Interdependent cascade failure of trade and technology innovation networks under United States export controls

Abstract

U.S. export controls significantly affect supply chain stability and technological innovation, particularly within the electronics and communication device (ECD) sectors. Using the 2011–2022 U.S. Commerce Control List and global patent cooperation data, we developed networks for ECD intermediate goods trade and technological innovation cooperation. Employing an interdependent cascading failure model, our analysis revealed how U.S. ECD export restrictions propagate shocks, leading to increased network fragmentation with trade networks exhibiting greater vulnerability than innovation networks. Notably, small, highly connected peripheral countries with uneven load distributions, particularly in Africa, Oceania, and Central America, are prone to failures due to overload. Additionally, the network's overall innovation capacity has been declining annually, with the most significant decreases observed in smaller countries in Africa and West Asia, such as Zimbabwe, Palestine, and Tajikistan. Network location emerges as a critical predictor of exposure to shocks and declining innovation capacity. Scenario analysis indicates that lower node capacity parameters significantly enhance network vulnerability, whereas higher parameters have minimal impact. These insights are crucial for enhancing the resilience of high-technology supply chains and innovation networks under major power technological blockades. JEL: C63; F10; O34

Full text

1
Interdependent cascade failure of trade and technology
1
innovation networks under United States export controls
1
2
Abstract: U.S. export controls significantly affect supply chain stability and
3
technological innovation, particularly within the electronics and communication device
4
(ECD) sectors. Using the 2011-2022 U.S. Commerce Control List and global patent
5
cooperation data, we developed networks for ECD intermediate goods trade and
6
technological innovation cooperation. Employing an interdependent cascading failure
7
model, our analysis revealed how U.S. ECD export restrictions propagate shocks,
8
leading to increased network fragmentation with trade networks exhibiting greater
9
vulnerability than innovation networks. Notably, small, highly connected peripheral
10
countries with uneven load distributions, particularly in Africa, Oceania, and Central
11
America, are prone to failures due to overload. Additionally, the network's overall
12
innovation capacity has been declining annually, with the most significant decreases
13
observed in smaller countries in Africa and West Asia, such as Zimbabwe, Palestine,
14
and Tajikistan. Network location emerges as a critical predictor of exposure to shocks
15
and declining innovation capacity. Scenario analysis indicates that lower node capacity
16
parameters significantly enhance network vulnerability, whereas higher parameters
17
have minimal impact. These insights are crucial for enhancing the resilience of high-
18
technology supply chains and innovation networks under major power technological
19
blockades.
20
Keywords: Commercial Control List (CCL); cascade failure; multiplex network;
21
robustness
22
JEL: C63; F10; O34
23
1
Electronic and Communication Device (ECD)
Global Electronic and Communication Device Intermediate Goods Trade (ETN)
Global Electronic and Communication Device Technology Innovation Cooperation Network (ICN)
Commerce Control List (CCL)

2
1. Introduction
24
The wave of globalisation has led to the rapid development of international trade
25
and technological innovation cooperation (Betz & Hein, 2023). However, in recent
26
years, as geopolitical tensions and the struggle for technological hegemony have
27
intensified, especially with the U.S. imposing export controls on certain key high-
28
technology areas through “control lists” and “licensing systems” (Bown, 2020;
29
Crosignani et al., 2023). The global high-technology product supply chain and
30
technological innovation cooperation network are facing unprecedented challenges. In
31
2018, the U.S. passed a new version of the “Export Control Reform Act (ECRA)”,
32
further expanding the scope of technology export controls, including 14 new “emerging
33
and basic technologies” on the basis of the original scope of controls
2
. The most
34
representative export control incident is the escalation of trade friction between the U.S.
35
and China, the U.S.’s export control measures against China continue to increase, the
36
trade imbalance and the contradictory focus of science and technology protection
37
triggered the U.S. and China to carry out a new type of game of “trade decoupling” and
38
“technology decoupling” (Cai et al., 2022). Under this background, the policy change
39
of export control poses a double challenge to a single country and the global high-
40
technology industry. On the one hand, the competition among core countries is
41
intensifying day by day, and the means of scientific and technological sanctions are
42
more flexible and targeted (Danilin, 2022), which has a negative impact on the
43
economic growth and technological progress of countries with low economic
44
development level and political countries (Wu, 2020). On the other hand, the bottleneck
45
phenomenon of “stuck neck” technology is getting worse and worse, which leads to the
46
potential fragility of key technology links. The supply restriction of high-technology
47
2
The 14 technologies include: biomedical; artificial intelligence (AI) and machine learning technologies;
positioning, navigation, and timing (PNT) technologies; microprocessor technologies; advanced computing
technologies; data analytics technologies; quantum information and sensing technologies; logistics technologies;
additive manufacturing; robotics; brain-machine interfaces; hypersonic aerodynamics; advanced materials; and
advanced surveillance technologies. See documents published on the website of the Bureau of Industry and Security
(BIS) of the U.S. Department of Commerce.

3
products has reshaped the competitive pattern and innovation ecology of global high-
48
technology industries (Hao & Wang, 2023; Jie et al., 2022). Then, under the increasing
49
trend of “de-globalization”, what cascade influence will the technological blockade of
50
core countries bring to the global high-technology product supply chains and
51
technological innovation cooperation? How can countries resist supply chain risks and
52
build their own networks of relational innovation cooperation? It is important to explore
53
in depth the impact of U.S. export controls, especially the effect on technological
54
innovation, which needs to be clarified.
55
Most of the existing studies focused on the direct impact of economic and trade
56
frictions, ignoring the imbalances in supply and demand and shortages of key materials
57
in the market for high-technology products, which trigger structural shocks in supply
58
chains and reshape the global technological innovation network. Early studies on
59
economic and trade frictions focused on the impact of policies such as tariffs (Benguria
60
et al., 2022), anti-dumping and countervailing (Shen et al., 2021), and “entity lists” (Liu
61
& Li, 2023), which have important implications for trade structure (Zhu et al., 2022)
62
and international cooperation (Nguyen et al., 2022) at the macro level as well as for the
63
innovation and development of domestic firms at the micro level (Amiti & Konings,
64
2007). In addition, part of the literature explored the impact of external shocks on
65
national innovation capabilities under open conditions, mainly focusing on the
66
mechanism of trade liberalization and FDI (Lu et al., 2017; Shin & Balistreri, 2022).
67
Some studies believed that the decrease of trade policy uncertainty will promote firm
68
innovation through import channels and FDI (Choi et al., 2021), and the decrease of
69
intermediate goods tariff will promote R&D innovation of firms and industrial clusters
70
through “learning-by-doing” and cost reduction (Yang et al., 2024); However, some
71
studies found that the import of intermediate products may make firms rely on foreign
72
products. When the external environment is exposed to the supply side, it will inhibit
73
their independent innovation and the emergence of innovation vitality of industrial
74
clusters (Gao & Li, 2024). The existing research only focuses on the influence of trade
75

4
friction policy on the innovation activities of a single country, ignoring the multiple
76
correlations between countries. In particular, there is a lack of research on the
77
interdependence between high-technology product supply chain and technological
78
innovation cooperation under the background of complex network, and the dynamic
79
process of how trade shocks induce chain reaction under the effect of export control,
80
which is difficult to provide theoretical guidance for international cooperative
81
innovation practice under the trend of global decoupling. Therefore, in order to
82
maintain the smooth operation of technological innovation cooperation network under
83
the risk of supply chain breakage, it is necessary to study the dynamic process of its
84
chain reaction.
85
Cascade failure, as a complex network dynamic process, is a dynamic process in
86
which after a node fails, the neighboring nodes fail through the coupling relationship
87
between nodes, resulting in a chain reaction that leads to the paralysis of the whole
88
network (Vespignani, 2010). The existing research focuses on single networks, such as
89
power network (Seo et al., 2015), trade network (Liu et al., 2023) and R&D network
90
(Song et al., 2017). However, due to the growing interdependence between networks,
91
the supply within networks is limited by the inter-layer network relationship. As a result,
92
cascading failures in interdependent network systems have become a research hotspot
93
in recent years (Buldyrev et al., 2010; Min et al., 2014; Qi et al., 2023). As a complex
94
network of multi-agent cooperation, the cooperation network of trade and technological
95
innovation, on the one hand, in the face of environmental uncertainty interference, any
96
country withdraws or the cooperation relationship breaks down, which may induce
97
other countries to fail one after another (Helbing, 2013), leading to the dissipation of
98
the whole network, similar to the process of cascading failure of complex networks. On
99
the other hand, the relationship between them is not isolated and linear. The stability of
100
the supply chain is very important to maintain the international innovation dynamics
101
and synergy (Ranganathan & Rosenkopf, 2014; Wang & Hu, 2020). Once the supply
102
chain is interrupted, the shortage of raw materials or the increase of costs will directly
103

5
affect domestic firms that rely on these inputs (McKinsey & Company, 2000), and then
104
it will be reflected that the country adjusts its innovation cooperation strategy to adapt
105
to the new supply conditions. Therefore, the failure of any country will be transmitted
106
through the connection relationship between networks, which will affect the
107
interdependent countries, and then lead to the cascade failure spreading in multiplex
108
networks. However, at present, the cascading failure process of a single network is not
109
enough to reveal the operation mechanism of multiplex network chain reactions.
110
To address the gaps in existing research, we constructed global electronic and
111
communication device (ECD) intermediate trade (ETN) and the ECD technology
112
innovation cooperation networks (ICN) from 2011 to 2022, using intermediate products
113
of ECD as a high-technology product supply chain under the export restriction scenario
114
represented by the U.S. Commercial Control List (CCL). Based on the interdependence
115
between ETN and ICN, a multi-network architecture with coupling dependence was
116
formed. Considering the load and capacity of nodes, the load of edges and the failure
117
mechanism of intra-layer and inter-layer networks, a interdependent cascade failure
118
model of multiplex networks was constructed. On this basis, we explored the dynamic
119
process of chain reaction under the export control of ECD from the U.S., and its
120
influence mechanism on topological structure and innovation ability. We used the
121
interdependent cascade failure model to answer the following four key research
122
questions: (1) Does U.S. export control have an impact on ICN? (2) What is the process
123
of interdependent cascading failures of ETN and ICN caused by U.S. export controls?
124
(3) Which countries’ innovation ability will be negatively affected by U.S. export
125
controls? (4) How can the vulnerability of other countries under U.S. export controls
126
be measured and mitigated?
127

6
2. Methodology
128
2.1 Data sources and network construction
129
CCL are one of the most representative lists in U.S. export control system (Liu &
130
Li, 2023). In order to safeguard its comprehensive interests, including economic,
131
political, military and diplomatic interests, the U.S. takes protective measures for the
132
export of high-technology products and strategic technologies that may improve the
133
comprehensive strength of other countries through CCL. The U.S. export control legal
134
system consists of two parts: controls on military goods and defence services, and
135
technical controls on civilian and dual-use products. For export control of “dual-use”
136
items, the Bureau of Industry and Security (BIS) of the U.S. Department of Commerce
137
(DOC) has established the CCL, which assigns a specific number, the Export Control
138
Classification Number (ECCN), to each item and provides a detailed description. With
139
its wide coverage and significant impact, the CCL is one of the key instruments in U.S.
140
foreign science and technology sanctions. Since the ECCN does not directly match any
141
of the current methods in international trade statistics, it is not possible to know the
142
specific trade flows of the various products regulated by the CCL. We manually
143
collected the CCL published by the U.S. Department of Commerce in different years,
144
and matched the product descriptions in the lists with the descriptions of products in
145
the customs codes to obtain the customs codes of various products in the CCL. At the
146
same time, we conducted a comprehensive inspection manually to finally got the HS
147
codes of products affected by the CCL.
148
Taking the intermediate products included in the CCL as the research object, the
149
data of the global intermediate products trade network we built mainly comes from
150
CEPII BACI database, which covers the bilateral import and export flows of 251
151
countries (or regions) from 1995 to 2022. When selecting high-technology products as
152
the research object, we chose according to the high-technology product catalogue of
153
CCL and OECD. According to the comparison between R&D expenditure and SITC
154

7
Rev.3, OECD has determined the catalogue of high-technology products and the
155
corresponding SITC codes. In particular, we chose the category of “electronic and
156
communication device (ECD)” as the research object for the following reasons: First,
157
ECD are the key components to promote global technological innovation. According to
158
the World Bank’s Global Economic Outlook Report (2023), the value of the global
159
electronic product market is estimated to exceed US$ 2 trillion in 2023, accounting for
160
more than 10% of the global manufacturing output value. In addition, R&D investment
161
in this field accounts for more than 20% of the global R&D expenditure, highlighting
162
that the innovation and production of ECD is one of the most active high-technology
163
industries in modern economies. Second, U.S. export controls on ECD are particularly
164
stringent, with over 300 different export controls in place, which is among the product
165
categories with the highest number of controls on the CCL (Seyoum, 2017). In order to
166
accurately capture the role of these products in the global supply chain, we chose the
167
trade data of intermediate products, which act as value carriers in GVCs connecting
168
production networks across countries. When the supply intensity of a particular type of
169
intermediate good from one country to another declines, firms in other countries that
170
are dependent on that country for the production of that good face an international
171
supply shock. The identification of intermediate products follows the United Nations
172
Classification by BEC, which is the most widely used and recognised classification of
173
intermediate productions (Baldwin & Freeman, 2022).
174
In conjunction with the previous section, the steps for the construction of ETN are
175
as follows: First, we identified the SITC codes of high-technology intermediate
176
products according to the high-technology product catalogue provided by OECD, from
177
which we extracted the SITC codes of ECD. Second, through the comparison table of
178
BEC classification and SITC codes provided by the United Nations, the SITC codes of
179
ECD intermediate product was found out correspondingly according to the codes
180
representing intermediate product in BEC classification; Third, in order to ensure
181
consistency with the codes in BACI database, we used the comparison table of SITC
182

8
(Rev.3) and HS96 customs codes provided by the United Nations to search the HS6-
183
digit customs codes of ECD intermediate products. Finally, the bilateral trade flow data
184
of ECD intermediate products between countries from 2011 to 2022 were screened out
185
through BACI database. Accordingly, ETN can be represented as 
186
󰇛󰇜, where
P
Nt
denotes the set of countries with the number of nodes N and
187
attribute P in the trade network in period t,
W
Et
denotes the set of trade links with the
188
number of directed edges E and weight W in period t, t=2011,..., 2022, whose weight
189
W is determined by the amount of trade from exporting country i to importing country
190
j in metric tonnes, where i, j
∈
{1.....
P
Nt
}.
191
The data of ICN mainly comes from the USPTO database. We used Python crawler
192
technology to obtain the information of patent text to identify the cross-border patent
193
cooperation relationship. The patent text contains a wealth of information such as patent
194
number, patent filing time, applicant, filing city, patent classification number, etc.,
195
which can fully support our research topic. First, all patent texts with the number of
196
patent inventors greater than or equal to 2 from 2011 to 2022 were collected from the
197
USPTO patent text database with the help of a search formula, and the relevant text
198
information was crawled. Next, information on the countries to which the patent
199
inventors belong were extracted, removing those patent texts that refer to only a single
200
country. For patents where the country to which the patent inventors belong were
201
greater than or equal to 2, permutations were performed to generate all possible bilateral
202
country combinations. Third, each patent was identified by a unique patent number, and
203
the IPC classification number in the patent text was extracted for distinguishing the
204
technical field to which the patent belongs, so as to build a database of bilateral relations
205
on cross-border cooperation on patents at the national level. We extracted 583,911
206
patent text information articles in the database, involving 194 countries and 100,191
207
IPC classification numbers worldwide. Finally, ECD product-related IPCs were
208
selected from database to construct ICN. ICN can be represented as 
209

9
󰇛󰇜, where
P
Nt
denotes the set of countries with the number of nodes N and
210
attribute P in the innovation network in period t,
W
Et
denotes the set of trade links with
211
the number of undirected edges E and weight W in period t, t=2011,..., 2022, whose
212
weight W is determined by the number of patent cooperations between country i and
213
country j, where i, j
∈
{1.....
P
Nt
}.
214
In addition, we obtained data on the country’s economy and innovation from the
215
World Development Indicators database (WDI), including GDP, GDP per capita
216
(GDPPC), R&D inputs (R&D), labour participation rate (LPR).
217
2.2 Interdependent cascade failure model
218
We applied an improved cascade failure model focused on analysing shock
219
propagation in ETN and ICN (Reis et al., 2014; Wang et al., 2018; Zhou et al., 2023).
220
Under the background of economic globalization, countries in the ETN have the ability
221
of self-adjustment, and can adjust their trade strategies in the face of potential overload,
222
rather than completely failing. However, when the activity level is lower than the
223
underload threshold, such as sudden interruption of technological innovation
224
cooperation or insufficient supply of key materials, it will directly threaten the stability
225
of the economic system and may trigger cascading failures. In our cascading failure
226
model, the initial load and capacity of each node need to be defined first. In order to
227
simulate the influence of U.S. export controls, the U.S. is regarded as the initial failure
228
node. In the load distribution stage, both node and edge loads will be considered. In the
229
process of distribution, we consider the dependence between nodes, and redistribute the
230
export volume of the U.S. to other import trade connections with the target importing
231
country according to the initial load ratio. In addition, a capacity cap per connection has
232
been introduced to prevent overloading of any single connection. If the load on any
233
connection exceeds the set upper limit, the remaining volume will be distributed to trade
234
relationships in other countries with which the target importing country has a trade
235

10
relationship. Subsequent cascade overloads may then occur. After distributing the
236
remaining capacity of all other available nodes, the undistributed trade volume will be
237
deleted directly. Eventually, nodes that cannot meet the minimum capacity limit will
238
trigger a failure rule that prohibits imports or exports, thus prohibiting further ECD
239
trade. The specific modelling process is as follows:
240
In step 1, we used the country’s export volume to measure the initial load. Each
241
country’s output has an capacity upper and lower limits.
242
󰇛󰇜 (1)
243
󰇛󰇜 󰇛󰇜 (2)
244
󰇛󰇜 󰇛󰇜 
 (3)
245
Where, 󰇛󰇜 denotes the upper limit of capacity of node i. 󰇛󰇜
246
denotes the lower limit of capacity of node i. 󰇛󰇜 denotes the initial load of node i.
247
 denotes the export volume of ECD of node i.  denotes the import
248
volume of ECD of node i.  denotes the fixed upper limit parameter of capacity of
249
node i.  denotes the fixed lower limit parameter of capacity of node i.
250
In steps 2 and 3, once the U.S. interrupts ECD exports, their export volume are
251
distributed to the edges of other exporting countries that have trade relationships with
252
the target importing country in proportion to their initial load.
253
  


 (4)
254
where e denotes the initial failed node, i denotes the country that imports ECD
255
from the U.S.. j denotes the country other than the U.S. that exports ECD to the target
256
importing country i. n denotes the number of nodes from node j to i.  denotes
257
the initial edge load from node j to i.  denotes the initial edge load from node e
258
to i. 

 denotes the sum of initial edge loads of all nodes h to i. 
259
denotes the edge loads of nodes j to i after load redistribution.
260
In step 4, secondary load distribution occurs, and the upper limit of node capacity
261
will be distributed to each edge in proportion to its initial load.
262

11
 󰇛󰇜 


 (5)
263
 󰇛󰇜


  (6)
264
  
  (7)
265
    


  
＜

  (8)
266
Where  denotes the upper limit of edge load from node j to i.  denotes
267
the upper limit of edge load from node m to i.  denotes the edge load from node
268
m to i after the secondary distribution. Here, when the export of country j to i exceeds
269
its edge load, this excess load needs to be distributed to other country m, i.e., country
270
m is exporting to i but has not yet reached its own capacity limit in the current
271
distribution time step.
272
Finally, after the remaining capacity distribution to other valid nodes is completed,
273
any capacity that cannot be distributed will be directly deleted. If the export volume of
274
the exporting country or the import volume of the importing country is lower than the
275
minimum capacity of node m in ETN, the failure rule will be triggered. This will cause
276
the failed node m to be prohibited from importing or exporting ECD.
277
At the same time, cascading failures in ETN can have an impact on ICN through
278
interdependencies. Assuming that the capacity of node m in either ETN or ICN is below
279
the limit of the coupling value q (0<q<1), then whatever the current capacity of node m
280
is, it will result in the simultaneous failure of node m in the two dependent networks
281
and a redistribution of neighbouring node loads and edge loads in ICN. Consistent with
282
the steps in ETN, the number of technological innovation cooperation of node m is
283
distributed to the edges of other countries that have cooperative relationships with the
284
target cooperative country in proportion to its initial load.
285
  


 (9)
286
Where m denotes the initial failure node. t denotes the node with which node m
287
has technological innovation cooperation relationship. k and b respectively denote the
288

12
node with technological innovation cooperation relationship with target node t except
289
m and the set of such countries.  denotes the initial edge load of country k to t.
290
 denotes the initial edge load of country m to t. 

 denotes the sum of
291
initial edge loads from all nodes g to t.  denotes the edge load from country k to
292
t after load redistribution. Similarly, the node t receives the extra load from the failed
293
node m, and carries out the secondary distribution by adjusting the load of nodes and
294
edges. If the node load falls below the minimum of its capacity, a failure rule will be
295
triggered, generating a cyclic iterative process on ICN. The failure nodes in ICN can
296
also lead to changes in trade relationships within ETN, propagating through
297
interdependencies and generating a dynamic process of load redistribution. Eventually,
298
the cascade failure propagation process ends with a devastating effect on the dependent
299
ETN and ICN, leading to network collapse or partial self-organised equilibrium.
300
2.3 Vulnerability assessment
301
As assessment indicators for cascading failures in complex networks, existing
302
studies mainly used failure scale (Pei et al., 2021), betweenness centrality (Huang et al.,
303
2014) and connectivity (Wang et al., 2014). Among them, the failure scale is described
304
only from the perspective of the number of failed nodes in the network, which lacks the
305
distinction between the network structure and the failed nodes; the betweenness
306
centrality is a measure of the propagation ability of the network nodes as mediators,
307
however, this indicator is based on the local characteristics of the nodes, which ignores
308
the influence of the overall network topology. More importantly after cascade failure
309
propagation, the assessed efficacy of this indicator is missing if the network splits into
310
multiple connected branches with full connectivity between internal nodes. The
311
connectivity coefficient is the reciprocal of the product of the number of connected
312
components and the weighted average of the shortest paths within each component.
313
Fewer components and shorter average shortest paths within each component indicate
314
higher network connectivity. This indicator is not only an important measure of the
315

13
influence on the network structure from the perspective of the global network, but also
316
can avoid the failure of the evaluation of betweenness centrality. Therefore, the formula
317
of connectivity considering both ETN and ICN was selected for comprehensive
318
evaluation perspective and efficiency.
319
󰆒






 (10)
320
Where  and  are the number of connected components of ETN and
321
ICN.  and  are the number of nodes in the corresponding components.  and
322
 are the lengths of the shortest paths in the connected components, with the range of
323
󰆒. The larger the value of this indicator, the higher the effectiveness of the
324
structure. When both ETN and ICN are fully connected, 󰆒 takes the maximum value
325
of 1. Therefore, in order to measure the degree of its structural failure, let S=1-󰆒, which
326
is the value of the loss of network structural connectivity. The smaller the network
327
connectivity, the greater the structural loss, and the stronger the effect of cascade failure
328
on both network topologies.
329
In addition to portraying the collapse of the network structure as a result of the
330
failure, it is also an important manifestation of the destructive result of cascade failure
331
that the country embeds in ICN to weaken its innovation ability. This is reflected in the
332
fact that when an innovation partnership is terminated, the sharing of knowledge
333
resources between countries is blocked, leading to a significant reduction in the degree
334
of heterogeneity of knowledge resources accessible to a country, which will directly
335
inhibit the country’s innovation ability. The change of national innovation ability is
336
influenced by three factors: First, the stronger the initial innovation ability of the
337
country, it shows that the country’s own knowledge base and R&D ability can resist
338
certain degree of uncertainty interference; Second, the proportion and influence degree
339
of the nodes with the nearest neighbor failure. The more failed nodes in the
340
neighborhood, the greater the influence on the target node, and the more obvious the
341
attenuation of innovation ability; Third, depending on the number of high- technology
342
intermediate trading countries and their influence, the greater the number of nodes that
343

14
the failed node is dependent on, the richer the source of vulnerability of its supply
344
structure. Based on the above analyses, we proposed a formula for calculating national
345
innovation ability.
346
󰆒󰇛󰇛󰇜

󰇜 (11)
347
󰇛󰇜󰇛󰇜 (12)
348
Where i and m are nodes in ETN; j is a node in ICN. i and j are dependent nodes,
349
ki, kj and km are the degree values of the nodes.  and  are the number of
350
neighboring nodes that fail after the failure of nodes i and j, and  and
351
 .  and  are the set of nearest neighbour nodes and the set of
352
neighbour failure nodes of node i in ETN.  and  are the set of nearest neighbour
353
nodes and the set of neighbour failure nodes of node j in ICN. α and β are adjustable
354
parameters, which control the average degree of load distribution of nodes in ETN and
355
ICN respectively. The smaller α and β values, the more uniform the distribution. 
356
represents the initial innovation ability of node i. and 󰆒 represents the innovation
357
ability of node i at the termination stage of cascading failure propagation. When all
358
neighboring nodes and dependent neighboring nodes fail, 󰆒=0, then node i becomes a
359
completely failed node, and its innovation ability only comes from itself. Therefore, let
360
I denote the ratio of the innovation ability of the network after cascade failure to the
361
initial innovation ability, the smaller the ratio I is, the larger the proportion of innovation
362
ability decay is, indicating that the cascade failure has a stronger impact on the
363
innovation ability of ICN.
364
󰆓



 (13)
365
2.4 Regression analysis
366
To better understand the relationship between a country’s position in the network
367
and the simulation results, we ran a series of regressions where the dependent variable
368
was the demand deficit (in metric tonnes) that each country j faced when country i was
369

15
hit by a shock. The independent variables include the country’s network location,
370
economic attributes, and innovation attributes. Given the large number of zeros in the
371
dependent variable, we refered to Ahern et al. (2015) and used panel Tobit regression
372
instead of standard OLS:
373
      (14)
374
Where  ,  ,  and  respectively
375
represent the network position indicators of the source country i and the destination
376
country j in ETN and ICN, including centrality, strength and transitivity indicators.
377
Currently, the main indicators used to measure the network centrality of nodes include
378
degree, betweenness, eigenvector and closeness centrality. We chose the country’s
379
degree as the indicator of network centrality, and used the closeness as an alternative
380
indicator for robustness testing (Liu et al., 2023). In addition, since betweenness
381
measures the role of the country as a “bridge”, some scholars have also used it as a
382
proxy for structural holes in their studies (Shi et al., 2019). With reference to such
383
practices, we used structural holes as an alternative indicator of betweenness for
384
robustness testing.  and  respectively represent the economic attributes and
385
innovation attributes of source country i and destination country j, including GDP, per
386
capita GDP, R&D input and labor participation rate.
387
Furthermore, we focused on the vulnerability of countries to external shocks from
388
the perspective of “destination countries”. The dependent variable is the decline of
389
innovation ability of each country affected by a shock. The independent variable is
390
consistent with equation (14). The decline of countries’ innovation ability may be due
391
to individual heterogeneity or time effects rather than the variables we are interested in,
392
so we used a two-way fixed model to control for these factors:
393
       
394
 (15)
395
where  and  denote the position indicators of country i in
396
ETN and ICN respectively, which are consistent with equation (14)., ,
397

16
 and  denote the GDP, GDP per capita, R&D input and labour
398
participation rate (LPR) of country i respectively.
399
3. Results
400
3.1 Network visualization
401
We visualized the ETN and ICN in 2011 and 2022 respectively. The results are
402
shown in Figures 1 and 2. Since the edges of the original ETN are too dense and it is
403
easy to overlook the key topological information in the network, we extracted the top
404
one import and export trading partners and their trading relationships for each country
405
in ETN to map the backbone of the ETN. The nodes in the network represent different
406
countries, and the size of the nodes is directly proportional to their total trade volume
407
or the number of innovative cooperation relationships; The color of the node represents
408
the region where the country is located; The label size of nodes is proportional to the
409
export volume; The size of the edge represents the flow between counrties; The edges
410
in the ETN network have arrows, indicating the direction of trade.
411
In 2011, ETN was characterised by few-core clustering, as shown by Germany,
412
the Netherlands, the U.S. and China leading the way with a network density of 0.0406.
413
This suggests that fewer countries are located at the centre of the network and that there
414
is a more frequent and efficient flow of information and resources through the network,
415
creating a dense trade network. The total trade volume of ECD intermediate products
416
of Germany, the Netherlands, the U.S. and China is in the top four. Developing to 2022,
417
ETN was characterised by multi-core dispersion and the focus of trade has gradually
418
shifted eastward, manifested by the rapid growth of trade volumes in Malaysia and
419
Thailand in Southeast Asia, which are leading ECD trade in intermediate products
420
alongside the U.S., Germany, China and the Netherlands. At this time, the network
421
density dropped to 0.0247, indicating that the relationship between countries is more
422
diversified and more dependent on globalization and cross-border cooperation and
423

17
competition. But it may also be affected by greater market volatility and global
424
economic changes.
425
426
427
Figure 1 Visualisation of the ETN pattern
428
In 2011, ICN showed a clear star-shaped structure centred on the U.S., with
429
countries such as the UK, Germany and Canada at the sub-centre for technological
430
innovation cooperation. The network density of 0.0216 indicates that there is more
431
communication, interaction and cooperation between the various players in the
432
economic system. The U.S. ranked first with 4,093 cooperations, demonstrating its
433
absolute dominance and influence in the field of ECD product innovation. Followed by
434
Germany and the UK, its main technological innovation partners are European
435
countries such as Switzerland, France and Austria, showing obvious characteristics of
436
regional innovation cooperation clusters. Canada and China are close behind and their
437
main innovation partners are not only the traditional developed countries in Europe, but
438
also Asian countries such as Indian and Japanese. Developing to 2022, ICN still be a
439
star structure with the U.S. as the core, with Indian, German, China and Canadian in
440
the sub-core position. The decrease in network density to 0.0078 indicates a decrease
441
in technological innovation cooperation and exchanges between countries, with links
442
between collaborators becoming looser or unidirectional, which is not conducive to the
443
sharing and transfer of knowledge. The focus of U.S. cooperation has shifted to Asia.
444

18
Germany ranks second, with its technology innovation partners still mainly located in
445
the Europe. As two Asian countries, China and India have strong influence in ICN.
446
447
448
Figure 2 Visualisation of the ICN pattern
449
3.2 Impact of export controls on ETN and ICN
450
In order to assess how U.S. export controls on ECD can bring about interdependent
451
cascading shocks across GVCs and technology innovation cooperation network, we
452
simulated equations (1)-(9). When export control-induced failures terminate at some
453
point, both ETN and ICN connectivity decreases and certain regional communities are
454
broken up into blocks. To this end, we measured network connectivity using the
455
network’s transitivity, average path length, assortativity, and maximum component size,
456
and normalise them separately to observe evolutionary trends between years. The
457
results are shown in Figures 3 and 4.
458
The transitivity indicates the tendency of countries in the network to form closed
459
triangles. It can be seen that the overall transitivity shows an increase trend from year
460
to year, suggesting that there is a greater number of friends of friends connecting with
461
each other, with ETN and ICN forming a degree of cluster effect. However, the
462
transitivity after experiencing the shock is significantly lower than the original network,
463
and the decline is greater for the ETN. This suggests that when shocks from the U.S.
464

19
propagate the network, they lead to less interaction and cooperation between nodes in
465
the supply chain, reducing the efficiency of flows. The average path length represents
466
the average length of edges that need to pass between any two reachable nodes in the
467
network. It can be seen that the average path lengths of the original ETN and ICN
468
remain around 1.7 and 2 respectively. However, shocks reduce the average path length
469
suggesting that trade flows and knowledge-sharing paths between countries are blocked,
470
leading to an increase in the number of isolated countries. It means that these countries
471
are more closed, and information, resources and opportunities may be more likely to
472
flow locally, making it difficult to reach out to new external opportunities for
473
cooperation. The assortativity indicates the convergence of connections between nodes,
474
i.e., the tendency of nodes with similar degree. Both ETN and ICN are characterised by
475
negative assortativity, suggesting that there is a phenomenon of preference attachment
476
in the network, i.e., small, peripheral countries tend to cooperate with large core
477
countries. After the shocks, the preference attachment of ETN and ICN persists, and the
478
tendency of ETN is more pronounced. Initial evidence suggests that the first to fail in
479
trade networks are the small, peripheral countries that cooperate less with the core
480
powers and fail to form a solid supply chain system through trade channels. These small
481
countries are not able to diversify their risks sufficiently when they receive shocks from
482
outside, resulting in a drop in capacity below the threshold. However, for ICN, due to
483
the absolute centrality of the U.S., the failure of the rest of the countries has less impact
484
on the whole network, so it has certain self-stability. The maximum component size is
485
the number of nodes in the largest interconnected subgroup or community. It can be
486
seen that the connectivity and stability of both ETN and ICN have been severely
487
disrupted by the shocks. The entire network was broken down from a highly connected
488
whole to a fragmented block structure, resulting in increased systemic production risks
489
and impeded progress in knowledge dissemination.
490

20
491
492
Figure 3 Impact of ECD export control shocks from the U.S. on ETN topology
493
494
495
Figure 4 Impact of ECD export control shocks from the U.S. on ICN topology
496

21
3.3 Coupled effect of export controls on ETN and ICN
497
In order to show the coupling effect of ECD export control shocks from the U.S.
498
on ETN and ICN, we used equation (10) to obtain a composite assessment metric of
499
network connectivity. The results are shown in Figure 5.
500
The damage caused by the shock to connectivity shows a decreasing trend as the
501
year increases. In 2015, the shock has the least negative impact on the overall
502
connectivity of ETN and ICN. Then it experienced a period of twists and turns, and the
503
level in 2022 was basically the same as that in 2013. To improve the resilience of ICN
504
to the shock, it is necessary to adjust from three aspects: countries, the relationships
505
between countries and the dependence between networks. First, enhance the country’s
506
adaptability and strengthen its capacity for independent innovation to consolidate the
507
domestic industry chain’s competitive advantage. This approach optimizes the
508
threshold for risk tolerance, reducing the likelihood of supply chain disruptions
509
inducing dependent ICN cascading failures. Second, the determination of the optimal
510
threshold depends not only on the country itself, but is also influenced by neighbouring
511
nodes. A country needs to regulate the balance between dependence and independence
512
with its partners. Finally, the interdependence between the supply chain network and
513
the technological innovation cooperation network is adjusted in time during the
514
innovation interaction process. On the one hand, real-time monitoring and response to
515
the variation and failure of key nodes in the supply chain to prevent the impact on the
516
normal innovation activities of domestic firms. On the other hand, ensuring inter-
517
country independence from third parties in terms of feedstock supply, technology
518
agreements, technology processes and knowledge acquisition can minimize the impact
519
of failure on the whole ICN.
520

22
521
Figure 5 Impact of U.S. ECD export control shocks on the connectivity of ETN and ICN
522
From a higher-order dependency perspective, it can be observed that a large
523
proportion of countries fail with capacity below the threshold. And most of the failed
524
nodes are characterised by a high degree of connectivity but uneven loading. Combined
525
with geographic location information, small countries at the periphery of the network
526
are more prone to failure (Figures 6 and 7). In 2011, a total of 87 countries were
527
destroyed, including 48 countries in ETN and 38 countries in ICN. In 2022, a total of
528
94 countries were destroyed, including 56 countries in ETN and 37 countries in ICN.
529
In ETN, the countries that failed in 2011 are mainly distributed in Africa, Oceania and
530
Central America, such as Benin, Eritrea, Seychelles, Zimbabwe, Marshall Islands and
531
Kiribati. In 2022, the number of failed countries from Asia and Europe increased, such
532
as Azerbaijan, Kyrgyzstan, Laos and Macedonia, but the countries that are most
533
vulnerable to failure are still concentrated in Africa and Oceania. In ICN, the countries
534
that failed in 2011 are concentrated in West Asia, South America and Africa, such as
535
Ghana, Israel, Kuwait, Peru and Nigeria. In 2022, the number of failed countries in
536
Europe increased significantly, but most of them were small countries in Europe, such
537
as Croatia, Luxembourg, Lithuania and Latvia.
538

23
539
540
Figure 6 Distribution of failure countries affected by the shock in 2011
541
542
543
Figure 7 Distribution of failure countries affected by the shock in 2022
544
We also evaluated the typical impact process of ECD export control shocks from
545
the U.S. on ETN and ICN. The shock from these failed nodes lead to a process of
546
secondary distribution of flow in the network. Figures 8 and 9 shows the top twenty
547
flow changes during shock propagation. What is clear is that ETN has a number of core
548
countries and that the distance between them is more diffuse. Thus the shock from the
549
U.S. spread further and affected a wider range of countries. ICN, on the other hand,
550
presents a star-shaped network structure with the U.S. at its absolute core, which makes
551
it easier to influence countries directly linked to the U.S.. Therefore, the indirect impact
552
reflects a low degree of high-order dependence.
553
In ETN, the shock from U.S. export controls on ECD in 2011 would significantly
554

24
reduce trade flows between countries in East Asia, Europe and the Americas. Examples
555
include trade flows from China to Japan, Germany to Poland, Argentina to Spain and
556
Indonesia to Italy. Reductions due to the shock from the U.S. in 2022 are then gradually
557
shifted eastwards. Trade flows concentrated in South-East Asia and East Asia, such as
558
China and Malaysia. As well as a small number of trade flows from the European, such
559
as Malaysia to Thailand, China to the Netherlands and China to Brazil. In ICN, the
560
shock from the U.S. in 2011 would significantly reduce the number of cooperations
561
between the U.S. and Canada, China and the UK. and the shock would also ripple
562
through the European region, e.g., cooperations between Switzerland and France,
563
Germany and the UK, and France and the UK. The shock also spread to Europe, such
564
as Switzerland and France, Germany and Britain, and France and the UK. The shock
565
from the U.S. in 2022 significantly reduce the amount of cooperation between the U.S.
566
and India, China, and Canada. The shock with larger negative impacts do not spread to
567
more distant regions.
568
569

25
570
Figure 8 Shock propagation process for ETN and ICN under US export controls. Note: the decreases
571
in the top 20 trade flows and technological innovation cooperation flows are presented.
572
(a) 2011 ETN; (b) 2022 ETN; (c) 2011 ICN; (d) 2022 ICN
573
In the ETN, the shock from the U.S. in 2011 significantly increase flows from
574
Europe, West Asia and Central America, causing small countries in these regions to
575
compensate for their losses by deepening their level of cooperation with other trading
576
partners. For example, flows from Austria to Croatia, Austria to Bosnia and
577
Herzegovina, the UAE to Jordan and Guatemala to Barbados. The increase due to the
578
shock from the U.S. in 2022 occurs more among small countries in Europe and Africa.
579
For example, flows from Germany, Belgium, Spain and the UK to Mauritius and flows
580
from the Czech Republic, Austria and Bulgaria to Montenegro. The higher density of
581
ICN in 2011 and the strong innovation cooperation between countries means that the
582
impact of the shock from the U.S. leading to an increase in the number of cooperations
583
spreads farther. The shock significantly increases the number of cooperations between
584
the Americas and the European. For example, the UK cooperates with France, Italy and
585
Spain, and Canada with Finland and China. ICN density declines in 2022, with inter-
586
country partnerships becoming fragmented and characterised by a rise in localised
587
agglomeration, leading to greater vulnerability to the shock within the communities of
588
countries that work closely with the U.S.. The impact from the U.S. will therefore
589
significantly increase the amount of cooperation in Europe and Asia. For example,
590
between the US and China, India, Israel, Sweden and Ireland, and between Russia and
591

26
France and India.
592
593
594
Figure 9 Shock propagation process for ETN and ICN under US export controls. Note: the increases
595
in the top 20 trade flows and technological innovation cooperation flows are presented. (a)2011
596
ETN; (b)2022 ETN; (c) 2011 ICN; (d) 2022 ICN
597
3.4 Impact of interdependent cascade failure on innovation ability
598
According to equation (13), we calculated the decline ratio of the network
599
innovation capability from 2011 to 2022. When the innovation ability decay ratio is
600
smaller, it indicates that the network innovation ability is more affected by the impact
601
of the shock, and the innovation ability declines more. As can be seen from Figure 10,
602
there is an overall growth trend before 2018. The impact of U.S. export controls on
603
ECD made the network’s overall innovative ability volatile, reaching its lowest point in
604

27
2015. After a short period of growth that peaked in 2018, innovation ability has been
605
declining ever since.
606
607
Figure 10 Dynamics of network innovation ability
608
Considering the attenuation of each country’s innovative ability, we further
609
calculated the initial innovative ability and the post-shock propagation innovative
610
ability of each country based on equations (11) and (12). Further calculations yield an
611
indicator of the country’s declining innovative ability. We visualised this on a map
612
based on shades of colours, where darker represents more innovation ability decay and
613
the labels indicate the top ten countries with the most innovation ability decay. The
614
results are shown in Figures 11 and 12. As can be seen, when the shock from U.S. ECD
615
export controls propagate through the network, most of the countries with the highest
616
declines in innovation ability in 2011 are located in Europe and North America, as well
617
as in Japan. The countries with the most declining innovation ability in 2022 shift
618
eastwards. China, Russia and India, as BRICS countries, have strengthened their ties
619
with the rest of the world and have gradually climbed up the ranks of the international
620
market, leading to a faster decline in their innovative ability and making them the more
621
affected countries. We further showed this for the top ten countries with the largest
622
declines in innovation ability from 2011-2022. The results are shown in Figure 13. It
623
can be seen that, on average, it is the countries in the African that are most affected by
624

28
innovation ability.
625
626
Figure 11 Decline in innovation ability in countries affected by the shock in 2011
627
628
Figure 12 Decline in innovation ability in countries affected by the shock in 2022
629

29
630
Figure 13 Top 10 countries with declining innovative ability
631
3.5 Vulnerability analysis
632
In order to better understand the relationship between the location and attributes
633
of a country in the network and the simulation results, we used equation (14) to conduct
634
a series of regression analysis on the panel data generated by the simulation results of
635
the dependent cascade failures of ETN and ICN. The dependent variable is the demand
636
deficit from source country i received by destination country j. The results are shown
637
in Table 1.
638
From a source-country perspective, the more the number of import and export
639
trade relationships and the greater the trade volume, the less the shock of its spread. It
640
suggests that a symmetrically homogeneous trade structure makes it more resilient to
641
risk, and that a large trading country at the centre is less likely to become a failure point
642
and propagate negative shocks. When a source country are connected to multiple core
643
countries, they may have faster access to information and respond to global market
644
movements, thus managing and adjusting their export strategies more effectively and
645
reducing potential negative shocks to other countries. When a source country is in a
646
high agglomeration community, it is less likely to spread negative shocks. The large
647
number of interconnections that exist in dense clusters diversifies the channels through
648

30
which the country can disperse shocks, thus reducing the risk of disproportionately
649
negative shocks on individual countries. Considering the innovation cooperation
650
relationship of exporting countries, the more innovation partners there are in a country,
651
the shock of its diffusion is higher. However, when the number of its innovation
652
cooperation is larger, the shock of its diffusion is lower. This suggests that
653
asymmetrically nonhomogeneous innovation patterns can help a country withstand
654
shocks from the supply chain. While a country’s linkages with more partners increase
655
the broader impact of its shocks, as the number of cooperations continues to grow, these
656
shocks are absorbed and dispersed across the network, reducing the impact on any
657
single node. When a country is more located in the shortest path of trade links with
658
other countries, it shows that it has more space to disperse or transmit the shock it faces.
659
From the perspective of the destination country, a country’s import trade is more
660
significant in terms of the intensity of the external shocks it receives than its export
661
trade. The larger a country’s import trading partners and volume of import trade, the
662
smaller its exposure to shocks, in line with our intuition. When the country is more
663
connected to the core countries, it receives fewer external shocks. It suggests that the
664
core countries are more resilient to risk, thus making their trade relations more stable.
665
When the country acts more as an “intermediary” between other trade cooperating
666
countries, it is more likely to diversify the trade shocks it receives, thereby reducing
667
risk. And when the number of trade relationships around the country is denser, it
668
receives fewer shocks. Considering innovation partnerships in importing countries, a
669
country receives a smaller degree of shock when it has a larger number of innovation
670
partnerships. The closer the country cooperates with a large innovative country, the
671
smaller the degree of shock it receives. And the greater the density of innovation
672
partnerships around it, the greater the degree of shock it receives. Combined with the
673
country’s performance in the trade network, this suggests that the high concentration of
674
destination countries in the trade network mitigates shocks through risk diversification.
675
However, the high concentration in the innovation network is closely related to the
676

31
specific technology and the market, which increases the sensitivity and influence to the
677
impact.
678
Table 1 Vulnerability analysis: Demand deficit of bilateral trade flows
679
Variables
Demand distribution
Demand distribution
Origin countries
Trade_Pagerank
-14.8019***
(4.1010)
-15.1492***
(4.1465)
Innov_Pagerank
-9.5619**
(4.8253)
-5.2285
(4.9229)
Trade_Outstrength
-5.1778***
(1.0186)
-5.6024***
(1.0276)
Trade_Instrength
0.7312
(0.7169)
1.0848
(0.6993)
Innov_Strength
-5.4480***
(1.3021)
-6.0216***
(1.1919)
Trade_Transitivity
-27.6227*
(16.4826)
7.9433
(16.6200)
Innov_Transitivity
1.1455
(1.8494)
4.0363**
(1.8077)
LPR
-10.3830***
(1.8394)
-15.0792***
(1.8506)
GDP
-4.0094
(3.4311)
-2.4452
(3.4346)
GDPPC
4.7391***
(1.2524)
6.3887***
(1.2806)
R&D
5.9605***
(1.3188)
4.7954***
(1.4101)
Trade_Outdegree
-30.6183***
(3.1338)
Trade_Indegree
-6.5034***
(2.3383)
Innov_Degree
7.0576***
(0.9869)
Trade_Betweeness
66.8092
(60.2494)
Innov_Betweeness
-3.2046***
(0.6927)
Trade_Outcloseness
-70.6473***
(8.7203)
Trade_Incloseness
1.6769

32
(8.4583)
Innov_Closeness
-23.6130***
(3.6050)
Trade_Hole
113.3186
(130.3369)
Innov_Hole
-16.6476***
(2.0884)
Destination countries
Trade_Pagerank
-14.8019***
(4.1010)
-15.1492***
(4.1465)
Innov_Pagerank
-9.5619**
(4.8253)
-5.2285
(4.9229)
Trade_Outstrength
0.0275
(0.7721)
1.2649*
(0.7365)
Trade_Instrength
-3.5258***
(1.2722)
-4.4634***
(1.2449)
Innov_Strength
-0.8606*
(0.4769)
-1.1576**
(0.5483)
Trade_Transitivity
-103.3886***
(19.9816)
-79.6097***
(20.3639)
Innov_Transitivity
4.7161**
(1.8573)
8.6211***
(1.8261)
LPR
-8.1056***
(1.9948)
-12.5615***
(1.9417)
GDP
3.4437***
(1.1927)
6.1069***
(1.1461)
GDPPC
-2.3903**
(1.2107)
-1.4216
(1.2042)
R&D
2.5151*
(1.4176)
1.9682
(1.4150)
Trade_Outdegree
2.4915
(2.4153)
Trade_Indegree
-23.5396***
(2.7850)
Innov_Degree
0.5399
(0.9078)
Trade_Betweeness
-195.0284***
(66.1336)
Innov_Betweeness
0.2627
(0.6162)
Trade_Outcloseness
-27.8328***
(7.7386)

33
Trade_Incloseness
-29.9974***
(8.0956)
Innov_Closeness
-2.7848
(3.6654)
Trade_Hole
-254.7590
(163.2208)
Innov_Hole
-5.3711**
(2.3003)
Constant
90.9041***
(9.9953)
96.0866***
(16.0267)
TE
√
√
FE
√
√
N
14171
14171
F
19.9205
18.1597
R2
0.7994
0.7978
Note: Values in parenthesis are robust standard errors. *, ** and *** indicate statistically significant
680
coefficient at 10%, 5% and 1%, respectively.
681
In order to analyse the impact of the shock from the U.S. on the vulnerability of a
682
country’s innovation ability, we set the dependent variable as the decay of the country’s
683
innovation ability under the shocks. It can be predicted that the technological innovation
684
cooperation under the shock affect a country’s innovation ability more than trade
685
relationships. The results are shown in Table 2.
686
From ETN perspective, when a country has a large number of export trade
687
relationships or a high volume of export trade, the less its innovative ability decays.
688
However, when a country’s import trade is low, the more its innovative ability decays.
689
This suggests that when a country has a more symmetrical and homogeneous trade
690
structure, it is more resilient to supply chain shocks and thus maintains its innovative
691
capacity. From ICN perspective, when a country has an asymmetric and
692
nonhomogeneous innovation pattern, that is, it has more technological innovation
693
partners or fewer cooperation relationships, the more its innovation capability decays.
694
As a country acts more as a mediator in technological innovation partnerships, the less
695
its innovation ability decays. The significantly positive coefficient on the eigenvector
696
centrality variable suggests that when a country cooperates with another large
697
innovative country with a high number of partnerships, it is more likely to lead to a
698

34
decline in innovation ability when the shock hits. The significant positive coefficient
699
on the transitivity variable suggests that a country has formed relatively dense small
700
communities with its partners, resulting in a more closed country with limited access to
701
new external information, resources and opportunities. The negative impacts of the
702
shock, on the other hand, are more likely to circulate internally and fail to spread
703
effectively to the wider country, thus increasing stress within communities.
704
In terms of the country’s own non-network attributes, a country’s GDPPC and
705
R&D can significantly affect its innovation ability, i.e., the higher the GDPPC, the less
706
the country’s innovation ability decays. And the higher the R&D, the more the country’s
707
innovative ability decays. Although counter-intuitive at first glance, this result is due to
708
the fact that countries with high R&D are more involved in international innovation
709
activities and are therefore more likely to be exposed to the shock from different sources.
710
At the same time, they tend to engage in more cooperative relationships.
711
Table 2 Vulnerability analysis: Decline of innovation ability
712
Variables
Reduced innovative capacity
Reduced innovative capacity
Trade_Pagerank
0.2678
(0.2096)
0.6827***
(0.2532)
Innov_Pagerank
0.9804*
(0.5293)
0.6561
(0.7719)
Trade_Outstrength
0.0715***
(0.0267)
0.0153
(0.0254)
Trade_Instrength
-0.0148*
(0.0077)
-0.0298*
(0.0156)
Innov_Strength
-0.1897***
(0.0447)
-0.0793**
(0.0320)
Trade_Transitivity
-0.0092
(0.2210)
0.0197
(0.2065)
Innov_Transitivity
0.3296***
(0.0992)
0.4371***
(0.0964)
LPR
-0.0723
(0.0898)
-0.1024
(0.0900)
GDP
-0.0329
(0.0812)
0.0404
(0.0665)
GDPPC
-0.2155**
(0.0843)
-0.2168***
(0.0799)

35
R&D
0.3445***
(0.0919)
0.3179***
(0.0874)
Trade_Outdegree
-0.5917***
(0.2009)
Trade_Indegree
-0.1504
(0.1506)
Innov_Degree
0.9691***
(0.0477)
Trade_Betweeness
4.1351
(4.0776)
Innov_Betweeness
-0.3249***
(0.0635)
Trade_Outcloseness
-0.9014***
(0.1954)
Trade_Incloseness
-1.2326***
(0.2930)
Innov_Closeness
-2.3727***
(0.2130)
Trade_Hole
0.0874
(0.1630)
Innov_Hole
0.3789***
(0.1207)
Constant
-0.4013***
(0.1224)
0.5924***
(0.2181)
TE
√
√
FE
√
√
N
2688
2688
F
79.6980
41.6711
R2
0.5452
0.5512
Note: Values in parenthesis are robust standard errors. *, ** and *** indicate statistically significant
713
coefficient at 10%, 5% and 1%, respectively.
714
3.6 Scenarios design
715
In order to observe the wide range of uncertainties and corresponding realities
716
associated with U.S. export controls, we set up scenario simulations based on the
717
selected parameters and value assignments. Four types of interdependent cascade
718
failure modes were modelled using different node load and capacity change rules. The
719
previous parameters were set to  = 1.5 and  = 0.5 as the reference scenario (S1),
720

36
and three other scenarios were analysed: The upper limit parameter  of the node
721
capacity was set to 1.5 and 2.0, and the lower limit parameter  of the node capacity
722
was set to 0.5 and 0.1, respectively, subject to the preconditions that 0 <  < 1 and 
723
≥ 1. The scenario parameters are set as shown in Table 3. We used the same
724
methodology as in the previous section to estimate the negative impact of dependent
725
cascade failures on the network under the diffusion of trade shocks. The results of the
726
estimation are shown in Figures 14-16.
727
Table 3 Parameter setting of dependent cascade failure model
728
Scene Setting
Upper limitation of node
load (󰇜
Lower limitation of node
load 󰇛)
 : Resource constraints,
efficient management (baseline
model)
=1.5
=0.5
 : Resource constraints,
inefficient management
=1.5
=0.1
: Adequate resources, efficient
management
=2.0
=0.5
 : Adequate resources,
inefficient management
=2.0
=0.1
The lower parameter of node capacity has a dominant effect on network
729
vulnerability, while the upper parameter of node capacity has a limited impact. This
730
further confirms that small countries that are overly dependent on imports or innovative
731
foreign technological cooperation are more likely to struggle in the face of global
732
shocks. Therefore special attention needs to be paid to these vulnerable nodes to
733
maintain the stability of the overall network. By comparing S1 and S2, S3 and S4, it can
734
be seen that as the lower limit of the node capacity increases, its sensitivity to external
735
shocks is higher. The ETN and ICN topologies show a decreasing trend in
736
interdependent connectivity, with the lower parameter growing from 0.1 to 0.5 from S2
737
to S1 and from S4 to S3, with a decrease in interdependent connectivity of 0.0005 and
738
0.0008 respectively. Comparison of S1 and S3, S2 and S4 shows that as the upper limit
739
of the node capacity decreases, its ability to carry redistributed load decreases. Both
740
types of networks have the same trend of decreasing interdependent connectivity. On
741

37
average, from S3 to S1 and S4 to S2, the upper parameter decreases from 2.0 to 1.5, and
742
the interdependent connectivity decreases by 0.0004 and 0.0007 respectively. The
743
reductions are all smaller than the fluctuations caused by the change in the lower limit
744
of node capacity.
745
746
Figure 14 Dynamics of ETN and ICN dependent connectivity under scenario analysis
747
From ETN perspective, comparing S1 and S2, S3 and S4 shows that the sensitivity
748
of the nodes increases as the lower limit of node capacity increases. Both the
749
connectivity and the completeness of the network have been more severely disrupted.
750
In addition, when the upper limit of the node capacity is low, the node have low load
751
capacity. The high sensitivity and low load capacity of the node make the negative
752
impact more obvious. This is shown by the fact that when the load capacity of the node
753
is low, accompanied by an increase in the lower parameter of node capacity from 0.1 to
754
0.5, there is a tendency for the average values of both the transitivity and the
755
assortativity of the network show an increasing trend. This suggests a gradual
756
withdrawal of peripheral countries from the network. Partnerships around core
757
countries are more stable and more likely to survive in a volatile external environment.
758
The mean values of shortest path length and maximum connected component size show
759
a decreasing trend. This shows that the gap between countries’ cooperative relationships
760
has increased, and some countries are even isolated. When the traffic carrying capacity
761

38
of a node is increased due to the growth of the upper limit parameter, the network
762
metrics vary less compared to when the carrying capacity is low. When the load
763
capacity of nodes is improved due to the increase of the upper limit parameters, the
764
variation range of network indicators is smaller than that when the load capacity is low.
765
766
767
Figure 15 Dynamics of ETN connectivity under scenario analysis
768
From the perspective of ICN, except for the assortativity, the trend of other
769
network indicators is basically consistent with ETN. However, overall, ICN is less
770
sensitive to changes in node capacity parameters. When the load capacity of the node
771
is low, it is accompanied by an increase in the node capacity lower limit parameter from
772
0.1 to 0.5, which decreases all the indicators except for the transitivity, which shows an
773
increasing trend. It suggests that the process of flow redistribution caused by the shock
774
makes those countries that do not directly cooperate with the U.S. fail, which leads to
775
an increase in the transitivity presented by the U.S.-centred star structure.
776
It suggests that the process of flow redistribution due to shocks invalidates
777
countries that do not work directly with the US, leading to an increase in the
778

39
agglomeration presented by the US-centred star structure. The center of the whole
779
network is more inclined to the core power and the partnership cannot be extended
780
much further. When the load capacity of the node is high, the network stability is
781
enhanced, which is reflected in the decrease in the variation of the network indicators.
782
783
784
Figure 16 Dynamics of ICN connectivity under scenario analysis
785
4. Conclusions
786
4.1 Research findings
787
This study extends the traditional single inter-country cooperation network by
788
constructing an interdependent cascade failure model consisting of the supply chain
789
sub-network, the innovation cooperation sub-network and the supply chain-innovation
790
cooperation binary network, both at the macro-whole network and micro-national
791
individual level. Specifically analyses the phenomenon of cross-network collapses
792
caused by the shock of high- technology product supply chain from U.S. CCL and
793

40
examines the vulnerability effects of the shock on the innovative capacity of individual
794
countries.
795
The results of the study are as follows: (1) There is a threshold of load capacity in
796
the country. If it is less than this threshold, ETN and ICN will have a dependent cascade
797
collapse when the shock comes. And with the change of years, ETN and ICN structures
798
tend to be decentralized and multi-core, which makes the sock spread more widely in
799
the network. It is embodied in the decoupling of the network from a highly connected
800
overall structure to a fragmented block structure. (2) Failed nodes lead to secondary
801
propagation of shocks through the network, and these failed nodes tend to be located in
802
peripheral regions of the network and are characterised by more cooperative
803
relationships but uneven traffic distribution. The countries in ETN that are affected by
804
decreasing flows are gradually shifting eastwards, while the countries that are affected
805
by increasing flows are shifting from the Central America to the Africa. ICN’s flow
806
distribution is more concentrated within the U.S. neighbourhood. (3) The supply chain
807
shock from the U.S. affects the innovation ability of the whole network. Countries
808
whose innovation ability is vulnerable to the shock are gradually shifting from the
809
Europe to emerging markets. The countries with the weakest innovation ability are
810
concentrated in Africa and West Asia, such as Kuwait, Tajikistan, Gabon and Guinea.
811
(4) Countries are more robust when they are in a symmetrically homogeneous and
812
preference attachment supply chain structure. And when it is located in a dense
813
community, the demand deficit from the supply chain is more easily dispersed through
814
diversified channels. (5) In the face of the shock, technological innovation cooperation
815
has a more significant impact on a country’s innovation ability than supply chain
816
relationships. Countries show significant resilience to the supply chain shock in their
817
innovation ability when they are in an asymmetrically nonhomogeneous, non-preferred
818
attachment, and relationally mediated innovation model. This model reduces the
819
potential negative impact of the failure of core powers, and maintains the stability of its
820
own supply chain by spreading the demand deficit to a wider range of partners, thus
821

41
maintaining the continuity of technological progress and industrial upgrading. (6) The
822
lower parameter of node capacity plays a dominant role in the country’s supply chain
823
security and innovation ability, while the upper parameter has a limited impact. When
824
the country has a lower upper capacity limit and a higher lower capacity limit it exhibits
825
high sensitivity to the shock and low carrying capacity, which leads to varying degrees
826
of damage to both ETN and ICN structures after the interdependent cascades failure.
827
4.2 Policy implications
828
This study proposes the following policy implications: (1) Countries needs to
829
improve its independent innovation ability and promote regional partnership economic
830
relations, adjust the optimal threshold of allowable parameters of anti-risk capability,
831
and reduce the possibility of cascading failure of high-technology product supply chain
832
and technological innovation network induced by the shock. (2) International
833
organisations should pay more attention to small countries in Africa and Asia to deal
834
with these risks. Simply increasing their own capacity will not increase the dynamism
835
and resilience of the overall innovation pattern. On the contrary, we should encourage
836
complementary cooperation and technical exchange between small countries and large
837
countries, and by seeking the optimal coupling strength of the supply chain of high-
838
technology products and international innovation cooperation to achieve a real global
839
innovation balance. (3) Countries should make trade breadth (diversification of trading
840
partners) and trade depth (growth in trade volume) go hand in hand so as to enhance
841
the homogeneity of the supply chain network for high-technology products. This
842
reduces reliance on a single source of supply to prevent a widespread cascading effect
843
if one node fails. (4) Core countries need to actively build trade relationships with
844
peripheral countries in high-technology products, while peripheral countries need to
845
pay attention to the dynamics of high-technology product trade in core countries. Two-
846
way adjustment creates preference attachment in international markets and enhances
847
supply chain resilience. At the same time, developed countries and international
848

42
organisations are encouraged to increase technology transfer and financial support to
849
vulnerable countries to help them enhance their self-recovery and adjustment capacity.
850
(5) Peripheral countries should make full use of bilateral economic, cultural and
851
political relationships to establish robust trade clusters for high-technology products,
852
thereby dissipating demand deficits and creating stable supply flows. The density of
853
innovation cooperation within neighbourhoods should also be further enhanced to
854
facilitate the emergence of new innovative dynamics. (6) Policymakers should pay
855
more attention to increasing the affordability of vulnerable countries under minimum
856
capacity conditions. Increase investment in infrastructure in less economically
857
developed countries to ensure that they can maintain their basic trade and productive
858
capacities in the face of external economic shocks.
859
There are still some shortcomings that need to be improved in future studies. The
860
analyses we carried out are based on physical flow and mass conservation equations
861
and do not take into account the role of international political relationships on the
862
dynamic process of cascade failures. External influences such as politics and the order
863
of international relationships should be quantified and modelled in future studies. In
864
addition, we were concerned with the impact of cascading effects on the countries at
865
the end of the value chain, that is, the consumer markets for the products. However
866
when a country imports a large amount of ECD while also exporting a larger amount of
867
ECD, then the country adds value in the process. Therefore, in the future, a refinement
868
of the dynamic process of shock propagation in the context of global value chains could
869
be considered.
870
871
Reference
872
Ahern, K. R., Daminelli, D., & Fracassi, C. (2015). Lost in translation? The effect of cultural values on mergers
873
around the world. Journal of Financial Economics, 117(1), 165–189.
874
https://doi.org/10.1016/j.jfineco.2012.08.006
875
Amiti, M., & Konings, J. (2007). Trade liberalization, intermediate inputs, and productivity: Evidence from
876

43
indonesia. American Economic Review, 97(5), 1611–1638. https://doi.org/10.1257/aer.97.5.1611
877
Baldwin, R., & Freeman, R. (2022). Risks and global supply chains: What we know and what we need to know.
878
Annual Review of Economics, 14(Volume 14, 2022), 153–180. https://doi.org/10.1146/annurev-economics-
879
051420-113737
880
Benguria, F., Choi, J., Swenson, D. L., & Xu, M. (2022). Anxiety or pain? The impact of tariffs and uncertainty on
881
Chinese firms in the trade war. Journal of International Economics, 137, 103608.
882
https://doi.org/10.1016/j.jinteco.2022.103608
883
Betz, J., & Hein, W. (2023). Globalization and technological development: Production, transport and communication.
884
In J. Betz & W. Hein, Globalization: Prerequisites, Effects, Resistances (pp. 21–42). Springer Fachmedien.
885
https://doi.org/10.1007/978-3-658-41717-8_2
886
Bown, C. (2020). Export controls: America’s other national security threat. Duke Journal of Comparative &
887
International Law, 30(2), 283–308.
888
Buldyrev, S. V., Parshani, R., Paul, G., Stanley, H. E., & Havlin, S. (2010). Catastrophic cascade of failures in
889
interdependent networks. Nature, 464(7291), 1025–1028. https://doi.org/10.1038/nature08932
890
Cai, Zhong-hua, Chen Hong, M. A. Huan, & China National Intellectual Property Administration. (2022). Sino-US
891
tech decoupling:Research and response in the field of technology. Studies in Science of Science, 1.
892
https://doi.org/10.16192/j.cnki.1003-2053.20210304.002
893
Choi, S., Furceri, D., & Yoon, C. (2021). Policy uncertainty and foreign direct investment. Review of International
894
Economics, 29(2), 195–227. https://doi.org/10.1111/roie.12495
895
Crosignani, M., Han, L., Macchiavelli, M., & Silva, A. F. (2023). Geopolitical risk and decoupling: Evidence from
896
U.S. export controls. Social Science Research Network. https://www.semanticscholar.org/paper/Geopolitical-
897
Risk-and-Decoupling%3A-Evidence-from-Crosignani-Han/39cff0d01ca7f12741f1bf5f6147ede25c8de675
898
Danilin, I. (2022). From technological sanctions to tech wars: Impact of the U.S. - China confl ict on sanctioning
899
policies and the high-tech markets. Journal of the New Economic Association, 55(3), 212–217.
900
https://doi.org/10.31737/2221-2264-2022-55-3-13
901
Gao, S., & Li, Z. (2024). Trade policy uncertainty, financing constraints, and firm innovation: Evidence from China.
902
Journal of the Knowledge Economy. https://doi.org/10.1007/s13132-024-02246-8
903
Hao, S., & Wang, M. (n.d.). Walking on eggshells: Politicizing sino-ROK semiconductor technological ties in the
904
shadow of sino-US rivalry. Pacific Review, 1–30. https://doi.org/10.1080/09512748.2024.2368223
905
Helbing, D. (2013). Globally networked risks and how to respond. Nature, 497(7447), 51–59.
906
https://doi.org/10.1038/nature12047
907
Huang, Ying-yi, Jin, Chun, & Rong, Li-li. (2014). Cascading failure model on logistics network based on the overall
908
importance of nodes. Operations Research and Management Science, 6.
909
https://kns.cnki.net/KCMS/detail/detail.aspx?dbcode=CJFD&dbname=CJFDLAST2015&filename=ycgl2014
910
06017
911
Jie, H., Leong, Y. C., & Fang, F. (2022). The US-China competition: Industrial chain and the future of systemic
912
great power. International Journal of Management, Accounting, Governance and Education, 83–100.
913
Liu, Bin & Li, Qiujing. (2023). US export control to China and china enterprise innovation. Journal of Finance and
914
Economics, 12. https://doi.org/10.16538/j.cnki.jfe.20230617.202
915
Liu, L., Wang, W., Yan, X., Shen, M., & Chen, H. (2023). The cascade influence of grain trade shocks on countries
916
in the context of the Russia-Ukraine conflict. Humanities and Social Sciences Communications, 10(1), 449.
917
https://doi.org/10.1057/s41599-023-01944-z
918

44
Lu, Y., Tao, Z., & Zhu, L. (2017). Identifying FDI spillovers. Journal of International Economics, 107, 75–90.
919
https://doi.org/10.1016/j.jinteco.2017.01.006
920
Min, B., Yi, S. D., Lee, K.-M., & Goh, K.-I. (2014). Network robustness of multiplex networks with interlayer
921
degree correlations. Physical Review E, 89(4), 42811. https://doi.org/10.1103/PhysRevE.89.042811
922
Nguyen, H. T. T., Larimo, J., & Ghauri, P. (2022). Understanding foreign divestment: The impacts of economic and
923
political friction. Journal of Business Research, 139, 675–691. https://doi.org/10.1016/j.jbusres.2021.10.009
924
Pei, J., Liu, Y., Wang, W., & Gong, J. (2021). Cascading failures in multiplex network under flow redistribution.
925
Physica A: Statistical Mechanics and Its Applications, 583, 126340.
926
https://doi.org/10.1016/j.physa.2021.126340
927
Qi, M., Chen, P., Wu, J., Liang, Y., & Duan, X. (2023). Robustness measurement of multiplex networks based on
928
graph spectrum. Chaos: An Interdisciplinary Journal of Nonlinear Science, 33(2).
929
https://doi.org/10.1063/5.0124201
930
Ranganathan, R., & Rosenkopf, L. (2014). Do ties really bind? The effect of knowledge and commercialization
931
networks on opposition to standards. Academy of Management Journal, 57(2), 515–540.
932
https://doi.org/10.5465/amj.2011.1064
933
Reis, S. D. S., Hu, Y., Babino, A., Andrade Jr, J. S., Canals, S., Sigman, M., & Makse, H. A. (2014). Avoiding
934
catastrophic failure in correlated networks of networks. Nature Physics, 10(10), 762–767.
935
https://doi.org/10.1038/nphys3081
936
Seo, J., Mishra, S., Li, X., & Thai, M. T. (2015). Catastrophic cascading failures in power networks. Theoretical
937
Computer Science, 607, 306–319. https://doi.org/10.1016/j.tcs.2015.08.021
938
Seyoum, B. (2017). Export controls and international business: A study with special emphasis on dual-use export
939
controls and their impact on firms in the US. Journal of Economic Issues, 51(1), 45–72.
940
https://doi.org/10.1080/00213624.2017.1287483
941
Shen, Haomin, Cheng, Xiaoke, & W. A. N. Qing. (2021). Does anti-dumping against China inhibit corporate
942
innovation? Finance
＆
Trade Economics, 4. https://doi.org/10.19795/j.cnki.cn11-1166/f.20210406.009
943
Shi, J., Yang, J., & Li, Y. (2019). Supply network position and firm performance: Evidence from Chinese listed
944
manufacturing companies. Journal of Business Economics and Management, 20(6), Article 6.
945
https://doi.org/10.3846/jbem.2019.10743
946
Shin, S., & Balistreri, E. J. (2022). The other trade war: Quantifying the korea–japan trade dispute. Journal of Asian
947
Economics, 79, 101442. https://doi.org/10.1016/j.asieco.2022.101442
948
Song, Y., Yang, N., Zhang, Y., & Wang, J. (2017). Modeling and simulation of cascading failure on R&D network
949
based on different node states under attack strategies. 2017 IEEE International Conference on Industrial
950
Engineering and Engineering Management (IEEM), 1286–1290. https://doi.org/10.1109/IEEM.2017.8290100
951
Vespignani, A. (2010). The fragility of interdependency. Nature, 464(7291), 984–985.
952
https://doi.org/10.1038/464984a
953
Wang, C., & Hu, Q. (2020). Knowledge sharing in supply chain networks: Effects of collaborative innovation
954
activities and capability on innovation performance. Technovation, 94–95, 102010.
955
https://doi.org/10.1016/j.technovation.2017.12.002
956
Wang, Jianwei, Jiang, Chen, & Sun, Enhui. (2014). Study on cascading failures’ model of edge in coupled networks.
957
Journal of Management Science, 6.
958
https://kns.cnki.net/KCMS/detail/detail.aspx?dbcode=CJFD&dbname=CJFDLAST2015&filename=jcjj20140
959
6012
960

45
Wang, X., Li, H., Yao, H., Zhu, D., & Liu, N. (2018). Simulation analysis of the spread of a supply crisis based on
961
the global natural graphite trade network. Resources Policy, 59, 200–209.
962
https://doi.org/10.1016/j.resourpol.2018.07.002
963
Wu, X. (2020). Technology, power, and uncontrolled great power strategic competition between China and the
964
united states. China International Strategy Review, 2(1), 99–119. https://doi.org/10.1007/s42533-020-00040-0
965
Yang, F., Wang, Y., & Whang, U. (n.d.). Impact of technical barriers to trade measures on innovation – evidence
966
from Chinese manufacturing firms. Economics of Innovation and New Technology, 1–19.
967
https://doi.org/10.1080/10438599.2024.2365316
968
Zhou, X.-Y., Lu, G., Xu, Z., Yan, X., Khu, S.-T., Yang, J., & Zhao, J. (2023). Influence of Russia-Ukraine War on
969
the Global Energy and Food Security. Resources, Conservation and Recycling, 188, 106657.
970
https://doi.org/10.1016/j.resconrec.2022.106657
971
Zhu, Z., Zheng, H., & Zhu, Z. (2022). Analysis on the economic effect of sino-US trade friction from the perspective
972
of added value. Environment, Development and Sustainability, 24(1), 180–203. https://doi.org/10.1007/s10668-
973
021-01390-4
974
975