Journal of Forecasting最新文献

We analyze the impact of investor sentiment on forecasting daily return volatility across various international Real Estate Investment Trust (REIT) indices. Notably, we propose that economic policy uncertainty plays a significant role in shaping investor sentiment and enhances its predictive power regarding REIT volatility. To address the mixed-frequency nature of the involved variables, we utilize the GARCH-MIDAS framework, which effectively mitigates the issues of information loss associated with data aggregation, as well as the biases resulting from data disaggregation. Our findings provide compelling evidence of improved forecasting in models that incorporate investor sentiment, demonstrating significant in-sample predictability. This suggests that heightened expressions of sentiment in investor behavior tend to amplify risks linked to international REITs. Further analysis indicates that economic policy uncertainty may enhance the forecasting capacity of investor sentiment for out-of-sample REIT volatility predictions. Consequently, it is crucial to monitor global economic policy uncertainty and recognize its potential effects on investor sentiment for optimal investment decision-making.

In this work, we explore the forecasting ability of a recently proposed normalizing and variance-stabilizing (NoVaS) transformation with the possible inclusion of exogenous variables in GARCH volatility specification. The NoVaS prediction method, which is inspired by a model-free prediction principle, has generally shown more accurate, stable and robust (to misspecifications) performance than that compared with classical GARCH-type methods. We derive the NoVaS transformation needed to include exogenous covariates and then construct the corresponding prediction procedure for multiple exogenous covariates. We address both point and interval forecasts using NoVaS type methods. We show through extensive simulation studies that bolster our claim that the NoVaS method outperforms traditional ones, especially for long-term time aggregated predictions. We also exhibit how our method could utilize geopolitical risks in forecasting volatility in national stock market indices. From an applied point-of-view for practitioners and policymakers, our methodology provides a distribution-free approach to forecast volatility and sheds light on how to leverage extra knowledge such as fundamentals- and sentiments-based information to improve the prediction accuracy of market volatility.

We propose a novel methodology for transforming financial time series into a geometric format via a sequence of point clouds, enabling richer modeling of nonstationary behavior. In this framework, volatility serves as a spatial directive to guide how overlapping temporal windows become connected in an adjacency tensor, capturing both local volatility relationships and temporal proximity. Spatial expansion then interpolates points of different connection strengths while gap filling ensures a regularized geometric structure. A subsequent relevance-weighted attention mechanism targets significant regions of each transformed window. To further illuminate underlying dynamics, we integrate the largest Lyapunov exponents directly into each point cloud, embedding a chaotic signature that quantifies local predictability. Unlike canonical CNN, RNN, or Transformer pipelines, this geometry-based representation makes it easier to detect abrupt changes, volatility clusters, and multiscale dependencies via explicit geometric and topological cues. Finally, an architecture incorporating graph-inspired components—along with point-cloud encoders and multihead attention—learns both short-term and long-term dynamics from the spatially enriched time series. The method's ability to harmonize volatility-driven structure, chaotic features, and temporal attention improves predictive performance in empirical testing on stock and cryptocurrency data, underscoring its potential for versatile financial analysis and risk-based applications.

In the financial sector, credit risk represents a critical issue, and accurate prediction is essential for mitigating financial risk and ensuring economic stability. Although artificial intelligence methods can achieve satisfactory accuracy, explaining their predictive results poses a significant challenge, thereby prompting research on interpretability. Current research primarily focuses on individual interpretability methods and seldom investigates the combined application of multiple approaches. To address the limitations of existing research, this study proposes a two-stage interpretability model that integrates SHAP and counterfactual explanations. In the first stage, SHAP is employed to analyze feature importance, categorizing features into subsets according to their positive or negative impact on predicted outcomes. In the second stage, a genetic algorithm generates counterfactual explanations by considering feature importance and applying perturbations in various directions based on predefined subsets, thereby accurately identifying counterfactual samples that can modify predicted outcomes. We conducted experiments on the German credit datasets, HMEQ datasets, and the Taiwan Default of Credit Card Clients dataset using SVM, XGB, MLP, and LSTM as base classifiers, respectively. The experimental results indicate that the frequency of feature changes in the counterfactual explanations generated closely aligns with the feature importance derived from the SHAP method. Under the evaluation metrics of effectiveness and sparsity, the performance demonstrates improvements over both basic counterfactual explanation methods and prototype-based counterfactuals. Furthermore, this study offers recommendations based on features derived from SHAP analysis results and counterfactual explanations to reduce the risk of classification as a default.

This paper introduces a novel Bayesian semiparametric multivariate GARCH framework for modeling returns and realized covariance, as well as approximating their joint unknown conditional density. We extend existing parametric multivariate realized GARCH models by incorporating a Dirichlet process mixture of countably infinite normal distributions for returns and (inverse-)Wishart distributions for realized covariance. This approach captures time-varying dynamics in higher order conditional moments of both returns and realized covariance. Our new class of models demonstrates superior out-of-sample forecasting performance, providing significantly improved multiperiod density forecasts for returns and realized covariance, as well as competitive covariance point forecasts.

We propose a default prediction framework that incorporates imbalance handling and feature selection. For imbalance handling, we determine the optimal ratio of non-default to default firms by minimizing the Type-II error of the majority voting deep fully connected network (MV-DFCN) model. For feature selection, we design a two-stage process that first eliminates highly correlated and redundant features, and then refines the feature set using backward selection. Experimental results show that the DFCN model within the proposed framework outperforms baseline models in terms of G-Mean and AUC and achieves the lowest Type-II error rate. Furthermore, the framework outperforms eight baseline combinations of imbalance handling and feature selection strategies. Additionally, SHAP values are used to assess feature contributions, and nine features with statistically significant impacts are identified.