How Does Supply Network Structure Influence Firms’ Financial Performance During Disruption?

In a post pandemic world, supply chain mapping has become a key concern for almost any supply chain operation. A single focal supply chain network might have hundreds of suppliers including those that supply both goods and services. These increase exponentially as one traces the network downstream. Using a pair of networks from the auto industry (Ford and General Motors) as a basis for the analysis, we apply a data mining technique to filter for significant supply chain relationships in each of these networks over the period spanning 2018–2022. We then examine the structural metrics for each of these networks and compare these measures with applicable financial measures to elucidate how supply chain network structure affects financial performance. Our regression analyses reveal that supplier network structures significantly affect customers’ market value. Specifically, Tobin’s Q is negatively related to the number of suppliers they source from and the importance of the customer (e.g., Ford) to its suppliers, but it is positively associated with the tightness of supplier customer connection and the density of the network.

Introduction

In earlier research, Orenstein and Tang (2021) collected data from a cross-section of companies spanning three years (2013–2015) and a sample of forty-four companies. The key results from this analysis showed that the average degree–the ratio of edge count to node count, when examined across all tiers, exhibited significance across several financial performance measures. The authors found a negative association (proportionally smaller gain) between the average degree and firm performance. In other words, as the number of suppliers increases, there is a threshold beyond which the effect of increased connectivity decreases and starts to impact firm performance.

In this work, we have narrowed the focus to two data sets, Ford (F) and General Motors (GM), over the period 2018–2022 to understand how disruption (caused by a shock such as COVID-19) affects the financial performance of the network. The impact of risk and how it propagates through the network of connections is often referred to as supply chain fit. Consequently, we selected F and GM as a starting point for this study since they are competing firms, yet they exhibit different network structures. Hence, the fit of each firm is what is of interest. Unlike the prior study, which included all supplier relationships, this study focuses only on the product value stream, thereby filtering the network for significant supply chain relationships and simplifying the networks. By reducing the network to its key players, we hope to reveal how network structure impacts firm performance.

The main contribution of this work is to establish a deeper understanding of supply chain fit and firms’ financial performance specifically pre and post disruption. An event such as Covid-19 represents fertile ground for studying the impact of these shocks. In this paper, we have also expanded the selection of topological metrics, specifically, we included the eigenvector centrality measure to examine the strength of the supplier connections relative to the firm performance. We explore how changes in the topological structure of the supply chain which occur in real-time impact customers’ financial performance. Specifically, we would like to understand the impact of a disruption (such as COVID-19) in this context.

Therefore, the key objectives for this paper are:

• To develop and visualize a comprehensive dataset describing multi-tiered filtered financial relationship data for a pilot of two sample supply chain networks.
• To extract significant statistical relationships between the dynamic topological structure and specific financial metrics.
• To understand the link between structural and financial parameters in the context of this pilot data set.

Literature Review

A modern supply chain network can be visualized as a large system of tiered, connected nodes (or firms) that are constantly evolving and changing. The dynamic nature of these systems lends itself to the need for a new approach to supply chain modeling. In earlier work, Orenstein (2021) provides a historical overview of research papers that focus on modeling approaches for tackling supply chain structure. An overall framework for modern-day supply chain topology emerges from the discussion.

The goal of this paper is to assume this structure yet place an emphasis on the financial implications of the inter-relationships. Effectively, we would like to examine the impact of risk and how it propagates through the network of connections. This concept is often referred to as supply chain fit.

In order to model the constantly evolving network of inter-relationships, several researchers have used data from Bloomberg and Factset. Bloomberg data is static in nature, whereas Factset can be employed to derive dynamic relationships.

To begin the discussion, we point out two important works: Wu and Birge (2014), who examine supply chain fit by exploring the relationship between network structure and stock returns across different industries, and likewise, Carnovale and Yeniyurt (2014), who consider how network structure affects financial performance by examining specific financial metrics (return on equity, return on asset and return on sales). However, the networks described in this pair of papers were limited to static relationships in a single industry and are somewhat dated.

Jussaet al. (2015) used a mix of Bloomberg and FactSet data to introduce the idea that a single shock at one company can be transmitted to other connected firms. They also develop a metric to identify significant players in the supply network. Supply chain fit is the theme of Wu (2015) and a follow-up study by Wu (2016). In this pair of papers, the author uses Factset data first to demonstrate the notion that the stock performance of ‘supplier central’ portfolios tends to predict the movements of the overall stock market and that the revenue of a particular firm at the local level can impact the revenue of firms up to multiple connections away. Jussaet al. (2015), Carnovale and Yeniyurt (2015), and Wu and Birge (2014) evaluate the network beyond the first tier and establish relationships that, in order to assess the impact of risk accurately, the entire network needs to be considered.

There is fundamental research by Arora and Brintrup (2021), which develops measures to characterize the embeddedness of individual firms in a supply network. These are centrality, tier position, and triads. They point out that centrality impacts individual performance through a diminishing returns relationship. These results pertain to a static network and relate to a single time slice. Another related study by Seileret al. (2020) finds some evidence that profitability is related to connectedness and market share.

The contribution by Orenstein and Tang (2021) further considers supply chain fit by examining the impact of structural metrics and how these influence firms’ performance and, importantly, it applies to a dynamic network. However, because the networks that were used were not filtered for significant supply chain relationships, the effect of supply chain network structure on performance was found to be weak, although they were in line with Arora and Brintrup (2021).

Firms competing in the same environment (e.g., industry) often adopt different supply chain structures based on their own needs and resources. Wagneret al. (2012), Christopher and Ryals (2014), Frankel and Mollenkopf (2015), Srinivasan and Swink (2015), and Tateet al. (2015) all present analytical frameworks that establish a strong link between supply chain fit and firms’ financial performance.

Our article is most closely related to the growing literature that emphasizes the role of production networks as a mechanism for propagation and amplification of shocks Our preliminary results are in line with the notion that the transmission of shocks within the supply chain networks affects financial performances of suppliers (Wu & Birge, 2014). By exploring the impact of network structures on GM and Ford performance during and after the Covid crisis, we extend research such as Carvalhoet al. (2021), who use market disruptions resulting from the Great East Japan Earthquake of 2011, to study the resilience of supply chain network in the economy. Other research in this area includes the building on the multisector model of Long and Plosser (1983), papers such as Acemogluet al. (2012), Acemogluet al. (2017), and Baqaee and Farhi (2019) who characterize the conditions under which propagation of microeconomic shocks over input-output linkages can translate into sizable aggregate fluctuations.

The contribution of this research is to expand further on supply chain fit, first by building supply chain networks which have been filtered for significant supply chain relationships. We also include some additional structural metrics to explain the strength of the supplier connections. We then compare the financial performance of two pilot firms via financial metrics to quantify the influence of supply network structure on firms’ performance using Covid-19 as the disruption. We would like to explore the degree to which our network structure variables can serve as measures of resilience for a given supply network and their roles in transmitting shocks along the supply chain.

Research Methodology

To trace out the supply chain networks, we first obtained financial performance data for two pilot companies in the auto industry (Ford and GM). Next, we pruned the data for significant relationships only, thereby eliminating companies that were not contributing to the supply chain structure (for example, payroll, IT, and the like). We then arranged the data by calendar quarter and tier for the period spanning 2018–2022.

We collected the data for each company and period from WRDS, focusing on the FactSet database, FACTSET Research Systems (2023a, 2023b). The WRDS output was then filtered using a customized tool which removed any non-essential suppliers. This allows us to look at the core network for a company. For example, if we are pulling the data for Ford, we want to look at the chip and raw material suppliers and remove any noise from G&A suppliers (i.e., accounting software, cleaning services, etc.).

We then constructed supply chain maps for each company. The visualizations were created using Gephi (Bastianet al., 2009), an open-source visualization software. In our initial review, we looked at Ford and GM over the period from Q1 2018 to Q1 2022. The visualizations for the respective networks of GM and Ford in 2019 were constructed, showing the tiered information in visual form, which includes the focal company and the key suppliers and their associated labels.

A sample visualization for the GM network is shown in Figs. 1a–1c. A similar map (not shown) is available for Ford. As more information is added to the visualization, it is harder to identify the critical suppliers on the map. However, this information is stored in the network data frame and is available for analysis.

Fig. 1. GM Tier 1 suppliers, 2019 (a), GM Tier 1–2 suppliers, 2019 (b), and GM Tier 1–3 suppliers, 2019 (c).

Each company’s visualizations were then analyzed by quarter and tier. Using Gephi, we collected the structural metrics for each supply chain visualization. Corresponding to each time period, we collected several known financial metrics. Using the structural and financial metric data, we then applied a series of regression analyses to elucidate those key structural parameters that showed improved financial performance of the entire tiered supply network.

To characterize the topology of the supply network, we determined the values of several leading structural metrics (Table I). These metrics include node count, edge count, average degree, graph density, and eigenvector centrality. Likewise, we used key financial parameters (stock market valuation and accounting return) to characterize the supply network overall financial performance. The metrics we used were Tobin’s Q and return on assets (ROA). The definitions for the structural and financial metrics are provided in Table I. In the results section, we provide a series of summary statistics and regression analyses that we have conducted using the the empirical data.

Table I. A Summary of Network Structural Metrics and Financial Metrics

Metric name	Definition
Structural metrics
Node & edge counts	The size of a given network is defined as the number of nodes and links/edges and characterizes the overall scale of the network.
Average degree	Provides a measure of how many connections a firm has. A high average degree implies strong inter-connectivity among the firms in the network.
Average weighted degree	Assigns weights to the edges in the network. In a supply chain, the weighted degree of a node is the sum of all its weighted edges. The average weighted degree is then calculated as the average of these weighted degrees across all nodes in the network.
Average clustering coefficent	In graph theory, a clustering coefficient is a measure of the degree to which nodes in a graph tend to cluster together.
Graph density	The density is a measure of how well-connected or interdependent the nodes are. It is calculated as the ratio of the actual number of edges to the total possible number of edges.
Eigenvector centrality	Measures firm’s influence in the network by considering the influence of its neighbors. A lower eigenvector centrality implies that there are fewer nodes with big connections, lower impact if a supplier connection fails.
Financial metrics
Tobin’s Q	Often referred to as market to book ratio or market value over book value of total assets = (atq − ceqq + prccq * cshoq)/atq.
ROA	Return on assets = 100 * niq/atq (denoted in percent).

Results

Structural Results

The overall pattern of the GM vs. Ford supply chain is similar (Fig. 2). Both exhibit more activity in the first two quarters, followed by a drop off in Q3–Q4; The cycle of the supply chain is noticeable in the data (Q1, Q2 peak followed by lower production in subsequent quarters). This follows the production cycle of the auto industry. Note that for GM, the node and edge count declines rapidly during the COVID-19 pandemic, suggesting a reduction in supply chain operation. Ford exhibits more resilience in this area, as indicated by the corresponding node/edge counts.

Fig. 2. Supply chain node and edge counts for GM (a) and Ford (b), 2018–2021. COVID-19 window extends from Q3 2020 to Q1 2021 across all tiers.

We examined the top ten suppliers in each of GM and Ford across 2018–2021. GM had the same suppliers showing up regularly in its list whereas we noticed some turnover in Ford’s list.

This phenomenon is reflected in the eigenvector centrality statistic, which remains steady in GM (except for one peak), whereas, for Ford, we see more variability (Fig. 3). This might be an indication that Ford is less reliant on specific suppliers, which appears to be reflected in the node and edge counts. It seems like Ford has a more resilient pattern, especially during COVID-19.

Fig. 3. Eigenvector centrality for GM vs. Ford (2018–2021).

When the eigenvector centrality is low, this implies fewer nodes with big connections–or less reliance on suppliers. Both GM and Ford have overall low centrality (<0.04) but in 2020 Q2 GM has a noticeable spike (0.19). This suggests slightly more reliance on a few nodes with higher connections (GM vs. Ford).

But unlike conventional scale-free networks, we do not have “boundless” growth, and the hubs are limited in size–hence the eigenvector centrality numbers are small.

Financial Performance Results

We construct two performance measures, namely, Tobin’s Q, which is the ratio of a firm’s stock market valuation to its book value of assets, and ROA, which is the ratio of net income to a firm’s book value of assets. We conduct OLS panel regression analyses using firms’ quarterly financial data and their network measures based on the following specification:

where i, t index firm, and year-quarter, respectively. Network stands for one of the five network structure measures, i.e., average degree, average weighted degree, average clustering coefficient, graph density, and eigenvector centrality. X stands for a vector of firm specific controls that could affect financial performance, including capital expenditure, asset tangibility, leverage ratio, firm size, and R&D intensity. We use firm and year-fixed effects to control for latent firm factors and time-varying factors. Tables II–V presents the estimation results of (1) for GM and Ford, respectively.

(1) $$\begin{array}{rl}RO{A}_{i,t}\left(orTobi{n}^{\prime }s{Q}_{i,t}\right)& =\propto +{\beta }_0\ast Networ{k}_{i,t-1}+\delta X_{i,t-1}\\ & +{\gamma }_t+{\eta }_i+{\epsilon }_{i,t}\end{array}$$

Table II. Network Structure and Firms’ Financial Performance, GM, (ROA)

Panel A: GM	ROA
	(1)	(2)	(3)	(4)	(5)
Average degree	−0.124
	(0.106)
Average weighted degree		−0.054
		(0.072)
Average clustering coefficient			0.027*
			(0.016)
Graph density				0.147
				(0.425)
Eigenvector centrality					0.254
					(0.223)
Controls	Included	Included	Included	Included	Included
Observation	100	100	100	100	100
Adjusted R2	0.730	0.728	0.735	0.726	0.726

Table III. Network Structure and Firms’ Financial Performance, GM-Tobin’s Q

Panel A: GM	Tobin’s Q
	(1)	(2)	(3)	(4)	(5)
Average degree	−0.012** (0.006)
Average weighted degree		−0.007* (0.004)
Average clustering coefficient			0.002* (0.001)
Graph density				−0.011 (0.023)
Eigenvector centrality					0.008 (0.066)
Controls	Included	Included	Included	Included	Included
Observation	100	100	100	100	100
Adjusted R2	0.961	0.960	0.960	0.959	0.958

Table IV. Network Structure and Firms’ Financial Performance, Ford-ROA

Panel B: Ford	ROA
	(1)	(2)	(3)	(4)	(5)
Average degree	−0.02 (0.157)
Average weighted degree		−0.017 (0.095)
Average clustering coefficient			−0.536 (1.249)
Graph density				−0.037 (0.907)
Eigenvector centrality					−1.272 (2.804)
Controls	Included	Included	Included	Included	Included
Observation	72	72	72	72	72
Adjusted R2	0.973	0.973	0.973	0.973	0.973

Table V. Network Structure and Firms’ Financial Performance, Ford-Tobin’s Q

Panel B: Ford	Tobin’s Q
	(1)	(2)	(3)	(4)	(5)
Average degree	−0.018** (0.008)
Average weighted degree		−0.009* (0.005)
Average clustering coefficient			−0.071 (0.067)
Graph density				0.029*** (0.040)
Eigenvector centrality					−0.29* (0.146)
Controls	Included	Included	Included	Included	Included
Observation	72	72	72	72	72
Adjusted R2	0.948	0.947	0.944	0.962	0.947

In Panels A and B, we find that the network structures have significant predictive power for Tobin’s Q but not for ROA for both GM and Ford. Specifically, we show that both the average degree and the average weighted degree are negatively associated with Tobin’s Q, indicating that when GM or Ford are connected to more suppliers, they receive lower stock market valuations.

However, the clustering coefficient is positively associated with Tobin’s Q, indicating that the tightness of supplier customer connections helps increase firms’ market value. This coefficient is marginally positive in the ROA regression for GM, indicating the strength of the network connection has some positive effect of GM’s accounting performance. For Ford, we find a highly significant and positive relation between graph density and Tobin’s Q and a marginally negative association between eigenvector centrality and Tobin’s Q.

Conclusions and Further Work

In this preliminary analysis of two networks from the auto industry pre and post-COVID-19, we examined the supply chain operation and found that both networks exhibited more activity in the initial part of Covid disruption (i.e., 2020) followed by a drop-off in Q3–Q4. We also observed that reliance on specific suppliers may result in less resilience because of a catastrophic event such as the pandemic. This observation needs to be examined further by exploring additional supply networks across several industries.

Our regression analyses reveal that supplier network structures significantly affect customers’ market value. Specifically, Tobin’s Q is negatively related to the number of suppliers they source from and the importance of the customer (e.g., Ford) to its suppliers, but it is positively associated with the tightness of supplier customer connection and the density of the network. Using the COVID pandemic as an exogenous shock allows us to explore whether customers can rely on their existing supply chain relations to minimize the large negative impact of such a crisis on their operations. Our preliminary analyses show some significant differences in network structures between GM and Ford and distinct network effects of firms’ performance. Our results indicate that stock market prices influence supply chain relationships.

In one body of further work, we hope to link the results of this quantitative analysis on GM and Ford with a qualitative analysis of key events in these companies to further understand the overall supply chain network performance.

Another future outcome of this research is to explore the degree to which our network structure variables can serve as measures of resilience for a given supply network and their roles in transmitting shocks along the supply chain. This avenue of research will be extended to additional supply networks from several industries (auto, technology, healthcare), for which we have already collected structural and financial metric data to measure the impact of structure on firms’ performance both in and outside the industry. It will be interesting to observe the extent to which this study mirrors the larger, more extensive analysis and if any other behaviours will emerge for specific industries.