58
Iran 46 34 38 144 135 114
Czech Republic 33 25 25 123 107 99
South Africa 32 33 44 43 31 22
Hungary 31 29 24 50 30 13
Romania 34 28 29 51 65 79
Saudi Arabia 23 23 33 92 63 64
Israel 25 26 27 79 91 83
Chile 25 30 31 93 103 98
Pakistan 37 41 38 131 139 128
Bulgaria 48 42 37 60 25 31
Argentina 27 36 35 96 105 115
Venezuela 39 38 32 159 156 156

Notes: Countries ranked according to the 1997 node importance results
using world trade share as the IV. One percent threshold refers to the
fact that a link between two countries was present if and only if the
exports of country i to country j, out of the total exports of country
i, were greater than 1% of the total as explained at the beginning of
Section IIA. The columns for Total Trade to GDP show how these
countries rank in this category. IV, intrinsic value.

TABLE 2
Top 35 Countries Affected by Shocks to Source Country according to the
Maximum Flow Measure

SOURCE COUNTRY

 Mexico (1994) Venezuela (1995)

 Bahamas 16 Bahamas 8
 Cuba 16 Cuba 8
 Mali 16 Mali 8
A Togo 16 Togo 8
 Algeria 15 Algeria 8
F Argentina 15 Argentina 8
 Australia 15 Australia 8
F Austria 15 Austria 8
 Azerbaijan 15 Azerbaijan 8
E Bahrain 15 Bahrain 8
 Belarus 15 Belarus 8
C Brazil 15 Brazil 8
 Bulgaria 15 Bulgaria 8
T Burundi 15 Burundi 8
 Chile 15 Chile 8
E China HK SAR 15 China HK SAR 8
 Colombia 15 Colombia 8
D Cyprus 15 Cyprus 8
 Denmark 15 Denmark 8
 Ecuador 15 Ecuador 8
C Egypt 15 Egypt 8
 Estonia 15 Estonia 8
O Finland 15 Finland 8
 Germany 15 Germany 8
U Greece 15 Greece 8
 Hungary 15 Hungary 8
N Iceland 15 Iceland 8
 India 15 India 8
T Indonesia 15 Indonesia 8
 Iran 15 Iran 8
R Israel 15 Israel 8
 Italy 15 Italy 8
Y Japan 15 Japan 8
 Jordan 15 Jordan 8
 Kazakhstan 15 Kazakhstan 8

 Thailand (1997) Russia (1998)

 Brazil 24 Jordan 25
 India 24 Korea Republic 24
 Jordan 24 India 23
A Russian Federation 24 South Africa 23
 South Africa 24 Saudi Arabia 23
F Germany 23 Brazil 22
 Iran 23 Chile 22
F Sweden 23 Pakistan 22
 Argentina 22 Germany 21
E Chile 22 Sweden 21
 Egypt 22 Finland 21
C Finland 22 Indonesia 21
 Korea Republic 21 Greece 21
T New Zealand 21 Iran 20
 Pakistan 21 Argentina 20
E Peru 21 New Zealand 20
 United Kingdom 21 Peru 20
D Ukraine 21 United Kingdom 20
 Sudan 21 Ukraine 20
 Togo 20 Vietnam 20
C Indonesia 20 Australia 20
 Israel 20 Ecuador 20
O Saudi Arabia 20 Panama 20
 Switz.Liecht 20 Japan 20
U Tanzania 20 Israel 19
 Turkey 20 Switz.Liecht 19
N United States 20 Mali 19
 Vietnam 20 Austria 19
T Australia 19 Zimbabwe 19
 Bahrain 19 Tajikistan 19
R Colombia 19 Zambia 19
 Ecuador 19 Iraq 19
Y Greece 19 Thailand 19
 Italy 19 Egypt 18
 Malawi 19 Togo 18

 Argentina (2000) Brazil (1998)

 Chile 8 Chile 22
 Peru 8 Argentina 20
 Ecuador 8 Peru 20 20
A Panama 8 Ecuador 20 20
 Mali 8 Panama 20 20
F Jordan 8 Mali 19
 Korea Republic 8 Jordan 18
F India 8 Korea Republic 18
 South Africa 8 India 18
E Saudi Arabia 8 South Africa 18
 Pakistan 8 Saudi Arabia 18
C Germany 8 Pakistan 18
 Sweden 8 Germany 18
T Finland 8 Sweden 18
 Indonesia 8 Finland 18
E Greece 8 Indonesia 18
 Iran 8 Greece 18
D New Zealand 8 Iran 18
 United Kingdom 8 New Zealand 18
 Ukraine 8 United Kingdom 18
C Vietnam 8 Ukraine 18
 Australia 8 Vietnam 18
O Japan 8 Australia 18
 Israel 8 Japan 18
U Switz.Liecht 8 Israel 18
 Austria 8 Switz.Liecht 18
N Zimbabwe 8 Austria 18
 Tajikistan 8 Zimbabwe 18
T Zambia 8 Tajikistan 18
 Iraq 8 Zambia 18
R Thailand 8 Iraq 18
 Egypt 8 Thailand 18
Y Togo 8 Egypt 18
 Tanzania 8 Togo 18
 United States 8 Tanzania 18

TABLE 3
List of Country Crises (Crisis Windows)

 Forbes(2001)

Mexico December 19, 1994, to December 25, 1994; January 16,
 1995, to January 29, 1995; February 27, 1995, to
 March 05, 1995; March 13, 1995, to March 19, 1995

Ecuador (1) January 23, 1995, to February 12, 1995; October 30,
 1995, to November 05, 1995

Argentina March 06, 1995, to March 12, 1995

Venezuela (1) December 11, 1995, to December 17, 1995; April 15,
 1996, to April 12, 1996

Venezuela (2) May 12, 1997, to May 18, 1997

Czech Republic May 19, 1997, to May 25, 1997

Thailand June 30, 1997, to July 06, 1997

Philippines July 07, 1997, to July 13, 1997; September 29, 1997,
 to October 05, 1997; August 11, 1997, to August 17,
 1997; August 25, 1997, to August 31, 1997; September
 29, 1997, to October 05, 1997

Indonesia December 08, 1997, to December 14, 1997; January 19,
 1998, to January 25, 1998; March 02, 1998, to March
 08, 1998; May 18, 1998, to May 24, 1998

Korea December 29, 1997, to January 04, 1998

India January 19, 1998, to January 25, 1998

Russia May 18, 1998, to May 31, 1998; July 06, 1998, to
 July 12, 1998; August 10, 1998, to September 06,
 1998; September 14, 1998, to September 20, 1998

Venezuela June 15, 1998, to June 21, 1998; September 14, 1998,
 to September 20, 1998

Slovak Republic September 28, 1998, to October 04, 1998

Ecuador (2) October 19, 1998, to October 25, 1998; January 11,
 1999, to January 17, 1999; March 01, 1999, to March
 07, 1999

Brazil January 11, 1999, to January 17, 1999

 Glick and Rose (1998)

Bretton Woods 1971
 collapse
Collapse of 1973 (winter)
 Smithsonian
 Agreement
EMS crisis 1992-1993
Mexican meltdown 1994-1995
 (Tequila effect)
Asian flu 1997-1998

 Rigobon (2003)

Mexican crisis December 19, 1994, to
 March 31, 1995
Asian crisis June 10, 1997, to January
 (Thailand) 30, 1998
Russian crisis August 3, 1998, to
 November 23, 1998

 Kali and Reyes

Mexican crisis December 24, 1994, to
 (Tequila crisis) March 18, 1995
Venezuelan crisis December 16, 1995, to
 April 20, 1996
Thailand crisis July 05, 1997, to January
 (Asian flu) 03, 1998
Russian crisis May 23, 1998, to January
 (Russian virus) 16, 1999
Argentinean crisis June 02, 2001, to April
 06, 2002

TABLE 4
Regression Results Using Node Importance and Centrality Indices
as the Network Indicators (Weekly Abnormal Stock Market Returns
as the Dependent Variable)

 1 2

 Node Importance Node Importance
 IV: World Trade IV: World Trade
 Share (OLS) Share (GLS)

Epicenter network -68.837 * (-4.3738) -44.884 * (-5.9316)#
 indicator
Country i network 5.293 ** (2.2092)# 3.912 * (3.1758)#
 indicator
Epicenter
 centrality
Country i
 centrality
Constant 0.0282 (0.0531)# -0.4863 ** (-2.1768)#
Bank reserves to -0.0039 (-0.2424)# -0.0052 (-0.4602)#
 assets ratio
Current account 0.0246 (1.3525)# 0.0217 ** (2.0575)#
 (% GDP)
Government 0.0316 *** (1.9683)# 0.030 9 * (3.1261)#
 expenditures
 (% GDP)
Private capital -0.0043 (-1.3156)# 0.0002 (0.1204)#
 flows (% GDP)
Overall budget -0.0288 (-1.0007)# -0.0566 * (-3.6589)#
 (% GDP)
GDP growth rate -0.0998 * (-2.6836)# -0.0662 * (-3.0622)#
Total trade (% 0.4981 (1.3533)# 0.3884 ** (2.0761)#
 GDP)
Inflation rate -0.0034 (-0.3600)# 0.0028 (0.4749)#
 (yearly)
Trade with -13.6729 (-1.5636)# -5.2725 (-1.6034)#
 epicenter
Adjusted .3030 .4726
 [R.sup.2]

 3 4

 Node Importance Node Importance
 IV: Centrality IV: Centrality
 1% (OLS) 1% (GLS)

Epicenter network -19.6106 ** (-2.3803)# -8.1457 ** (-2.3642)#
 indicator
Country i network 3.7962 (1.3972)# 2.1410 (1.5998)#
 indicator
Epicenter
 centrality
Country i
 centrality
Constant -0.2598 (-0.5396)# -0.3846 (-1.5347)#
Bank reserves to -0.0069 (-0.4288)# -0.0137 (-1.2664)#
 assets ratio
Current account 0.0282 (1.4826)# 0.0260 ** (2.6092)#
 (% GDP)
Government 0.0315 *** (1.8568)# 0.0259 ** (2.5055)#
 expenditures
 (% GDP)
Private capital -0.0045 (-1.3430)# -0.0010 (-0.6690)#
 flows (% GDP)
Overall budget -0.0157 (-0.5241)# -0.0428 * (-2.9784)#
 (% GDP)
GDP growth rate -0.0724 *** (-1.9372)# -0.0513 * (-2.6425)#
Total trade (% 0.4314 (1.1755)# 0.0897 (0.4285)#
 GDP)
Inflation rate -0.0051 (-0.4604)# 0.0016 (0.2655)#
 (yearly)
Trade with -14.9240 (-1.6005)# -8.8809 ** (-2.2729)#
 epicenter
Adjusted .2382 .3608
 [R.sup.2]

 5 6

 Node Centrality Node Centrality
 Threshold 0% Threshold 0%
 (OLS) (GLS)

Epicenter network
 indicator
Country i network
 indicator
Epicenter -0.0235 * (-3.3302)# -0.0106 * (-3.0938)#
 centrality
Country i 0.0137 ** (2.2429)# 0.0084 ** (2.4439)#
 centrality
Constant -0.0282 (-0.0353)# -0.6036 (-1.5744)#
Bank reserves to -0.0001 (-0.0095)# -0.0078 (-0.7216)#
 assets ratio
Current account 0.0256 (1.3805)# 0.0271 ** (2.5826)#
 (% GDP)
Government 0.0331 *** (1.9627)# 0.0307 * (2.9657)#
 expenditures
 (% GDP)
Private capital -0.0046 (-1.3779)# -0.0015 (-0.8230)#
 flows (% GDP)
Overall budget 0.0021 (0.0718)# -0.0279 *** (-1.7802)#
 (% GDP)
GDP growth rate -0.0772 ** (-2.1331)# -0.0419 ** (-2.0954)#
Total trade (% 0.3821 (1.1502)# 0.1306 (0.6466)#
 GDP)
Inflation rate -0.0027 (-0.2462)# 0.0031 (0.4673)#
 (yearly)
Trade with -12.320 (-1.4020)# -5.4503 (-1.2441)#
 epicenter
Adjusted .2832 .3351
 [R.sup.2]

 7 8

 Node Centrality Node Centrality
 Threshold 1% Threshold 1%
 (OLS) (GLS)

Epicenter network
 indicator
Country i network
 indicator
Epicenter -0.0196 ** (-2.3793)# -0.0081 ** (-2.3623)#
 centrality
Country i 0.0038 (1.3949)# 0.0021 (1.5972)#
 centrality
Constant -0.2598 (-0.5396)# -0.3842 (-1.5331)#
Bank reserves to -0.0069 (-0.4292)# -0.0137 (-1.2669)#
 assets ratio
Current account 0.0282 (1.4828)# 0.0260 ** (2.6090)#
 (% GDP)
Government 0.0315 *** (1.8569)# 0.0259 ** (2.5053)#
 expenditures
 (% GDP)
Private capital -0.0045 (-1.3431)# -0.0011 (-0.6703)#
 flows (% GDP)
Overall budget -0.0157 (-0.5240)# -0.0428 * (-2.9770)#
 (% GDP)
GDP growth rate -0.0724 *** (-1.9372)# -0.0513 * (-2.6433)#
Total trade (% 0.4313 (1.1751)# 0.0894 (0.4272)#
 GDP)
Inflation rate -0.0051 (-0.4605)# 0.0016 (0.2652)#
 (yearly)
Trade with -14.925 (-1.6004)# -8.8852 ** (-2.2738)#
 epicenter
Adjusted .2381 .3608
 [R.sup.2]

 9 10

 Node Centrality Node Centrality
 Threshold 2% Threshold 2%
 (OLS) (GLS)

Epicenter network
 indicator
Country i network
 indicator
Epicenter -0.0181 (-1.6469)# -0.0096 ** (-2.3454)#
 centrality
Country i 0.0051 *** (1.7913)# 0.0026 ** (2.0352)#
 centrality
Constant -0.4071 (-0.8481)# -0.4091 *** (-1.6636)#
Bank reserves to -0.0066 (-0.4085)# -0.0141 (-1.3649)#
 assets ratio
Current account 0.0295 (1.5449)# 0.0271 * (2.8433)#
 (% GDP)
Government 0.0308 *** (1.8265)# 0.0255 ** (2.4948)#
 expenditures
 (% GDP)
Private capital -0.0042 (-1.3189)# -0.0008 (-0.5281)#
 flows (% GDP)
Overall budget -0.0194 (-0.6401)# -0.0466 * (-3.5341)#
 (% GDP)
GDP growth rate -0.0621 (-1.6535)# -0.0450 ** (-2.5581)#
Total trade (% 0.4268 (1.1216)# 0.0787 (0.3686)#
 GDP)
Inflation rate -0.0057 (-0.5229)# 0.0015 (0.2570)#
 (yearly)
Trade with -17.0973 *** (-1.8097)# -9.5069 * (-2.7493)#
 epicenter
Adjusted .2262 .3871
 [R.sup.2]

Notes: Number of observations for all regressions: 142. t
Statistics reported in italics. IV, intrinsic value. * 1%; **
5%; *** 10%.

Note: t Statistics reported in italics is indicated with #.

TABLE 5
Estimated Effects on the Weekly Abnormal Stock Market Return of a
One-Standard Deviation Increase in the Network Indicator

 Effects in Terms of
 Standard Deviations

 OLS GLS

Epicenter importance (IV: world trade share) -0.275 -0.187
Country i importance (IV: world trade share) 0.114 0.084
Epicenter importance (IV: centrality 1%) -0.171 -0.071
Country i importance (IV: centrality 1%) 0.077 0.043
Epicenter centrality (0% threshold) -0.239 -0.108
Country i centrality (0% threshold) 0.177 0.108
Epicenter centrality (1% threshold) -0.171 -0.071
Country i centrality (1% threshold) 0.077 0.043
Epicenter centrality (2% threshold) -0.116 -0.061
Country i centrality (2% threshold) 0.088 0.044

 Effects in Terms of
 Percentage Points

 OLS GLS

Epicenter importance (IV: world trade share) -0.351 -0.239
Country i importance (IV: world trade share) 0.146 0.108
Epicenter importance (IV: centrality 1%) -0.218 -0.091
Country i importance (IV: centrality 1%) 0.098 0.055
Epicenter centrality (0% threshold) -0.305 -0.137
Country i centrality (0% threshold) 0.226 0.138
Epicenter centrality (1% threshold) -0.218 -0.091
Country i centrality (1% threshold) 0.098 0.055
Epicenter centrality (2% threshold) -0.148 -0.078
Country i centrality (2% threshold) 0.113 0.057

 Accumulated Effects over
 a Period of 10 Wk (%)

 OLS GLS

Epicenter importance (IV: world trade share) -3.509 -2.392
Country i importance (IV: world trade share) 1.455 1.076
Epicenter importance (IV: centrality 1%) -2.180 -0.906
Country i importance (IV: centrality 1%) 0.979 0.552
Epicenter centrality (0% threshold) -3.051 -1.373
Country i centrality (0% threshold) 2.256 1.380
Epicenter centrality (1% threshold) -2.180 -0.905
Country i centrality (1% threshold) 0.978 0.552
Epicenter centrality (2% threshold) -1.475 -0.783
Country i centrality (2% threshold) 1.129 0.566

Notes: This table uses the descriptive statistics for the
dependent and independent variable data for the calculation of
these results. Basically, it multiplies the estimated coefficient
by the standard deviation of the independent variable for the
estimation of the effects in terms of standard deviations and
then uses the standard deviation of the dependent variable to
compute the effects in terms of percentage points. IV, intrinsic
value.

TABLE 6
Epicenter and Country i Effects for the Five Crisis Windows
Considered Using the Estimated Coefficients in Table 4 (in
Percentage Points of Weekly Abnormal Stock Market Returns)

 Node Importance

 IV: World Trade Share IV: Node Centrality 1%

Mexican crisis
 Epicenter country -0.759 -0.131
 Brazil 0.038 0.042
 Canada 0.148 0.080
 Chile 0.011 0.018
 Ecuador 0.004 0.005
 Greece 0.014 0.030
 India 0.027 0.044
 Indonesia 0.036 0.027
 Italy 0.162 0.165
 Korea 0.091 0.078
 Malaysia 0.059 0.033
 Mexico 0.066 0.034
 Russia -- --
 Thailand 0.046 0.057
 United States 0.576 0.190
 Venezuela 0.011 0.011
Venezuelan crisis
 Epicenter country -0.133 -0.047
 Brazil 0.041 0.054
 Canada 0.141 0.076
 Chile 0.012 0.017
 Ecuador 0.004 0.005
 Greece 0.014 0.033
 India 0.028 0.052
 Indonesia 0.036 0.038
 Italy 0.165 0.158
 Korea 0.099 0.083
 Malaysia 0.063 0.037
 Mexico 0.059 0.015
 Russia 0.026 0.000
 Thailand 0.049 0.064
 United States 0.549 0.196
 Venezuela 0.012 0.012
Asian crisis
 Epicenter country -0.505 -0.234
 Brazil 0.043 0.053
 Canada 0.147 0.083
 Chile 0.013 0.016
 Ecuador 0.004 0.005
 Greece 0.013 0.033
 India 0.029 0.050
 Indonesia 0.037 0.036
 Italy 0.155 0.153
 Korea 0.098 0.084
 Malaysia 0.061 0.039
 Mexico 0.078 0.030
 Russia 0.056 0.073
 Thailand 0.044 0.062
 United States 0.582 0.194
 Venezuela 0.013 0.011
Russian crisis
 Epicenter country -0.562 -0.248
 Brazil 0.042 0.037
 Canada 0.152 0.087
 Chile 0.012 0.016
 Ecuador 0.004 0.012
 Greece 0.014 0.031
 India 0.029 0.050
 Indonesia 0.031 0.027
 Italy 0.161 0.156
 Korea 0.080 0.066
 Malaysia 0.052 0.031
 Mexico 0.087 0.030
 Russia 0.049 0.065
 Thailand 0.036 0.049
 United States 0.603 0.195
 Venezuela 0.012 0.018
Argentinean crisis
 Epicenter country -0.175 -0.033
 Brazil 0.037 0.047
 Canada 0.161 0.084
 Chile 0.011 0.014
 Ecuador 0.003 0.005
 Greece 0.012 0.034
 India 0.031 0.060
 Indonesia 0.032 0.033
 Italy 0.143 0.161
 Korea 0.105 0.074
 Malaysia 0.061 0.034
 Mexico 0.110 0.038
 Russia 0.049 0.058
 Thailand 0.042 0.059
 United States 0.633 0.201
 Venezuela 0.014 0.016

 Node Centrality

 1% Threshold 2% Threshold

Mexican crisis
 Epicenter country -0.131 -0.077
 Brazil 0.042 0.041
 Canada 0.080 0.040
 Chile 0.018 0.012
 Ecuador 0.005 0.003
 Greece 0.030 0.016
 India 0.044 0.038
 Indonesia 0.027 0.015
 Italy 0.165 0.152
 Korea 0.078 0.063
 Malaysia 0.033 0.031
 Mexico 0.034 0.021
 Russia -- --
 Thailand 0.057 0.043
 United States 0.190 0.210
 Venezuela 0.011 0.006
Venezuelan crisis
 Epicenter country -0.047 -0.033
 Brazil 0.054 0.044
 Canada 0.076 0.034
 Chile 0.017 0.010
 Ecuador 0.005 0.003
 Greece 0.033 0.019
 India 0.052 0.034
 Indonesia 0.038 0.019
 Italy 0.158 0.150
 Korea 0.082 0.066
 Malaysia 0.037 0.028
 Mexico 0.015 0.007
 Russia 0.000 0.000
 Thailand 0.064 0.043
 United States 0.196 0.210
 Venezuela 0.012 0.009
Asian crisis
 Epicenter country -0.234 -0.171
 Brazil 0.053 0.029
 Canada 0.082 0.035
 Chile 0.016 0.010
 Ecuador 0.005 0.004
 Greece 0.033 0.019
 India 0.050 0.040
 Indonesia 0.036 0.016
 Italy 0.153 0.145
 Korea 0.084 0.067
 Malaysia 0.039 0.028
 Mexico 0.030 0.015
 Russia 0.073 0.062
 Thailand 0.062 0.045
 United States 0.193 0.211
 Venezuela 0.011 0.007
Russian crisis
 Epicenter country -0.2418 -0.220
 Brazil 0.037 0.025
 Canada 0.087 0.041
 Chile 0.016 0.013
 Ecuador 0.012 0.004
 Greece 0.031 0.019
 India 0.050 0.041
 Indonesia 0.027 0.018
 Italy 0.156 0.145
 Korea 0.066 0.054
 Malaysia 0.031 0.022
 Mexico 0.030 0.010
 Russia 0.065 0.059
 Thailand 0.049 0.034
 United States 0.194 0.214
 Venezuela 0.018 0.009
Argentinean crisis
 Epicenter country 0.033 -0.022
 Brazil 0.047 0.029
 Canada 0.084 0.041
 Chile 0.014 0.010
 Ecuador 0.005 0.003
 Greece 0.034 0.022
 India 0.060 0.047
 Indonesia 0.033 0.018
 Italy 0.161 0.144
 Korea 0.074 0.067
 Malaysia 0.034 0.028
 Mexico 0.038 0.023
 Russia 0.058 0.048
 Thailand 0.059 0.037
 United States 0.201 0.226
 Venezuela 0.016 0.010

TABLE 7
Regression Results Using the Maximum Flow Measure as the Network
Indicator (Weekly Abnormal Stock Market Returns as the Dependent
Variable)

 1 2

 Threshold 0% (OLS) Threshold 0% (GLS)

Maximum Flow e to i 0.0003 (0.0402#) -0.0018 (-0.9747#)

Constant -1.5092 (-1.1856#) -1.0616 * (-4.0477#)

Bank reserves to 0.0023 (0.1308#) -0.0022 (-0.1870#)
assets ratio

Current account 0.0509 ** (2.4486#) 0.0445 * (4.8450#)
(% GDP)

Government 0.0510 * (2.6370#) 0.0419 * (4.8385#)
expenditures
(% GDP)

Private capital -0.0001 (-0.0324#) 0.0016 (1.3204#)
flows (% GDP)

Overall budget -0.0609 *** (-1.6749#) -0.0781 * (-5.6444#)
(% GDP)

GDP growth rate -0.0604 (-1.3840#) -0.0464 ** (-2.4114#)

Total trade (% GDP) 0.6554 *** (1.7051#) 0.4370 ** (2.4997#)

Inflation rate -0.0015 (-0.1505#) 0.0016 (0.2246#)
(yearly)

Adjusted [R.sup.2] .1379 .5386

 3 4

 Threshold 1% (OLS) Threshold 1% (GLS)

Maximum Flow e to i -0.0241 (-0.8496#) -0.0143 ** (-2.1535#)

Constant -1.1175 (-1.1138#) -0.9708 * (-5.2713#)

Bank reserves to 0.0035 (0.2012#) -0.0045 (-0.3973#)
assets ratio

Current account 0.0498 ** (2.1637#) 0.0398 * (4.2798#)
(% GDP)

Government 0.0519 ** (2.3819#) 0.0393 * (4.4502#)
expenditures
(% GDP)

Private capital 0.0000 (-0.0123#) 0.0018 (1.1982#)
flows (% GDP)

Overall budget -0.0577 (-1.5102#) -0.0711 * (-5.1946#)
(% GDP)

GDP growth rate -0.0630 (-1.2662#) -0.0447 ** (-2.3238#)

Total trade (% GDP) 0.6000 (1.5155#) 0.3404 ** (2.0349#)

Inflation rate -0.0033 (-0.3427#) 0.0030 (0.4085#)
(yearly)

Adjusted [R.sup.2] .1491 .5049

 5 6

 Threshold 2% (OLS) Threshold 2% (GLS)

Maximum Flow e to i -0.0351 (-1.0448#) -0.0112 (-1.2733#)

Constant -1.2885 (-1.6384#) -1.2212 * (-9.7512#)

Bank reserves to 0.0031 (0.1778#) -0.0060 (-0.5293#)
assets ratio

Current account 0.0493 ** (2.1455#) 0.0420 * (4.5953#)
(% GDP)

Government 0.0517 ** (2.3644#) 0.0405 * (4.7101#)
expenditures
 (% GDP)

Private capital 0.0000 (-0.0124#) 0.0019 (1.3854#)
flows (% GDP)

Overall budget -0.0558 (-1.4463#) -0.0768 * (-5.7897#)
(% GDP)

GDP growth rate -0.0615 (-1.2712#) -0.0371 ** (-1.8657#)

Total trade (% GDP) 0.6505 *** (1.7208#) 0.4380 ** (2.5411#)

Inflation rate -0.0026 (-0.2655#) 0.0044 (0.5752#)
(yearly)

Adjusted [R.sup.2] .1474 .6768

Notes: Number of observations for all regressions: 146.
t Statistics reported in italics

* 1%; ** 5%; *** 10%.

Note: t Statistics reported in italics is indicated with #.

TABLE 8
Regression Results Using Alternative Specifications and Alternative
Definition for the Weekly Abnormal Stock Market Return

Alternative Specification Using Weekly Abnormal Stock Market Return
as the Dependent Variable

 1 2

 Node Importance IV: Node Importance
 World Trade Share IV: Centrality 1%

Epicenter network -22.8937 * (-2.8019#) -7.6602 ** (-2.2780#)
indicator

Country i network 4.3235 * (3.0697#) 4.7991 * (4.1350#)
indicator

Epicenter centrality

Country i centrality

Constant -0.7965 * (-3.0964#) -0.6583 * (-2.8760#)

Bank reserves to -0.0005 * (-0.0497 -0.0039 (-0.4865#)
assets ratio

Current account (% 0.0305 * (3.1176#) 0.0228 ** (2.5515#)
GDP)

Government 0.0490 * (4.3042#) 0.0354 * (3.3391#)

expenditures (% GDP)

Private capital flows -0.0029 ** (-2.4819#) -0.0033 * (-3.1151#)
(% GDP)

Overall budget (%
GDP)

GDP growth rate

Total trade (% GDP)

Inflation rate
(yearly)

Trade with epicenter -12.6108 * (-3.8955#) -13.0643 * (-4.2602#)

Adjusted [R.sup.2] .3402 .3989

 3 4

 Node Centrality Node Importance IV:
 Threshold: 1% World Trade Share

Epicenter network -55.198 * (-12.2064#)
indicator

Country i network 5.832 * (5.2553#)
indicator

Epicenter centrality -0.0077 ** (-2.2776#)

Country i centrality 0.0048 * (4.1339#)

Constant -0.6582 * (-2.8754#) 0.0998 (0.6824#)

Bank reserves to -0.0039 (-0.4870#)
assets ratio

Current account (% 0.0228 ** (2..5508#)
GDP)

Government 0.0354 * (3.3380#)

expenditures (% GDP)

Private capital flows -0.0033 * (-31156#)
(% GDP)

Overall budget (% -0.0646 * (-4.3904#)
GDP)

GDP growth rate -0.0712 * (-3.2327

Total trade (% GDP) 0.5701 * (4.1892#)

Inflation rate -0.0025 (-0.5232#)
(yearly)

Trade with epicenter -13.065 * (-4.2606#) -8.0104 ** (-2.3027

Adjusted [R.sup.2] .3989 .7232

 5 6

 Node Importance Node Centrality
 IV: Centrality 1% Threshold: 1%

Epicenter network -10.1355 * (-3.4395#)
indicator

Country i network 4.9535 * (4.2450#)
indicator

Epicenter centrality -0.0101 * (-3.4386#)

Country i centrality 0.0050 * (4.2439#)

Constant -0.0746 (-0.5173#) -0.0744 (-0.5157#)

Bank reserves to
assets ratio

Current account (%
GDP)

Government

expenditures (% GDP)

Private capital flows
(% GDP)

Overall budget (% -0.0302 ** (-2.4257#) -0.0302 ** (-2.4227
GDP)

GDP growth rate -0.0584 * (-3.1480#) -0.0585 * (-3.1493#)

Total trade (% GDP) 0.3127 ** (2.3293#) 0.3125 ** (2.3276#)

Inflation rate -0.0059 (-0.9315#) -0.0059 (-0.9317
(yearly)

Trade with epicenter -9.5255 ** (-2.2177#) -9.5258 ** (-2.2176#)

Adjusted [R.sup.2] .3186 .3186

 7/1 8/1

 Node Importance IV: Node Importance
 World Trade Share IV: Centrality 1%

Epicenter network -52.898 * (-6.0751#) -15.7774 * (-4.0633#)
indicator

Country i network 3.475 ** (2.4906#) 1.3995 (1.0369#)
indicator

Epicenter centrality

Country i centrality

Constant -0.4002 (-1.5609#) -0.3854 (-1.5031#)

Bank reserves to 0.0(#)77 (0.5497#) -0.0006 (-0.0464#)
assets ratio

Current account (% 0.0322 ** (2.6075#) 0.0400 * (3.5238#)
GDP)

Government 0.0374 * (3.3742#) 0.0332 * (2.9008#)

expenditures (% GDP)

Private capital flows 0.0035 *** (1.6766#) 0.0032 (1.4361#)
(% GDP)

Overall budget (% -0.0347 ** (-1.9841#) -0.0282 *** (-1.9226#)
GDP)

GDP growth rate -0.0703 * (-2.9771#) -.0.0388 ** (-2.3199#)

Total trade (% GDP) 0.2274 (1.1042#) 0.0001 (0.0003#)

Inflation rate -0.0028 (-0.4089#) -0.0032 (-0.4396#)
(yearly)

Trade with epicenter -5.6760 (-1.6424#) -9.3390 (-3.1005#)

Adjusted [R.sup.2] .4674 .5014

 9/1 10

 Node Centrality Node Importance IV:
 Threshold 1% World Trade Share

Epicenter network -47.9507 * (-8.5640#)
indicator

Country i network 2.5209 ** (2.4872#)
indicator

Epicenter centrality -0.0158 * (-4.0607#)

Country i centrality 0.0014 (1.0351#)

Constant -0.3852 (-1.5024#) 0.2598 (1.5135#)

Bank reserves to -0.0006 (-0.0470#) -0.0048 (-0.6580
assets ratio

Current account (% 0.0400 * (3.5239#) 0.0096 (0.9712#)
GDP)

Government 0.0332 * (2.9005#) 0.0090 (1.0914#)

expenditures (% GDP)

Private capital flows 0.0032 (1.4350#) -0.0037 *** (-1.7670
(% GDP)

Overall budget (% -0.0282 *** (-1.9225#) -0.0192 (-1.4069#)
GDP)

GDP growth rate -0.0388 ** (-2.3201#) -0.0474 * (-3.4814#)

Total trade (% GDP) -0.002 (-0.0008#) 0.4162 * (2.9172#)

Inflation rate -0.0032 (-0.4395#) -0.0041 (-4.5919#)
(yearly)

Trade with epicenter -9.3437 * (-3.1029#) -6.7956 ** (-2.2431#)

Adjusted [R.sup.2] .5015 .4334

 11 12

 Node Importance Node Centrality
 IV: Centrality 1% Threshold 1%

Epicenter network -16.2528 * (-4.6569#)
indicator

Country i network 2.9755 * (2.8075#)
indicator

Epicenter centrality -0.0163 * (-4.6557#)

Country i centrality 0.0030 * (2.8062#)

Constant 0.1440 (0.6995#) 0.1441 (0.7003#)

Bank reserves to -0.0091 (-1.1730#) -0.0091 (-1.1738#)
assets ratio

Current account (% 0.0126 (1.3102#) 0.0126 (1.3102#)
GDP)

Government 0.0042 (0.5200 0.0042 (0.5195#)

expenditures (% GDP)

Private capital flows -0.0036 ** (-2.0182#) -0.0036 ** (-2.0191#)
(% GDP)

Overall budget (% -0.0109 (-1.1652#) -0.0109 (-1.1657#)
GDP)

GDP growth rate -0.0307 ** (-2.1042#) -0.0307 ** (-2.1038#)

Total trade (% GDP) 0.3303 ** (2.1914#) 0.3301 ** (2.1902#)

Inflation rate -0.0023 ** (-2.4231#) -0.0023 ** (-2.4229#)
(yearly)

Trade with epicenter -6.2261 *** (-1.7988#) -6.2278 *** (-1.7990

Adjusted [R.sup.2] .3338 .3337

 13 14

 Node Importance IV: Node Importance
 World Trade Share IV: Centrality 1%

Epicenter network -32.1211 * (-4.3243#) -10.1211 * (-3.7543#)
indicator

Country i network 2.6763 *** (1.9387#) 3.4743 * (3.5674#)
indicator

Epicenter centrality

Country i centrality

Constant 0.1528 (0.7176#) 0.1013 (0.6100#)

Bank reserves to -0.0187 ** (-2.4863#) -0.0175 * (-2.7631#)
assets ratio

Current account (% 0.0159 *** (1.9030 0.0204 ** (2.5690
GDP)

Government 0.0150 *** (1.7844#) 0.0073 (1.0592#)

expenditures (% GDP)

Private capital flows -0.0052 * (-6.7327#) -0.0051 * (-7.7680#)
(% GDP)

Overall budget (%
GDP)

GDP growth rate

Total trade (% GDP)

Inflation rate
(yearly)

Trade with epicenter -10.0586 * (-3.9384#) -9.8769 * (-3.3558#)

Adjusted [R.sup.2] .3580 .3908

 15 16

 Node Centrality Node Importance IV:
 Threshold: 1% World Trade Share

Epicenter network -56.523 * (-19.7835#)
indicator

Country i network 3.099 * (5.1514#)
indicator

Epicenter centrality -0.0101 (-3.7539#)

Country i centrality 0.0035 * (3.5672#)

Constant 0.1014 (0.6103#) 0.5049 * (4.8686#)

Bank reserves to -0.0175 * (-2.7634#)
assets ratio

Current account (% 0.0204 ** (2.5690
GDP)

Government 0.0073 (1.0587#)

expenditures (% GDP)

Private capital flows -0.0051 * (-77698#)
(% GDP)

Overall budget (% -0.0205 *** (-1.8551#)
GDP)

GDP growth rate -0.0603 * (-4.8081#)

Total trade (% GDP) 0.3342 * (3.5208#)

Inflation rate -0.0050 * (-5.9415#)
(yearly)

Trade with epicenter -9.8766 * (-3.3554#) -8.1938 * (-2.6997#)

Adjusted [R.sup.2] .3909 .7633

 17 18

 Node Importance Node Centrality
 IV: Centrality 1% Threshold: 1%

Epicenter network -17.1017 * (-6.4823#)
indicator

Country i network 3.7949 * (4.7729#)
indicator

Epicenter centrality -0.0171 * (-6.4803#)

Country i centrality 0.0038 * (4.7705#)

Constant 0.0881 (0.6521#) 0.0881 (0.6523#)

Bank reserves to
assets ratio

Current account (%
GDP)

Government

expenditures (% GDP)

Private capital flows
(% GDP)

Overall budget (% -0.0232 * (-3.8225#) -0.0232 * (-3.8215#)
GDP)

GDP growth rate -0.0419 * (-3.6318#) -0.0419 * (-3.6309#)

Total trade (% GDP) 0.3578 * (3.1265#) 0.3577 * (3.1259#)

Inflation rate -0.0033 * (-3.7981#) -0.0033 * (-3.7977#)
(yearly)

Trade with epicenter -5.7028 *** (-1.7094#) -5.7036 (-1.7093#)

Adjusted [R.sup.2] .4865 .4861

Notes: Number of observations: 158 (1-3), 147 (4-6), 156 (7-9), 174
(10-12), and 161 (13-15). t Statistics reported in italics.
/1 indicates those regressions that include the Argentinean crisis
window, and these results are discussed in the Section V. IV,
intrinsic value.

* 1%; ** 5%; *** 10%.

Note: t Statistics reported in italics is indicated #.

TABLE 9
Regression Results Using the GDP-Based Network Indicators with
Alternative Specifications and Alternative Definition for the Weekly
Abnormal Stock Market Return

GDP-Based Network Indicators Using the Weekly Abnormal Stock Market
Return as the Dependent Variable

 1 2

 Node Importance IV: Node Importance
 World Trade Share IV: Centrality 1%

Epicenter network -44.9216 * (-5.9292#) -10.1070 ** (-2.1607#)
indicator

Country i network 3.9145 * (3.1751#) 3.0183 ** (2.0844#)
indicator

Epicenter
centrality

Country i
centrality

Constant -0.4864 ** (-2.1768#) -0.4201 *** (-1.7206#)

Bank reserves to -0.0052 (-0.4603#) -0.0136 (-1.3303#)
assets ratio

Current account (% 0.0217 ** (2.0577#) 0.0270 * (2.8340#)
GDP)

Government 0.0309 * (3.1262#) 0.0253 ** (2.4790#)
expenditures (%
GDP)

Private capital 0.0002 (0.1208#) -0.0007 (-0.4573#)
flows (% GDP)

Overall budget (% -0.0566 * (-3.6587#) -0.0453 * (-3.4421#)
GDP)

GDP growth rate -0.0662 * (-3.0620#) -0.0430 ** (-2.5248#)

Total trade (% GDP) 0.3884 ** (2.0759#) 0.0716 (0.3444#)

Inflation rate 0.0028 (0.4750#) 0.0014 (0.2348#)
(yearly)

Trade with -5.2715 (-1.6030#) -9.8782 * (-2.8918#)
epicenter

Adjusted [R.sup.2] .4725 .3941

 3 4

 Node Centrality Node Centrality IV:
 Threshold 1% World Trade Share

Epicenter network -22.914 * (-2.8020#)
indicator

Country i network 4.326 * (3.0697#)
indicator

Epicenter -0.0101 ** (-2.1666#)
centrality

Country i 0.0030 ** (2.0901#)
centrality

Constant -0.4188 *** (-1.7175#) -0.7964 * (-3.0964#)

Bank reserves to -0.014 (-1.3256#) -0.0005 (-0.0497#)
assets ratio

Current account (% 0.0270 * (2.8367#) 0.0305 * (3.1176#)
GDP)

Government 0.0253 ** (2.4751#) 0.0490 * (4.3043#)
expenditures (%
GDP)

Private capital -0.0007 (-0.4622#) -0.0029 ** (-2.4814#)
flows (% GDP)

Overall budget (% -0.045 * (-3.4512#)
GDP)

GDP growth rate -0.043 ** (-2.5220#)

Total trade (% GDP) 0.0704 (0.3391#)

Inflation rate 0.0013 (0.2292#)
(yearly)

Trade with -9.8623 * (-2.8893#) -12.6103 * (-3.8953#)
epicenter

Adjusted [R.sup.2] .3945 .3402

 5 6

 Node Importance Node Centrality
 Threshold 1% Threshold 1%

Epicenter network -11.539 * (-2.6126#)
indicator

Country i network 6.566 * (4.6654#)
indicator

Epicenter -0.012 * (-2.6217#)
centrality

Country i 0.0066 * (4.6664#)
centrality

Constant -0.7046 * (-3.1556#) -0.703 * (-3.1496#)

Bank reserves to -0.0027 (-0.3545#) -0.0027 (-0.3526#)
assets ratio

Current account (% 0.0249 * (2.9012#) 0.0248 * (2.8984#)
GDP)

Government 0.0376 * (3.7122#) 0.0376 * (3.7057#)
expenditures (%
GDP)

Private capital -0.0029 * (-2.7904#) -0.003 * (-2.8059#)
flows (% GDP)

Overall budget (%
GDP)

GDP growth rate

Total trade (% GDP)

Inflation rate
(yearly)

Trade with -14.1647 * (-3.7507#) -14.1492 * (-5.7404#)
epicenter

Adjusted [R.sup.2] .4874 .4879

 7 8

 Node Importance IV: Node Importance
 World Trade Share IV: Centrality I%
 (GLS) (GLS)

Epicenter network -55.241 * (-12.1875#) -11.9355 * (-2.9849#)
indicator

Country i network 5.837 * (5.2554#) 6.0303 (4.3755#)
indicator

Epicenter
centrality

Country i
centrality

Constant 0.0997 (0.6813#) -0.1015 (-0.7213#)

Bank reserves to
assets ratio

Current account (%
GDP)

Government
expenditures (%
GDP)

Private capital
flows (% GDP)

Overall budget (% -0.0646 * (-4.3900#) -0.0319 * (-2.7095#)
GDP)

GDP growth rate -0.0712 * (-3.2326#) -0.0514 * (-2.9253#)

Total trade (% GDP) 0.5702 * (4.1891#) 0.3196 ** (2.4075#)

Inflation rate -0.0025 (-0.5230#) -0.0066 (-1.0585#)
(yearly)

Trade with -8.0089 ** (-2.3023#) -10.8750 * (-2.6840#)
epicenter

Adjusted [R.sup.2] .7220 .3280

 9 10

 Node Centrality Node Importance IV:
 Threshold 1% World Trade Share
 (GLS)

Epicenter network -47.9977 * (-83598#)
indicator

Country i network 2.5225 ** (2.4861#)
indicator

Epicenter -0.0120 * (-2.9952#)
centrality

Country i 0.0060 * (4.3838#)
centrality

Constant -0.1018 (-0.7239#) 0.2596 (1.5126#)

Bank reserves to -0.0048 (-0.6577#)
assets ratio

Current account (% 0.0096 (0.9714#)
GDP)

Government 0.0090 (1.0920#)
expenditures (%
GDP)

Private capital -0.0037 *** (-1.7664#)
flows (% GDP)

Overall budget (% -0.0320 * (-2.7172#) -0.0192 (-1.4065#)
GDP)

GDP growth rate -0.0513 * (-2.9180#) -0.0474 * (-3.4815#)

Total trade (% GDP) 0.3194 ** (2.4066#) 0.4163 * (2.9179#)

Inflation rate -0.0067 (-1.0590#) -0.0041 * (-4.5918#)
(yearly)

Trade with -10.8274 * (-2.6765#) -6.7930 ** (-2.2423#)
epicenter

Adjusted [R.sup.2] .3280 .4333

 11 12

 Node Importance Node Centrality
 IV: Centrality 1% Threshold 1%

Epicenter network -20.1186 * (-4.3026#)
indicator

Country i network 4.1077 * (3.1562#)
indicator

Epicenter -0.0201 * (-43051#)
centrality

Country i 0.0041 * (3.1527#)
centrality

Constant 0.1060 (0.5274#) 0.1073 (0.5346#)

Bank reserves to -0.0117 (-1.5652#) -0.0117 (-1.5664#)
assets ratio

Current account (% 0.0104 (1.1260#) 0.0105 (1.1290#)
GDP)

Government 0.0026 (03338#) 0.0026 (0.3315#)
expenditures (%
GDP)

Private capital -0.0025 (-1.5564#) -0.0025 (-1.5602#)
flows (% GDP)

Overall budget (% -0.0139 (-1.5180#) -0.0140 (-1.5228#)
GDP)

GDP growth rate -0.0283 *** (-1.9736#) -0.0283 *** (-1.9767#)

Total trade (% GDP) 0.2969 *** (1.9122#) 0.2960 *** (1.9068#)

Inflation rate -0.0021 ** (-2.1895#) -0.0021 ** (-2.1957#)
(yearly)

Trade with -7.6226 ** (-2.2932#) -7.6148 ** (-2.2918#)
epicenter

Adjusted [R.sup.2] .3518 .3519

 13 14

 Node Importance IV: Node Importance
 World Trade Share IV: Centrality 1%

Epicenter network -32.1474 * (-4.3244#) -13.1062 * (-3.2947#)
indicator

Country i network 2.6778 *** (1.9387#) 4.6996 * (3.8313#)
indicator

Epicenter
centrality

Country i
centrality

Constant 0.1527 (0.7174#) 0.0952 (0.6039#)

Bank reserves to -0.0187 ** (-2.4864#) -0.0187 * (-3.0156#)
assets ratio

Current account (% 0.0159 *** (1.9032#) 0.0176 ** (2.2835#)
GDP)

Government 0.0150 *** (1.7847#) 0.0072 (1.1076#)
expenditures (%
GDP)

Private capital -0.0052 * (-6.7320#) -0.0048 * (-7.8231#)
flows (% GDP)

Overall budget (%
GDP)

GDP growth rate

Total trade (% GDP)

Inflation rate
(yearly)

Trade with -10.0576 * (-3.9381#) -11.5137 * (-4.0544#)
epicenter

Adjusted [R.sup.2] .3579 .4394

 15 16

 Node Centrality Node Importance IV:
 Threshold 1% World Trade Share

Epicenter network -56.579 * (-19.7513#)
indicator

Country i network 3.101 * (5.1473#)
indicator

Epicenter -0.0131 * (-3.3018#)
centrality

Country i -0.0047 * (-3.8284#)
centrality

Constant 0.0962 (0.6110#) 0.5048 * (4.8665#)

Bank reserves to -0.0188 * (-3.0187#)
assets ratio

Current account (% 0.0176 ** (2.2849#)
GDP)

Government 0.0071 (1.1036#)
expenditures (%
GDP)

Private capital -0.0048 * (-7.8453#)
flows (% GDP)

Overall budget (% -0.0205 *** (-1.8547#)
GDP)

GDP growth rate -0.0603 * (-4.8084#)

Total trade (% GDP) 0.3343 * (3.5221#)

Inflation rate -0.0050 * (-5.9414#)
(yearly)

Trade with -11.5155 * (-4.0566#) -8.1905 * (-2.6989#)
epicenter

Adjusted [R.sup.2] .4403 .7625

 17 18

 Node Importance Node Centrality
 IV: Centrality 1% Threshold 1%

Epicenter network -20.4260 * (-5.3842#)
indicator

Country i network 5.3643 * (5.2565#)
indicator

Epicenter -0.0205 * (-3.6044#)
centrality

Country i 0.0054 * (5.2632#)
centrality

Constant 0.0074 (0.0587#) -0.0073 (0.0582#)

Bank reserves to
assets ratio

Current account (%
GDP)

Government
expenditures (%
GDP)

Private capital
flows (% GDP)

Overall budget (% -0.0245 * (-3.7323#) -0.0246 * (-3.7495#)
GDP)

GDP growth rate -0.0383 * (-3.2846#) -0.0382 * (-3.2795#)

Total trade (% GDP) 0.3674 * (3.1341#) 0.3672 * (3.1330#)

Inflation rate -0.0033 * (-3.2024#) -0.0033 * (-3.2002#)
(yearly)

Trade with -7.2006 ** (-2.2006#) -7.1723 ** (-2.1967#)
epicenter

Adjusted [R.sup.2] .4440 .4452

Notes: Number of observations: 142 (1-3), 158 (4-6), 147 (7-9),
155 (10-12), 174 (13-15), and 161 (16-18). t Statistics reported
in italics. IV, intrinsic value.

* 1%; ** 5%; *** 10%.

Note: t Statistics reported in italics

TABLE 10
Regression Results Using a Fixed Time Period Effects Specification,
Node Importance, and Centrality Indices as the Network Indicators
(Weekly Abnormal Stock Market Returns as the Dependent Variable)

 1 2

 IV: World Trade Share IV: Centrality 1%

Country i 3.6691 * (2.9441) 1.6842 (1.0503)
importance

Country i
centrality

Constant -1.2111 * (-5.1765) -1.0435 * (-5.3488)

Dummy Venezuela 0.7728 * (5.6581) 0.7573 * (5.6346)

Dummy Thailand 0.2905 ** (2.1183) 0.2812 ** (2.1837)

Dummy Russia 0.2640 *** (1.8777) 0.2480 *** (1.8811)

Dummy Argentina 0.2571 *** (1.7251) 0.2487 *** (1.7080)

Bank reserves to -0.0003 (-0.0257) -0.0026 (-0.2367)
assets ratio

Current account (% 0.0295 * (2.6509) 0.0312 * (2.7183)
GDP)

Government 0.0309 * (3.1121) 0.0306 * (3.1429)
expenditures (%
GDP)

Private capital 0.0014 (0.8194) 0.0005 (0.2525)
flows (% GDP)

Overall budget (% -0.0442 * (-2.6795) -0.0375 ** (-2.2766)
GDP)

GDP growth rate -0.0736 * (-3.0157) -0.0800 * (-3.1902)

Total trade (% GDP) 0.3224 (1.5649) 0.2441 (1.2088)

Inflation rate 0.00004 (0.0067) -0.0010 (-0.1646)
(yearly)

Trade with -5.0634 (-1.2478) -5.7433 (-1.3404)
epicenter

Adjusted [R.sup.2] .4342 .4777

 3 4

 Threshold: 0% Threshold: 1%

Country i
importance

Country i 0.0064 ** (2.07718) 0.0017 (1.04702)
centrality

Constant -1.5191 * (-4.5087) -1.0432 * (-5.3472)

Dummy Venezuela 0.7633 * (5.7709) 0.7573 * (5.6340)

Dummy Thailand 0.2653 *** (1.9764) 0.2812 ** (2.1837)

Dummy Russia 0.2345 *** (1.7125) 0.2480 *** (1.8813)

Dummy Argentina 0.2693 *** (1.8051) 0.2486 *** (1.7075)

Bank reserves to 0.0001 (0.0109) -0.0026 (-0.2370)
assets ratio

Current account (% 0.0283 ** (2.4129) 0.0312 * (2.7185)
GDP)

Government 0.0321 * (3.2972) 0.0306 * (3.1431)
expenditures (%
GDP)

Private capital -0.0001 (-0.0746) 0.0005 (0.2521)
flows (% GDP)

Overall budget (% -0.0294 *** (-1.7687) -0.0375 ** (-2.2764)
GDP)

GDP growth rate -0.0732 * (-2.8514) -0.0800 * (-3.1909)

Total trade (% GDP) 0.2277 (1.1613) 0.2440 (1.2083)

Inflation rate 0.0000 (0.0017) -0.0010 (-0.1649)
(yearly)

Trade with -4.8018 (-1.1025) -5.7448 (-1.3407)
epicenter

Adjusted [R.sup.2] .4426 .4778

 5

 Threshold: 2%

Country i
importance

Country i 0.0024 (1.53782)
centrality

Constant -1.0739 * (-5.6257)

Dummy Venezuela 0.7576 * (5.7608)

Dummy Thailand 0.2823 ** (2.1784)

Dummy Russia 0.2455 *** (1.8396)

Dummy Argentina 0.2552 *** (1.7648)

Bank reserves to -0.0030 (-0.2770)
assets ratio

Current account (% 0.0313 * (2.7995)
GDP)

Government 0.0309 * (3.1407)
expenditures (%
GDP)

Private capital 0.0006 (0.3062)
flows (% GDP)

Overall budget (% -0.0388 ** (-2.3871)
GDP)

GDP growth rate -0.0776 * (-3.0945)

Total trade (% GDP) 0.2732 (1.3278)

Inflation rate -0.0007 (-0.1071)
(yearly)

Trade with -5.6116 (-1.3145)
epicenter

Adjusted [R.sup.2] .4646

Notes: Number of observations for all regressions: 142.
t Statistics reported in italics. IV, intrinsic value.

* 1%; ** 5%; *** 10%.

Note: t Statistics reported in italics is indicated with #.