Network effects of a shock to the semiconductor industry

The supply chain of semiconductors – popularly called chips – is vulnerable to shocks due to the high market concentration and strong global interdependencies. Five economies account for more than half of total trade in the semiconductor supply chain. And despite regional initiatives such as the CHIPS and Science Act from the US and the European Chips Act from the EU, both of which are intended to strengthen their respective positions within the supply chain, it will be nearly impossible for a single region to control chip production from start to finish. The main reasons for this are the high capital expenditures, R&D investment costs and accumulated expertise in the development and production of chips.

Chips are strategically important in that they are indispensable to so many aspects of our digital society, from cell phones to advanced weapons. Recently, we saw new distortions in the chips supply chain, as, among others the Netherlands imposed new export controls on ASML chip-making machines to China. China has just retaliated with export controls on germanium and gallium, two important raw materials in the production of chips and other critical products. But how will this and other shocks to the semiconductor industry affect other parts of the economy? In this comment we approach that question by applying network analysis.

Sectors around the world deliver to and receive inputs from each other. These flows form connections, which can be analysed by methods based on network analysis, which in turn has its origins in mathematical graph theory (Euler, 1736).

In figure 1, we show different types of network structures based on Foerster en Choi (2017). These networks have different density measures: the ratio between the actual connections and the possible connections, including intra-sectoral connections (self-loops). In the schematic networks there are 16 possible connections. The density of the first network is therefore 4/16 = 0.25.

How a sectoral shock affects other sectors depends on the type of network and a sector’s position within that network. As we can see in the first network, the non-connected network, a shock to a sector will not immediately impact other sectors as there are no connections between them.

This is different in the second network, the strong value-chain network. Imagine a supply-driven shock, such as export restrictions. A shock to sector 1 has a stronger initial effect on the entire chain than a shock to sector 4, which is at the end of the chain. For a demand shock this works the other way around.

In the third economy, the hub-network, we clearly see one central sector, so shocks that spread through sector 1 will have a large impact. And in the final economy, the fully interconnected network, a shock to any sector will likely have an equal initial impact on all sectors.

If we apply network analysis to an economy, we make use of input-output tables to understand the connections between sectors. For the following analysis we use the most recent Inter-Country-Input-Output (ICIO) table from the OECD. We will only focus on the intersectoral connections, the flows to final consumers are out of scope for this analysis.

In figure 1 it is very easy to visually determine the centrality of a sector. However, the input-output table we use in our analysis consists of 67 economies/regions and 45 industries, totalling 3015 sectors or nodes and with many links or edges between these nodes. Therefore, we need a method to establish the centrality of a sector. Network analysis provides various methods to calculate a sector’s centrality. In our case we chose to use the eigenvector centrality measure.

Our chosen centrality measure focuses either on the inflows and in-edges (so, from the perspecitive of a node, those values it receives from other nodes), or the outflows or out-edges (the value of the flows from a node to other nodes), of a sector. To get a good understanding of how a shock to a sector affects the rest of the economy, we need to look at both the in- and the out-centrality.

To tackle this issue we do the following. We first calculate the in-centrality. This gives us a ranking of the sectoral centrality, where the most central sector gets the rank of 1. Then we calculate the out-centrality, which also gives us a ranking. To get the final rank we take the average of the two rankings per sector. For example: sector A ranks 5 in the in-centrality and 10 in the out-centrality, therefore its average rank is 7.5.

The most central sector in the world economy, perhaps unsurprisingly, is the financial sector of the US. The Global Financial Crisis of 2008 reminds us what a shock to that sector could mean for the world economy. The fact that the financial sector of the US turns out to be the most central sector in our analysis is a good check on the validity of using the eigenvector centrality to measure the importance of a sector to the world economy.

The ICIO tables from the OECD provide detailed input-output information for 45 industries across 67 economies, but unfortunately there is not a specific sector corresponding to semiconductors. We know from previous analysis that several sectors are needed to produce semiconductors. Accordingly, we combine the following sectors which are involved in the production process: 26 (Computer, electronic and optical equipment), 27 (Electrical equipment) and 28 (Machinery and equipment, not elsewhere classified). In this report, we dissect the production chain of semiconductors in detail.

Of course it would be more accurate if the semiconductor industry was a separate sector instead of a more general, composite, sector that includes products not used in the semiconductor industry. While that data is not available, we believe that the general importance of the semiconductor itself in the electronics sector suggests that this more generalized analysis remains relevant.

In figure 2 we show the average eigenvector centrality rank (of in- and out-edges) of the combined electronics sector as defined above, per economy on the horizontal axis. The lower the rank, the higher the centrality and thus the larger the impact of a sectoral shock.

On the vertical axis we show the output size of an economy, to put the centrality measure in perspective. Because, of course, while the electronics sectors of the US and China are very central to the world economy, they are also very large economies. More interesting perhaps is the cluster of countries in the orange circle. These economies do not stand out in output size, but have relatively central electronics sectors. There are 3015 sector-economy combinations in total, so a rank between 100 and 300 implies a relatively high centrality. Taiwan and South Korea, both relatively small economies, have rather central electronics sectors. We also know that they are important to the semiconductor industry. In fact, about 40 percent of total chip fabrication capacity is located in Taiwan and South Korea (Varas et al., 2021).

The Netherlands also has quite a central electronics sector (rank of 600) given the size of its economy. That undoubtably has much to do with ASML, a maker of machines necessary to produce chips, and the cluster of firms around it.

From figure 2 we can conclude that important economies in the semiconductor industry, like the US, China, Taiwan, Japan and South Korea, appear to also have relatively central electronics sectors. This becomes more visible when we compare the maps in figures 3 and 4. And not only the two largest economies in the world, the US and China, but also other countries that are highly involved in semiconductor-related trade, have relatively central electronical sectors.

We know that a shock to a central sector is more likely to affect the rest of the world when compared to shocks occuring in other industries. In our previous study we pointed out that the production chain of the semiconductor industry is vulnerable due to its high concentration and specialisation. In this publication we explain that, due to the strong connections between the electronics sector and other sectors, a shock in the supply chain is likely to impact other parts of the economy.

The recently imposed export restrictions on materials needed for the production of semiconductors by the Netherlands and China will probably not only impact the semiconductor industry itself – it is likely that through network effects, these shocks will spill over to other parts of the economy.

Foerster, A., Choi, J., (2017). The changing input-output network structure of the U.S economy. Economic Review (QII), 23,49.

Varas, A., Varadarajan, R., Goodrich, J., & Yinug, F. (2021). Strengthening the Global Semiconductor Supply Chain in an Uncertain Era. BCG and SIA

Euler, Leonhard (1736). "Solutio problematis ad geometriam situs pertinentis". Comment. Acad. Sci. U. Petrop 8, 128–40.