We construct a climate risk (CR) index based on the autoencoder, a deep learning method to merge the features from multiple proxies, for examining the interaction among green assets (GAs) under varying market states via the QVAR model from the time-frequency perspective. We further introduce the news sentiment (NS) and economic policy uncertainty (EPU) as moderators to investigate the role of instant information and persistent policy in shaping the effect. We find that (i) in the time-domain aspect, CR reduces the interaction among GAs in the normal market, with NS and EPU exacerbating its impact, while in the bear (bull) market, CR dampens (strengthens) their co-movement, with NS and EPU acting as the accelerator and decelerator, respectively; (ii) from the frequency-domain viewpoint, CR's negative impact becomes positive from the short to long term, indicating that the information lasting for more than five days promotes the interaction, and moreover, NS acts as an accelerator in the short run and weakens CR's divergent effect, whereas EPU functions as a stabilizer in the long run and enhances (attenuates) the tightness of GAs in the bear (bull) market; and (iii) individually, the green cryptos (stocks) primarily appear to be the risk senders (receivers), whereas the green bonds and carbon allowances remain relatively isolated, and CR boosts (curbs) the interaction between green cryptos and green stocks in bear (bull) markets.

The momentum effect is a pricing anomaly that is widely observed in financial markets but not promised in the Chinese stock market. We explore the interaction between investor attention and momentum effects to strengthen momentum-based strategies' profitability by transforming the inherent noise of investor attention into valuable signals. Applying the conditional autoencoder (CAE) asset pricing model, we extract signals from noisy information to estimate stock returns that reflect the expected price adjustments driven by collective attention. Results yield four key conclusions. (i) The signal derived from investor attention acts as a catalyst that significantly enhances momentum strategies' performance, and the attention-based momentum (AttMOM) strategy consistently outperforms the conventional momentum (MOM) strategy in various formation periods. (ii) Although pricing anomalies, such as firm size, influence both strategies' returns, the attention-driven signal enables AttMOM to achieve higher and more stable returns. (iii) Investor attention helps AttMOM to maintain stable profits during market downturns. (iv) Investor attention reinforces the AttMOM strategy's resilience during turbulence, improving its hedging capabilities. Overall, our findings highlight the pivotal role of investor attention in boosting momentum returns, offering valuable insights for investment decision-making.

Cryptocurrency is a remarkable financial innovation that has affected the financial system in fundamental ways. Its increasingly complex interactions with the conventional financial market make precisely forecasting its volatility increasingly challenging. To this end, we propose a novel framework based on the evolving multiscale graph neural network (EMGNN). Specifically, we embed a graph that depicts the interactions between the cryptocurrency and conventional financial markets into the predictive process. Furthermore, we employ hierarchical evolving graph structure learners to model the dynamic and scale-specific interactions. We also evaluate our framework’s robustness and discuss its interpretability by extracting the learned graph structure. The empirical results show that (i) cryptocurrency volatility is not isolated from the conventional market, and the embedded graph can provide effective information for prediction; (ii) the EMGNN-based forecasting framework generally yields outstanding and robust performance in terms of multiple volatility estimators, cryptocurrency samples, forecasting horizons, and evaluation criteria; and (iii) the graph structure in the predictive process varies over time and scales and is well captured by our framework. Overall, our work provides new insights into risk management for market participants and into policy formulation for authorities.

We propose a cross-market volatility forecasting framework by applying attention-based spatial-temporal graph convolutional network model (ASTGCN) to forecast future volatility of stock indices in 18 financial markets. In our work, we construct cross-market volatility networks to integrate interrelations among financial markets and the corresponding features of each market. ASTGCN combines the spatial-temporal attention mechanisms with the spatial-temporal convolutions to simultaneously capture the dynamic spatial-temporal characteristics of global volatility data. Compared with competitive models, ASTGCN exhibits superiority in multivariate predictive accuracies under multiple forecasting horizons. Our proposed framework demonstrates outstanding stability through several robustness checks. We also inspect the training process of ASTGCN by extracting spatial attention matrices and find that interrelations among global financial markets perform differently in tranquil and turmoil periods. Our study levitates empirical findings in financial networks to practical application with a novel forecasting method in the deep learning community.

We integrate three enterprise networks, i.e., the stock return network, risk spillover network, and market transaction network to predict the credit risk in supply chain finance (SCF) by applying the explainable GraphSAGE model. We construct the aforementioned networks to comprehensively illustrate the relationships among enterprises, train the GraphSAGE model to classify the nodes in the graph structure, and use GNNExplainer to analyze the explainability of model's predictions. We find that (i) GraphSAGE significantly outperforms the baseline models and achieves the highest scores in terms of all performance metrics in predicting credit risk; (ii) GNNExplainer is able to identify the financial indicators (reflecting the profitability, liquidity and leverage of enterprises) that have significant impacts on the predictions; and (iii) the influential neighbors of risky enterprises tend to be risky themselves, while those of the non-risky enterprises are often non-risky, thus demonstrating a credit risk alignment among enterprise relationships. Our findings offer market participants valuable insights into enhancing credit risk prediction by utilizing advanced graph-based models, identifying the critical financial indicators, and assessing credit risk based on enterprise networks.

Precisely forecasting carbon price helps to make comprehensive plans for promoting green development. However, the carbon price is affected by many factors that are not isolated but influences each other, including energy, international carbon allowance, stock, foreign exchange and metal price. The common multivariate forecasting methods assume that each factor plays an equally important role, but they fail to (i) dynamically distinguish the relative importance of these factors; (ii) timely capture the time-varying interactions among factors; and (iii) selectively aggregate the information from different types of factors. To overcome this obstacle, we present a novel multi-factor spatial-temporal GNN framework integrating Graph WaveNet and self-attention mechanism, which incorporates factor interactions. In our empirical analysis, we take the Hubei emission allowances (HBEA) price as predictive target, and investigate how the carbon price is affected by various factors. We find that, on the one hand, our framework significantly performs better than the baseline models, and the aforementioned interactions obviously change when major events occur; on the other hand, European Union Allowance (EUA), the steel rebar futures and natural gas futures exert considerable influence on HBEA, while the foreign exchange rate and stock index are not crucial factors that explain the variation in the carbon price.

Using high-frequency data from the Chinese stock market, we investigate whether order imbalances predict intraday returns. We build the multivariate predictive models and demonstrate that: (i) order imbalances positively predict stock returns from 5 to 30 minutes; (ii) the predictive relation between order imbalances and stock returns reverses to negative from 60 to 120 minutes; (iii) the reversals do not exist in large-size stocks, high-turnover stocks, and specially treated stocks. Additionally, we add an interacting variable of lagged order imbalances and the liquidity dummy into models and demonstrate that excessive liquidity inhibits the predictability of order imbalances on intraday returns.

We explore the co-movement between two climate-related uncertainty indices and five carbon market returns in China from a time-frequency perspective based on wavelet coherence analysis. The co-movements are fragmented and time-varying, exhibiting complex intercorrelations. Several occasions of high correlations between CU and the two market returns (Hubei and Shanghai) are observed in the medium- and long-term, while those between CPU and the three market returns (Shenzhen, Guangdong and Hubei) are mainly in the short- and medium-term. The Shenzhen and Beijing markets present a climate policy-oriented feature after 2018, as a consistently strong co-movement between CPU and them is revealed. In addition, we find that the increases in climate (policy) uncertainty tend to reduce carbon market returns, mainly showing a negative correlation.

We propose a novel aggregate economic policy uncertainty (EPU) index, which is constructed using an autoencoder to extract the relevant component from eight news-based EPU proxies, for examining the impact of EPU on the stock market returns. We find that the autoencoder-based aggregate EPU index (i) exhibits the strong in-sample and out-of-sample forecasting power, and outperforms the existing EPU measures as well as well-known macroeconomic variables; (ii) generates the considerable economic value for the mean-variance investors in terms of portfolio optimization; (iii) derives its predictive ability primarily from the cash flow channel; and (iv) displays the asymmetric return predictability, with heightened performance in the low-sentiment periods.

By combining the time-varying parameter vector autoregression (TVP-VAR) model and the marginal spillover analysis, we investigate the dynamic return spillovers among digital, green, and traditional financial assets while considering the uncertainty exemplified by unprecedented events, including technical security threats, trade policy turbulences, public health emergencies, and geopolitical upheavals. Further, we disclose the impact of underlying macroeconomic and financial uncertainty factors under different market conditions using the quantile regression approach. Our empirical results illustrate that (i) the digital and green assets gradually incorporate into a risk-sharing community together with traditional assets; (ii) the direction and magnitude of spillovers are highly sensitive to uncertainty shocks, and the green assets provoke dramatic risk spreading during the COVID-19 pandemic, while the digital ones show enhancing directional spillovers during the cryptocurrency hacking, the Sino-US trade friction, and the Russia-Ukraine conflict; and (iii) investor sentiment, geopolitical risks, and trade and monetary policy uncertainties significantly drive the spillovers in turbulent periods, and the importance of cryptocurrency policy uncertainty is highlighted under all market conditions. Overall, our research provides insightful implications for risk management and policy formulation in times of uncertainty.

We investigate the impact of venture capital (VC) on the technological innovation of the firms listed on China's National Equities Exchange and Quotations (NEEQ) and identify the role of the VC strategy in this process, by using the data related to VC investment, firm innovation, and human capital. The empirical results demonstrate that (i) VC effectively promote the technological innovation of target firms, especially in the high-tech industry; (ii) as two parts of the VC strategy in the preinvestment phases, the entry time affects the technological innovation, and VC investing in the mature stage of target firms generally implements the short-termism strategy, appearing to inhibit the technological innovation, while the shareholding ratio affects the innovation in a nonlinear form, with a pivotal point at an ratio of 21%; and (iii) in the postinvestment phases, the intervention of VC on optimizing the target firms' human capital structure forms its strategy, promoting a significant increase in the proportion of R&D personnel or skilled labor, and promotes the technological innovation. On the whole, our findings provide an insight of how VC influences technological innovation, underscoring the relationship among investment strategy, firm characteristics, and innovation trajectory.

The global financial crisis not only highlights the important role of financial network connectedness, but also urges regulators pay more attention to systemic risk. We study whether network connectedness helps systemic risk prediction by using machine learning techniques, including k-nearest neighbor (KNN), support vector regression (SVR), extreme gradient boosting (XGB), and feedforward neural network (FNN). Based on prediction results, we use the fingerprint model to deliver interpretability evaluations of financial variables and network connectedness. We find that (1) the linear and nonlinear effects of stock market volatility and network connectedness effectively interpret systemic risk; (2) the interaction effects between stock market volatility and other drivers have a stronger ability to predict system risk in SVR, KNN, and XGB approaches, while network connectedness performs more profoundly in FNN approach; and (3) the interaction effect between stock market volatility and network connectedness is evident in all approaches.

We explore the tail risk spillover in the stock and foreign exchange (forex) markets along the G20 countries based on the tail-event driven network (TENET) method. To effectively capture the risk spillover mechanism from the tail-dependence perspective, we analyze the network connectedness in four aspects (namely system, market, region, and country) and explore it at the major emergencies. We find that (i) the system-level risk spillover peaks at the US subprime crisis, and subsequently undergoes several cyclical volatility in relatively high level in the period of major emergencies; (ii) the cross-market risk spillover from the stock markets to the forex markets plays a dominant role, while that from the forex markets to the stock markets is small; and (iii) the stock and forex markets in Europe and America transmit the large tail risk spillovers to other regions, and the forex markets in these regions are sensitive to the major emergencies.

We study systemic risk drivers of FinTech and traditional financial institutions under normal and extreme market conditions. We use machine learning (ML) techniques (i.e. random forest and gradient boosted regression trees) to evaluate the role of macroeconomic variables, firm characteristics, and network topologies as systemic risk drivers and perform the ML-based interpretation by Shapley individual and interaction values. We find that (i) the feature importance in driving systemic risk depends on market conditions; namely, market volatility (MVOL), individual stock volatility (IVOL), and market capitalization (MC) are positive drivers of systemic risk under extreme (downside and upside) market conditions, while under normal market conditions, institutions with high price-earnings ratio, large MC, and low IVOL play an essential role in stabilizing markets; (ii) macroeconomic variables are the most important extreme systemic risk drivers, while firm characteristics are more important under normal market conditions; and (iii) the interaction between IVOL and MC or MVOL is the significant source of extreme systemic risk, and MC is the most crucial interaction attribute under normal market conditions. The interactions between macroeconomic variables are the most prominent in systemic risk under different market conditions.

In this study, we investigate the impact of introducing enterprise relationship to train machine learning (ML) algorithms on improving its ability of forecasting small and medium-sized enterprises' (SMEs) credit risk in supply chain finance (SCF). First, we incorporate attributes of SME node in the CEs-SMEs network, a bipartite network constructed with transaction data in the Chinese automobile industry, into financial indicators to train ML algorithms for improving forecasting performance. Second, we employ the weighted one-mode projection approach to extract a projected-SMEs network that reveals the relationship among SMEs from the CEs-SMEs network and examine whether the ML algorithms' forecasting performances is improved when considering attributes in the projected-SMEs network. Third, we analyze the correlation between SME's credit risk and individual network attribute to specifically understand the impact of enterprise relationship on credit risk. Overall, the empirical results indicate that the forecasting performance is obviously better when enterprise transaction relationship is considered, and it is further enhanced after we apply weighted one-mode projection approach. Meanwhile, the variable importance analysis and the binomial logistic regression demonstrate how significantly each network attribute is correlated with credit risk, and the discussion on partial dependence plot shows that SMEs with large degree are non-risky generally.

This paper pioneers exploring the risk contagion attributes of emerging NFT markets, characterized by considerable volatility, through a time-frequency risk spillover lens within the integrated Carbon-NFT-Stock system. Our findings are multifaceted. Firstly, NFT acts as the risk transmitter in extreme upside condition and receiver in extreme downside condition. Secondly, in extreme downside condition, the destructiveness of risk contagion remains unabated in the long term, and major events amplify the total risk spillover. Thirdly, risk spillovers of NFT and EUA are susceptible to crude oil. Fourthly, carbon market regulators should remain vigilant about risks from the US stock market.

We investigate the interactions between innovative financial assets (e.g., FinTech-related stocks, green bonds, and cryptocurrencies) and traditional ones (e.g., global equities, gold, crude oil, U.S. dollar, and government bonds) at short-, medium-, and long-time scales by adopting the multiscale entropy-based approach, through which the dominant influencers in information flowing are identified. Also, we discuss the impact of unprecedented events exemplified by the COVID-19 pandemic. Further, we propose the high-transfer-entropy trading strategy that considers the dominant influencers. The empirical results show that (i) the information exchange is heterogenous at three time scales, with the greatest intensity at short-time scale; (ii) no isolated dominant influencer is discovered, and the traditional assets are stronger influencers in long term while the innovative assets are more influential in short and medium terms; (iii) the COVID-19 pandemic alters the magnitude and direction of information transfer, during which the influence of innovative assets relative to traditional assets enhances evidently; and (iv) the high-transfer-entropy strategy can effectively provide the investor with excess returns.

We investigate whether corporates’ environmental, social, and governance (ESG) performance affects their systemic risk. Based on 284 publicly-listed Chinese firms over the period 2011–2020, we construct a tail risk spillover network for measuring their connectedness and systemic risk and use a panel regression model to examine the influence of corporate ESG performance on systemic risk. Network connectedness is an essential channel for risk contagion, and the energy, industry, and finance sectors occupy a significant position in the system. The ESG performance has a significant negative impact on systemic risk, both in terms of systemic vulnerability and systemic risk contribution, i.e., the ESG performance can dampen the two-way transmission of shocks between individual firms and the system. The results are robust to proxy measures of systemic vulnerability and systemic risk contribution, as well as to winsorize all variables and lag the core explanatory variables. Our study provides a new angle from the ESG performance for regulating systemic risk.

We construct correlation-based networks linking 86 assets (stock indices, bond indices, foreign exchange rates, commodity futures, and cryptocurrencies) and analyze the impact of asset selection on portfolio optimization using different centrality measures (including degree, eigenvector, eccentricity, betweenness, PageRank, and hybrid centralities). In times of a global crisis, peripheral assets located in cross-market networks are more suitable for investment. By comparing portfolio performance based on different centrality measures, we find that (i) hybrid, eigenvector, and PageRank centralities can best improve portfolio performance; (ii) degree centrality is suitable for larger portfolios; and (iii) eccentricity and betweenness centralities are unsuitable for network optimization portfolios. In response, we explain them based on the construction principle of centrality measures. Additionally, our optimal portfolios suggest that investors pay more attention to the role of emerging countries, which are less exposed to external shocks and whose financial markets are more likely to remain stable.