Predict Resilience in Market Turmoil: Applying Centrality Measures to the Supply Chain Network

Supply chain risk has become an increasing concern for investors, and in this article we analyze whether a unique factor called the centrality measure has predictive capabilities for stock price performance during disruptions.

During the COVID-19 pandemic, global supply chain disruptions led to widespread shortages of various products, significantly impacting businesses. Even after the initial crisis, several events continued to challenge the resilience of the supply chain, such as geopolitical events and tariff disruptions.

Figure 1 illustrates the number of searches for the keyword "supply chain risk," sourced from Google Trends for the U.S. over the past 10 years. The chart implies that the number of searches began to rise in 2020, reaching its peak at the end of the most recent month. It may indicate a growing interest and concern regarding the risks associated with the supply chain.

Figure 1

For equity investments, some fundamental factors such as earnings quality and stability can measure resistance to the supply chain disruptions. In this article, we step beyond the traditional factor analysis and examine a unique factor derived from the supply chain network topology: the centrality measure. We calculate the measure for all entities in the FactSet Supply Chain Relationships database and create portfolios based on their exposure to this factor. We then backtest these portfolios to examine the performance of the centrality measure during the periods of increased supply chain disruptions.

HITS Algorithm and Good Suppliers and Customers

The centrality measure estimates the importance of nodes/edges in a supply chain network. Centrality measures are essential for understanding the relative importance of suppliers or customers (nodes) and their interactions within the supply chain. These measures help identify the most crucial nodes in a network, which can significantly impact the performance of the entire supply chain.

Various algorithms and centrality measures exist for this purpose. One such algorithm is Hyperlink-Induced Topic Search (HITS), which identifies central hubs and authorities based on the network structure. This algorithm is often used to analyze websites, operating under the assumption that an important hub website will have numerous links to other important websites. At the same time, the important website (the authority) is defined by the number of links from the important hub websites.

We apply this algorithm to the supply chain network, where important customers (the hub) source products/materials from many key suppliers, and key suppliers (the authority) provide products to numerous important customers. We assume these key suppliers and customers, who have strong connections with one another, have more resilience to the supply chain disruption and lead to better stock performance during times of supply chain turmoil.

Analyzing Factor Performance for the Global Market

To analyze our hypothesis, we first created five equal-weighted portfolios from the global market index constituents based on the centrality measures derived from the FactSet Supply Chain Relationships database. These portfolios were rebalanced monthly. We then compared their performance over the last decade, assuming no transaction fees.

Figure 2 shows the cumulative returns of the five portfolios based on the centrality of important customers derived by the HITS algorithm. H5 consists of companies with the highest centrality of important customers, and H1 consists of companies with the lowest centrality of important customers.

The chart displays the performance differences among customer groups based on their importance in the supply chain network. The red dotted line (H5-H1) represents the cumulative spread return between the top group (H5) and the bottom group (H1).

Figure 2

On the other hand, Figure 3 shows the cumulative returns of the five portfolios based on the centrality of important suppliers. Therefore, the chart exhibits the performance differences among supplier groups based on their importance.

While Figures 2 and 3 display the performances of portfolios based on different groups, the cumulative returns of the portfolios show very similar results. Figure 2 exhibits a slightly higher spread return for the entire analysis period compared with Figure 3. (107.76 vs 86.48)

Figure 3

As we anticipated, the spread return has been widening since the COVID-19 pandemic. Figure 4 illustrates the cumulative spread returns from Figures 2 and 3, highlighting some of the major iconic incidents that have affected the global supply chain. Some events are not immediately apparent when they begin, and others are either a cause or a result of other global events, such as shipping cost surges, inflation, and global interest rate hikes.

At the beginning of most events, the spread return initially shrank but then widened afterwards. It appears that stocks among companies with high centrality are sold during times of market turmoil and subsequently bought back as the market stabilizes, while companies with low centrality fail to recover, even after stability returns.

Figure 4

The next question is whether the performance related to these factors originates from the factors themselves or whether it actually correlates with other factors, reflecting their performance instead.

Comparison with Alpha/Risk Factors

Let's compare the centrality measures with other conventional alpha/risk factor values.

Figure 5 displays a (Spearman) correlation matrix of several key factors retrieved from the FactSet Quant Factor Library. Both Important Customer and Supplier centrality measures show positive correlation with Market Value (Size factor) to some extent, which makes sense as large companies tend to expand as they play a more significant role in the supply chain network. Additionally, Important Customer and Supplier centralities are positively correlated with each other, which also makes sense because central customers often serve as central suppliers for their own downstream customers.

Other factors show little correlation with either centrality measure, as centrality is solely derived from the supply chain network without any additional fundamental information. Important Customer Centrality shows very similar correlation with Important Supplier Centrality (0.5) and with Market Value (0.48), while the correlation between Important Supplier Centrality and Market Value is lower, at 0.24.

Figure 5

On the other hand, Centrality measures exhibit a stronger correlation in spread returns compared to Market Value. Figure 6 shows a correlation matrix for the spread returns across each pair of factors. The spread returns of centrality measures show a correlation of 0.8, whereas the spread return correlation between centrality measures and market value is 0.6. It is noteworthy that the correlation in spread returns between the two centrality measures is higher, even though the overall factor correlation between centrality measures and Market Value is quite similar, meaning that the centrality measures, regardless of customers or suppliers, may capture stock price signals that are not captured by the size of companies.

Figure 6

In fact, backtest results show a significant difference between centralities and the size factor. Figure 7 highlights several key metrics from the backtesting results. While centrality measures show some correlation with the size factor, the factor performances exhibit significant differences. The Important Customer and Supplier Centralities exhibit annualized spread returns of 6.08% and 4.67%, respectively. In contrast, the annualized spread return for Market Value is 1.44%.

Among the sample factors, EBIT Estimate Revision has the highest annualized spread return and Sharpe ratio. However, it also has the highest turnover rates in its fractile portfolios. On the other hand, centralities demonstrate relatively high spread returns and Sharpe ratios while maintaining very low turnover in their fractile portfolios.

Figure 7

Conclusion and Further Research Ideas

We have studied a unique factor derived from supply chain relationships known as the centrality measure. Our findings indicate that this measure may have predictive power regarding stock price performance, particularly during supply chain disruptions. Additionally, our analysis suggests that the centrality measure shares some similarities with the size factor, but its overall performance as a factor is superior to that of the size factor.

While our analysis primarily focuses on the application of a centrality measure as an alpha signal, it has more potential for broader use cases. For example, centrality may be more suited for predicting the volatility of companies during market turbulence, serving as a key risk factor. In addition, utilizing other centrality algorithms may provide additional valuable insights into a company's business that are not readily apparent from conventional factors.

This blog post is for informational purposes only. The information contained in this blog post is not legal, tax, or investment advice. FactSet does not endorse or recommend any investments and assumes no liability for any consequence relating directly or indirectly to any action or inaction taken based on the information contained in this article.