Climate Volatility and Inward Spillover Exposure: Topology, Sector Heterogeneity, and Persistence

Technical Summary: Climate Volatility and Inward Spillover Exposure

Problem and Research Gap
This paper investigates whether climate-related uncertainty systematically intensifies inward spillover exposure—the vulnerability of financial nodes to shocks transmitted from other parts of an interconnected network. While existing literature extensively documents how climate risk affects asset prices, volatility, and total systemic risk, a critical gap remains regarding the directional nature of these risks. Much of the connectedness literature focuses on outward transmission (who transmits stress) or total spillovers, paying less attention to inward exposure (who absorbs stress). Furthermore, prior studies often treat climate risk as a uniform, system-wide correlate or focus on long-run climate levels and policy uncertainty, rather than short-run climate volatility as a driver of instability.

The central research question is conditional: Under what specific network and sector conditions does climate volatility become financially relevant for inward spillover exposure? The paper posits that climate risk may not operate through a uniform pooled effect but rather through interactions with network topology, sector heterogeneity, and the persistence of prior spillover states.

Methodology
The study utilizes a monthly panel dataset from 2011 to 2025, covering eight sector-region nodes: Banking, Insurance, Energy, and Utilities in both North America and Europe.

• Data Sources: Financial data comprises daily adjusted closing prices of ETFs (converted to monthly series) from Yahoo Finance. Climate data consists of monthly global temperature anomalies from NASA GISTEMP. Macro-financial controls include policy rates, unemployment, industrial production, the VIX, and oil prices from the Federal Reserve Bank of St. Louis (FRED).
• Dependent Variable: The primary measure is GARCH-based inward spillover exposure (SpilloverIN), constructed from rolling Vector Autoregression (VAR) forecast-error variance decompositions (FEVD) applied to the log conditional variances of sector-region returns. A realized-return spillover measure serves as a robustness check.
• Key Independent Variable: The primary climate regressor is temp_vol12_z, the standardized 12-month rolling volatility of global temperature anomalies. Alternative indicators include temperature surprise and an entropy-weighted climate index.
• Network Metric: The study constructs a conditioning variable, Net_sd_centrality_L1, representing the one-period lag of the cross-sectional standard deviation of node-level net directional spillovers. This measures the unevenness of the network structure (i.e., the dispersion between net transmitters and net receivers).
• Empirical Strategy:
  • Baseline: Fixed-effects panel models with node-specific intercepts.
  • Climate Treatment: Climate volatility is modeled using an orthogonalized distributed-lag (ODL) framework (Almon lag structure) to capture dynamic adjustments over a 12-month horizon while mitigating multicollinearity.
  • Inference: Standard errors are calculated using the Driscoll–Kraay method to account for cross-sectional dependence and serial correlation.
  • Extensions: The analysis includes sector-specific estimations, interaction terms between climate volatility and network dispersion, regime splits (high vs. low climate volatility), placebo tests (using outward spillovers), persistence-augmented models (including lagged SpilloverIN), and a nonlinear XGBoost benchmark.

Key Results

1. Weak Pooled Association: In the baseline pooled models, the distributed-lag block for climate volatility is only weakly significant. The cumulative implied association is small and positive, suggesting that climate volatility is not a uniform, stand-alone driver of inward spillover exposure across all nodes and periods.
2. Sectoral Heterogeneity (H1): Significant heterogeneity exists across sectors.
   • Banking and Insurance: Show positive and significant joint associations with climate volatility.
   • Energy: Exhibits a weaker but still significant joint test (F=2.82, p=0.03) with a negative cumulative implied association (-0.02). However, under Driscoll–Kraay inference, none of the individual climate coefficients for Energy are statistically significant, indicating that while the sectoral block is jointly significant, the specific timing and sign of the climate association differ from the financial sectors.
3. Network Topology Conditioning (H2): The association between climate volatility and inward spillover exposure is stronger when network dispersion is higher. The interaction term between lagged climate volatility and Net_sd_centrality is positive and highly significant. This implies that climate uncertainty translates into greater vulnerability specifically when spillover positions are unevenly distributed (i.e., when the network is more fragmented or concentrated).
4. Volatility vs. Levels (H4): The volatility-based climate proxy (temp_vol12_z) demonstrates stronger explanatory power than alternative indicators (temperature surprise or composite indices), which are largely insignificant in pooled specifications.
5. Regime Dependence (H5): The relationship is state-dependent. Climate volatility is positively significant in low-volatility regimes but becomes insignificant in high-volatility regimes, suggesting its signal is absorbed by broader market stress during turbulent periods.
6. Dominance of Persistence and Structure: When models explicitly control for persistence (lagged inward spillover exposure), the direct climate association largely disappears. The lagged inward spillover term is overwhelmingly dominant (coefficient ~0.88–0.91), followed by network dispersion. The XGBoost analysis confirms this ordering, with lagged spillover being the primary predictor, followed by network structure, while climate lags add only marginal predictive value.

Key Contributions

• Shift to Inward Exposure: The paper extends the connectedness literature by focusing on inward spillover exposure as a distinct dimension of systemic vulnerability, rather than solely on outward transmission or total connectedness.
• Conditional Mechanism: It provides evidence that climate-related financial vulnerability is conditional rather than uniform. The relevance of climate volatility depends on the configuration of the financial network (topology) and sector-specific sensitivities.
• Volatility as a Signal: It argues that climate volatility (instability) is a more informative indicator for spillover dynamics than climate levels or policy uncertainty, particularly in capturing short-run repricing and expectation revisions.
• Persistence as a Driver: The study highlights that the persistence of prior spillover states is a more stable determinant of current vulnerability than contemporaneous climate shocks.

Significance and Claims
The paper claims that climate-related financial risk should not be interpreted as a stand-alone warning signal. Instead, its relevance is network-conditioned. The findings suggest that climate volatility becomes financially consequential primarily when it interacts with:

1. Uneven network structures (high dispersion in spillover positions).
2. Sector-specific exposure channels (e.g., borrower quality for banks, underwriting for insurers).
3. Pre-existing states of vulnerability (persistence of inward spillovers).

For policy and practice, the paper argues that supervisory monitoring and stress testing should integrate climate indicators with measures of connectedness and spillover-position dispersion. A "network-aware" approach is necessary to identify where climate uncertainty is most likely to translate into incoming vulnerability, rather than relying on aggregate, pooled climate signals. The study concludes that climate risk is best understood as a process of how vulnerability is distributed across interconnected sectors, rather than a homogeneous system-wide shock.