The research, led by Li Zeng of Henan Polytechnic University, addresses a gap in climate finance. Most existing studies of climate risk examine individual assets, industries, or institutions, asking how climate exposure shows up in prices or in the risk of a single firm. Far less attention has been paid to how market signals from many firms aggregate into collective financial structures — and whether those structures reorganize when the external climate environment shifts. By moving the analysis to the city scale, the team created an intermediate level that links the behavior of individual companies to the system-wide organization of energy finance.

Building the network required a careful data pipeline. The researchers identified 248 listed energy companies spanning conventional energy production and processing, electricity and heat supply, gas distribution, renewable energy equipment, and related businesses. Each company was mapped to one of 103 cities, and within each city the firms were treated as a value-weighted portfolio, with weights set by total market capitalization. The daily return of each city portfolio then became the basic signal for that node. For every rolling window of 60 trading days, the team computed Pearson correlations among all city return series, ranked the positive correlations, and retained only the strongest 5 percent as the network’s backbone. The result is an undirected, weighted network in which edges represent statistical comovement in city returns rather than directional flows of money or contagion.

Over the sample’s 1,212 trading days, this procedure produced 57 consecutive rolling network states, each matched to the average level of a national daily climate stress index over the same window. That climate index was itself constructed from 52 constituent climate index series drawn from the Wind database, each standardized and direction-aligned so that higher values consistently indicate stronger stress, then combined with normalized weights into a composite measure. The authors emphasize that this index summarizes common variation across a national set of climate indices; it is not a single meteorological observation, and it does not capture differences in physical climate exposure between individual cities.

The headline result concerns connection strength. In baseline regressions that control for a linear time trend and use Newey-West heteroskedasticity and autocorrelation-consistent standard errors, mean climate stress was positively associated with mean edge weight (a coefficient of 0.1142, p below 0.001) and mean weighted degree (0.8206, p below 0.001). The association with mean clustering coefficient was also positive and significant, while network density showed a weaker positive association. Because the network construction rule fixes the proportion of retained edges in every window, an increase in mean edge weight is a particularly direct signal: the strongest city-to-city correlations genuinely intensified rather than merely multiplying in number.

A complementary comparison of climate states sharpened the picture. The researchers split the 57 rolling windows into 15 low-stress states (bottom quartile of the climate stress distribution), 15 high-stress states (top quartile), and 27 intermediate states. Mean edge weight and mean weighted degree were significantly higher in the high-stress group, and so were network centralization and the largest connected component ratio — the latter two measures, notably, had shown no significant association in the continuous regressions. This layered pattern suggests that connection strength adjusts more smoothly with climate stress, while some global topological properties change in a more state-dependent fashion, becoming visible only when distinct climate regimes are set side by side.

The network’s persistent architecture emerged clearly from the longer-term analysis. Pooling the annual top-5-percent networks from 2021 to 2025 produced an aggregate network of 610 unique city pairs whose largest connected component contained 91 of the 103 cities. Within that structure, a small group of cities combined high PageRank scores with strong weighted connections and dense local embedding, while a larger population of cities hung on the periphery through fewer or weaker links. A greedy modularity algorithm identified four communities of 45, 28, 10, and 8 cities, with an observed modularity of 0.3711 — well above the mean of 0.1992 obtained from 500 degree-sequence-preserving null networks, an empirical p value of 0.0020. Community assignments also proved stable under 200 repetitions of random node removal. K-core decomposition added a nested hierarchy: the innermost 16-core contained 17 cities, with the nested shells contracting progressively from 91 cities at the 1-core level down to 47 at the 11-core.

Position within this hierarchy mattered for how cities responded to climate stress. In each rolling network, cities in the upper quartile of the contemporaneous PageRank distribution were classified as core nodes and the rest as peripheral — a classification that shifted over time, with a mean Jaccard similarity of 0.450 between adjacent windows. Both groups showed significant positive associations between climate stress and weighted degree, but the core-periphery difference in weighted degree was itself significantly associated with climate stress (coefficient 2.3360, p equal to 0.0385), indicating deeper weighted embedding of core cities. The corresponding difference in clustering was not statistically significant, so the evidence supports heterogeneity in weighted embedding without establishing a general asymmetry in local organization. Notably, peripheral nodes also participated actively in network adjustment, with significant increases in their weighted connectivity and local clustering.

The findings survived an extensive battery of robustness checks. Changing the backbone threshold to the strongest 4 percent or 6 percent of positive correlations left the weighted measures significantly positive. So did shortening the rolling window to 40 trading days or lengthening it to 120. Restricting the city sample to cities with at least two, and then at least three, sampled energy firms — reducing coverage to 39 and 20 cities respectively — preserved the significant positive associations for mean edge weight and mean weighted degree, though density and clustering became less stable, suggesting those measures are more sensitive to node composition. Fixed absolute correlation thresholds of 0.6183, 0.6683, and 0.7183 yielded similar conclusions, and a Benjamini-Hochberg false discovery rate adjustment continued to support the associations for edge weight, weighted degree, and clustering, while weakening the evidence for centralization and the largest connected component ratio.

Several caveats frame the results. Controlling for the return and volatility of the Shanghai Composite Index did not eliminate the climate association — mean edge weight retained a coefficient of 0.0862 and mean weighted degree 0.5580, both with p values below 0.001 — but the authors stress that their design identifies conditional associations, not causal effects. No external instrument or source of exogenous climate variation was used, and omitted variable concerns cannot be fully excluded. A joint model including lagged climate stress found no stable delayed associations, pointing to a contemporaneous relationship, and a comparison of extreme climate windows (the top 10 percent of stress, only six observations) found positive but statistically insignificant differences across all six network measures, likely reflecting limited statistical power. The authors also note that city returns in cities with few listed firms may be dominated by individual companies, and that the national climate index cannot capture city-specific climate exposure. Future work, they suggest, could incorporate directed network methods, localized climate indicators, and longer time series to test whether nonlinear network changes under extreme conditions might serve as early warning signals. For now, the study offers a clear empirical message: climate stress leaves its mark not only on the risk of individual energy assets but on the collective architecture of the financial system itself.

Subject of Research: Association between climate stress and the topology of China's urban energy financial network

Article Title: Collective coupling and topological reorganization under climate stress: evidence from China’s urban energy financial network

Article References: Zeng, L., Tang, N., Zhao, L., Liu, X., & Li, R. (2026). Collective coupling and topological reorganization under climate stress: evidence from China’s urban energy financial network. Frontiers in Physics, 14, 1980569. https://doi.org/10.3389/fphy.2026.1980569

Image Credits: AI Generated

DOI: 10.3389/fphy.2026.1980569

Keywords: climate stress, energy finance, complex networks, China, systemic risk, topological reorganization, urban financial network, econophysics, stock returns, network coupling, climate risk, financial stability