In the context of global climate governance, green technology innovation is essential for urban sustainable development. To address the limitations of traditional linear econometric models, this study investigates the impact of the National Innovative City Pilot Policy on green innovation using a novel framework combining a Multi-period Difference-in-Differences model and Deep Learning Counterfactual Prediction. Analyzing panel data from 100 eastern Chinese cities between 2004 and 2023, the research reveals that the policy significantly and robustly enhances innovation levels in pilot cities. Furthermore, the policy operates through a dual-track synergistic governance mechanism, successfully combining government scientific and technological support with environmental regulation. Additionally, heterogeneity analysis reveals that the policy exerts a more pronounced driving effect on green innovation in small-to-medium-sized cities and regions with lower industrial upgrading levels. Finally, deep learning counterfactual trajectories demonstrate that the policy dividend exhibits a non-linear, long-term cumulative effect that expands over time—a dynamic that traditional linear models often underestimate. Ultimately, this study provides solid empirical evidence that a combined governance system of incentives and constraints effectively promotes innovation-driven, sustainable urban transitions.

This paper explores how green finance policies influence the sustainability and resilience of urban areas, particularly those operating under tight fiscal budgets. By treating China’s Green Finance Reform and Innovation Pilot Zone as a quasi-natural experiment, we examine a comprehensive panel of 299 prefecture-level cities between 2011 and 2024. The analytical framework integrates traditional difference-in-differences (DID) estimations with machine-learning-driven text analysis. To quantify governmental foresight regarding sustainability risks, the authors employ Word2Vec semantic expansion and NLP tokenization. Additionally, a TextCNN classifier is utilized to group cities by their digital-economy readiness, allowing for a nuanced assessment of heterogeneous policy impacts. The results demonstrate that implementing green finance pilots not only tangibly boosts sustainable productive capacities but also noticeably sharpens official textual focus on sustainability governance. Furthermore, mechanism analyses suggest that these positive outcomes are driven by structural industrial upgrades, enhanced technological expenditures, and the easing of local fiscal burdens. The study’s conclusions—reinforced by rigorous robustness checks like PSM-DID, CS-DID, and spatio-temporal mapping—indicate that green finance functions as a vital governance mechanism capable of stabilizing local finances and driving long-term, low-carbon transitions.

Financial time series in turbulent markets exhibit complex long-memory dynamics and multifractal features that traditional deep learning models fail to capture due to inherent exponential forgetting mechanisms. To address this, we propose Frac-Attn-GL, a novel Fractional-order Spatiotemporal Attention-based GRU-LSTM framework. Grounded in the Fractal Market Hypothesis, the model embeds Grünwald–Letnikov fractional-order operators into a dual-channel architecture (FracLSTM and FracGRU) to characterize long-range memory with rigorous power-law decay priors. Furthermore, an extreme-aware asymmetric loss function is designed to drive a dynamic spatiotemporal routing mechanism, enabling adaptive shifts between long-term macro trends and short-term micro shocks. Empirical tests on major U.S. stock indices reveal three significant findings. First, the Frac-Attn-GL framework substantially reduces prediction errors, achieving up to a 93.1% RMSE reduction on the highly volatile NASDAQ index compared to standard baselines. Second, the adaptively learned fractional-order parameters exhibit a consistent quantitative alignment with the market’s empirical multifractal singularity spectrum, supporting the physical interpretability of the model’s endogenous memory mechanism. Finally, hybrid residual multifractal diagnostics indicate that the framework effectively captures deep long-range correlations, reducing the Hurst exponent of the prediction residuals from ~0.83 to approximately 0.50, a level consistent with the absence of significant long-range dependence.

Forecasting extreme-tail risk in energy equities involves temporal-memory mismatch, quantile crossing, and scarce extreme observations. This study proposes a multi-scale entropy-gated deep quantile learning model, EntropyFusion-VaR. Weighted permutation entropy (WPE) characterizes local ordinal disorder, Rényi entropy describes concentration in the full binned return distribution, and transfer entropy (TE) describes directional predictive information. These state features drive a gate that coordinates a causal dilated TCN and a Patch Transformer. An incremental cumulative output layer ensures ordered joint quantiles at the 1%, 2.5%, and 5% levels. An exponential aggregation of pinball losses gives greater optimization weight to large forecast errors, while a differentiable coverage penalty regularizes exception frequencies. Experiments on EOG Resources (EOG), Marathon Petroleum (MPC), and Schlumberger (SLB) show lower reported average tail loss and zero quantile crossing. The model attains the best mean rank at each studied horizon and ranks first in eight of nine stock–horizon combinations. The evidence supports state-dependent tail-risk forecasting for these three U.S. energy-sector equities, with coverage assessed separately from loss improvement.

Against the backdrop of global climate change, the increasing frequency and intensity of extreme weather events pose unprecedented challenges to the protection of historical landmarks and the insurance industry. To address this issue, this study aims to develop a comprehensive risk assessment model to assist insurance companies in formulating underwriting strategies in regions prone to extreme weather. We first constructed a future climate prediction model based on a Bidirectional Long Short-Term Memory (Bi-LSTM) network to forecast future anomalies in key climate indicators for specific regions, such as temperature, precipitation, and evaporation. Subsequently, we innovatively improved the CRITIC (Criteria Importance Through Intercriteria Correlation) method to objectively determine the weights of various climate factors and, based on this, calculated a comprehensive climate hazard level (L). By simplifying the insurance company’s profit formula, we established a mathematical model for the relationship between the comprehensive hazard level and insurance revenue, deriving a clear decision threshold: insurance companies should cease underwriting when the comprehensive hazard level (L) exceeds the ratio of claim costs to premium income. We applied this model to historical and predicted data for Xigaze, China, and Cairo, Egypt, using visual charts to intuitively display the monthly comprehensive hazard levels for each region. This provides insurance companies with a scientific, quantitative basis for deciding whether to underwrite policies in specific areas and time periods.