Financialization involves changes in both aggregate financial activity and the relative roles of individual submarkets. Static measures and networks organized around institutions or markets provide limited information about the temporal structure of these changes. This paper develops a dynamic-weight visibility-graph framework that combines a monthly financialization index ( FI ) with endogenous temporal segmentation. Time-varying weights derived from responses of the consumer price index ( CPI ) to submarket innovations in a time-varying parameter vector autoregressive (TVP-VAR) model aggregate activity across six financial markets. The resulting index is mapped into a network whose nodes represent calendar months. Applied to China from January 2002 to September 2024, the framework identifies heterogeneous temporal connectivity and a persistent sequence of financialization regimes. The broad segmentation is retained under alternative index constructions and community-detection specifications, although several estimated boundary dates vary. Historical decomposition identifies differences in the market composition of these phases: equity, real-estate, and credit contributions fluctuate more strongly during several volatile intervals, while bond, insurance, and foreign-exchange contributions are generally smoother and become more pronounced in selected later periods. This chronology traces how shifts in submarket contributions accompany changes in the temporal structure of financialization. Dynamic weighting and visibility graphs trace financialization over time. Monthly network nodes identify endogenous regimes in financialization. Main temporal structure is robust to alternative index and network specifications. Submarket contributions to financialization vary substantially across regimes.
This paper develops a technique for an optimal portfolio allocation that explicitly accounts for different market phases. The basic idea is to distinguish market phases with respect to the dynamics of a sentiment index and to accordingly interpolate among bull, bear, and neutral input portfolios. The algorithm we design is based on optimal transport theory and exploits the notion of Entropic Gromov-Wasserstein discrepancy to generate optimal asset allocation. The resulting optimal portfolio is obtained as the barycenter of the input portfolios with barycentric coordinates that vary according to the different market conditions. Through simulation studies and empirical applications, we demonstrate the effectiveness of this approach. Comparative analysis against alternative portfolios indicates that our proposed technique delivers superior risk-adjusted performance over the full evaluation window, while its year-by-year performance varies, as expected from a dynamic strategy.
To address the persistent challenges of misaligned incentives and value loss during green innovation commercialization, this paper develops a two-stage, five-agent evolutionary game model. Incorporating the government, chain leaders, affiliated enterprises, academic institutions, and consumers, the model systematically explores the sequential dynamic impact of initial value co-creation on subsequent value diffusion. Furthermore, a dual-value-linked revenue-sharing contract is embedded to mathematically align economic returns with environmental objectives. Analytical and simulation results reveal three core findings: (1) Dynamic sales subsidies anchored to environmental benefits significantly enhance multi-agent R D synergy. (2) The incentive efficacy of revenue-sharing contracts is strictly contingent on a cost-coverage threshold; increasing the sharing ratio below this threshold paradoxically crowds out collaborative willingness. (3) Government penalties for substandard products exhibit a dual-threshold effect. Although appropriate penalties are essential for constraining low-quality product provision during value diffusion, excessive regulatory pressure can increase anticipated commercialization losses and weaken the collaborative willingness of upstream innovation participants. Ultimately, this study proposes an integrated governance framework featuring bounded floating penalties and dynamic multiplier incentives, providing operational parameters to mitigate coordination tension and minimize value dissipation during the innovation-to-market transformation.
This study develops a dynamic portfolio framework for multinational enterprises investing in green manufacturing, analyzing how subsidies interact with supply chain completeness (SC) to shape location, production, sales, and reinvestment. We extend a four-stage Stackelberg game to an infinite-horizon stochastic game solved via deep reinforcement learning (PPO), complemented by Bayesian sequential Monte Carlo and generalized random forest. Using 120 MNE projects across 12 countries (2018–2023), we find investment subsidies (> 15
We study daily volatility forecasting and uncertainty quantification for liquid equity and cryptocurrency markets using a conservative hybrid econometric–machine-learning framework. Because the analysis is conducted at the daily frequency, the forecasting target is a daily variance proxy rather than an intraday realized-volatility measure in the high-frequency sense. Specifically, we model the logarithm of the squared-return proxy and combine a heterogeneous autoregressive (HAR) benchmark, which captures multi-horizon persistence through daily, weekly, and monthly components, with an extreme gradient boosting (XGBoost) model built on the same core volatility features and additional return-based covariates. The two point forecasts are then combined through a convex stacking weight estimated on a separate calibration segment. To quantify predictive uncertainty, we construct split-conformal prediction bands around the hybrid forecast and compare two calibration schemes: a standard benchmark based on independent and identically distributed (i.i.d.) absolute residual scores, and a blockwise alternative based on maxima of absolute residuals over non-overlapping short blocks. The latter is designed to improve reliability over consecutive trading days under volatility clustering and serial dependence. Using daily data for six assets—XOM, EFA, XLB, FDX, BRK-B, and BTC–USD—over 2015–2024, we find that HAR is already a strong benchmark and that the stand-alone XGBoost model does not outperform it in average root mean squared error (RMSE). The hybrid HAR+XGBoost specification nevertheless attains the lowest RMSE, or ties the lowest RMSE up to rounding, across all six assets, although the gains are modest (for example, from 2.204 to 2.193 for EFA and from 2.541 to 2.538 for BTC–USD on the log-variance-proxy scale). The main empirical gains arise in the uncertainty-quantification layer. Standard split-conformal bands deliver pointwise coverage broadly near the nominal level (0.914–0.956), but their blockwise coverage over five-day periods is much lower (0.610–0.780). By contrast, block-conformal calibration raises pointwise coverage to 0.970–0.994 and blockwise coverage to 0.860–0.970, albeit at the cost of substantially wider bands. A complementary AR(1)–GARCH(1,1) simulation study confirms the same width–reliability trade-off: block-conformal calibration materially improves blockwise coverage relative to the standard split-conformal benchmark under realistic dependence. Overall, the results show that combining a transparent econometric benchmark with a flexible nonlinear correction and a block-maxima conformal calibration layer yields an auditable framework for daily variance-proxy forecasting that is especially useful when short-run reliability is more important than interval sharpness. Replication code and data access instructions are publicly available in an online repository .
