CAIRN-INT.INFO : International Edition
The opinions expressed in this paper are those of the author(s) and do not necessarily reflect the position of AFD. It is therefore published under the sole responsibility of its author(s).

1 – Introduction

1Achieving the climate-related objectives of the Paris Agreement (UNFCCC, 2016) will require a substantial decline in the global production and consumption of fossil fuels (SEI et al., 2020). Considering the obstacles in the implementation of demand-side climate policies (e.g carbon taxes), several authors have argued in favour of introducing complementary supply-side policies aimed at limiting the extraction of fossil fuels (Harstad, 2012; Asheim et al., 2019; Erickson et al., 2018).

2Modern economic systems are still heavily reliant on fossil fuels (IEA, 2020b). Many productive processes use raw or refined fossil fuels directly as a material input or to produce heat. While competitive low-carbon alternatives exist in some activities such as electricity generation (Lazard, 2020), high-carbon incumbent technologies still represent the most convenient option in a large range of economic sectors, e.g. transport, chemicals, steel (IEA, 2020a). The problem is exacerbated by the existence of a substantial amount of long-lived capital stock (e.g. coal/gas electricity plants and blast furnaces). Considering the greenhouse gas emissions that would result from the full utilisation of these physical assets until their natural end of life, respecting a temperature ceiling of 1.5-2°C would involve partly stranding them, i.e using them at a lower capacity utilisation rate, prematurely decommissioning them, or paying for a costly technological conversion (Tong et al., 2019; Cui et al., 2019; Johnson et al., 2015; Pfeiffer et al., 2018; IEA, 2020a).

3It is reasonable to expect supply-side transition-related disruptions not be limited to sectors directly employing fossil fuels in their productive processes. These activities provide in turn indispensable intermediate inputs to more downstream sectors producing consumption goods and services. Through production network linkages, the defossilisation process could cause substantial disruptions to the entire economic system. The relevance of sectoral disruptions in triggering macroeconomic fluctuations via production networks is being thoroughly investigated in the economics literature (see Gabaix, 2011; Acemoglu et al., 2012; Carvalho and Tahbaz-Salehi, 2019; Joya and Rougier, 2019, among others). However, production network analysis has been so far largely excluded from the expanding literature trying to assess the risk of asset stranding and the wider macro-financial implications of moving to a carbon-free economy (Caldecott, 2018; van der Ploeg and Rezai, 2020; Semieniuk et al., 2021). Most of the contributions on the topic have focused on the stranding of fossil reserves (McGlade and Ekins, 2015; Mercure et al., 2018), on knock-on financial effects (Battiston et al., 2017) or on the role of policies and institutions in mitigating climate-related risks (Campiglio et al., 2018). Sectoral dependencies have been incorporated in a limited number of works (e.g. Vermeulen et al., 2018; Allen et al., 2020) but without considering physical capital stocks. As a result, we currently do not have methods to study how a decrease in fossil fuel inputs would strand capital stocks in the rest of the productive system.

4We contribute to filling this research gap by providing a systemic perspective on the supplyside risk of physical capital stranding. We do so by developing a simple methodology rooted in input-output analysis and applying it to a multi-regional production network database. This allows us to compute cross-sectoral and cross-boundary ‘marginal stranding multipliers’. These multipliers provide a monetary estimate of the exposure of sectoral capital stocks to the risk of becoming unutilised due to a marginal loss of primary inputs employed in the fossil sector of a country, including both direct and indirect effects. Following Blöchl et al. (2011), Cahen-Fourot et al. (2020) and others, we treat input-output linkages (or stranding linkages, in our case) as the edges of a directed weighted network. Through a disaggregation of the multipliers into distinct rounds of effects, we construct cascading networks to study how fossil stranding propagates within the international production system.

5Our results offer several interesting insights [1]. First, we compare the stranding multipliers of fossil industries with the ones of other productive sectors. We find that, while some other sectors (e.g. real estate and waste) exhibit higher total marginal multipliers due to their high sectoral capital intensity, the fossil sector is the one with the highest stranding potential on other sectors. Second, assuming global productive sectors, we study how fossil stranding propagates within the production network and rank productive activities according to their exposure to it. Among the most exposed sectors we find some that are not significantly affected by direct stranding links from the mining sector, but are instead exposed to indirect cascading effects. These results support our intuition regarding potential systemic effects driven by defossilisation involving also more downstream activities. Third, we analyse national marginal stranding multipliers and disaggregate them according to their destination. Given the nature of the shock we assume, the international ranking of stranding multipliers does not depend on the absolute relevance of the country as a fossil producer or exporter. Rather, results are driven by i) for what sectors extracted fossil fuels are used (e.g. domestic use vs export); ii) their capital intensity. We find France, Australia and Slovakia to have the highest external marginal stranding multipliers; and USA, Italy and China to have the lowest. Finally, we flip the perspective and study the extent to which countries are exposed to (rather than creating) capital stranding risks. We perform a more detailed analysis of the exposure for a selection of countries (USA, China and Germany), isolating the origins and network transmission channels affecting the most exposed country sectors. We find the US to be exposed to a limited number of very strong stranding links towards Canada and Mexico, with a particularly relevant stranding channel affecting the US public administration via the US coke and refinery industry. Chinese exposure is slightly lower in absolute terms and much more diversified, although Australia and Taiwan stand out as the most relevant origins of risk. Finally, Germany is exposed to several other European countries, and their refinery industries in particular.

6The remainder of the paper is structured as follows. Section 2 describes our methodology and the source of the data we use. Section 3 presents a first set of results where we consider only global sectors. Section 4 adopts a more granular approach to focus on cross-boundary stranding. Section 5 performs a more detailed exposure analysis for USA, China and Germany, who are among the most exposed countries to supply-side fossil stranding risk. Section 6 discusses limitations and future avenues of research. Finally, Section 7 concludes.

2 – Methodology and data

7This section presents our methodological approach. First, we explain the method to compute our matrix of ‘marginal stranding multipliers’. Second, we discuss how we distinguish among rounds of effects and create ‘stranding cascades’. Finally, we present our data sources.

2.1 – The matrix S of sectoral stranding multipliers

8Figure 1 offers a stylised representation of a multi-regional Input-Output (MRIO) table (Miller and Blair, 2009). Each sector of each country appears twice in the table. First, it appears as a producer of goods and services (on the rows). Goods and services can be then purchased by other firms to be used as intermediate inputs in further production processes, or be consumed by households, firms or governments as final demand items. Second, it appears as a user of inputs (on the columns). These inputs can take the form of intermediate inputs purchased from other firms or of value added items such as compensation of employees, consumption of fixed capital and gross operating surplus. The square matrix recording all the transactions of goods and services among industrial sectors is the inter-industry matrix Z (in grey in Figure 1). The set of column vectors f represents final demand, while the set of row vectors v represents value added items.

Figure 1

A stylised multi-regional input-output table

Figure 1
Country 1 Country 2 Inter-industry matrix (Z) Sector A.1 Sector B.1 Sector A.2 Sector B.2 Final demand (f) Total use (x) Country 1 Sector A.1 A.1 products used by A.1 A.1 products used by B.1 A.1 products used by A.2 A.1 products used by B.2 Consumption of A.1 products Sector B.1 B.1 products used by A.1 B.1 products used by B.1 B.1 products used by A.2 B.1 products used by B.2 Consumption of B.1 products Intermediate production + Final demand Country 2 Sector A.2 A.2 products used by A.1 A.2 products used by B.1 A.2 products used by A.2 A.2 products used by B.2 Consumption of A.2 products Sector B.2 B.2 products used by A.1 B.2 products used by B.1 B.2 products used by A.2 B.2 products used by B.2 Consumption of B.2 products Value added (v) Value added in A.1 Value added in B.1 Value added in A.2 Value added in B.2 Total supply (xT) Intermediate consumption + Value added

A stylised multi-regional input-output table

9For the input-output table to be balanced, total supply of each industry xT = iT Z + v must be equal to its total use x = Zi + f, where i is a column vector of 1’s of the same dimension of Z[2] and T denotes the matrix transposition. In other words, the sum of all flows over a row (total industry output broken down by type of use, i.e. intermediate use and final consumption) must equal the sum over the corresponding column (total industry input broken down by ‘source’, i.e. other industries and value added items).

10Sectoral dependencies in IO tables are often studied via the Leontief model (Leontief, 1951). However, the Leontief demand-side approach does not fit well with the supply-side nature of our research question (i.e. what is the stranding effect of a reduction of fossil inputs). We hence choose to employ the supply-side model first proposed by Ghosh (1958) instead. While the Leontief model calculates the matrix A of technical input coefficients, the Ghosh model defines a matrix figure im2 of output allocation coefficients, whose elements represent the allocation of the output of a sector to all other sectors. In other words, each element bij quantifies the share of industry i’s output that is used by industry j. The Ghosh matrix G is then defined as G = (I B)1.

11For convenience, we transpose G to be able to read the effects of changes in sectoral primary inputs over the columns of GT (similarly to the Leontief matrix L). Each element gi,j of GT describes the change in output x in sector i that would result from a marginal change of primary inputs flowing into sector j. In other words, an increase (decrease) of one monetary unit of primary inputs contributing to production in sector j will increase (decrease) the output of sector i by an amount equal to gi,j, where gi,j includes both direct and indirect effects. ‘Primary inputs’ refers to any item appearing on the rows below the inter-industry matrix. Traditionally, this has been meant to represent compensation of employees (and thus labour input) but, more generally, it can be used to represent any form of societal effort put in producing the output of a specific sector, as represented by factor payments.

12We then combine GT with sectoral data of physical capital stocks k. We define κi = ki/xdi as the capital intensity of sectori, where xd is the domestic output of the sector. By multiplying the diagonalised form of the vector of capital intensities by GT, we find the matrix S of asset stranding multipliers figure im3[3]. Figure 2 offers a stylised representation of the S matrix. Each element sij represents the change in the utilisation of capital in sector i triggered by a marginal change of primary inputs used by sector j. For our purposes, the elements of S define the amount of capital stock of a sector i that could become stranded because of a marginal decrease in the primary inputs used in the production of goods and services of another sector j (e.g. fossil fuels). The column sum of matrix S gives a measure of the total amount of stranded physical assets resulting from a marginal reduction of primary inputs in a sector j. Similarly, we can interpret the sum of the rows of S as the exposure of a sector i to stranding risk (i.e. the loss in capital utilisation resulting from a marginal loss in primary inputs used in all productive sectors).

Figure 2

A stylised representation of the S matrix of stranding multipliers

Figure 2
Country 1 Country 2 Matrix S Sector A.1 Sector B.1 Sector A.2 Sector B.2 Stranding exposure Country 1 Sector A.1 sA1,A1 sA1,B1 sA1,A2 sA1,B2 Total A.1 stranding exposure Sector B.1 sB1,A1 sB1,B1 sB1,A2 sB1,B2 Total B.1 stranding exposure Country 2 Sector A.2 sA2,A1 sA2,B1 sA2,A2 sA2,B2 Total A.2 stranding exposure Sector B.2 sB2,A1 sB2,B1 sB2,A2 sB2,B2 Total B.2 stranding exposure Stranding multiplier Total stranding from A.1 Total stranding from B.1 Total stranding from A.2 Total stranding from B.2

A stylised representation of the S matrix of stranding multipliers

13The Ghosh approach is unsuited to analyse the causal effects of large-scale supply bottlenecks (e.g. in the aftermath of natural disasters). Limitations include the assumptions of perfect elasticity of demand in reacting to changes in supply and of perfect substitutability among input factors (Oosterhaven, 1988; Dietzenbacher, 1997; Galbusera and Giannopoulos, 2018). However, it can be usefully employed to describe and compare economic structures and the relative economic/environmental importance of sectors (see for instance Zhang, 2010; Antràs et al., 2012; Aldasoro and Angeloni, 2015; Piñero et al., 2019; Zhang et al., 2020; Cahen-Fourot et al., 2020). Our results should thus be interpreted uniquely as stranding effects at the margin, i.e. as the exposure of countries and sectors to the risk of physical capital stranding following a marginal shock in the fossil sector. They do not offer causal predictions on what the dynamic stranding effects of a low-carbon transition will be. Rather, they provide valuable insights on the productive structure of nations at a point in time and, more specifically, on the relevance of fossil sectors in keeping downstream capital stocks in operation. Our approach is, therefore, a diagnostic methodology and does not aim at being predictive.

2.2 – Cascade networks

14The stranding multipliers in S contain all direct and indirect stranding effects resulting from a change in primary inputs of a productive sector. To better understand these results, it is useful to investigate how stranding propagates through the economic system, distinguishing the direct effects of the initial impulse from the following indirect inter-industry responses. For this purpose, we make use of the fact that the G matrix can be approximated by a power series: G = (I B)1 = limn→∞(I + B + B2 +…+ Bn) (Miller and Blair, 2009). Each element of the series can be interpreted as one round of inter-industry production responses resulting from an exogenous supply change. These rounds should not be interpreted as taking place at real time intervals. Rather, they indicate the successive steps through which initial impulses propagate within the productive system via sectoral interdependencies. The analysis of the power series is common in the input-output literature and is otherwise known as production layer decomposition (Lenzen and Crawford, 2009). This sequential perspective has been for instance applied to the allocation of environmental pressure responsibility in global value chains (Piñero et al., 2019). We instead perform a stranding layer decomposition analysis, where the power series steps should be understood as stranding allocation steps. We focus on the first few rounds of effects, which are the most likely to actually take place in the short term, before longer-term dynamic adjustments trigger structural changes our static framework is not able to capture. This allows us to partially offset the limitation of the Ghosh model described in the previous section. By constraining the indirect effects to a few rounds, we assume partial adjustment to the original shock, and hence do not consider supply bottleneck effects at their full scale.

15We hence rewrite figure im5 to disentangle the direct and subsequent indirect stranding effects caused by a fossil supply shock. This decomposition allows us to analyze the distribution of stranding effects over the individual rounds and to identify which sectors are more directly exposed to fossil stranding disruptions and which are affected indirectly via production network adjustments (i.e. in later stages of the power series).

16Bearing in mind that stranding multipliers in S are the sum of individual stranding channels cascading gradually through the economic system, the power series perspective allows us to increase the resolution of our analysis even further. In particular, we can identify the most important stranding channels by sequentially isolating the strongest linkages from round to round. This can be used to construct graphs that can be interpreted as weighted directed networks. We employ this approach to generate two kinds of networks. First, we study how an initial shock in the fossil sector propagates through the economic system by isolating the dominant stranding cascades it creates. We refer to these graphs as ‘cascade networks’. Second, we look at the opposite direction and investigate the exposure of particular sectors to fossil stranding by identifying the most important direct and indirect stranding channels affecting them. The resulting graphs are called ‘exposure networks’.

17For the cascade networks, we start by placing the fossil sector at the origin of the network, assuming a marginal unitary decrease in its primary inputs as the initial stranding shock. Then we identify the sectors most directly affected by this shock (i.e. in the first round of the stranding power series), given by the highest values of the originating fossil sector’s column in the figure im6 matrix. Only the top n sectors are retained and placed in the first layer of the network, with the edge weights corresponding to the value of the direct stranding link. The next layer is obtained by repeating this procedure for each sector of the first layer, taking into account that the input loss in these sectors will be lower than one and a function of their relation to the originating sector, as given by the B matrix of output allocation coefficients [4]. This is done by simply re-weighting the respective sector’s direct stranding links (i.e its column in the figure im7 matrix) by its loss of intermediate inputs from the fossil sector according to the corresponding fixed allocation coefficient in the BT matrix [5].

18If a sector in the resulting second layer appears in the top n stranding links of more than one sector of the previous layer, it consequently has multiple incoming edges and its input loss is the sum of input losses resulting from all incoming stranding channels [6]. We also add the value of total stranding taking place in each sector in a specific round (corresponding to the sum of incoming edges for the case in which n is set equal to the total number of sectors) to the node labels. This makes it possible to compare the size of the displayed (dominant) stranding links to the sum of all possible links.

19This procedure can be repeated for an arbitrary number of layers, each one corresponding to a round of the power series. However, the size of effects diminishes with each round. The values of the power series matrices typically become insignificant after seven or eight rounds, and - as will be shown later - most of the effect is captured by the first few rounds (Miller and Blair, 2009). In our representation of the networks we will focus on the first few rounds and set n sufficiently low to isolate the most important stranding channels and ensure readability of the results. We will also exclude self-loops (i.e. direct stranding from a sector to itself) in order to better investigate inter-industry propagation of effects. Our second type of network, the exposure network, aims at capturing the main sources and channels that cause certain sectors to be particularly exposed to fossil stranding. Here we again make use of the power series, but construct the network from a different starting point and employ a different selection approach. We first place the sectors of a country that are most exposed to fossil stranding (according to the S matrix) at the bottom of the network. For each of those sectors, we then identify the m international fossil sectors they are most exposed to directly (i.e. in the first round of stranding), place them on top of the network and connect them with edge weights corresponding to the direct stranding effect. In the next step we identify the m most important two-step fossil exposure channels (i.e. indirect incoming stranding linkages originating in a fossil sector with one intermediate step), again add the originating fossil sectors to the top layer and the intermediate sectors to an intermediate layer. Similar to the cascade networks, this procedure could theoretically be repeated for stranding channels of any length. However, we in turn limit our analysis to the first few steps, as the most important transmission channels can be expected within the first few rounds of stranding. This approach allows us to capture the most dominant network origins and transmission channels that create transition risks in vulnerable sectors by means of relatively simple and clear-cut network graphs.

2.3 – Data

20Our main source of data is the World Input-Output Database (WIOD) (Timmer et al., 2015) [7], which has been used in the past for a variety of research purposes (see Voigt et al., 2014; Marin and Vona, 2019; Klimek et al., 2019; Chen et al., 2019 among others). WIOD is a multiregional input-output database comprising 43 countries plus a Rest of the World (ROW) region [8]. Most of the countries in the sample have high per capita income, but the dataset also includes several relevant emerging economies (Table 1 lists the whole sample). The sample of non-ROW countries in WIOD allows us to have disaggregated coverage for around 48.3% of oil production; 61.1% of gas production; 89.8% of coal production; and 69.0% of extraction-based CO2 emissions [9].

Table 1

WIOD regions

Income groupCountry
High-incomeAustralia (AUS); Austria (AUT); Belgium (BEL), Canada (CAN), Switzerland (CHE), Cyprus (CYP), Czechia (CZE), Germany (DEU), Denmark (DNK), Spain (ESP), Estonia (EST), Finland (FIN), France (FRA), Great Britain (GBR), Greece (GRC), Croatia (HRV), Hungary (HUN), Ireland (IRL), Italy (ITA), Japan (JPN), South Korea (KOR), Lithuania (LTU), Luxembourg (LUX), Latvia (LVA), Malta (MLT), Netherlands (NLD), Norway (NOR), Poland (POL), Portugal (PRT), Slovakia (SVK), Slovenia (SVN), Sweden (SWE), Taiwan (TWN), United States of America (USA)
Upper-middleBulgaria (BGR), Brazil (BRA), China (CHN), Mexico (MEX), Romania (ROU), Russia (RUS), Turkey (TUR)
Lower-middleIndonesia (IDN), India (IND)

WIOD regions

21The 56 productive sectors present in WIOD are classified using NACE level 2 categories (Eurostat, 2008). Table 2 in the Appendix lists all sectors included in the analysis. We create new sector codes to make our results easier to understand. The first three upper-case letters of each sector code reflect the NACE level 1 category (e.g. MAN for manufacturing), while the following three lower-case letters reflect the NACE level 2 category (e.g. MANche for manufactured chemical products). When discussing a NACE level 1 sector, or in the case of NACE level 1 sectors for which no further disaggregation is available, we use a + sign at the end of the code, to signify that several sub-activities are included there (e.g. MAN+ is the equivalent of the entire NACE C level 1 sector). The most important sector for our analysis is denominated MINfos, as it records the activities of mining and extraction of fossil fuels, themselves a part of the larger mining sector.

22WIOD offers values for the mining sector as a whole (NACE sector B). This is where fossil fuels are extracted, hence at the core of our analysis. However, other materials are also included in the B sector, such as metal ores, stone, sand, clay and numerous other minerals. Hence, using the whole B sector as the core of our analysis will not accurately represent a supply shock in fossil fuels, and thus bias the results. A more detailed disaggregation of the mining sector can be found in the OECD Inter-Country Input-Output (ICIO) database (OECD, 2018). ICIO covers 66 countries (plus a Rest of the World region) and 36 sectors [10]. Even though its overall sectoral resolution is coarser compared to WIOD, the mining sector is disaggregated into three sub-sectors, namely “Mining and extraction of energy producing products” (NACE sectors B05 & B06), “Mining and quarrying of non-energy producing products” (NACE sectors B07 & B08) and “Mining support service activities” (NACE sector B09). Sectors B05 (“Mining of coal and lignite”) and B06 (“Extraction of crude petroleum and natural gas”) contained in the first ICIO mining sub-sector represent the core activities of the fossil extraction industry. Mining support services (specialised support provided to the extraction industry on a fee or contract basis, such as exploration services) are largely part of the fossil complex as well, but they do not directly produce fossil fuels.

23We employ this database to split the mining sector in WIOD into three sub-sectors and therefore isolate the fossil extraction industry as our sector of interest [11]. First, we aggregate regions in the 2014 ICIO table so to match WIOD regional disaggregation. From this table, we compute (element-wise) ratios to split every mining element in WIOD according to the relative size of the three corresponding sub-sector elements in ICIO [12]. Final demand is disaggregated as a column vector and value added (plus taxes less subsidies and transport margins) as a row vector. Second, the resulting WIOD table with a disaggregated mining sector is balanced using a two-stage RAS (TRAS) procedure (Gilchrist and St. Louis, 1999, 2004) [13]. This method allows us to ensure consistency between the new mining sub-sectors and the original aggregate WIOD mining sector, while keeping the original WIOD values for all cells unaffected by the mining disaggregation [14]. The result of implementing the TRAS algorithm is a revised WIOD table with three new mining sub-sectors and all other elements identical to the original table [15]. Finally, we split the capital stock of each country’s mining sector using the ratios obtained from ICIO total output data. This is made necessary by the unavailability of capital stock data at the sub-sectoral level.

24The procedure of splitting and rebalancing an input-output database involves making limiting assumptions and risks altering the underlying data. However, the potential benefits of disaggregating heterogeneous sectors for the accuracy of input-output multipliers – even if based on incomplete information - outweigh the risks, as emphasised by Lenzen (2011). For our specific research purposes, we believe that disaggregating the mining sector so to isolate the stranding effects of the fossil fuel industry adds significant value to our analysis, while providing plausible estimates.

3 – Global fossil stranding

25We first analyse the results aggregating S so to obtain a new matrix SW showing the monetary exchanges among global productive sectors. The fossil stranding multipliers of the SW matrix provide an estimate of the exposure of capital stock to the risk of remaining unutilised due to a marginal shock in the global fossil sector (MINfos). While limited by definition by the lack of regional disaggregation, this analysis is useful to introduce some general implications of fossil stranding. We will relax this limitation in Section 4.

26We start by noticing that, among all global productive sectors, the fossil fuel sector is the productive sector with the strongest external stranding potential. Table 3 in the Appendix ranks sectors according to: i) their total potential stranding; ii) their external potential stranding; iii) their total stranding exposure; and iv) their exposure to stranding from other sectors. In the first column, MINfos ranks sixth with a stranding multiplier of 4.636. This means that a marginal reduction in primary inputs of $1 in the global fossil sector causes $4.636 of capital to be stranded in the whole economic system. However, most of the stranding risk originating in a sector concerns the sector itself. If we abstract from internal stranding, MINfos appears as the sector capable of creating the largest stranding effect on the rest of the economic system, with a multiplier of 2.4. It is followed by the waste sector (WATwst) and financial services (FINser). The two final columns give an interesting estimates of the total and external exposure of a sector to a scenario with a marginal shock taking place in all sectors (e.g. a generalised drop in economic activity). Due to their high capital intensity and their large use of intermediate inputs, the real estate (RES+) and public administration (PUB+) sectors are by far the most exposed to such a scenario.

27Figure 3 focuses on the stranding cascade originating in the global mining sector [16]. The mining sector is at the top of the pyramid by choice. The numerical value inside the mining node represents the stranding strength of the initial marginal shock we assume: $2.099 worth of capital become immediately stranded in the fossil sector due to the $1 shock in its primary inputs. We then identify the sectors most affected by the lack of intermediate fossil inputs [17] and place them in the first layer. The numerical value attached to the network links represents the strength of that specific stranding relation, while the values inside the nodes reflect the value of the total stranding taking place in the sector in a specific round. The most affected sectors in the first round are the power (PWR+), coke and refined petroleum products (MANref) and basic metals (MANmet). This is unsurprising, as the power and refinery industries require fossil fuels as direct inputs in their production, while the metal industry uses fossil fuels to generate heat (e.g. in blast furnaces). The second layer of the network is composed by the sectors most affected by the stranding taking place in the first-layer sectors. The most relevant stranding links here include the ones connecting MANref with the land transport services sector, which includes transport via pipelines (TRAinl), and the public administration sector (PUB+); and the one linking PWR+ back to the mining fossil sector. The stranding in MANmet cascades down to industries using metallic products, but its strength is less pronounced. Finally, the third layer is composed of the sectors most affected by the stranding originating from the second-layer sectors. Several more downstream sectors appear here. The strength of the single stranding links are lower than in upper layers (the strongest being the one connecting PWR+ to RES+) but, due to the multiple active stranding links, the overall stranding in these rounds is still relevant, especially in the PWR+ (0.056) and RES+ (0.055).

Figure 3

Stranding cascade from the global fossil mining sector (n = 3)

Figure 3

Stranding cascade from the global fossil mining sector (n = 3)

28Figure 4 shows more aggregate results where all stranding impacts coming from different sectors within a certain round have been summed up. We distinguish first-, second- and third-round effects, and aggregate all remaining rounds in the ‘Further rounds’ category. The overall length of the bar corresponds to the fossil stranding multiplier present in matrix SW. We report the results for the top 10 sectors by their overall fossil stranding multiplier. We exclude the initial shock impact in the fossil industry, equal to 2.099 and much larger than the stranding impacts on other sectors. For PWR+ and MANref, the two most affected sectors, first-round effects are the most relevant. For the following sectors (RES+, PUB+, TRAinl and MANche), the opposite seems to be the case. The second-, third- and further-round effects are much larger than first-round effects, and strong enough to move their fossil stranding multiplier above the one of MANmet, which has the third largest direct stranding effect as seen in Figure 3. These results support our initial hypothesis that stranding in downstream sector might be relevant, and possibly as relevant as the stranding taking place in upstream sectors.

Figure 4

Top 10 global sectors by their exposure to fossil stranding

Figure 4

Top 10 global sectors by their exposure to fossil stranding

4 – Cross-country fossil stranding

29We now relax our previous assumption of globally integrated sectors to explore the more granular results offered by the full S matrix. Before doing so, it is worth reminding that the shock we investigate — a marginal reduction of primary inputs employed in the domestic production of fossil fuels — is applied equally to all countries, irrespective of their absolute amounts of fossil fuel extraction. Hence, the resulting stranding results do not depend on the relevance of the country as a producer or exporter of fossil fuel, but rather on: i) how concentrated are monetary outflows from the fossil industry towards specific sectors; and ii) how capital intensive are the sectors receiving fossil products. A marginal shock in a country where the entire amount of fossil products flows to a single sector with high capital intensity will cause large capital stranding effects, even if fossil fuel extraction levels are very low.

30We start by discussing countries’ total fossil stranding potential. The left column of Table 4 in the Appendix ranks countries by their overall global stranding potential. The numerical values listed indicate the monetary value of the physical capital remaining idle due to the fossil shock. The results are strongly shaped by the capital intensity of domestic fossil sectors, where the initial marginal shock takes place. Indeed, the countries in the top 3 of the ranking (Slovakia, Brazil and Australia) are also in the top 3 of countries ranked by the capital intensity of their fossil sectors. The explanation for having Luxembourg in fourth place is different.

31The Luxembourg fossil sector is not particularly capital intensive, but almost 96% of its (very small) MINfos production goes to its capital-intensive electricity and gas sector. A similar explanation applies to South Korea, in fifth position.

32To abstract from domestic stranding, we look at the ranking of countries according to the stranding they create externally to other countries. Figure 5 shows the top 10 countries by their external stranding multiplier, with a disaggregation of the most affected countries [18]. As with total stranding rankings, low levels of fossil extraction do not contribute in shaping the results. France, a marginal producer of fossil fuels, is at the top the ranking, mainly due to the very high proportion of production being exported (96%) and the high capital intensity of its major importing sectors (especially Slovakian PWR+ and MANref sectors). Australia is in second place, mainly due to the strong stranding effect created on the Japanese PWR+ sector and to a lesser extent the PWR+ sectors of China, South Korea and Taiwan. Slovakia ranks third, predominantly due to its high stranding effects on the Czech economy, which also creates relevant stranding links back to Slovakia (in seventh place). These results suggest the permanence of a strong integration between the two productive systems after the separation of the two countries in 1993.

Figure 5

Top 10 countries for external marginal stranding multipliers

Figure 5

Top 10 countries for external marginal stranding multipliers

33It should also be noted how USA and China - the two largest fossil fuel producers in our sample - are at the bottom of the external stranding ranking. While this might seem counter-intuitive, this is clearly explained by two facts. First, the large majority of their fossil fuel production (92.5% for USA and 99.5% for China) is consumed internally; therefore most of the stranding effects are felt internally. Indeed, both countries rank much higher in the total stranding ranking that includes domestic stranding (15th position for US; 22nd for China). Second, they export fossil fuels to a large number of countries (i.e. their fossil outflows have a lower concentration) whose sectors have relatively lower capital intensity.

34Finally, Figure 6 ranks the top 10 countries according to their exposure to fossil stranding coming from abroad [19]. The total exposure values listed represent the monetary value becoming stranded in the country in the scenario of a generalised drop of external fossil fuel extraction (i.e. a marginal shock is assumed to take place in all the countries of the sample, except the country for which we analyse exposure). We are also able to disaggregate the exposure to stranding originating in a specific country. The USA is by far the most exposed country (with $6.76 of capital stranding in the case of a generalised drop of external fossil production), with Canada (1.75) and Mexico (0.98) being the most relevant origins of stranding risks (followed by Lithuania, ROW and Finland, with much lower values). China is the second most exposed country (with a coefficient of 4.69). In the Chinese case, potential origins of stranding risk are more diversified than for US. Australian (0.61) and Taiwanese (0.45) fossil sectors are the ones with the highest stranding effect on China, closely followed by ROW (0.40), South Korea (0.34), Russia (0.30), Brazil (0.29) and others. Japan is in third place of the ranking with a total exposure coefficient of 3.99, originating predominantly in Australia (1.08) and Spain (0.66).

Figure 6

Top 10 countries for exposure to fossil stranding risk

Figure 6

Top 10 countries for exposure to fossil stranding risk

5 – Fossil stranding exposure

35We now move to analysing more in depth how countries are exposed to stranding links, looking at the network origins and transmission channels. We do so by constructing exposure networks representing the most relevant one-step (i.e. direct), two-step and three-step (i.e. indirect with respectively one and two intermediate steps) stranding links affecting the most overall exposed sectors of the selected country. We focus on three countries: USA, China and Germany. These are among the countries most exposed to supply-side external stranding risk [20], as shown in Table 4 and Figure 6. USA and China are also the largest economies in the world and the largest producers and consumers of fossil fuels in our country sample. We exclude internal stranding exposure (to the domestic fossil sector) from the analysis in order to focus on cross-boundary (external) exposure linkages.

36Figure 7 shows the exposure network for USA. The most exposed US sectors, according to the marginal stranding multipliers found in matrix S, are public administration, electricity and gas, and real estate. We place them at the bottom of the network and add their total external exposure values to the node labels. We then look for their strongest incoming one-step, two-steps and three-steps stranding links to understand where their exposure originates and how it reaches them through the production network. The choice of the number of steps is arbitrary, but most of the dominant stranding cascades can be expected to take place within the first few steps. By setting m = 2 we select the two most important stranding channels for each of the three cascade lengths.

Figure 7

Main exposure one-step (blue), two-step (red) and three-step (yellow) links for USA (m = 2)

Figure 7

Main exposure one-step (blue), two-step (red) and three-step (yellow) links for USA (m = 2)

37As already observed in Figure 6, US sectors appear heavily exposed to the Canadian and Mexican MINfos sectors. Significant direct stranding links exist between them and all of the US sectors at the bottom of the network, with the one linking CAN MINfos and USA PWR+ being the strongest. Indeed, shocks originated in Canada, the main fossil exporter in our sample, affect the US economy more than the Canadian economy itself. Several important two-step stranding links also exist, with the most relevant connecting CAN MINfos and MEX MINfos to the US refining and coke industry, and then from there to USA PUB+ and, to a lesser extent, USA PWR+. Three-step stranding chains follow similar channels, with USA MANref affecting the land transport, constructions and power industries, which then in turn affect the sectors at the bottom. In addition, we spot a three-step stranding cascade originating in the Taiwan fossil industry, passing through the Taiwanese refinery and air transport industry, and finally affecting the USA PUB+ sector. This cascade draws its strength primarily from the substantial linkage between TWN MINfos and TWN MANref, arising from the fact that almost all of the Taiwanese fossil extraction is used by the domestic refinery industry [21]. This strong impulse then trickles through to USA PUB+ via the TWN TRAair industry, which serves as a major trans-pacific cargo service provider and is used by the USA PUB+ sector to import manufactured goods. However, due to the very limited production volume of the Taiwanese fossil industry, this cascade is of little practical relevance. Rather, it again shows that sectors with a concentrated use structure typically have high marginal stranding potential.

38Figure 8 shows the results for China. The Chinese sectors with the highest overall exposure multipliers are electricity and gas (PWR+), the art, entertainment and recreation sector (ART+), and the chemical industry (MANche). We find that the most relevant originating fossil sectors are the Australian, Brazilian and ROW ones for what concerns direct stranding links, with the one going from AUS MINfos to CHN PWR+ being the strongest. Two-step cascades also originate from Australia and Brazil, passing through the Chinese PWR+ and MANref sectors, with an additional cascade originating in the Korean MINfos sector and affecting the Chinese chemical sectors via Korean MANref. Three-step cascades identify three alternative transmission channels: i) a cascade mainly internal to Australia (AUS MINfos to AUS MANref and back to the mining industry) and affecting CHN PWR+; ii) a variation on the Korean cascade passing through KOR MANche; and iii) a cascade originating in Taiwan, passing through TWN MANref and TWN MANche, and landing on both CHN ART+ and CHN MANche.

Figure 8

Main exposure links for China (m = 2)

Figure 8

Main exposure links for China (m = 2)

39We notice how the stranding links of the networks explain only a minor proportion of the overall external stranding in the bottom sectors (value reported inside the node). This is due to the exposure of China to a multiplicity of countries, although with lower singular effects, as already shown by Figure 6. This is in contrast to the US situation, which is instead mainly exposed to two very strong stranding links.

40Australia is also a major fossil exporter, and towards the top of countries ranked according to their external stranding potential (see Table 4). It is worth mentioning that, while China is indeed one of the most affected countries by a marginal shock originating in the Australian mining sector, the Japanese economy is even more exposed to it than China.

41Finally, Figure 9 represents the exposure of the German economy. The high number of European fossil fuel sectors from which stranding originates illustrates well the integration of Germany into European energy value chains. It is also indicative of the strong integration of European economies. The most affected sectors in terms of overall stranding multipliers are the real estate sector (RES+), electricity and gas (PWR+), and public administration sector (PUB+). The most relevant 1-step stranding links originate in the Dutch fossil sector and affect the PWR+ and PUB+ sectors in Germany. Significant direct links originate also in Norway, Slovakia and Italy. Relevant two-step stranding cascades originate in the Netherlands, Belgium and Switzerland, through the German construction and power sectors, and through the Belgian and Swiss coke and refinery industries. Three-step stranding cascades originate from Lithuania, Belgium and Switzerland; pass through their respective coke and refinery sectors in the first stranding layer; touch the German CNS+ and TRAair sectors, and the Dutch MANche sector; before landing on all three bottom sectors.

Figure 9

Main exposure links for Germany (m = 2)

Figure 9

Main exposure links for Germany (m = 2)

6 – Limitations and future avenues of research

42Our analysis is not exempt of limitations. First, it is limited by data availability and granularity. We are not able to disaggregate among specific types of capital stocks (e.g. machinery vs dwellings), despite they are probably exposed to stranding risks to different extents. We have a good coverage of global fossil extraction but we miss disaggregated data for several important fossil producers/exporters (e.g. Saudi Arabia, Venezuela, South Africa). Sectoral capital stock might not be entirely precise due to the need of balancing the MRIO table (which is done at two stages; once by WIOD researchers, and once by us after splitting the mining sector). The use of industry-by-industry tables, rather than product-by-product, can also give rise to imprecision in our calculations. Second, we rely on a number of limiting assumptions, mostly inherited by the use of input-output methods and the Ghosh supply-driven framework. As a consequence, our results are valid only for marginal shocks and should not be mistaken as an estimate of actual stranding impacts in a low-carbon transition scenario, as the latter would involve price adjustments, substitution of inputs, technological progress and change in preferences. Rather, our analysis offers insights on the interdependent nature of international economic sectors and the importance of fossil fuel extraction for downstream sectors.

43We also hope to contribute to opening further research avenues building on the methodology and results presented here. Some of these avenues could be pursued with relatively limited modifications. For instance, replacing capital stock data with sectoral employment values could offer insights on the transition implications on labour, as in Bastidas and Mc Isaac (2019) and Perrier and Quirion (2018). Instead of adopting a supply-side approach focused on fossil sectors, it would be possible to use the Leontief model to study the stranding implications of sectoral demand constraints depending on their carbon intensity. Other research avenues would instead require more work. We highlight two. First, dynamic effects could be included by inserting the MRIO analysis into a macroeconomic modelling framework (e.g. using computable general equilibrium or stock-flow consistent models). Second, the analysis of production networks could be linked to the ongoing study of financial networks to create a multi-layer network analysis capable of offering a more complete perspective on how the macro-financial system would react to a low-carbon transition.

7 – Conclusion

44The systemic risks of transitioning to a low-carbon society under the current technological conditions are complex and still largely unknown. We contribute to filling this research gap by developing a simple methodology to calculate the monetary value of capital stocks becoming stranded as a consequence of a marginal loss of primary inputs employed in the fossil sector, taking into consideration the network of economic inter-dependencies. We apply the methodology to a revised version of the WIOD multi-regional database to compute i) the marginal stranding multipliers of countries’ fossil sectors; ii) the exposure of national economic systems to the risk of capital stranding coming from abroad.

45We obtain several interesting results. First, among all productive sectors, fossil industries exhibit the strongest potential to create capital stranding in other sectors. Second, we show that, while some sectors (e.g. energy and manufacturing) are directly exposed to the fossil shock, several other productive activities are mainly affected by indirect effects. Indeed, when taking into account all rounds of stranding, service sectors like real estate and public administration rank amongst the most affected ones by a global fossil stranding. This supports the intuition that the whole global productive system would be affected by the decarbonisation process, and not only heavy industry sectors using fossil fuels as direct inputs.

46Third, we rank countries according to their external stranding potential, finding France, Australia and Slovakia at the top of the ranking, and USA, Italy at its bottom. These results can seem surprising but they are explained by the proportion of the exported fossil production of the top countries and the capital intensity of the sectors importing this production. To the contrary, major producers whose production is largely consumed onshore, such as the USA and China, are at the bottom of the external stranding ranking. Our results therefore show the counter-intuitive fact that a country’s stranding power does not correlate automatically with its importance in global fossil fuels production. We also show how the capital stocks of USA, China and Japan are the most exposed to the risk of stranding due to a generalised drop of fossil production. Finally, we zoom in a selection of countries (USA, China and Germany) to provide a more granular understanding of how their productive systems are exposed to direct or indirect stranding risk. In all three countries the main stranding cascades end in secondary and tertiary sectors. It might be surprising that sectors like public administration are particularly affected. However, this can be explained by several factors, such as public administration buildings requiring heating and military activities employing fossil fuels (military activities rank very high in terms of greenhouse gas emissions (Crawford, 2019)).

47Despite the limits mentioned above, our analysis still offers meaningful results with important implications for policy-makers. There has been a strong expansion of research contributions trying to assess the macro-financial implications of a low-carbon transition (Allen et al., 2020; Vermeulen et al., 2018). This is of particular interest to central banks, financial supervisors and other institutions interested in mitigating climate-related financial risks. Including a systemic view on capital stranding through the representation of production networks, as we do for the first time in this article, might contribute to the definition of more sophisticated and comprehensive risk assessment methods.

Acknowledgements

The research leading to these results has received funding from the Swedish Foundation for Strategic Environmental Research (Mistra) and the Oesterreichische Nationalbank (OeNB Jubiläumsfonds project number: 17641). E.C. gratefully acknowledges funding from the European Research Council (ERC) under the European Union’s Horizon 2020 Research and Innovation Programme (Grant agreement No. 853050). The authors are grateful to Hanspeter Wieland and Martin Bruckner for useful comments. The usual disclaimer applies.

Appendix

A – Sector codes and descriptions

Table 2

NACE level 2 sectors [22]

NACECodeSector description
AAGR+Agriculture, forestry and fishing
 A01AGRagrCrop and animal production, hunting and related service activities
 A02AGRforForestry and logging
 A03AGRfisFishing and aquaculture
BMIN+Mining and quarrying
 B05-06MINfosMining and extraction of energy producing products
 B07-08MINothMining and quarrying of non-energy producing products
 B09MINsupMining support service activities
CMAN+Manufacturing
 C10-12MANfooFood, beverages and tobacco products
 C13-15MANtexTextiles, wearing apparel, leather and related products
 C16MANwooWood and products of wood and cork, except furniture
 C17MANpapPaper and paper products
 C18MANpriPrinting and reproduction of recorded media
 C19MANrefCoke and refined petroleum products
 C20MANcheChemicals and chemical products
 C21MANphaBasic pharmaceutical products and pharmaceutical preparations
 C22MANplaRubber and plastic products
 C23MANminOther non-metallic mineral products
 C24MANmetBasic metals
 C25MANfmpFabricated metal products, except machinery and equipment
 C26MANcomComputer, electronic and optical products
 C27MANeleElectrical equipment
 C28MANmacMachinery and equipment n.e.c.
 C29MANmotMotor vehicles, trailers and semi-trailers
 C30MANtraOther transport equipment
 C31_32MANfurFurniture and other manufactured goods
 C33MANrepRepair and installation services of machinery and equipment
DPWR+Electricity, gas, steam and air conditioning
EWAT+Water supply; sewerage; waste management and remediation
 E36WATwatNatural water; water treatment and supply services
 E37-39WATwstSewerage services; sewage sludge; waste collection, treatment and disposal services
FCNS+Constructions and construction works
GTRD+Wholesale and retail trade; repair of motor vehicles and motorcycles
 G45TRDmotWholesale and retail trade and repair services of motor vehicles and motorcycles
 G46TRDwhoWholesale trade, except of motor vehicles and motorcycles
 G47TRDretRetail trade services, except of motor vehicles and motorcycles
HTRA+Transportation and storage
 H49TRAinlLand transport and transport via pipelines
 H50TRAwatWater transport
 H51TRAairAir transport
 H52TRAwarWarehousing and support activities for transportation
 H53TRAposPostal and courier activities
IFD+Accommodation and food service activities
JCOM+Information and communication
 J58COMpubPublishing activities
 J59_60COMvidMotion picture, video and television production, sound recording, broadcasting
 J61COMtelTelecommunications
 J62_63COMcomComputer programming, consultancy; Information service activities
KFIN+Financial and insurance activities
 K64FINserFinancial services, except insurance and pension funding
 K65FINinsInsurance, reinsurance and pension funding services, except compulsory social security
 K66FINauxActivities auxiliary to financial services and insurance services
LRES+Real estate activities
MPRO+Professional, scientific and technical activities
 M69_70PROlegLegal and accounting services; Activities of head offices; management consultancy activities
 M71PROengArchitectural and engineering activities; technical testing and analysis
 M72PROsciScientific research and development
 M73PROadvAdvertising and market research
 M74_75PROothOther professional, scientific and technical activities; Veterinary activities
NADM+Administrative and support service activities
OPUB+Public administration and defence; compulsory social security
PEDU+Education
QHEA+Human health and social work activities
R_SART+Arts, entertainment and recreation
UHOU+Activities of households as employers

NACE level 2 sectors [22]

B – Sectoral global stranding

Table 3

Top 15 global sectors by stranding and exposure

Table 3
Total stranding External stranding Total exposure External exposure 1 RES+ (9.971) MINfos (2.399) RES+ (22.512) RES+ (12.972) 2 WATwat (6.922) WATwst (2.349) PUB+ (11.768) PUB+ (8.603) 3 COMvid (5.701) FINser (2.309) PWR+ (7.493) PWR+ (4.134) 4 PWR+ (5.214) ADM+ (2.209) WATwat (5.939) TRAinl (2.674) 5 WATwst (4.922) TRApos (2.122) COMvid (5.602) CNS+ (2.592) 6 MINfos (4.636) PROleg (2.045) ART+ (4.729) ART+ (2.505) 7 MINsup (4.355) MINoth (2.033) TRAinl (4.573) MANfoo (1.888) 8 MINoth (4.168) MANpri (2.01) MINfos (3.979) HEA+ (1.798) 9 TRAwar (3.91) PROadv (1.942) COMtel (3.58) MINfos (1.743) 10 PUB+ (3.389) MANref (1.922) AGRagr (3.29) MANmet (1.686) 11 ADM+ (3.311) PWR+ (1.855) TRAwar (3.287) AGRagr (1.684) 12 AGRfor (3.25) TRAwar (1.825) WATwst (3.15) MANche (1.639) 13 COMtel (3.237) MINsup (1.823) MINoth (3.012) COMtel (1.543) 14 TRAinl (3.194) MANpap (1.801) CNS+ (2.998) MANmot (1.29) 15 TRApos (3.178) AGRfor (1.798) MINsup (2.8) FD+ (1.257)

Top 15 global sectors by stranding and exposure

C – Country stranding multipliers and exposure

Table 4

Ranking of countries by stranding and exposure

Total strandingExternal strandingExternal exposure
1SVK (8.802)FRA (2.961)USA (6.762)
2BRA (7.394)AUS (2.638)CHN (4.695)
3AUS (7.273)SVK (2.598)JPN (3.992)
4LUX (7.055)ESP (2.494)DEU (3.586)
5KOR (6.213)SVN (2.241)KOR (3.081)
6FRA (5.972)CAN (2.052)ITA (2.688)
7CZE (5.847)CZE (2.015)CZE (2.66)
8SVN (5.834)MLT (1.882)SVK (2.657)
9HUN (5.485)NLD (1.864)GBR (2.367)
10CAN (5.43)DEU (1.814)AUT (2.037)
11IND (5.366)AUT (1.58)ESP (1.936)
12EST (5.338)DNK (1.57)FRA (1.77)
13TWN (5.201)NOR (1.555)RUS (1.669)
14DNK (5.079)BEL (1.551)HUN (1.423)
15USA (4.892)FIN (1.525)SWE (1.386)
16PRT (4.871)RUS (1.515)TUR (1.205)
17LVA (4.844)LVA (1.509)IND (1.164)
18ESP (4.644)SWE (1.489)BRA (1.069)
19RUS (4.575)LTU (1.471)TWN (1.037)
20ROU (4.51)MEX (1.421)BEL (1.02)
21CYP (4.47)IDN (1.36)FIN (0.969)
22CHN (4.437)TWN (1.165)NLD (0.876)
23FIN (4.244)BGR (1.155)IDN (0.764)
24TUR (4.178)KOR (1.144)ROU (0.663)
25BGR (4.174)HUN (1.132)POL (0.628)
26AUT (4.117)EST (1.097)DNK (0.589)
27IRL (4.039)CHE (1.049)HRV (0.578)
28MEX (4.007)POL (0.934)AUS (0.543)
29NLD (3.893)GBR (0.901)NOR (0.54)
30MLT (3.855)GRC (0.899)CHE (0.46)
31JPN (3.818)PRT (0.887)BGR (0.424)
32SWE (3.755)CYP (0.851)CAN (0.374)
33GBR (3.697)LUX (0.774)PRT (0.328)
34NOR (3.681)HRV (0.757)MEX (0.309)
35CHE (3.466)TUR (0.677)LVA (0.298)
36IDN (3.452)BRA (0.647)SVN (0.19)
37LTU (3.422)ROU (0.449)GRC (0.152)
38DEU (3.352)IRL (0.437)LTU (0.133)
39ITA (3.345)IND (0.408)EST (0.13)
40HRV (3.026)JPN (0.371)IRL (0.118)
41GRC (2.971)USA (0.29)LUX (0.075)
42BEL (2.893)ITA (0.224)CYP (0.048)
43POL (2.828)CHN (0.212)MLT (0.021)

Ranking of countries by stranding and exposure

Notes

  • [1]
    The code to replicate our results and charts is available at https://github.com/ecampiglio/capital_stranding_cascades.
  • [2]
    Pre-multiplying a matrix by iT returns its column sum; post-multiplying a matrix by i returns its row sum.
  • [3]
    This definition of stranding multipliers assumes a linear relationship between production and capital utilisation: for a marginal output loss of $1, the stranded capital stock corresponds to the capital intensity of the respective sector. In other words, we assume constant returns to scale.
  • [4]
    Due to the assumption of perfect input substitutability in the Ghosh model, any input loss - may it come from primary or intermediate inputs - corresponds directly to an output loss of the same size, as all other inputs remain unchanged. Thus, the fossil sector at the origin of a stranding cascade is the only sector in the network that changes its primary inputs, while all other sectors experience only losses in intermediate inputs according to fixed output allocation coefficients.
  • [5]
    If i is the affected sector and j the originating sector, the allocation coefficient is given by element bij of the BT matrix.
  • [6]
    This also means that if n was set equal to the total number of sectors of the IO table (i.e. each layer contains all sectors of the economy), the sum of all incoming edges of a particular sector in a certain layer l would correspond exactly to the total stranding that this sector receives in the lth round of the power series and therefore to the respective sector’s element in the fossil mining column of the figure im16 matrix. Thus, the network simply depicts subprocesses of the matrix multiplication of the power series. Setting n equal to the total number of sectors means that all possible stranding channels originating in the fossil sector (i.e. all sub-processes of the matrix multiplication leading to the fossil column of the transposed power series matrices) are considered. By defining the parameter n for filtering the top direct stranding linkages of each sector in each layer, we simply extract the most pronounced stranding channels.
  • [7]
    A few other MRIO datasets exist, such as EORA (Lenzen, 2011) or EXIOBASE (Stadler et al., 2018). However, to the best of our knowledge, WIOD is the only one offering sector-specific values for physical capital stocks, a necessary component of our analysis. The WIOD database is available at http://www.wiod.org.
  • [8]
    One small adjustment was made to the WIOD capital stock data set prior to the analysis: A negative capital stock value for the Portuguese MANrep sector, likely the result of a negative price deflator, was set to a positive value of the same absolute magnitude.
  • [9]
    These values are calculated using BP (2020) for oil, gas and coal production; and SEI et al. (2020) for extractionbased CO2 emissions
  • [10]
    The ICIO database lacks sectoral capital stock data. For this reason, despite its more granular mining classification, it cannot be used as the main data source in our analysis.
  • [11]
    The approach we take here is similar to the one used in building the Global Trade Analysis Project (GTAP) database, where the agricultural sectors of some countries’ national I-O tables are further disaggregated using more detailed agricultural I-O data from other sources (McDougall, 2009).
  • [12]
    Certain sectors have a more granular representation in WIOD than in ICIO. For instance, ICIO has an aggregate sector for agriculture, forestry and fishing, while WIOD has three separates subsectors (A01 to A03). In these cases we use the mining ratios of the corresponding parent sector in ICIO to disaggregate their transactions with mining industries, and apply this to all corresponding WIOD subsectors. For the NACE sector U (“Activities of extraterritorial organisations and bodies”), which is not included in ICIO, we split transactions with the B sector (if existing) into three equal parts.
  • [13]
    In an RAS procedure, a technical matrix (A) is pre-multiplied and post-multiplied by diagonal matrices R and S to derive a new technical matrix with specified row and column sums. TRAS is an extension of this procedure, additionally allowing for constraints on arbitrary subsets of matrix cells.
  • [14]
    More specifically, every non-mining cell is constrained by its original value and every aggregate of the three mining sub-sectors (i.e. a block of 3x1 cells in mining rows, 1x3 cells in mining columns or 3x3 cells in mining intraindustry trade) is constrained by the original value of the B sector.
  • [15]
    Two small preparatory adjustments are made to ease the convergence of the algorithm. First, zeroes resulting from the mining disaggregation process are replaced by small positive values. Second, several negative values that are naturally contained in the WIOD final demand and value added vectors are masked during the balancing procedure. Once these adjustments are complete, the TRAS algorithm proceeds with two steps in each iteration: i) a rescaling of rows and columns in a RAS step; and ii) a rescaling of the known cells and cell aggregates in a TRAS step. The algorithm stops when all row, column and cell rescaling factors converge to unity (given a certain tolerance value).
  • [16]
    Edge values smaller than 0.001 are not displayed to improve readability of the graph.
  • [17]
    The choice of n is, admittedly, arbitrary. We choose n = 3 so to create a readable network with enough depth.
  • [18]
    The entire ranking is available in the middle column of Table 4.
  • [19]
    See Table 4 for a full ranking
  • [20]
    We prefer Germany to Japan, despite the latter exhibits a higher exposure multiplier than the former, to have a more diverse regional representation.
  • [21]
    This link is further exacerbated by a negative inventory change of the TWN MINfos sector in 2014, which effectively reduces its total output of the same year and consequently increases the sector’s output allocation coefficients.
  • [22]
    See Eurostat (2008) for a more detailed description of NACE codes.
English

The aim of this article is to assess the exposure of economic systems to the risk of physical capital stranding following a reduction of fossil fuel production and use. We calculate cross-sectoral and cross-country ‘marginal stranding multipliers’ for 43 regions, and study how supply-side capital stranding might propagate via international production networks. We show how the fossil industry has the potential of creating significant stranding cascades affecting downstream sectors and the economic system as a whole. We then focus on cross-country stranding impacts and rank countries according to their external stranding potential and to their exposure to external stranding risk. Finally, we analyse more in depth the origins and transmission channels of the stranding links affecting the most exposed countries (US, China and Germany). Our results confirm the relevance of including multi-regional production networks and physical capital stranding into the ongoing effort to assess the macro-financial implications of a low-carbon transition.

  • low-carbon transition
  • asset stranding
  • production networks
  • capital stocks
  • fossil fuels
  • supply-side policies
Français

L’objectif de cet article est d’évaluer l’exposition des systèmes économiques au risque d’échouage du capital physique suite à une réduction de la production et de l’utilisation des combustibles fossiles. Nous calculons des “multiplicateurs marginaux d’échouage” intersectoriels et transnationaux pour 43 régions, et nous étudions comment l’échouage du capital du côté de l’offre pourrait se propager via les réseaux de production internationaux. Nous montrons comment l’industrie fossile a le potentiel de créer d’importantes cascades d’échouages affectant les secteurs en aval et le système économique dans son ensemble. Nous nous concentrons ensuite sur les impacts d’échouage entre pays et classons les pays en fonction de leur potentiel d’échouage externe et de leur exposition au risque d’échouage externe. Enfin, nous analysons plus en profondeur les origines et les canaux de transmission des liens d’échouage affectant les pays les plus exposés (États-Unis, Chine et Allemagne). Nos résultats confirment la pertinence d’inclure les réseaux de production multirégionaux et l’échouage du capital physique dans l’effort en cours pour évaluer les implications macrofinancières d’une transition vers une économie à faible émission de carbone.
JEL Classification: C67; E22; L71; O10; Q32

  • transition bas-carbone
  • actifs échoués
  • réseaux de production
  • stocks de capitaux
  • combustibles fossils
  • politiques de l’offre

References

  • Acemoglu, D., Carvalho, V. M., Ozdaglar, A., and Tahbaz-Salehi, A. (2012). The network origins of aggregate fluctuations. Econometrica, 80(5):1977–2016.
  • Aldasoro, I. and Angeloni, I. (2015). Input–output-based measures of systemic importance. Quantitative Finance, 15(4):589–606.
  • Allen, T., Dées, S., Boissinot, J., Caicedo Graciano, C. M., Chouard, V., Clerc, L., de Gaye, A., Devulder, A., Diot, S., Lisack, N., Pegoraro, F., Rabate, M., Svartzman, R., and Vernet, L. (2020). Climate-Related Scenarios for Financial Stability Assessment: An Application to France. Working Paper 774, Banque de France, Paris.
  • Antràs, P., Chor, D., Fally, T., and Hillberry, R. (2012). Measuring the Upstreamness of Production and Trade Flows. American Economic Review, 102(3):412–416.
  • Asheim, G. B. (2013). A distributional argument for supply-side climate policies. Environmental and Resource Economics, 56(2):239–254.
  • Asheim, G. B., Fæhn, T., Nyborg, K., Greaker, M., Hagem, C., Harstad, B., Hoel, M. O., Lund, D., and Rosendahl, K. E. (2019). The case for a supply-side climate treaty. Science, 365(6451):325–327.
  • Baldwin, E., Cai, Y., and Kuralbayeva, K. (2020). To build or not to build? Capital stocks and climate policy. Journal of Environmental Economics and Management, 100:102235.
  • Bastidas, D. and Mc Isaac, F. (2019). Reaching Brazil’s Nationally Determined Contributions: An assessment of the key transitions in final demand and employment. Energy Policy, 135:110983.
  • Battiston, S., Mandel, A., Monasterolo, I., Schütze, F., and Visentin, G. (2017). A climate stress-test of the financial system. Nature Climate Change, 7:283–288.
  • Blöchl, F., Theis, F. J., Vega-Redondo, F., and Fisher, E. O. (2011). Vertex centralities in inputoutput networks reveal the structure of modern economies. Physical Review E, 83(4):046127.
  • BP (2020). Statistical Review of World Energy 2020. Technical report, BP, London.
  • Cahen-Fourot, L., Campiglio, E., Dawkins, E., Godin, A., and Kemp-Benedict, E. (2020). Looking for the Inverted Pyramid: An Application Using Input-Output Networks. Ecological Economics, 169:106554.
  • Caldecott, B., editor (2018). Stranded Assets and the Environment: Risk, Resilience and Opportunity. Routledge Explorations in Environmental Studies. Routledge, Oxon.
  • Campiglio, E., Dafermos, Y., Monnin, P., Ryan-Collins, J., Schotten, G., and Tanaka, M. (2018). Climate change challenges for central banks and financial regulators. Nature Climate Change, 8(6):462–468.
  • Carvalho, V. M. and Tahbaz-Salehi, A. (2019). Production networks: A primer. Annual Review of Economics, 11:635–663.
  • Chen, G., Wu, X., Guo, J., Meng, J., and Li, C. (2019). Global overview for energy use of the world economy: Household-consumptionbased accounting based on the world input-output database (WIOD). Energy Economics, 81:835–847.
  • Crawford, N. C. (2019). Pentagon Fuel Use, Climate Change, and the Costs of War. Watson Institute, Brown University.
  • Cui, R. Y., Hultman, N., Edwards, M. R., He, L., Sen, A., Surana, K., McJeon, H., Iyer, G., Patel, P., Yu, S., Nace, T., and Shearer, C. (2019). Quantifying operational lifetimes for coal power plants under the Paris goals. Nature Communications, 10(1):1–9.
  • Dietzenbacher, E. (1997). In vindication of the Ghosh model: A reinterpretation as a price model. Journal of Regional Science, 37(4):629–651.
  • Erickson, P., Lazarus, M., and Piggot, G. (2018). Limiting fossil fuel production as the next big step in climate policy. Nature Climate Change, 8(12):1037– 1043.
  • Eurostat (2008). NACE Rev.2 Statistical classification of economic activities in the European Community. Technical report, Eurostat, Luxembourg.
  • Gabaix, X. (2011). The granular origins of aggregate fluctuations. Econometrica, 79(3):733–772.
  • Galbusera, L. and Giannopoulos, G. (2018). On input-output economic models in disaster impact assessment. International Journal of Disaster Risk Reduction, 30:186–198.
  • Ghosh, A. (1958). Inputoutput approach in an allocation system. Economica, 25(97):58–64.
  • Gilchrist, D. A. and St. Louis, L. V. (1999). Completing Input–Output tables using partial information, with an application to canadian data. Economic Systems Research, 11(2):185–194.
  • Gilchrist, D. A. and St. Louis, L. V. (2004). An algorithm for the consistent inclusion of partial information in the revision of input-output tables. Economic Systems Research, 16(2):149–156.
  • Harstad, B. (2012). Buy coal! A case for supply-side environmental policy. Journal of Political Economy, 120(1):77–115.
  • IEAa (2020a). Energy Technology Perspectives 2020. Technical report, International Energy Agency, Paris.
  • IEAb (2020b). Key world energy statistics 2020. Statistical Report, International Energy Agency, Paris.
  • IPCC (2014). Climate Change 2014: Mitigation of climate change. Technical report, Working Group III-Intergovernmental Panel on Climate Change.
  • Johnson, N., Krey, V., McCollum, D. L., Rao, S., Riahi, K., and Rogelj, J. (2015). Stranded on a low-carbon planet: Implications of climate policy for the phase-out of coal-based power plants. Technological Forecasting and Social Change, 90:89–102.
  • Joya, O. and Rougier, E. (2019). Do (all) sectoral shocks lead to aggregate volatility? Empirics from a production network perspective. European Economic Review, 113:77–107.
  • Klimek, P., Poledna, S., and Thurner, S. (2019). Quantifying economic resilience from input– output susceptibility to improve predictions of economic growth and recovery. Nature Communications, 10(1):1677.
  • Lazard (2020). Levelized cost of energy analysis-Version 14.0. Technical report, Lazard, New York.
  • Lenzen, M. (2011). Aggregation versus disaggregation in input– output analysis of the environment. Economic Systems Research, 23(1):73–89.
  • Lenzen, M. and Crawford, R. (2009). The path exchange method for hybrid LCA. Environmental science & technology, 43(21):8251–8256.
  • Lenzen, M., Moran, D., Kanemoto, K., and Geschke, A. (2013). Building Eora: A Global Multi-Region Input– Output Database at High Country and Sector Resolution. Economic Systems Research, 25(1):20–49.
  • Leontief, W. W. (1951). The Structure of American Economy, 1919-1939: An Empirical Application of Equilibrium Analysis. Oxford University Press, New York.
  • Marin, G. and Vona, F. (2019). Climate policies and skill-biased employment dynamics: Evidence from EU countries. Journal of Environmental Economics and Management, 98:102253.
  • McDougall, R. A. (2009). Disaggregation of input-output tables. In Narayanan, B. and Walmsley, T. L., editors, Global Trade, Assistance, and Production: The GTAP 7 Data Base. Center for Global Trade Analysis, Purdue University, Department of Agricultural Economics, Purdue University, West Lafayette, IN.
  • McGlade, C. and Ekins, P. (2015). The geographical distribution of fossil fuels unused when limiting global warming to 2C. Nature, 517(7533):187–190.
  • Meinshausen, M., Meinshausen, N., Hare, W., Raper, S. C. B., Frieler, K., Knutti, R., Frame, D. J., and Allen, M. R. (2009). Greenhouse-gas emission targets for limiting global warming to 2C. Nature, 458(7242):1158–1162.
  • Mercure, J.-F., Pollitt, H., Viñuales, J. E., Edwards, N. R., Holden, P. B., Chewpreecha, U., Salas, P., Sognnaes, I., Lam, A., and Knobloch, F. (2018). Macroeconomic impact of stranded fossil fuel assets. Nature Climate Change, 8(7):588–593.
  • Miller, R. E. and Blair, P. D. (2009). Input-Output Analysis: Foundations and Extensions. Cambridge University Press.
  • Mules, T. J. (1983). Some simulations with a sequential input-output model. In Papers of the Regional Science Association, volume 51, pages 197–204. Springer.
  • OECD (2018). OECD inter-country input-output (ICIO) tables. Technical report, Organisation for Economic Co-operation and Development, Paris.
  • Oosterhaven, J. (1988). On the Plausibility of the Supply-Driven Input-Output Model. Journal of Regional Science, 28(2):203–217.
  • Perrier, Q. and Quirion, P. (2018). How shifting investment towards low-carbon sectors impacts employment: Three determinants under scrutiny. Energy Economics, 75:464–483.
  • Pfeiffer, A., Hepburn, C., Vogt-Schilb, A., and Caldecott, B. (2018). Committed emissions from existing and planned power plants and asset stranding required to meet the Paris Agreement. Environmental Research Letters, 13(5):054019.
  • Piñero, P., Bruckner, M., Wieland, H., Pongrácz, E., and Giljum, S. (2019). The raw material basis of global value chains: Allocating environmental responsibility based on value generation. Economic Systems Research, 31(2):206–227.
  • Roncoroni, A., Battiston, S., Escobar Farfàn, L. O. L., and Martinez Jaramillo, S. (2019). Climate risk and financial stability in the network of banks and investment funds. Available at SSRN 3356459.
  • Rozenberg, J., Vogt-Schilb, A., and Hallegatte, S. (2020). Instrument choice and stranded assets in the transition to clean capital. Journal of Environmental Economics and Management, 100:102183.
  • Schnabel, I. (2020). When markets fail – the need for collective action in tackling climate change. Speech by Isabel Schnabel, Member of the Executive Board of the ECB, at the European Sustainable Finance Summit, European Central Bank, Frankfurt.
  • SEI, IISD, ODI, Climate Analytics, CICERO, and UNEP (2019). The Production Gap: The discrepancy between countries’ planned fossil fuel production and global production levels consistent with limiting warming to 1.5°C or 2°C. Technical report, Stockholm Environment Institute.
  • SEI, IISD, ODI, E3G, and UNEP (2020). The Production Gap Report: 2020 Special Report. Technical report, Stockholm Environment Institute.
  • Semieniuk, G., Campiglio, E., Mercure, J.-F., Volz, U., and Edwards, N. R. (2021). Low-carbon transition risks for finance. WIREs Climate Change, 12(1):e678.
  • Stadler, K., Wood, R., Bulavskaya, T., Södersten, C.-J., Simas, M., Schmidt, S., Usubiaga, A., Acosta-Fernández, J., Kuenen, J., Bruckner, M., Giljum, S., Lutter, S., Merciai, S., Schmidt, J. H., Theurl, M. C., Plutzar, C., Kastner, T., Eisenmenger, N., Erb, K.-H., de Koning, A., and Tukker, A. (2018). EXIOBASE 3: Developing a Time Series of Detailed Environmentally Extended Multi-Regional Input-Output Tables. Journal of Industrial Ecology, 22(3):502–515.
  • Timmer, M. P., Dietzenbacher, E., Los, B., Stehrer, R., and de Vries, G. J. (2015). An Illustrated User Guide to the World Input–Output Database: The Case of Global Automotive Production. Review of International Economics, 23(3):575–605.
  • Tong, D., Zhang, Q., Zheng, Y., Caldeira, K., Shearer, C., Hong, C., Qin, Y., and Davis, S. J. (2019). Committed emissions from existing energy infrastructure jeopardize 1.5 °C climate target. Nature, 572(7769):373– 377.
  • UNFCCC (2016). Report of the Conference of the Parties on its twenty-first session, held in Paris from 30 November to 13 December 2015. Addendum. Part two: Action taken by the Conference of the Parties at its twenty-first session. Technical Report FCCC/CP/2015/10/Add.1, United Nations Framework Convention on Climate Change, Paris.
  • van der Ploeg, F. and Rezai, A. (2020). Stranded Assets in the Transition to a Carbon-Free Economy. Annual Review of Resource Economics, 12(1):281–298.
  • Vermeulen, R., Schets, E., Lohuis, M., Kölbl, B., Jansen, D.-J., and Heeringa, W. (2018). An energy transition risk stress test for the financial system of the Netherlands. Occasional Studies 16-7, De Nederlandsche Bank, Amsterdam.
  • Voigt, S., De Cian, E., Schymura, M., and Verdolini, E. (2014). Energy intensity developments in 40 major economies: Structural change or technology improvement? Energy Economics, 41:47–62.
  • Zhang, Y. (2010). Supply-side structural effect on carbon emissions in China. Energy Economics, 32(1):186–193.
  • Zhang, Z., Zhang, Z., and Zhu, K. (2020). Allocating carbon responsibility: The role of spatial production fragmentation. Energy Economics, 87:104491.
Louison Cahen-Fourot
Vienna University of Economics and Business
Emanuele Campiglio
University of Bologna
Antoine Godin
Agence Française de Développement
Eric Kemp-Benedict
Stockholm Environment Institute
Stefan Trsek
University of Bologna
This is the latest publication of the author on cairn.
This is the latest publication of the author on cairn.
This is the latest publication of the author on cairn.
This is the latest publication of the author on cairn.
Uploaded on Cairn-int.info on 22/10/2021
Cite
Distribution électronique Cairn.info pour Agence française de développement © Agence française de développement. Tous droits réservés pour tous pays. Il est interdit, sauf accord préalable et écrit de l’éditeur, de reproduire (notamment par photocopie) partiellement ou totalement le présent article, de le stocker dans une banque de données ou de le communiquer au public sous quelque forme et de quelque manière que ce soit.
keyboard_arrow_up
Chargement
Loading... Please wait