{"title":"A Deep Learning Model for Earnings Management Prediction","authors":"Trong-Hiep Hoang, Toan Nguyen-Mau, Van-Nam Huynh","doi":"10.1002/for.70198","DOIUrl":"https://doi.org/10.1002/for.70198","url":null,"abstract":"<div>\u0000 \u0000 <p>This paper proposes a novel firm embedding long short-term memory (LSTM) model for panel data analysis, which integrates firm-level embeddings with an LSTM network. The model jointly addresses unobserved heterogeneity and temporal dependence, providing a high-performing solution for dynamic financial prediction tasks. To achieve this, each firm is represented by a trainable low-dimensional embedding vector, concatenated with financial inputs and fed into an LSTM network. This architecture is trained end-to-end to jointly capture firm characteristics and temporal financial patterns. This integration significantly enhances predictive performance by leveraging both cross-sectional firm information and temporal financial dynamics. Experiments on earnings management prediction were conducted using data from 334 Vietnamese listed firms over a 7-year period (2014–2020), with discretionary accruals estimated via the Modified Jones model. In our setting, standalone machine learning models underperform their econometric benchmark, whereas the proposed embedding LSTM framework achieves superior performance.</p>\u0000 </div>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3474-3491"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149312172","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A New Implementation of Network GARCH Model for Stock Volatility and Co-Volatility Forecasting","authors":"Peiyi Zhou","doi":"10.1002/for.70191","DOIUrl":"https://doi.org/10.1002/for.70191","url":null,"abstract":"<p>Volatility clustering and spillovers are key features of financial time series with many cross-sectional assets. While network analysis links similar or correlated stocks and helps trace volatility spillovers, contemporary multivariate ARCH-GARCH formulations struggle to represent structured network dependence and remain parsimonious. We introduce the generalized network GARCH model under the DCC framework (DCC-GNGARCH), that embeds the GARCH dynamics within the generalized network autoregressive (GNAR) framework to capture an asset's volatility driven by its own history and by neighboring assets in a constructed network. DCC-GNGARCH also extends existing network GARCH formulations by adapting neighboring volatility persistence, dynamic conditional covariance updates, and allowing higher order neighboring effects beyond immediate neighbors. This paper provides the model derivation, vectorization and conversion, and an extension by incorporating threshold coefficients to capture leverage effects. We show that the DCC-GNGARCH is a valid volatility model satisfying the stylized facts of financial return series through simulation. Parameter estimation is then performed by using squared returns as variance proxy and minimizing the negative log-likelihood (NLL) loss function. We apply our model on 75 of the most active US stocks under a constructed network and highlight the model's ability in volatility estimation and forecast, whereas robustness checks under alternative network constructions, different proxy choices, error specifications, and sparsity levels confirm the stability of the main empirical findings.</p>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3403-3426"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/for.70191","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149311644","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A Novel Text-Based Framework for Forecasting Carbon Prices","authors":"Christian Oliver Ewald, Yaoyu Li","doi":"10.1002/for.70202","DOIUrl":"https://doi.org/10.1002/for.70202","url":null,"abstract":"<p>This study proposes a text-based framework for predicting EU carbon prices. Using weekly data from 2020 to 2024, we construct a multivariate dataset combining financial indicators, commodity prices, Google Trends measures, and news-based sentiment extracted using FinBERT. To address potential noise and redundancy among predictors, principal component analysis (PCA) is applied for dimensionality reduction. The empirical results suggest that deep learning models generally achieve lower forecast errors than traditional models. In particular, the CNN-LSTM model with PCA-based inputs achieves the best predictive performance, reducing RMSE by approximately 33% relative to the corresponding multivariate CNN-LSTM specification. These findings provide evidence that combining deep learning architectures with dimensionality-reduction techniques can improve forecasting performance in carbon markets.</p>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3507-3529"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/for.70202","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149310836","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hanbyeol Park, Sunghyun Sim, Kikun Park, Yongjae Lee, Eunhee Park, Hyerim Bae
{"title":"Hybrid Temporal Autoencoder and Similarity Matching for Low Aggregation Level Long Time Series Forecasting","authors":"Hanbyeol Park, Sunghyun Sim, Kikun Park, Yongjae Lee, Eunhee Park, Hyerim Bae","doi":"10.1002/for.70194","DOIUrl":"https://doi.org/10.1002/for.70194","url":null,"abstract":"<p>Deep learning-based long time series forecasting (LTSF) has achieved high accuracy by effectively capturing the underlying trends, seasonality, and temporal dependencies within time series data. However, at the individual entity level, termed the low aggregation level (LAL), intermittency, irregularity, and data sparsity undermine the inductive biases of neural networks, leading to a significant decline in performance. To address the challenges of LAL-LTSF, this study introduces a similarity matching (SM)-based method that leverages historical time series patterns from analogous entities. Recognizing that the efficacy of the SM algorithm in LTSF is highly dependent on embedding quality, we present a hybrid temporal autoencoder (HTAE) architecture tailored to effectively capture LAL-specific time series features. The HTAE integrates conventional self-attention mechanisms with elements derived from a newly developed contextual multihead self-attention module, facilitating the creation of robust representations, even in noisy and intermittent LAL scenarios. The proposed approach outperformed 13 baseline models, including state-of-the-art techniques, across two real-world time series datasets. This study offers a promising solution for enhancing LAL-LTSF accuracy and has the potential to improve long-term demand forecasting and service delivery in real-world industrial applications.</p>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3565-3596"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/for.70194","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149311664","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Forecasting Financial Assets Using Recursive Clustering Approach","authors":"Jeongwoo Kim","doi":"10.1002/for.70200","DOIUrl":"https://doi.org/10.1002/for.70200","url":null,"abstract":"<p>Machine learning-based forecasting has been gaining attention for its flexibility and adaptability. Given the dynamics of temporal data, clustering can be an effective approach for improving forecasting accuracy. While clustering for classification typically considers all observations in the data to effectively divide the data into clusters, clustering for temporal data forecasting may differ. Instead of focusing on dividing the entire data, identifying a single cluster close to the future value may yield better forecasting accuracy. A method to determine the single cluster by dividing a cluster obtained from previous clustering is adopted in this study, thereby improving the forecasting accuracy of clustering methods. Various financial assets are used to validate this approach across different time intervals. The empirical validation demonstrates the improved forecasting accuracy of the method, suggesting its broad applicability in other forecasting contexts.</p>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3597-3633"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/for.70200","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149311773","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Forecasting Duration in High-Frequency Financial Data Using a Self-Exciting Flexible Residual Point Process","authors":"Kyungsub Lee","doi":"10.1002/for.70209","DOIUrl":"https://doi.org/10.1002/for.70209","url":null,"abstract":"<p>This paper presents a method for forecasting limit order book durations using a self-exciting flexible residual point process. High-frequency events in modern exchanges exhibit heavy-tailed interarrival times, posing a significant challenge for accurate prediction. The proposed approach incorporates the empirical distributional features of interarrival times while preserving the self-exciting and decay structure. This work also examines the stochastic stability of the process, which can be interpreted as a general state-space Markov chain. Under suitable conditions, the process is irreducible, aperiodic, positive Harris recurrent, and has a stationary distribution. An empirical study demonstrates that the model achieves strong predictive performance compared with several alternative approaches when forecasting durations in ultrahigh-frequency trading data.</p>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3710-3728"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/for.70209","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149312183","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Does Trading Volume Improve Long-Term Volatility Forecasts? Evidence from the MF2-GARCH Framework","authors":"Anjana Yatawara","doi":"10.1002/for.70199","DOIUrl":"https://doi.org/10.1002/for.70199","url":null,"abstract":"<div>\u0000 \u0000 <p>We augment the long-term volatility component of the MF2-GARCH model of Conrad and Engle (2025) with smoothed trading volume, yielding the MF2V-GARCH. The extension adds a single parameter \u0000<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>δ</mi>\u0000 <mo>≥</mo>\u0000 <mn>0</mn>\u0000 </mrow>\u0000 <annotation>$$ delta ge 0 $$</annotation>\u0000 </semantics></math>, preserves covariance stationarity, and nests the MF2-GARCH as a special case. Estimation on sixteen US assets (the S&P 500 index, nine sector ETFs, and six individual stocks) over January 2000 to March 2026 reveals that \u0000<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>δ</mi>\u0000 </mrow>\u0000 <annotation>$$ delta $$</annotation>\u0000 </semantics></math> is significant for the S&P 500 (\u0000<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>p</mi>\u0000 <mo>=</mo>\u0000 <mn>0</mn>\u0000 <mo>.</mo>\u0000 <mn>006</mn>\u0000 </mrow>\u0000 <annotation>$$ p&amp;amp;#x0003D;0.006 $$</annotation>\u0000 </semantics></math>), Apple, and financial-sector assets, but effectively zero for defensive sectors such as consumer staples and healthcare. Out-of-sample, the MF2V-GARCH produces lower QLIKE losses than the MF2-GARCH at all twelve forecast horizons for the S&P 500. All twelve one-sided Diebold-Mariano tests, computed with HAC variance estimates and small-sample corrections and confirmed by a stationary bootstrap, are significant at the 5% level under both the QLIKE and squared-error losses, and the relative root mean squared error at the three-month horizon falls by 23% on the Conrad-Engle evaluation window. Sub-period analysis shows that the volume advantage is largest during regime transitions: during COVID-19, the MF2-GARCH's eight-month forecasts are worse than a historical average while the MF2V-GARCH remains competitive (a descriptive comparison on the episode's small set of valid origins). Across the full cross-section, the MF2V-GARCH wins 151 of 192 horizon-level comparisons (79%) on the outlier-screened evaluation design of Conrad and Engle (2025); a new sensitivity analysis shows that on unfiltered forecast origins this advantage attenuates to statistical zero without ever significantly reversing. In a horse-race against alternative daily information-arrival proxies, volume outperforms Amihud illiquidity out of sample and is the most robust single proxy across horizons, while the Corwin–Schultz high–low spread fits best in-sample and performs comparably out of sample (marginally better at several monthly horizons); in characteristic-sorted portfolios of 238 stocks the volume effect is significant in thirteen of fifteen quintiles at the 5% level and strengthens with size.</p>\u0000 <","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3530-3564"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149311020","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A Data-Driven Investigation of How Forecast Horizon and Training Data Size Influence Machine Learning Performance in Solar Irradiance Forecasting","authors":"Fatemeh Keramati, Hamid Reza Mohammadi","doi":"10.1002/for.70208","DOIUrl":"https://doi.org/10.1002/for.70208","url":null,"abstract":"<div>\u0000 \u0000 <p>Accurate short- and medium-term solar irradiance forecasting is vital for integrating solar power into the grid, but high variability and weather dependence make it challenging. Machine learning (ML) models offer promise, but their performance often depends on forecast horizon and training data size. In this study, four models, including random forest (RF), support vector regression (SVR), gradient boosting regressor (GBR), and deep neural networks (DNN), are compared for 15-min, hourly, and day-ahead solar irradiance prediction. The models are trained with meteorological and irradiance data for the city of Sioux Falls, South Dakota, using datasets ranging from 1 to 8 years (2012–2019). A total of 96 models are developed by combining four algorithms across three horizons and eight training data sizes. Using the maximum relevance minimum redundancy (MRMR) technique, the best features are selected, and the results are evaluated using standard error metrics. The results indicate that RF provides the most stable accuracy prediction, SVR provides the best accuracy on day-ahead forecasts with limited data, GBR performs best at the hourly prediction level, and DNN accuracy is more sensitive to both data quality and overfitting. It emphasizes the importance of choosing models that align with the forecast horizon and the size of available data. In this paper, we provide an in-depth comparison of four state-of-the-art ML models across various horizons and training data sizes. Results show the importance of integrating MRMR, with guidelines for researchers and energy planners to achieve more accurate and efficient solar irradiance prediction.</p>\u0000 </div>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3666-3678"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149311836","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Centered-Innovation MA for Bayesian Dirichlet ARMA: Theoretical Equivalence and an Application to Bank-Asset Shares","authors":"Harrison Katz","doi":"10.1002/for.70205","DOIUrl":"https://doi.org/10.1002/for.70205","url":null,"abstract":"<div>\u0000 \u0000 <p>We study a minimal change to an observation-driven Bayesian Dirichlet ARMA (B–DARMA) for compositional time series: replace the raw additive log-ratio (ALR) residual in the moving-average block with a <i>centered</i> innovation that subtracts the Dirichlet conditional ALR mean, available in closed form via digamma identities. We prove a recursion-level first-order equivalence (in \u0000<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mn>1</mn>\u0000 <mo>/</mo>\u0000 <mi>ϕ</mi>\u0000 </mrow>\u0000 <annotation>$$ 1/phi $$</annotation>\u0000 </semantics></math>) between the centered specification and a digamma-link DARMA at fixed parameters, under explicit interior and lag-stability conditions. The result motivates expecting small recursion-induced predictive differences in high-precision settings but does not by itself govern the geometry of the Bayesian posteriors that re-estimation produces. On weekly Federal Reserve H.8 bank-asset shares (October 2015 through October 2025, \u0000<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>T</mi>\u0000 <mo>=</mo>\u0000 <mn>522</mn>\u0000 </mrow>\u0000 <annotation>$$ T&amp;amp;#x0003D;522 $$</annotation>\u0000 </semantics></math> weeks), paired comparisons across 104 rolling weekly origins detect no statistically significant mean differences between the Centered–MA and Raw–MA specifications on any predictive performance metric examined. Hamiltonian Monte Carlo divergent transitions are approximately an order of magnitude more frequent under the raw specification, driven by isolated rolling fits at which the raw posterior exhibits localized pathologies. A four-reference sensitivity analysis shows that no statistically significant mean difference is detected on any predictive metric at any of the four references, while the geometric advantage of centering varies with the prevalence of pathological raw fits, from a substantial reduction at the loans reference to parity at the cash reference. The main practical implication in this application is operational: Centering avoids the catastrophic raw-MA divergence spikes that occur at isolated rolling origins, which matters for production workflows in which posterior simulation feeds downstream stress tests. The adjustment is analytic and plug-ins and requires only a local change to the MA innovation calculation.</p>\u0000 </div>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3651-3665"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149311863","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Bao Doan, Dulani Jayasuriya, John B. Lee, Jonathan J. Reeves
{"title":"Beta Forecasting With Realized Beta Estimators and Machine Learning Algorithms","authors":"Bao Doan, Dulani Jayasuriya, John B. Lee, Jonathan J. Reeves","doi":"10.1002/for.70206","DOIUrl":"https://doi.org/10.1002/for.70206","url":null,"abstract":"<p>This paper applies machine learning algorithms to the modeling of realized betas for the purposes of forecasting stock systematic risk. Higher levels of beta forecast accuracy are demonstrated, relative to other studies in the literature. These improvements are also highly significant, both statistically and economically. The approaches include individual machine learning algorithms and forecast combinations with multiple machine learning algorithms and realized betas. Forecast combinations lead to the highest level of beta forecast accuracy. The machine learning algorithms employed are ridge regression, decision tree learning, adaptive boosting, gradient boosting, random forests, and neural networks.</p>","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"45 8","pages":"3679-3709"},"PeriodicalIF":2.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/for.70206","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149311804","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}