With the rapid development of e-commerce, impulse purchases have become a significant driver of revenue growth for e-commerce platforms. Traditional research often relies on questionnaire surveys for analysis, which are susceptible to memory bias and social desirability bias, and commonly conflate the number of items added to a shopping cart with purchase intent. Based on 778 valid purchase sessions from the Retail Rocket public dataset, this paper constructs the core metric of the "add-to-cart-to-view ratio" to identify impulse purchases from the perspective of behavioral conversion efficiency. It also introduces the average dwell time per product to avoid circular reasoning and compares the predictive performance of logistic regression, random forest, and XGBoost models. The results indicate that the "add-to-cart-to-view ratio" is a core predictive indicator of impulse purchases, while the number of times an item is added to the cart alone has no significant impact; the XGBoost model performed best (AUC = 0.7576). This study provides data support for e-commerce platforms to accurately identify impulse buyers and implement targeted marketing strategies.

Against the backdrop of intensifying global warming and frequent extreme weather events, the low-carbon transition of residents' consumption behavior has become an important micro-foundation for achieving the "dual carbon" goals. Based on 615 resident survey datasets collected from four cities in Zhejiang Province—Huzhou, Jiaxing, Quzhou and Jinhua—this study empirically examines the effect of extreme weather on residents' low-carbon consumption behavior and its underlying mechanisms using a structural equation model. The findings are as follows: (1) Extreme weather exerts a significant positive direct impact on residents' low-carbon consumption behavior (β = 0.312, p < 0.001). (2) Green cognition and environmental attention play partial mediating roles between extreme weather and low-carbon consumption behavior, yet the mediating effects are weaker than the direct effect. (3) Government environmental regulation intensity positively moderates the relationship between extreme weather and low-carbon consumption behavior (β = 0.228, p < 0.01). This study reveals the mechanism by which extreme weather drives residents' low-carbon consumption through the path of "risk perception—cognition improvement—behavior transformation", and provides empirical evidence for governments to formulate differentiated low-carbon consumption guidance policies under climate risk contexts.

Bitcoin has become an increasingly important asset for portfolio allocation, yet its diversification value and option-implied information remain difficult to evaluate. This paper examines Bitcoin risk from portfolio and option-implied perspectives. This study assesses whether Bitcoin improves the risk-return opportunity set with traditional assets and whether its diversification role remains stable during market stress. Option-implied measures, including the 25-delta Risk Reversal (RR25), smile curvature, and an at-the-money Implied-Volatility-minus-Realized-Volatility (IV-minus-RV) proxy, are then constructed to predict market conditions. Portfolio analysis shows that Bitcoin can improve risk-return tradeoffs but does not function as a stable minimum-variance asset or a reliable crisis hedge. Baseline regressions provide limited evidence that RR25 consistently predicts future realized volatility or returns. However, extreme negative short-dated RR25 is followed by higher future realized volatility, suggesting RR25 is more informative as a nonlinear stress-state indicator than as a continuous forecasting variable. Smile curvature captures the implied-volatility surface but provides weaker predictive information. Finally, the IV-minus-RV analysis shows that gradual RR25-based exposure scaling achieves a better risk-adjusted profile than a binary exposure rule. Overall, the findings indicate that Bitcoin's diversification benefits and option-implied information are state-dependent.

Changes in stock market liquidity affect transaction costs, price discovery, and risk premium adjustment. When volatility rises or trading depth declines, liquidity deterioration can amplify future return fluctuations through concentrated selling pressure, order book contraction, and investor rebalancing. Focusing on the relationship between liquidity shocks and return predictability, daily public trading data of Shanghai and Shenzhen A share listed companies from 2014 to 2024 are used to construct the Amihud illiquidity measure. Abnormal liquidity shocks are identified through residuals from a rolling AR(1) model. Based on this measurement, two way fixed effects models, threshold regressions, and quantile regressions are applied to examine transmission differences across market states and return distribution regions. The results show that negative liquidity shocks significantly suppress future excess returns. This effect is stronger in high volatility states, small capitalization stocks, and low return quantiles. Nonlinear models also show lower forecast errors and higher directional accuracy under stressed market conditions. These findings indicate that the predictive value of liquidity shocks is mainly concentrated in downside markets and tail risk regions. The evidence provides practical implications for risk warning, asset pricing, and portfolio adjustment.

Algorithm-driven automated decision-making has become the core operational paradigm of digital consumer credit. While significantly reducing transaction costs, it has also given rise to systemic risks of algorithmic discrimination. Taking the protection of the rights and interests of digitally disadvantaged groups as the analytical starting point, this paper integrates data from the 57th Statistical Report on China's Internet Development, the financial consumer complaint statistics of the People's Bank of China for 2024, and 120 effective court judgments, and conducts an empirical examination of the legal attributes, typological characteristics, and rights-infringement mechanisms of algorithmic discrimination in the field of digital consumer credit. The study shows that algorithmic discrimination has emerged with three interrelated features-concealment, precision, and amplification-and has reconstructed the stratification structure of the financial market through technological means, causing institutional infringement upon the equal credit rights, fair trading rights, and property rights of the elderly, rural residents, and low-income groups. At present, China's regulatory system for algorithmic discrimination suffers from structural deficiencies, including the absence of special legislation, outdated regulatory technology, imbalance in judicial proof, and malfunctioning mechanisms for consumer rights protection. Accordingly, this paper constructs a coordinated multi-dimensional regulatory framework integrating "legislative regulation—supervisory innovation—judicial relief—rights protection—technical governance", and introduces institutional tools such as algorithmic auditing, penetrating supervision, and reversal of the burden of proof, so as to promote the dynamic balance between algorithmic ethics and financial fairness, and to provide legal support for inclusive financial development in the digital economy era.

Asset price prediction is one of the most challenging research topics in finance. The inherent nonlinearity, low signal-to-noise ratio, and non-stationarity of Financial Time Series (FTS) make it difficult for traditional time series models to perform adequately. This paper reviews the evolution from traditional statistical models to Machine Learning (ML) and Deep Learning (DL), with a focus on comparing their performance in predictive accuracy, interpretability and robustness. Traditional models offer good interpretability and computational efficiency but are constrained by linear assumptions. ML models do not significantly outperform traditional methods, but exhibit conditional advantages that depend on asset characteristics and market conditions. DL models can capture complex temporal dependencies, but face challenges such as poor interpretability, overfitting and deployment difficulties. Based on current research, this review concludes that no universally optimal model exists, but only relatively suitable models exist for specific tasks, asset characteristics, and market environments. The future direction lies not in more complex architectures, but in reshaping prediction tools into interpretable decision support systems, which requires simultaneous advances in model architecture innovation, interpretability methods, and standardized evaluation frameworks.

This paper examines whether historical return correlations and Z-score signals help predict profitable pairs trades using high-frequency equity data. Stocks are matched by industry and market capitalization, and minute-by-minute prices are divided into separate lookback and trading periods of equivalent lengths. Historical correlations are calculated at several frequencies (5-minute, 10-minute, 1-hour, and 2-hour), while trading-period Z-scores are computed from spread parameters estimated during formation. The results provide insufficient evidence that stronger historical correlation predicts higher subsequent gross profit and loss at any of the frequencies examined. The current Z-score has a statistically significant association with gross P&L, but the regression explains almost none of the variation in trading outcomes. Most predicted returns also fall below the lowest transaction-cost threshold, leaving few potentially profitable opportunities under the assumed costs. Overall, these results suggest that historical convergence of stock pairs does not guarantee economic profitability. They also illustrate the reasons for statistically significant signals to be assessed concurrently with its predictability and associated costs. Within the sample studied, historical pair similarity and spread deviations alone offer limited guidance for identifying trades that remain profitable after execution costs.

Generation Alpha, the first cohort to grow up with artificial intelligence, is expected to become the largest consumer segment in history. However, academic research on their cognitive characteristics and consumption behaviour remains limited. This study aims to construct a multi‑dimensional cognitive profile of Generation Alpha and to examine how this profile drives retail business model innovation. A structured questionnaire survey was conducted with 316 parents of Generation Alpha children in China, covering four cognitive dimensions: behavioural patterns, product preferences, social concerns, and brand acceptance. Descriptive statistics, correlation analysis, and multiple regression analysis were used to analyse the data. The results show that Generation Alpha children exhibit moderate levels across all four cognitive dimensions. Product preference emerged as the strongest predictor of brand acceptance (β = 0.318, p < 0.001), followed by social concern (β = 0.218) and behavioural profile (β = 0.213). The model explained 31.8% of the variance in brand acceptance. These findings provide empirical support for three retail innovation pathways: product creativity enhancement, channel transformation, and communication upgrading. The study offers both theoretical contributions to generational consumer behaviour research and practical guidance for retailers seeking to engage Generation Alpha.