Business, Economics and Management

Abstract: During financial crises, markets do not only fall or become more volatile. They may also become more dynamically coupled, with the behaviour of one market becoming more recoverable from another. This study applies Convergent Cross Mapping, a state-space reconstruction method, to examine whether crisis periods strengthen nonlinear coupling between major US and European equity indices. Daily log returns for the Dow Jones Industrial Average, S&P 500, FTSE 100 and DAX are analysed across pre-crisis, crisis and post-crisis windows for the COVID-19 market shock and the Global Financial Crisis. Pairwise bidirectional Convergent Cross Mapping is used to estimate cross-map skill, convergence and directional asymmetry, with a focused lagged analysis of key trans-atlantic pairs during COVID-19. Cross-map skill is interpreted as the strength of the recoverable dynamical footprint between markets. The results show that nonlinear coupling increases during crisis phases. During COVID-19, mean late-library cross-map skill rises from the pre-crisis to the crisis period, and all tested directional relationships satisfy the convergence criterion. The Global Financial Crisis also shows increased cri-sis-period coupling, with stronger persistence into the post-crisis phase. Lagged COVID-19 results suggest a short contemporaneous to three-trading-day coupling hori-zon. The findings position Convergent Cross Mapping as a complementary mathematical modelling framework for identifying recoverable dynamical information between markets during financial stress.

Abstract: Renewable energy communities (RECs) are increasingly recognized as a strategic instrument for enhancing energy system resilience, promoting local renewable integration, and reducing consumer exposure to electricity market volatility. This study assesses the economic performance of RECs relative to individual consumers using high-frequency hourly data from 2021 to 2023, covering both the 2022 European energy crisis and the subsequent Italian regulatory reform of incentive mechanisms. An optimized REC configuration is developed to maximize shared photovoltaic generation and minimize external grid dependence. Through panel econometric analysis, we estimate the sensitivity of economic value to electricity price fluctuations and demonstrate that RECs exhibit significantly lower price dependence than standalone consumers. To complement these findings, machine learning techniques and SHAP (SHapley Additive exPlanations) analysis are employed to capture non-linear dynamics and quantify the relative contribution of electricity prices to value formation. Results consistently show that while market prices remain an important determinant, RECs substantially attenuate their impact, particularly during periods of extreme price stress. A policy counterfactual comparison between pre- and post-reform incentive structures further indicates that the revised regulatory framework introduces a more stable and counter-cyclical compensation mechanism, strengthening the protective role of RECs. Overall, the study provides robust empirical evidence that renewable energy communities function not only as instruments for renewable deployment but also as effective mechanisms for economic stabilization under volatile market conditions.

Abstract: Expected Shortfall (ES) is a tail functional whose estimation precision is governed by the effective tail sample size nα rather than by the nominal calibration size n. The resulting (nα)−1/2 information limit is well established, yet no practical framework exists for deciding whether two ES forecasts can be meaningfully distinguished over a finite calibration window. This paper converts the asymptotic rate into four operational diagnostics: a plug-in precision benchmark, a sample-size rule, a precision-fragile pairwise comparison screen, and a VaR-first diagnostic linking excess ES dispersion to first-stage quantile miscalibration. An empirical application to global financial assets and heterogeneous forecasters under standard regulatory tail parameters shows that roughly one in five pairwise ES comparisons is precision-fragile, with excess dispersion concentrated in cells with poor VaR calibration. The results suggest that ES forecast rankings at typical tail levels can be constrained by effective tail information rather than by model sophistication.

Abstract: This study investigates the dynamics of white maize price transmission across six regional markets in Mozambique (Manica, Gorongosa, Mutarara, Montepuez, Ribáuè, and Lichinga) and examines their relationship with global energy prices (oil and gas) between 2003 and 2020. The analysis applies wavelet coherence techniques, the Augmented Dickey–Fuller (ADF) test, Johansen cointegration, Vector Error Correction Models (VECM), and Granger causality tests to evaluate both short- and long-run market integration. The results indicate that all series are integrated of order one (I(1)) and that significant long-run cointegration relationships exist among the variables, with three cointegration vectors identified according to the FPE and AIC criteria. Wavelet coherence analysis reveals strong long-run synchronization (1.2–8 years) among regional white maize markets, while short-run coherence (0.1–1.2 years) remains weak, suggesting limited short-term integration. The findings further show that global energy prices do not exert a persistent structural influence on maize price dynamics in Mozambique. The VECM results identify Gorongosa as the main transmission hub of short-run price shocks, exerting positive effects on the other regional markets, whereas oil prices negatively affect Gorongosa and Ribáuè. Granger causality analysis confirms that Gorongosa, Mutarara, and Ribáuè act as the primary price transmitters within the system, while oil prices are mainly driven by exogenous factors. Overall, the findings demonstrate that Mozambican white maize markets are more strongly integrated in the long run than in the short run, highlighting the strategic role of Gorongosa in regional price transmission and suggesting that energy shocks have limited long-term effects on the domestic maize market system.

Abstract: Credit risk prediction is a significant challenge in modern financial systems due to the dynamic and nonlinear nature of borrower behavior. This study introduces a Bayesian-Optimised Hybrid DeepSurv–LSTM framework for dynamic credit risk forecasting, integrating survival analysis with temporal deep learning methodologies. The framework combines DeepSurv for hazard modeling and Long Short-Term Memory (LSTM) networks for analyzing borrower repayment behavior, using Bayesian optimization to identify optimal hyperparameters and enhance model generalization. It was assessed using borrower-level financial data, including demographic, behavioral, and transactional variables. Results showed that the Bayesian-optimised DeepSurv–LSTM model outperformed XGBoost, standalone DeepSurv, and standalone LSTM models across classification and survival-analysis metrics. The hybrid model achieved a C-index of 0.8617, ROC-AUC of 0.9726, accuracy of 94.83%, F1-score of 0.9197, and the lowest Integrated Brier Score of 0.1293. Statistical validation confirmed the significance of these improvements. The findings suggest that integrating survival-aware hazard modeling with temporal deep learning enhances credit default prediction and provides a robust framework for financial risk management and early credit risk monitoring in dynamic banking environments.

Abstract: Financial crises are usually identified through drawdowns, volatility and changes in returns, but these indicators do not fully describe changes in the underlying dynamical structure of markets. This study tests whether Laminarity, a measure derived from Recurrence Quantification Analysis, can provide a complementary indicator of financial market stress during the COVID-19 shock. Daily data for the Dow Jones Industrial Average, S&P 500 and NASDAQ Composite from 2018 to 2022 are analyzed using adjusted prices and log returns. Rolling-window Recurrence Quantification Analysis is applied across alternative window lengths and recurrence thresholds, and the resulting Laminarity measures are compared with conventional benchmarks including drawdown and rolling volatility. The results confirm that the COVID-19 crisis is clearly identified by conventional risk indicators, while Laminarity provides a more nuanced and parameter-sensitive signal. Price-based Laminarity generally increases during the COVID-19 stress period, suggesting a more persistent crisis trajectory, whereas return-based Laminarity produces mixed evidence, including some cases of Laminarity loss depending on index and window length. The findings indicate that Laminarity should not be interpreted as a universal or mechanical crash-warning signal, but as a complementary diagnostic measure that can help describe changes in market-regime structure during periods of acute stress.

Abstract: Emerging-market equity exchanges require regime forecasting systems that are continuous in time, robust to heavy-tailed distributions, and optimised against false alarms. No existing method addresses all three simultaneously, and no prior study has reported a crisis false alarm rate on JSE equities. We propose S-NODE-ANF-RRC: a Stochastic Neural ODE embedded within an Adaptive Neuro-Fuzzy Risk-Regime Clustering architecture, motivated by the Heston stochastic volatility framework and integrated by a Milstein scheme with Lyapunov-regularised dual-loss training. The system is evaluated as a one-step-ahead probabilistic forecaster (h = 1 trading day) on 2696 daily observations across 17 JSE securities (March 2015–March 2026). Gaussian mixture clustering on raw features (kurtosis 54.8) inflates ARI by 1.3×; log-transformation corrects this systematic artefact. Two operational profiles emerge after correction: the N-ODE-ANF-RRC achieves the lowest cost (10,350 bp, 65.1% below GMM), longest lead time (0.71 days), and best MCC (0.596); the S-NODE-ANF-RRC achieves the lowest false alarm rate among probabilistic architectures (FAR = 0.051, log-loss = 1.07), with a 42.0% cost reduction versus GMM (bootstrap 95% CI [5, 250, 19, 600] bp; McNemar p = 0.027). Ablation confirms drift, diffusion, and dual-loss as the minimum viable daily-frequency configuration. The interdisciplinary fusion of physics-informed SDE dynamics, time series forecasting, and fuzzy interpretability yields two complementary JSE risk tools: an early-warning forecaster (N-ODE) and a low-false-alarm crisis classifier (S-NODE). Code and data: https://doi.org/10.5281/zenodo.19787658.

Abstract: This paper analyzes volatility forecasting in the Spanish electricity spot market over the period 2021–2025, characterized by uncertainty, frequent price jumps, and the increasing occurrence of zero and negative prices. To accommodate these features, electricity prices are shifted to ensure welldefined logreturns, and predictable intraday and seasonal patterns are removed using the Ullrich demeaning procedure. Daily realized volatility measures are constructed from highfrequency data, including jumprobust and noiserobust estimators such as Median Realized Volatility and Realized Kernel. A broad set of volatility models, comprising GARCHtype specifications and multiple extensions of the Heterogeneous Autoregressive (HAR) framework, is evaluated using a coherent outofsample forecasting procedure. Model comparison is conducted through the Model Confidence Set methodology based on the QLIKE loss function, which identifies a Superior Set of Models with equal predictive ability. Conditional diagnostics, including OutofSample ROOS2measures and Mincer–Zarnowitz regressions, are subsequently used to characterize forecast accuracy, unbiasedness, and efficiency. The empirical results show that all GARCH models are systematically excluded from the superior set, while HARtype specifications based on realized volatility dominate. Within this set, a HAR model incorporating Median Realized Volatility, jump components, and dayoftheweek effects delivers the strongest economic performance, achieving an OutofSample ROOS 2close to 0.5 with unbiased forecasts. Overall, the findings highlight the importance of longmemory dynamics, discontinuous price movements, and residual weekly seasonality for volatility forecasting in modern electricity markets.

Abstract: The most widely used method for trend estimation in economics is the Ho-drick-Prescott (HP) filter. The HP filter has various disadvantages as the arbitrary and frequency-dependent choice of the smoothing parameter λ, boundary problems and difficult interpretation when linking to economic theory. We suggest an alternative method by improving some of these disadvantages using a purely data-driven, endog-enous nonparametric trend estimation. A simulation study and different applications demonstrate the advantages of the nonparametric trend compared to the HP filter. We identify optimal time windows supporting the momentary growth trend. Within this window economic fundamentals smoothly change and drive the trend.

Abstract: Achieving carbon neutrality has become a central policy objective for emerging economies, particularly the BRICS countries-BRICS (Brazil, Russia, India, China, and South Africa) which collectively account for a substantial share of global carbon emissions and energy consumption. The transition toward green energy, rapid technological innovation, and the expansion of green finance mechanisms are increasingly viewed as critical drivers of sustainable development and environmental improvement. However, empirical evidence integrating these three dimensions within a unified analytical framework for BRICS remains limited. This study examines the contribution of green energy transition, technological innovation, and green finance to achieving carbon neutrality in BRICS countries using a Pooled mean group auto regressive distributed Lag (PMG ARDL) framework and Dumitrescu–Hurlin panel causality analysis. The results indicate that green energy transition significantly reduces carbon emissions in both the long run (−0.45) and short run (−5.65), emphasizing the importance of shifting toward renewable energy sources. Technological innovation exerts a significant negative effect in the long run (−0.17), reflecting efficiency gains and cleaner production, although its short-run impact remains insignificant. Similarly, green finance improves environmental quality in the long run (−0.10) by supporting low-carbon investments, while short-run effects are statistically insignificant due to adjustment frictions. Economic growth increases emissions in the long run (0.43), confirming the scale effect, whereas trade openness reduces emissions (−0.87), indicating the role of technology diffusion. The error correction term (−0.76) confirms a strong convergence toward long-run equilibrium. The causality analysis reveals unidirectional causality from green energy transition, technological innovation, and green finance to carbon emissions, while bidirectional causality exists between economic growth and emissions, highlighting a feedback mechanism. Policy implications suggest that BRICS economies should strengthen green financial systems, accelerate renewable energy adoption, promote innovation-driven sustainability, and design growth strategies that decouple economic expansion from environmental degradation.

Abstract: This paper revisits and extends the machine learning framework for U.S. recession prediction introduced by Yazdani2020 by incorporating post-pandemic macroeconomic dynamics, an expanded predictor set and machine learning models. Using monthly data from January 1959 through December 2024, recession forecasting is formulated as an imbalanced binary classification problem. We use downsampling for static models and class-weighted loss functions for neural networks and evaluate model performance using classification metrics robust to rare events. We further examine structural stability across four economic regimes and assess economic value through a dynamic stock–bond allocation strategy. We observe that ensemble tree methods, particularly gradient boosting (XGBoost, LightGBM) and random forests, consistently deliver the strongest discrimination, with out-of-sample AUC above 0.99 and PR-AUC above 0.96. The Transformer achieves probability calibration, and Deep sequence models exhibit high discrimination, while performance deteriorates across model classes in the 2020–2024 regime, especially for linear specifications. We also examine risk-adjusted returns of models. Overall, ensemble trees and Transformers show high predictive power and emerge as complementary tools in macroeconomic recession forecasting.

Abstract: Urban eco-efficiency (UEE) is fundamental to achieving China's dual-carbon goals. However, literature has overlooked green space carbon sequestration, and linear models fail to capture complex nonlinear relationships. This study integrates green space carbon sinks into the evaluation framework, employing the global super-efficiency EBM model to measure the UEE of 108 cities in the Yangtze River Economic Belt (YREB) from 2012 to 2023. It combines XGBoost-SHAP with Geographically and Temporally Weighted Regression (GTWR) to examine UEE's spatiotemporal dynamics and driving mechanisms. The findings reveal that: (1) UEE in the YREB increased from 1.0760 in 2012 to 1.0990 in 2023, while spatial polarization became more pronounced. (2) Core driving factors exhibited significant nonlinear threshold and interactive effects. Specifically, fiscal decentralization's environmental dividend is contingent on active government intervention to circumvent localized "race to the bottom" behaviors. Furthermore, population density transitions from yielding scale dividends to inducing "crowding effects" beyond optimal capacities—a degradation advanced financial systems appear unable to mitigate. (3) A spatiotemporal misalignment was observed: fiscal decentralization unleashed green institutional dividends downstream (coefficients up to 0.0682), but caused a race to the bottom in middle and upper reaches (extending to -0.6548); excessive population agglomeration in megacities induced a crowding effect eroding early pollution control dividends. This study supports abandoning one-size-fits-all approaches and developing precise, spatiotemporally differentiated low-carbon policies.

Abstract: This paper derives closed-form expressions for the asymptotic covariance matrices of factor loading and uniqueness estimators obtained from several widely used factor extraction methods, including least squares, principal factor, iterative principal component, alpha factor, and image factor analysis. By treating factor solutions as implicitly defined estimators, the proposed framework characterizes the asymptotic behavior of factor loadings and uniquenesses as explicit functions of the asymptotic covariance matrix of the sample covariance or correlation matrix. This approach avoids reliance on likelihood-based information matrices, numerical differentiation, and resampling methods. Consequently, valid statistical inference is feasible under non-Gaussian sampling, serial dependence, and conditional heteroskedasticity, and can be implemented using heteroskedasticit- and autocorrelation-robust or other sandwich estimators of second moments. The framework naturally accommodates applications in which factor analysis is applied to residual covariance matrices arising from multivariate regressions, panel data models, or structural vector autoregressions (SVARs). Monte Carlo simulations demonstrate accurate finite-sample performance, and an empirical illustration shows how the proposed formulas can be implemented in practice. From an econometric perspective, the results are particularly relevant for settings in which factor structures serve as intermediate objects---such as dynamic factor models, factor-augmented regressions, and SVARs---allowing uncertainty in factor estimates to be coherently propagated into impulse response functions, forecast-error variance decompositions, and other nonlinear functionals used in structural inference.

Abstract: The purpose of this investigation was to analyze the effect of financial structure and fruit-fly control on the development of Small and Medium Enterprises (SMEs) of citrus in the Central Jungle of Peru. Using a quantitative design and a balanced sample of 54 observations, the analysis estimates complementary linear models with interaction terms and restricted cubic spline specifications to capture direct, synergistic, and nonlinear effects. The baseline results show that both financing structure and fruit-fly control exert positive and statistically significant effects on business growth. The interaction term is also positive and significant, indicating that the returns to improved financing rise when phytosanitary management is stronger, and that effective pest control becomes more productive when firms operate with more stable and diversified financial resources. Flexible spline estimates further reveal that these relationships are not constant across the explanatory range, but vary according to firms’ positions within the financial and technological space. Overall, the findings suggest that sustainable growth in citrus SMEs depends on the simultaneous strengthening of rural finance and phytosanitary capabilities under conditions of production risk and market constraints. The study contributes to the agricultural development literature by linking crop protection, farm-level managerial capacity, and enterprise performance in a single empirical framework.

Abstract: This study looks into the predictive performance of linear econometric and deep learning methodologies for the South African unemployment rate quarterly data. In this paper, the Autoregressive Integrated Moving Average with exogenous variables (ARIMAX) model was compared to the Long Short-Term Memory (LSTM) network using unemployment rate quarterly data. Exploratory Data Analysis (EDA) suggested that the unemployment rate series is non-stationary, with structural breaks around 2020 and time-varying volatility. Stationarity tests established the need for differencing, whereas diagnostic tests revealed the presence of autocorrelation and ARCH effects in the raw data. The ARIMAX model added labour market covariates, and the differenced Not Economically Active (NEA) variable was statistically significant, whereas Discouraged workers were not. Although the ARIMAX model provided a good in-sample fit, residual diagnostics showed deviations from normality. Out-of-sample forecast study revealed moderate predictive accuracy, with relatively substantial forecast errors and increasing prediction intervals over time. In contrast, the LSTM model showed significant learning capacity, with early convergence and well-behaved residuals that meet both independence and homoskedasticity criteria. The model achieved significantly lower forecast errors, with RMSE, MAE, and MAPE values much lower than those of the ARIMAX model. Comparative forecast analysis using Diebold-Mariano (DM) test and model confidence Set (MCS) method and bootstrap confidence intervals consistently demonstrated the statistical superiority of the LSTM model. The findings give strong evidence that the LSTM model outperformed the ARIMAX model for projecting South African unemployment rate. The findings emphasise the importance of nonlinear modelling approaches in capturing complex labour market dynamics while also demonstrating the limitations of classic linear models. These findings also emphasise the importance of using nonlinear machine learning algorithms in macroeconomic forecasting.

Abstract: The increasing use of large language models (LLMs) in enterprises creates a need for the effective selection between lower-cost models and more advanced ones. The aim of the article is to propose a multicriteria decision-making framework for prompt routing to LLMs in an enterprise environment, taking into account organizational preferences regarding cost, response quality, business risk, response time, standardization, and creativity. The study adopts a design-and-evaluation approach. In the design phase, a mechanism was developed in which prompts are assessed according to managerial routing criteria, weighted using the AHP method, and then directed to either a lower-cost or a more powerful model using the SAW method. In the evaluation phase, the solution was tested on a dataset of 100 business prompts and compared with two benchmark strategies: always cheap and always strong. The article’s contribution includes framing LLM routing as a managerial decision-support problem, operationalizing managerial routing criteria, and proposing evaluation metrics such as sufficiency rate, average cost per prompt, cost per sufficient response, and incremental cost of sufficiency gain. The results indicate that the proposed solution improves the cost–quality trade-off, while maintaining an acceptable level of response sufficiency and limiting the cost of query handling.

Abstract: The global copper market is experiencing a period of fundamental structural volatility, guided by supply chain realignments, geopolitical friend-shoring, and an increasing reliance on the circular economy. To accurately diagnose the current state of this critical mineral, this paper presents a strictly empirical, data-driven algorithmic pipeline, the Apex Empirical Model, applied to recent UN Comtrade transaction ledgers (2020-2025). By utilizing robust machine learning architectures, this research systematically identifies a phenomenon we term Stage-Specific Starvation (SSS) across the upstream, midstream, circular, and downstream stages of the value chain. Integrating Deep Autoencoders, Network Graph Analysis, Holt-Winters Time-Series Forecasting, and Risk-Parity Optimization, the model successfully isolates targeted capital flight via transfer mispricing and maps the exact flow-through volumes of global transshipment hubs. Furthermore, the framework applies network topology to assess systemic vulnerabilities, empirically confirming the existence of a geopolitical price premium, and engineers a continuous mass-balance metric to predict projected smelter capacity adjustments six months into the future. Finally, our resilience metrics mathematically prove the financial arbitrage and stability advantages of secondary scrap integration. Ultimately, this research leverages Causal Inference to introduce Circular Risk Parity (CRP), providing a prescriptive, optimized portfolio allocation that balances risk equally across the supply chain, allowing stakeholders to navigate exogenous supply shocks in the modern copper market.

Abstract: A locally parametric framework is proposed for Monte Carlo simulation of electricity prices that jointly reproduces the key stylized facts of power markets: mean-reversion, fat tails, asymmetry, and volatility clustering. Following a two-stage pipeline in which mean-reversion is estimated separately from the innovation distribution, the paper focuses on the second stage: simulating the residual innovations via topological conditioning on Natural Visibility Graphs (NVG) built on the observed innovation sequence. At each simulation step, the local structure of the graph is used to identify historically similar market states and to draw the next innovation from a locally fitted distribution. The key methodological contribution is that this topological conditioning mechanism simultaneously determines the local scale, skewness, and tail weight of the innovation distribution — three properties that parametric models such as GARCH must address through separate equations — without any assumption on regime dynamics or transitions. The framework is locally parametric: the number of model parameters grows with the sample size rather than being fixed in advance, and the specific distributional family used as a local working model can be replaced without altering the conditioning mechanism. Applied to two power markets with contrasting distributional characteristics — the Italian Power Exchange (PUN) and PJM West Hub (US) — the framework achieves simultaneous coverage of three distributional statistics (\( \hat\sigma \), \( \hat\gamma, \hat\kappa \)) and the first-order autocorrelation of squared innovations \( \hat\rho_1(\varepsilon_t^2) \) for both markets, with a single neighbourhood size k=10 and no market-specific re-calibration; more generally, k serves as the natural adaptation parameter for markets with substantially different distributional characteristics.

Abstract: Background: Small-scale retail kiosks commonly deploy transactional point-of-sale (POS) systems that capture sales data but lack integrated analytical and forecasting capabilities for operational decision support. This gap limits the ability of small-and-medium enterprise (SME) operators to respond proactively to demand fluctuations. Methods: This study presents the structured analysis, design, implementation, and evaluation of a cloud-deployed self-service kiosk system embedding interactive analytics and demand forecasting modules. The system integrates a Django-based backend, a PostgreSQL relational database, RESTful APIs, a structured demand simulation engine, and three forecasting models: Seasonal ARIMA (SARIMA), XGBoost Regressor, and Scikit-Learn Gradient Boosting Regressor. Forecasting performance was evaluated using rolling backtesting with Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). Results: The Gradient Boosting Regressor achieved the highest predictive accuracy (MAE = $93.74, RMSE = $112.65, MAPE = 8.9%), outperforming both XGBoost (MAPE = 10.0%) and SARIMA (MAPE = 10.2%). The proposed architecture demonstrates that Systems Analysis and Design principles can guide the development of an integrated decision-support platform for small retail environments. Machine learning ensemble models more effectively capture nonlinear demand patterns generated by growth and seasonality dynamics than classical time-series models. The system is deployed as a proof-of-concept cloud application accessible at the address listed in the Data Availability section.

Abstract: The flow of R&D factors serves as a crucial channel for linking regional collaborative innovation resources and plays a significant role in promoting spatial knowledge spillovers, making it an important engine for the in-depth implementation of innovation-driven development strategies. This study takes the Yangtze River Delta urban agglomeration as its research object, utilizing data from 2003-2023, and employs gravity models and dynamic spatial fixed effects models to analyse the impact of R&D factor flows on regional collaborative innovation, as well as the moderating role of intellectual property protection. The study revealed that both the flow of R&D personnel and R&D capital significantly promote regional collaborative innovation, with the flow of R&D capital playing a more prominent role. The intensity of intellectual property protection positively moderates the relationship between the flow of R&D factors and regional collaborative innovation, but a single threshold effect exists, where the moderating effect weakens after exceeding the threshold. The intensity of inter-city collaborative innovation continues to increase, with core cities such as Shanghai, Hangzhou, and Nanjing playing a significant leading role. The emergence of new central cities such as Nantong, Ningbo, and Jiaxing has driven the evolution of collaborative innovation toward a "star-shaped" structure. The mechanism for the flow of R&D factors should be optimized, the intensity of intellectual property protection should be balanced, collaborative innovation between core cities and emerging central cities should be strengthened, and regional innovation infrastructure construction should be enhanced to promote high-quality innovative development in the Yangtze River Delta region.