Credit Chain and Sectoral Co-movement: A Multi-Region Investigation

This paper empirically examines how sectoral co-movement are correlated with trade credit usage in a multi-region setting. Extending the models in Shea (2002) and Raddatz (2010), we develop a framework that captures the impact of trade credit usage on co-movement between sectors within a country and cross countries separately. Using the Multi-Regional Input-Output Table developed by Asian Development bank, we assemble a dataset consisting of 14 manufacturing industries for 53 economies. We provide empirical evidence that trade credit linkage is an influential channel for both the domestic and cross-border shocks to propagate and to create a more profound impact in industries around the globe. We find the impact of domestic credit chain on sectoral co-movement is twice as strong as the international ones. We further examine the time trend of this relationship, and find that from 2000 to 2018, the positive relationship between the intensity of trade credit usage and sectoral correlation decreases. We posit that this could be due to a more diversified global trade pattern changes during these two decades.


Introduction
As the volume of global trade nearly quadrupled over the last two decades, the global supply chain becomes increasingly interconnected. While such interconnection allows companies to lower cost, it also results co-movement across different sectors and regions. Consequently, a shock originates from one specific sector or one geographic region could cascade to different sectors across the globe.
For example, in 2011, Thailand experienced severe flooding during the monsoon season. In addition to hundreds of deaths, the floods also caused tremendous economic damages. The damages are mostly due to the disruption in the manufacturing industry, where several industrial estates were inundated in water as much as 3 meters deep during the flood. These industrial estates host a large amount of global production capacity in a few sectors. For example, as the second largest hard disk drives producer in the world, Thailand produced approximately a quarter of total hard drives in 1 the world. As many of these factories were flooded, it caused a global hard drive shortage, pushing up hard drive prices and affecting the profitability of downstream sectors. Such shocks propagated far beyond Thailand. As Japanese companies such as Canon, Honda, and Toyota all produce a large share of products and/or components in Thailand, the flood has negatively influenced not only these companies' profit, but also worker incomes in Japan.
Natural disaster is only one type of shocks that demonstrate how the global supply chain are intertwined. Other shocks, from the Global Financial Crisis in 2008 to the Covid-19 pandemic, all highlighted that sectors and regions are closely linked by the flow of physical goods. In addition to such physical flow, different sectors and regions also experience co-movement due to the financial flow in between. One of the most important financial linkage between companies within supply chains is trade credit. Trade credit (credit extended by sellers to buyers in supply chains) is one of the most important sources of external financing around the world (Rajan and Zingales 1995).
For example, in the United States, trade credit amounted to $5.4 trillion in 2019, with its growth outpacing that of GDP (Federal Reserve Board 2019). Thus, it is a financial instrument with macroeconomic significance.
Trade credit creates credit chains across firms in different sectors of the economy (Kiyotaki and Moore 1998). That is, different sectors in the economy are not only linked through business transactions (sellers and buyers of products and service), but also financially by providing and receiving trade credit from each other. An important implication of such credit chains is that trade credit could serve as an additional channel that links different sectors together. Thus, it is natural to hypothesize that the intensity of trade credit usage has a positive influence on sectoral correlation.
Put differently, if sector A (buyer) receives a lot of trade credit from sector B (supplier), then a shock in sector A could translate into a larger shock in sector B relative to those sector pairs with less trade credit in between.
The above hypothesis was first confirmed empirically by Raddatz (2010) by combining standard input-output matrix from US Bureau of Labor Statistics and firm financial data from WorldScope and CompuStat. However, due to data limitation, a few questions remains unanswered in the paper: first, without a multi-region input and output matrix, Raddatz (2010) studies exclusively transactions within each country. Thus, it is unclear whether such credit chains act differently for domestic and international trade. Second, Raddatz (2010) focuses on a single time period in the early 2000s. However, the development of global supply chain in the last two decades begs the question whether the dynamics between credit chain and sectoral co-movement has changed over time. Finally, as an increasingly important powerhouse of the global economy, do the Asia countries exhibit any special patterns?
Electronic copy available at: https://ssrn.com/abstract=3897212 To answer the above questions, in this paper, we first extend the model in Shea (2002) and Raddatz (2010) to decompose the impact of trade credit usage on sectoral co-movement by domestic and cross-border trade. Specifically, our model allows the the shock propagation via domestic credit chains has a different rate from that via cross-border chains. This allows us to empirically examine whether within-and cross-country credit chains have different impact on sectoral co-movement.
We then construct a new dataset to identify these two effects. Specifically, we triangulate the financial data from Compustat and Worldscope, the input- output table from  Further, we examine how the correlation between trade credit usage and sectoral co-movement evolves throughout time by using different MRIO data at 2000MRIO data at , 2007MRIO data at , 2013MRIO data at , and 2018. We find that both the positive relation between trade-credit usage intensity and correlations decreases. We posit that this could be due to a more dissected supply chain structures that enables industries to diversify their production as well as trade credit linkage intensity from a domestic based to foreign based. By ruling out a number of plausible explanations, we provide robust empirical support to our hypothesis.
Finally, similar to Raddatz (2010), we investigate the interaction between bank credit and trade credit, with a special focus on whether such interaction is different between Asian countries and the rest of the world. We find that while the availability of bank financing does not help mitigate risk propagation in the rest of the world, it does help when both parties in the transaction are located in Asia, including domestic trade within an Asian country, or cross-border trade between two Asian countries.
Our paper contributes to two strands of research: trade credit and sectoral co-movement. There is a rich literature on trade credit, both theoretical and empirical. The theoretical literature of trade credit focuses on explaining the existence of trade credit, that is, why buyers borrow money from sellers in the presence of specialized financial institutions. This literature dates back to at least in the 1950s, when Meltzer highlights that trade credit is used as an effective marketing tool. Following research identified various reasons why trade credit is adopted, such as suppliers having easier access to financing (Schwartz 1974), price discrimination (Brennan et al. 1988), quality assurance Electronic copy available at: https://ssrn.com/abstract=3897212 (Long et al. 1993, Babich andTang 2012), alleviating moral hazard (Burkart and Ellingsen 2004), facilitating relationship specific investment (Cuñat 2007), demand risk-sharing (Kouvelis andZhao 2012, Yang andBirge 2018), softening competition (Peura et al. 2017).
Relative to the theoretical literature, the empirical literature of trade credit is relatively new, which is largely attributable to data limitation. Earlier research in this field focuses on validating theories of trade credit and identifying determinants of trade credit, such as Petersen and Rajan (1997), Ng et al. (1999) and Giannetti et al. (2011). In a developing setting, McMillan and Woodruff (1999) document that trade credit is closely related to relationship building. Fisman and Love (2003) find that trade credit access could facilitate industry growth. Boissay and Gropp (2013) and Jacobson and Von Schedvin (2015) quantify how trade credit default cascades along the supply chain. Lee et al. (2018) and Chod et al. (2019) examine the interaction between trade credit usage and the horizontal relationship between firms. Most recently, several studies focused on identifying the causal relationship between trade credit and operational and financial performance. Barrot (2016) shows that limiting trade credit provision improves the financial strength of upstream firms; Breza and Liberman (2017) and Chen et al. (2020), on the other hand, find that restricting trade credit provision reduces transaction volume between different firms and negatively affect downstream firms' investment and revenue.
On the other hand, sectoral co-movement is an important topic in macroeconomics. Macroeconomics model are in general concerned with business cycles, both correlated movements in economywide output over time and co-movement between sectors (Lucas 1995). Long Jr and Plosser (1983) develop a multi-sector economy to capture shocks moving along input-output linkages. The model is further extended by Shea (2002). Lilien (1982) points out that labor movement could be another source of linkage between sectoral co-movement. Cooper and Haltiwanger (1990) develop a dynamic model to capture the sectoral co-movement with inventory and offer empirical evidence to support their theory.
Combining the above two streams of work, Kiyotaki and Moore (1998) is the first to formalize the idea of credit chain, that is, different sectors are linked financially. This model is further extended by Cardoso-Lecourtois (2004) and Boissay (2006). These literature highlight that through trade credit, sectoral shocks can move both from upstream to downstream, but also the other way around.
Finally, Raddatz (2010) provide empirical evidence that the usage of trade credit does have a material impact on sectoral co-movement.
Our paper extend Raddatz (2010) in three aspects: first, we extend the model in Raddatz (2010) to distinguish domestic and international trade. Second and relatedly, using the multi-regional input-output table developed by Asian Development Bank, we empirically quantify the impact of Electronic copy available at: https://ssrn.com/abstract=3897212 trade credit on sectoral co-movement for domestic and international transactions separately, revealing that such impact is stronger within country than cross-border trade. Finally, unlike Raddatz (2010), which documents results from one year due to data limitation, we conduct the empirical analysis over nearly two decades. This longitude analysis shows the aforementioned impact change over time, implying important trend in global supply chain.

The Model
This section extends the model in Raddatz (2010) that examines the intra-country sectoral comovement to analyze the sectoral co-movement not only within but also cross various countries.
To do so, we consider an economy comprised of J sectors (indexed by subscripts i and j) and N countries (indexed by superscripts n and m). Following the notation in Raddatz (2010), we represent sectoral output fluctuations for sectors 1 to J at countries 1 to N in the following reduced form: (1) In the above equation, λ is also a (JN ) × 1 vector consists of elements λ m i for i = 1, ..., J and m = 1, ..., N , where λ m i represents the sectoral shocks for sector i at country m. B is a (JN ) × (JN ) matrix with elements b mn ij representing the share of total demand faced by sector i in country m (the supplier) directly attributable to sector j in country n (the customer). Combined, y is a (JN ) × 1 vector with elements y m i for i = 1, ..., J and m = 1, ..., N , where y m i represents the sectoral output fluctuations for sector i at country m.
To build trade credit into the model, denote P mn ij ∈ [0, 1] as the fraction of direct demand b mn ij supplied by trade credit. Thus, we have: If trade credit has an additional effect on the transmission of shocks (let the impact coefficient be α mn ij ), the coefficient of direct linkages would be b mn ij (1 + α mn ij P mn ij ), in which we have a maximum of (JN ) 2 different α mn ij . To ensure that the model is tractable and that it could fits with data for estimation, we make the following two assumptions.
Assumption 1. P mn ij is constant across suppliers. That is, ∀ i and m, P mn ij = P n j .
Assumption 1 is similar to Raddatz (2010). Here, we assume that P mn ij is constant across suppliers regardless of which sector they are selling to and which country the customers are located at. This assumption allows us to allocate trade credit received by a customer to be proportionally allocated Electronic copy available at: https://ssrn.com/abstract=3897212 to all its suppliers, as a firm only report the aggregated trade credit (account payables) it receives from all suppliers. With this assumption, ∀i, ∀m P mn ij = P n j . After (A1), we have reduced P mn ij (and therefore α mn ij ) to JN . Assumption 2 states that we only distinguish between domestic (D) and cross-border (F ) trade credit. With Assumption 2, we can separate the effects of domestic and cross-border trade credit, that is, where B D captures domestic trade and B F captures cross-border trade. Let P be a diagonal matrix with elements P n j , we then can rewrite Eq.
(2) as: Define JN × JN matrix A(α, B, P ) as We can then write y as Taking a linear approximation to A around α D = 0 and α F = 0, we obtain: where If we assume that country-sector shocks λ n j are independent and identically distributed, the correlation between sector i in country m and sector k in country o is: in which a mn ij is the ((m, i), (n, j)) element of the matrix A. Taking a first-order approximation, Eq. (12) can be re-written as:   (16) Intuitively, if the use of trade credit along the chain linking (n, j) and other state pairs is higher than average and is important, then shocks to this (n, j) pair increases the correlation between pairs (m, i) and (o, k).
Finally, Eq. (13) also motivates us to test whether α D = 0 or α F = 0 based on the following: in which θ represent various fixed effects, 1 G mo ik is the physical linkage between (m, i) and (o, k), which is computed as the first term in Eq. (13), C Dmo ik and C F mo ik captures trade credit usage, and W mo ik represents other determinants of sectoral correlation.

Data and Variable Constructions
In order to test Eq. (18), we need three major data inputs: 1. the use of trade credit P ; 2. the input-output linkages B and D; 3. the sectoral correlations ρ.
We use data from Compustat and Worldscope for constructing trade credit usage P . To constructing B and D, we use the Multi-Region Input-Output (MRIO) Database developed by the Asian Development Bank (ADB). This database collects the input-output data for 35 industries and 62 countries. We referee the readers to As our unit of analysis is at the industry level and due to data availability, we follow the industry classification of the MRIO database and only include manufacturing industries in our analysis.
Next, we introduce the data used, our constructions of these three sets of variables, and brief summaries of the each set.

The Use of Trade Credit (Matrix P )
The intensity of trade credit usage is defined as the ratio of the average accounts payable at the end of years t and t − 1 to the cost of goods sold in year t. Similar to Raddatz (2010), we make the following assumption, due to data availability issue: Assumption 3. The ratio of an industry's use of trade credit to the average use in a country (P n j /P n ) is assumed constant across countries, so the elements of P in a given country can be expressed as the product of this ratio, P j , and the country's use of trade credit P n . Formally, This assumption enables us to leverage the extensive data coverage for U.S. firms without sacrificing countries with much less data coverage. We will construct P j based on the Compustat database using US data, and P n based on the Worldscope database. We detail our procedures as follows. First, we extract public firms' accounts payable and cost of goods sold from two databases, Compustat and Worldscope, in which the former one is used for US publicly listed firms, whereas the latter is used for non-US firms. We resort to the Compustat database for the industry-level financial data aggregation in the United States. We collect data from 1990 to 2019, and then map these publicly listed firms using their four-digit SIC (Standard Industry Classification) code to the 35 industries used in MRIO. To clean the data, we remove observations outside of the U.S.
(e.g., ADR and GDR), with missing or negative cost of goods sold, and belonging to Industry 28 (the financial intermediation industry) for it performs financial transactions differently from other industries. For observations with missing accounts payable (about 0.6% out of total observations), we treat them as zero. After the above data cleaning procedure, this data set presents 17,110 unique US firms, with 173,663 firm-year observations. In Table 1, we list the number of firms and the number of firms with more than 5 years of data for each US industry. 3 Next, we resort to Worldscope for financial information for the non-US firms in 54 countries.
Similarly to Compustat, we collect data from 1990 to 2019, and then map these firms using their four-digit SIC. After data cleaning, 4 we have 69,250 unique manufacturing and service firms from 32 industries across 55 countries from 1990 to 2019, 5 with a total of 808,288 observations. In Table   2, we list the number of firms, and the number of firms with more than 5 years of data for each 55 countries in Worldscope. We note that in this dataset, data coverage is limited amongdeveloping countries. We also analyze the data coverage at the country-industry level. The result summary are presented in Table 3, which indicates the low data coverage of Cambodia and Mongolia. Out of the 32 industries, Table 3 suggests that only 26 countries report more than 5 firms (each with at least 5 years of data) in more than 20 industries. 6 With these raw data, we next present the computation steps to obtain the use of trade credit (the matrix P ). For a given firm and year, the use of trade credit corresponds to the average of the accounts payable at the end of years t − 1 and t divided by the total cost of goods sold in year t, and we denote it as p i t . Thus, the first step is to construct the firm-level representative measure of payables by taking the median of p i t across time for each firm reporting data to the Worldscope database. Only firms with more than five years of (annual) data from 1990 to 2019 are kept in the sample to reduce the impact of cyclical fluctuation.
Next, within country n (except the United States), the median of the representative ratios of those firms located in n is used as a country-level representative value of payables financing (P n ).
For industry level trade credit usage P j , by using data from Compustat, we construct representative ratios for each industry j in the United States (P U S j ) by taking the median ratio across U.S. firms within the industry. Based on Assumption 3, we then can have: P j = P U S j /P U S , and P n j = P j × P n . In Table 4, we present the use of trade credit (columns with Payables Financing) and the use Electronic copy available at: https://ssrn.com/abstract=3897212 Electronic copy available at: https://ssrn.com/abstract=3897212 of bank credit (columns with Short-Term Debt to Payables) for countries listed in Worldscope. 7 In Table 5, we present these two for the manufacturing industries in the U.S. we reported these summary statistics for 2000, 2007, 2018, and 2018 due to the data availability from 2000 to 2018 as well as as a preparation for the time trend analyses later.

Input-Output Linkages (B and D)
Similar to World Input-Output Database (WIOD) and the input-output (IO)   To obtain both B and D, we follow Shea (2002). Specifically, We start by computing D. According to Miller and Blair (2009), the technical input coefficient matrix, β, is defined as: Value of input (m, i) bought by (n, j) producers Total value of (n, j) production .
Based on the multi-sectoral general equilibrium model developed by Shea (2002), we can estimate the fluctuations in industry i (denoted as q i ) as following:  (2002) is as follows: where β is a matrix whose ((m, i), (n, j)) element is the share of country m industry i's cost directly attributable to country n industry j (computed according to Eq. (21)), and where f k is industry k's final demand defined as the sum of purchases from consumption, government, and non-manufacturing industries. 9 As a result, to compute D and B, we start by computing technical input coefficient matrix β using Equation (21), and then compute the COST matrix using Equation (23  In the MRIO table, the "final uses" item contains five elements: final consumption expenditure by households, final consumption expenditure by non-profit organizations serving households (NPISH), final consumption expenditure by government, gross fixed capital formation, changes in inventories and valuables. We currently use the first three elements to approximate for the final demand.
10 In this step, we remove the five region-sectors that has a almost zero-column sums in COST or D. Originally, number of region-sector should be 53 × 14, which is 742. After moving these five region-sectors, our final matrix size is 737×737. 11 We use the 2020 edition of INDSTAT2 ISIC Revision 3.
As a result, our unit of analysis will be on the manufacturing industries. We choose CPI as our deflator when computing sectoral correlation, whereas Raddatz (2010) used producer price index (PPI). We use the CPI data from World Bank as it was the only deflator with historical data across regions and sufficiently good data coverage. After data cleaning, 14 we construct real value added equals nominal value added divided by CPI/100.
Next, we illustrate the steps to construct the sectoral co-movement across regions and across years. First, we compute the growth rate for the real value added g m it , which is the growth rate of industry i in country m between years t − 1 and t. Then, we compute the averageḡ m i across times, which is taking average of g m it for all t. Finally, the correlation across regions and years is computed as: See the summary statistics in Table 8, in which we report the summary statistics for overall correlations, domestic (i.e., within-country) correlations, and the cross-border (i.e., across-country) correlations for the entire data. We note that the domestic correlation is higher than the ones from Raddatz (2010), likely because of a higher level of aggregation in our setting. To further justify the needs to differentiate domestic from cross-border correlations for our country-industry setup, we also report the domestic and cross-border correlations using 20 years of prior data from the year specified in Table 9.
14 We remove observations with missing nominal value added, we map the 23 manufacturing sectors (based on 2-digit level of ISIC industry classification) from UNIDO to 14 manufacturing sectors from MRIO and aggregate values for these 14 sectors, and we merge the CPI data based on region × year while removing observations with missing or zero CPI. 15 Raddatz (2010) (footnote 15 on page 991) specified that "[w]ith N sectors and T observations (per sector), there are N (N − 1)/2 correlation coefficients to be estimated from N T observations. The order condition therefore requires that T > (N − 1)/2 for a full rank matrix. With 28 sectors, this requires 14 observations at a minimum. I allowed for one more than that." 16 We drop missing correlations, correlations with higher than 1 or smaller than −1; these cases are likely to occur when T mn ij is small. At the end, there are 416,469 region-sectors left.
Electronic copy available at: https://ssrn.com/abstract=3897212 One potential concern with the baseline measure is the use of a common deflator: in the presence of significant heterogeneity in the evolution of prices across industries, the correlations computed with a common deflator may be driven by the correlation of relative inflation rates instead of the correlation of real output growth. This concern can be addressed by using the correlation of the growth rates of the index of industrial production, also reported in UNIDO. Results obtained using this measure are not affected by the relative price problem, but results obtained using real value added are preferable because the production index data are of lower quality and smaller coverage than the value-added data. Nevertheless, this choice does not affect the results.

Base Results
With all variables constructed, our model specification is thus: in which W mo ik includes other determinants of sectoral correlation, and G mo ik is the physical linkage between (m, i) and (o, k), which is computed as the first term in Eq. (13). We consider two fixed effects combinations: (1) country fixed effects for input and output countries separately, and industry fixed effects for input and output industries separately, and (2) country-industry joint fixed effects for input and output country-industry pairs. We cluster standard errors based on the fixed effects combinations.
We present our results in Models 1 to 4 of Table 10. Models (1) and (2) consider only an aggregate credit linkage (i.e., we do not separate the within and across country credit linkages) for the two fixed effects combinations, and (3) and (4) consider the domestic and cross-border credit linkages, separately. In this specification, we use the 2018 MRIO table in computing the input-output linkage, and we compute the sectoral correlation using the prior 20 years INDSTAT and CPI data.
Two observations are notable. First, we echo the results shown in Raddatz (2010); the significant and positive coefficient of C mo ik in (1) and (2) support the hypothesis that the intensity of the use of trade credit increases the correlation between the two industries linked by the credit chain.
Electronic copy available at: https://ssrn.com/abstract=3897212 Second, we find that after separating the credit linkage, C mo ik , to the domestic component (or within country linkage), C Dmo ik , and the cross-border component (or cross-border linkage), C F mo ik , the results remain similar in that the intensity of the use of trade credit still increases the correlation between the two industries, regardless of these two industries are in the same country or different countries, as the coefficients of the two are positive and significant. Moreover, the comparison between the coefficients of C Dmo ik and C F mo ik suggest that the domestic credit linkage has higher impact on the correlation between two industries within the same country than that between two in two different countries.

Robustness
We note that one possible concern of the base specification is whether the results are robust to outliers. It could be possible that our results are driven by the strong correlations between a few strong countries and/or dominating industries. To address this concern, we follow two traditional approach to reduce the effect of outliers: winsorization or trimming, the former is to set the extreme values of sectoral correlations (higher or lower than a certain threshold) to the value at the threshold, whereas the latter is to drop these extreme values from the sample.
We report the results of the above two robustness tests in Table 11. The first three columns are the results after winsorizing the sectoral correlation based on the top 1, 3, and 5 percentile, and the last column shows the result after we trim the top 5% values. As we can observe from the coefficients of C Dmo ik and C F mo ik , these two remain positive and significant, thereby, again, supporting our hypothesis that the intensity of the use of trade credit increases sectoral correlation. The magnitude of the two also suggests that the influence of domestic trade intensity is higher than that of cross-border one; although the extreme values indeed have impact on how much these trade intensity influences correlation (as the gap between the two slightly reduced), the impact is minor.
To further examine the outlier issues, we winsorize and trim our sample based on the respective percentile cutoffs based on the country pairs instead of doing so for the entire sample and report the results in Table 12, and we again obtain qualitatively the same results.

Time Variation
In this section, we explore how the intensity of the use of trade credit influence sectoral correlation differently from 2000 to 2018. Specifically, we repeat the regression model in Eq. (26) but use the MRIO tables in year 2000MRIO tables in year , 2007MRIO tables in year , 2013MRIO tables in year , and 2018 In Table 13, we report our empirical results when the sectoral correlations are computed based on the prior 20 years of data. Table 13 Regression Table (Time window:  to 2018. This decreasing trend suggest that the trade linkage intensity is gradually reducing its impact on sectoral correlations, regardless of the domestic or cross-border ones. We note that a similar pattern can be observed when we compute sectoral correlation based on the prior 10 years of data (see Table 14 for the results), except that using only 10 years of data, correlations tend to respond to sudden change of patterns. Nonetheless, we still have a roughly decreasing trend with a dip at 2013. Standard errors in parentheses. Robust for heteroskedasticity. * p < 0.05, * * p < 0.01, * * * p < 0.001 One plausible explanation is the booming global trade credit linkages provide a risk diversification mechanism. We provide heatmaps of C Dmo We further support our argument using the domestic and cross-border trade share changes. See Figure 2 for the two plots as well as see Table 15 for the summary statistics for the domestic trade share. We find that in Figure 2(a), the domestic trade shares are shifting to the left when time increases, whereas in Figure 2(b), the cross-border trade shares are shifting to the right.
Similar pattern can also be found in Table 15 that the mean and the three percentiles are both Electronic copy available at: https://ssrn.com/abstract=3897212 This comparison also highlights the contribution of our analysis; while the global trading volume has reached a steady state, the distribution is not, thereby leading to a still changing magnitude of sectoral co-movement.
18 https://unctad.org/webflyer/world-investment-report-2020 Electronic copy available at: https://ssrn.com/abstract=3897212  Another plausible explanation for the decreasing trend that the intensity of the use of trade credit increases correlation can be that the global economies are gradually reducing their sectoral correlation by means of decoupling risks (e.g., better use of inventory management to buffer against uncertainties and shocks). To rule out this alternative, we plot overall correlations, the domestic correlations, cross-border correlations in Figures 3-5, respectively, for a time window of prior 10 years (a) 19 and a time window of prior 20 years (b). We also provide the summary statistics of these figures in Table 16.
Both the figures and the summary statistics reveal that comparing 2000 against 2018, while the overall correlation increases from 0.144 to 0.198, the domestic correlation decreases from 0.574 to 0.476 (use 20 years as our time window), whereas the cross-border correlation increases from 0.132 to 0.193. Decline in domestic sector correlations of US is also documented by Irvine and Schuh (2005), who compare US sector correlations between 1967-1983 and 1984-2001. The increasing trend of the cross-border correlation may be a result from many factors, such as the cross-border information barriers have be lifted from 2000 to 2018. These number changes again support our previous explanation that the diversification resulting from a more dissected value chain structure.
Moreover, in our regression specification, we used the overall correlation as our dependent variable.
Therefore, the increasing overall correlation helps rule out this explanation that the decreasing trend that the intensity of the use of trade credit increases correlation is a result of decreasing correlations. Electronic copy available at: https://ssrn.com/abstract=3897212    Next, we examine another plausible explanation that the decreasing trend is a result of global trends reducing its needs for payables financing. In this case, despite that the intensity heatmaps show a more diversified trade credit linkages, as the overall trends less relying on payables financing, shocks are less likely to propagate via the channel of value chains. To reject this explanation, we provide summary statistics of the country-level payables financing P n (which is the median of the representative ratios of publicly listed firms in country n) in Table 17. As suggested from the summary statistics, we find that the country-level payables financing is increasing, including the mean as well as the 25 and 75 percentiles, suggesting that firms are more relying on payables as one way of managing their cash flow, instead of reducing their reliance on payables.
A related explanation is that, although the overall country-level increases trade credit usage, such an increasing trend only exists within the same industry for firms in the same industry have better understanding of their peers and hence, higher confidence and trusts to use trade credit. In this case, the sector-level trade credit rate, P j then should decrease, and such reduction of trade Electronic copy available at: https://ssrn.com/abstract=3897212 credit rates drives our result. We report the sector-level trade credit ration in Table 18. Again, while there are some industries reduce their trade credit ratios, a good number of the sectors also rely more on trade credit, and even for the reducing ones, the difference is not high.    (2000) with P (2000). Although the magnitude slightly changes, the trend of decreasing are consistent.

Trade Credit Financing versus Bank Credit Financing
In this section, we compare trade credit and the bank financing, and examine how they influence sectoral co-movement. To do that, we follow a similar procedure in Section 3 to construct the credit linkage variables (domestic and cross-border components) of bank credit financing by replacing accounts payable by short-term debt, and denote them as C Dmo ik (debt to COGS) and C Dmo ik (debt to COGS), respectively. Also to better explanation the relative dependence of these two sources of financing, we also construct the related variables using the debt to payables ratios, and annotate them with debt to payables. Finally, with the newly constructed variables, we repeat our analysis and show the results in Table 20.
Model 1 in Table 20 serves as a benchmark as it includes the standard constructs in Eq. (26).
Model 2 replaces the main explanatory variables of credit linkage constructed using payables by those using short-term debt, whereas Model 3 include both. Similarly, Models 4 and 5 are the regression results with debt-to-payable related variables, without and with the controls of payables to COGS variables, respectively. We observe that without controlling for the payables financing, relying more on domestic bank credit (either considering the ratio to cost of goods sold or to payables) may seem to lead to a sizable reduction on sectoral correlations, whereas more on crossborder bank credit may have reduce sectoral correlation though the effect is much minor. However, after we control for the trade credit linkage variables, the effect of bank credit linkages disappears.
Electronic copy available at: https://ssrn.com/abstract=3897212  Raddatz (2010) assuming that IO linkages in the U.S. can be extrapolated to the rest of the countries, therefore, the effect of bank credit in our model might be different from that of the U.S. * p < 0.05, * * p < 0.01, * * * p < 0.001 While the results above suggest that bank credit financing does not help mitigate shock propagation, we suspect that this result could be due to the geographic differences. To examine whether the bank credit linkages have differential impacts for countries in Asia where the economic activities are more export-oriented and the banking industry is less mature, we consider three dummy variables  Standard errors in parentheses and are robust to heteroskedasticity. * p < 0.05, * * p < 0.01, * * * p < 0.001 We report the results in Table 21. Models 1 to 3 refers to the result with the debt-to-COGS related variable without Asia related interaction terms, with Asia interaction terms, and finally with additional controls of payables-to-COGS related variables. Models 4 to 6 follows a similar order, except that we do so for debt-to-payables related variables. We observe two interesting Electronic copy available at: https://ssrn.com/abstract=3897212 results. First, when trading within Asia countries, regardless of domestic trades within a Asia country or cross-border trades among Asia countries, relying more on bank credit than trade credit may help mitigate shock propagation, as the coefficient of C Dmo ik × Asia and C F mo ik × Both from Asia in Models 5 and 6 are both negative and significant. Second, using more bank credit mitigates shock propagation for domestic trades in Asia, as the coefficients of C Dmo ik × Asia is negative and significant in Models 2 and 3.

Policy Implications and Conclusion
Trade credit is a widely used source of short-term external financing and it connects buyers and sellers within a supply chain. In this paper, we develop a framework that captures the impact of credit chain on domestic and international trade separately. Leveraging on the Multi-Regional Input-Output (MRIO) Table developed by Asian Development Bank, we assemble a dataset including 14 manufacturing industries and 53 countries. We find that the use of trade credit indeed enhances sectoral co-movement, and the intensity of this impact is twice as strong for within country transactions as for international ones. We further find that over the last two decades, this intensity in general declined. Our final comparison on the use of trade credit and bank credit offers insights around the geographic use of the two financing tools.
Our work offers the following policy implications. First, similar to Raddatz (2010), trade credit linkage is indeed an effective channel for shocks to propagation via value chains. Our work extends Raddatz (2010) to differentiate domestic and cross-border trades as two different channels for shock propagation, though the domestic channel has a stronger impact than the cross-border one. In the case, supply chain finance programs that allow upstream suppliers to receive cash before trade credit maturity could be a valuable instrument that helps decompose such correlation, and possibly lower systemic risk.
Our results also imply that in the past two decades, the influence of these two channels on sectoral co-movements both reduce, possibly due to a higher level of risk diversification. Combining these results together, governments can encourage diversifying cross-border trades among firms in the manufacturing industries. Second, our final comparison between the use of trade credit and bank credit also suggests the needs for credit redistribution from financial institutions to manufacturing sector in Asia so as to reduce the degree of shock propagation.
This research can be extended along different directions. First, there is no existing theory that rationalizes the difference of credit chain impact between domestic and international trade. Further development in this direction could be promising. Second, as in Raddatz (2010), our results are not based on causal identifications. When adopting exogenous shocks for identifications, one could potentially identify how different types of shocks (e.g., natural disaster, financial crisis, global Electronic copy available at: https://ssrn.com/abstract=3897212 pandemic) are propagated differently through both physical and credit channels. Further study that using a specific shock on certain sector/geographic regions could be explored. Finally, due to data limitation, this work focus mainly on financial variables, such as payables and bank credit. Should reliable data become available, we could examine the impact of other factors, such as inventory, on sectoral co-movement.