|
on Forecasting |
| By: | Millend Roy; Vladimir Pyltsov; Yinbo Hu |
| Abstract: | Accurate electricity load forecasting is essential for grid stability, resource optimization, and renewable energy integration. While transformer-based deep learning models like TimeGPT have gained traction in time-series forecasting, their effectiveness in long-term electricity load prediction remains uncertain. This study evaluates forecasting models ranging from classical regression techniques to advanced deep learning architectures using data from the ESD 2025 competition. The dataset includes two years of historical electricity load data, alongside temperature and global horizontal irradiance (GHI) across five sites, with a one-day-ahead forecasting horizon. Since actual test set load values remain undisclosed, leveraging predicted values would accumulate errors, making this a long-term forecasting challenge. We employ (i) Principal Component Analysis (PCA) for dimensionality reduction and (ii) frame the task as a regression problem, using temperature and GHI as covariates to predict load for each hour, (iii) ultimately stacking 24 models to generate yearly forecasts. Our results reveal that deep learning models, including TimeGPT, fail to consistently outperform simpler statistical and machine learning approaches due to the limited availability of training data and exogenous variables. In contrast, XGBoost, with minimal feature engineering, delivers the lowest error rates across all test cases while maintaining computational efficiency. This highlights the limitations of deep learning in long-term electricity forecasting and reinforces the importance of model selection based on dataset characteristics rather than complexity. Our study provides insights into practical forecasting applications and contributes to the ongoing discussion on the trade-offs between traditional and modern forecasting methods. |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2505.11390 |
| By: | Altug Aydemir; Mert Gokcu |
| Abstract: | [EN] In recent years, machine learning-based techniques have gained prominence in forecasting crude oil prices due to their ability effectively handle the highly volatile and nonlinear nature of oil prices. The primary objective of this paper is to forecast monthly oil prices with the highest level of precision and accuracy possible. To do this, we propose a deepened and high-parametrized version of the deep neural network model framework that integrates widely adopted algorithms and a variety of datasets. Additionally, our approach involves the optimal architecture for deep neural networks used in oil price forecasting and offers forecasts that are repeatable and consistent. All the evaluation metrics values indicate that the proposed model achieves superior forecasting performance compared to some simple conventional statistical models. [TR] Son zamanlarda, makine ogrenimi tabanli yontemler, petrol fiyatlarinin son derece oynak ve dogrusal olmayan dogasi ile etkin bir sekilde basa cikma yetenekleri sayesinde ham petrol fiyatlarini tahmin etmede onem kazanmistir. Bu calismanin temel amaci, aylik bazda petrol fiyatlarini mumkun olan en yuksek hassasiyet ve dogrulukla tahmin etmektir. Bunu yapmak icin, ham petrol fiyat tahmini icin iyi bilinen algoritmalari ve cesitli veri kumelerini kullanan derin sinir agi modeli cercevesinin derinlestirilmis ve yuksek parametreli bir versiyonunu oneriyoruz. Ayrica, yaklasimimiz petrol fiyat tahmininde kullanilan derin sinir aglari icin en uygun mimariyi icermekte ve tekrarlanabilir ve tutarli tahminler sunmaktadir. Tum degerlendirme metrik degerleri, onerilen modelimizin geleneksel yontemlere kiyasla tahmin performansinda onemli bir iyilesmeye sahip oldugunu gostermektedir. |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:tcb:econot:2511 |
| By: | Anubha Goel; Puneet Pasricha; Martin Magris; Juho Kanniainen |
| Abstract: | Time series foundation models (FMs) have emerged as a popular paradigm for zero-shot multi-domain forecasting. These models are trained on numerous diverse datasets and claim to be effective forecasters across multiple different time series domains, including financial data. In this study, we evaluate the effectiveness of FMs, specifically the TimesFM model, for volatility forecasting, a core task in financial risk management. We first evaluate TimesFM in its pretrained (zero-shot) form, followed by our custom fine-tuning procedure based on incremental learning, and compare the resulting models against standard econometric benchmarks. While the pretrained model provides a reasonable baseline, our findings show that incremental fine-tuning, which allows the model to adapt to new financial return data over time, is essential for learning volatility patterns effectively. Fine-tuned variants not only improve forecast accuracy but also statistically outperform traditional models, as demonstrated through Diebold-Mariano and Giacomini-White tests. These results highlight the potential of foundation models as scalable and adaptive tools for financial forecasting-capable of delivering strong performance in dynamic market environments when paired with targeted fine-tuning strategies. |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2505.11163 |
| By: | Qingyu Li; Chiranjib Mukhopadhyay; Abolfazl Bayat; Ali Habibnia |
| Abstract: | Recent advances in quantum computing have demonstrated its potential to significantly enhance the analysis and forecasting of complex classical data. Among these, quantum reservoir computing has emerged as a particularly powerful approach, combining quantum computation with machine learning for modeling nonlinear temporal dependencies in high-dimensional time series. As with many data-driven disciplines, quantitative finance and econometrics can hugely benefit from emerging quantum technologies. In this work, we investigate the application of quantum reservoir computing for realized volatility forecasting. Our model employs a fully connected transverse-field Ising Hamiltonian as the reservoir with distinct input and memory qubits to capture temporal dependencies. The quantum reservoir computing approach is benchmarked against several econometric models and standard machine learning algorithms. The models are evaluated using multiple error metrics and the model confidence set procedures. To enhance interpretability and mitigate current quantum hardware limitations, we utilize wrapper-based forward selection for feature selection, identifying optimal subsets, and quantifying feature importance via Shapley values. Our results indicate that the proposed quantum reservoir approach consistently outperforms benchmark models across various metrics, highlighting its potential for financial forecasting despite existing quantum hardware constraints. This work serves as a proof-of-concept for the applicability of quantum computing in econometrics and financial analysis, paving the way for further research into quantum-enhanced predictive modeling as quantum hardware capabilities continue to advance. |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2505.13933 |
| By: | Tony Paul, Nitin |
| Abstract: | Traditional GARCH models, while robust, are deterministic and their long-horizon forecasts converge to a static mean, failing to capture the dynamic nature of real markets. Conversely, classical stochastic volatility models often introduce significant implementation and calibration complexity. This paper introduces GARCH-FX (GARCH Forecasting eXtension), a novel and accessible framework that augments the classic GARCH model to generate realistic, stochastic volatility paths without this prohibitive complexity. GARCH-FX is built upon the core strength of GARCH—its ability to estimate long-run variance—but replaces the deterministic multi-step forecast with a stochastic simulation engine. It injects controlled randomness through a Gamma-distributed process, ensuring the forecast path is non-smooth and jagged. Furthermore, it incorporates a modular regime-switching multiplier, providing a flexible interface to inject external views or systematic signals into the forecast’s mean level. The result is a powerful and intuitive framework for generating dynamic long-term volatility scenarios. By separating the drivers of mean-level shifts from local stochastic behavior, GARCHFX aims to provide a practical tool for applications requiring realistic market simulations, such as stress-testing, risk analysis, and synthetic data generation. |
| Keywords: | Stochastic Volatility Forecasting, GARCH Extensions, Regime-Switching Volatility, Gamma-Distributed, Volatility, Volatility Forecast Uncertainty, Nonlinear GARCH Models, Stochastic Vol Forecast, Financial Time Series, Heteroskedasticity Dynamics, Gamma Noise in Volatility |
| JEL: | C22 C53 C6 |
| Date: | 2025–07–10 |
| URL: | https://d.repec.org/n?u=RePEc:pra:mprapa:125321 |
| By: | Altug Aydemir; Cem Cebi |
| Abstract: | This study aims at forecasting the future behavior of budget variables, using Artificial Neural Network (ANN) and Deep Neural Network (DNN) techniques for Türkiye. Particularly, we focus on budget expenditures, tax revenues and their main components. Annual data were used and divided into two sub-periods: a training set (2002-2019) and a test set (2020-2022). Each fiscal item is estimated using relevant explanatory variables selected based on economic theory. We achieved good forecasting performance for main budget items using ANN and DNN methodologies. We found that most of the Mean Absolute Error (MAE) values fell within the acceptable range, an indicator of good prediction performance. Second, we see that the MAE values for public expenditures are lower than taxes. Third, estimating total tax revenues (aggregate data) performs better compared to subcomponents of taxes (disaggregated data). The opposite is the case for public expenditures. |
| Keywords: | Machine Learning, Deep Learning, Artificial Neural Network (ANN), Deep Neural Network (DNN), Budget Forecast, Government Spending, Tax Revenue |
| JEL: | C53 H20 H50 H68 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:tcb:wpaper:2509 |
| By: | Yingjie Kuang; Tianchen Zhang; Zhen-Wei Huang; Zhongjie Zeng; Zhe-Yuan Li; Ling Huang; Yuefang Gao |
| Abstract: | Accurately predicting customers' purchase intentions is critical to the success of a business strategy. Current researches mainly focus on analyzing the specific types of products that customers are likely to purchase in the future, little attention has been paid to the critical factor of whether customers will engage in repurchase behavior. Predicting whether a customer will make the next purchase is a classic time series forecasting task. However, in real-world purchasing behavior, customer groups typically exhibit imbalance - i.e., there are a large number of occasional buyers and a small number of loyal customers. This head-to-tail distribution makes traditional time series forecasting methods face certain limitations when dealing with such problems. To address the above challenges, this paper proposes a unified Clustering and Attention mechanism GRU model (CAGRU) that leverages multi-modal data for customer purchase intention prediction. The framework first performs customer profiling with respect to the customer characteristics and clusters the customers to delineate the different customer clusters that contain similar features. Then, the time series features of different customer clusters are extracted by GRU neural network and an attention mechanism is introduced to capture the significance of sequence locations. Furthermore, to mitigate the head-to-tail distribution of customer segments, we train the model separately for each customer segment, to adapt and capture more accurately the differences in behavioral characteristics between different customer segments, as well as the similar characteristics of the customers within the same customer segment. We constructed four datasets and conducted extensive experiments to demonstrate the superiority of the proposed CAGRU approach. |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2505.13558 |
| By: | Shaobo Li; Ben Sherwood |
| Abstract: | This paper considers equity premium prediction, for which mean regression can be problematic due to heteroscedasticity and heavy-tails of the error. We show advantages of quantile predictions using a novel penalized quantile regression that offers a model for a full spectrum analysis on the equity premium distribution. To enhance model interpretability and address the well-known issue of crossing quantile predictions in quantile regression, we propose a model that enforces the selection of a common set of variables across all quantiles. Such a selection consistency is achieved by simultaneously estimating all quantiles with a group penalty that ensures sparsity pattern is the same for all quantiles. Consistency results are provided that allow the number of predictors to increase with the sample size. A Huberized quantile loss function and an augmented data approach are implemented for computational efficiency. Simulation studies show the effectiveness of the proposed approach. Empirical results show that the proposed method outperforms several benchmark methods. Moreover, we find some important predictors reverse their relationship to the excess return from lower to upper quantiles, potentially offering interesting insights to the domain experts. Our proposed method can be applied to other fields. |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2505.16019 |
| By: | Mahdi Kohan Sefidi |
| Abstract: | Financial crises often occur without warning, yet markets leading up to these events display increasing volatility and complex interdependencies across multiple sectors. This study proposes a novel approach to predicting market crises by combining multilayer network analysis with Long Short-Term Memory (LSTM) models, using Granger causality to capture within-layer connections and Random Forest to model interlayer relationships. Specifically, we utilize Granger causality to model the temporal dependencies between market variables within individual layers, such as asset prices, trading values, and returns. To represent the interactions between different market variables across sectors, we apply Random Forest to model the interlayer connections, capturing the spillover effects between these features. The LSTM model is then trained to predict market instability and potential crises based on the dynamic features of the multilayer network. Our results demonstrate that this integrated approach, combining Granger causality, Random Forest, and LSTM, significantly enhances the accuracy of market crisis prediction, outperforming traditional forecasting models. This methodology provides a powerful tool for financial institutions and policymakers to better monitor systemic risks and take proactive measures to mitigate financial crises. |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2505.11019 |
| By: | Kubra Bolukbas; Ertan Tok |
| Abstract: | The goal of this study is to identify the most effective model for predicting credit risk, the likelihood a commercial loan defaults (become a non-performing loan) in the Turkish banking sector and to determine which firm and loan characteristics influence that risk. The analysis draws on an unbalanced dataset of 1.2 million firm-level observations for 2018–2023, combining financial ratios with detailed loan- and firm-specific information. Class imbalance is addressed through oversampling (including SMOTE) and multiple down-sampling schemes. Although the risk is assessed ex-ante, model performance is evaluated ex-post using the ROC-AUC metric. Within tested conventional econometric and machine learning approaches accompanied with different sampling techniques, Extreme Gradient Boosting (XGBoost) with oversampling delivers the best result with a ROC-AUC score of 0.914. Compared with logistic regression under the same sampling setup, a 4.9- percentage-point increase in test ROC-AUC is attained, confirming the model’s superior predictive performance over conventional approaches. Accordingly, the study finds that the industry and location in which a firm operates, its loan-restructuring status, loan cost and type (fixed vs. floating rate), the firm’s record of bad checks, and core ratios capturing profitability, liquidity and leverage to be the most influential predictors of credit risk. |
| Keywords: | Credit Risk, Machine Learning Techniques, Financial Ratios, Banking Sector, Macro-Financial Stability, Feature Importance |
| JEL: | C52 C53 C55 G17 G2 G32 G33 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:tcb:wpaper:2508 |
| By: | Elliot L. Epstein; Apaar Sadhwani; Kay Giesecke |
| Abstract: | In many financial prediction problems, the behavior of individual units (such as loans, bonds, or stocks) is influenced by observable unit-level factors and macroeconomic variables, as well as by latent cross-sectional effects. Traditional approaches attempt to capture these latent effects via handcrafted summary features. We propose a Set-Sequence model that eliminates the need for handcrafted features. The Set model first learns a shared cross-sectional summary at each period. The Sequence model then ingests the summary-augmented time series for each unit independently to predict its outcome. Both components are learned jointly over arbitrary sets sampled during training. Our approach harnesses the set nature of the cross-section and is computationally efficient, generating set summaries in linear time relative to the number of units. It is also flexible, allowing the use of existing sequence models and accommodating a variable number of units at inference. Empirical evaluations demonstrate that our Set-Sequence model significantly outperforms benchmarks on stock return prediction and mortgage behavior tasks. Code will be released. |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2505.11243 |
| By: | Paker, Meredith; Stephenson, Judy; Wallis, Patrick |
| Abstract: | Understanding long-run economic growth requires reliable historical data, yet the vast majority of long-run economic time series are drawn from incomplete records with significant temporal and geographic gaps. Conventional solutions to these gaps rely on linear regressions that risk bias or overfitting when data are scarce. We introduce “past predictive modeling, ” a framework that leverages machine learning and out-of-sample predictive modeling techniques to reconstruct representative historical time series from scarce data. Validating our approach using nominal wage data from England, 1300-1900, we show that this new method leads to more accurate and generalizable estimates, with bootstrapped standard errors 72% lower than benchmark linear regressions. Beyond just bettering accuracy, these improved wage estimates for England yield new insights into the impact of the Black Death on inequality, the economic geography of pre-industrial growth, and productivity over the long-run. |
| Keywords: | machine learning; predictive modeling; wages; black death; industrial revolution |
| JEL: | J31 C53 N33 N13 N63 |
| Date: | 2025–06–13 |
| URL: | https://d.repec.org/n?u=RePEc:ehl:lserod:128852 |
| By: | Yuke Zhang |
| Abstract: | This study introduces an interpretable machine learning (ML) framework to extract macroeconomic alpha from global news sentiment. We process the Global Database of Events, Language, and Tone (GDELT) Project's worldwide news feed using FinBERT -- a Bidirectional Encoder Representations from Transformers (BERT) based model pretrained on finance-specific language -- to construct daily sentiment indices incorporating mean tone, dispersion, and event impact. These indices drive an XGBoost classifier, benchmarked against logistic regression, to predict next-day returns for EUR/USD, USD/JPY, and 10-year U.S. Treasury futures (ZN). Rigorous out-of-sample (OOS) backtesting (5-fold expanding-window cross-validation, OOS period: c. 2017-April 2025) demonstrates exceptional, cost-adjusted performance for the XGBoost strategy: Sharpe ratios achieve 5.87 (EUR/USD), 4.65 (USD/JPY), and 4.65 (Treasuries), with respective compound annual growth rates (CAGRs) exceeding 50% in Foreign Exchange (FX) and 22% in bonds. Shapley Additive Explanations (SHAP) affirm that sentiment dispersion and article impact are key predictive features. Our findings establish that integrating domain-specific Natural Language Processing (NLP) with interpretable ML offers a potent and explainable source of macro alpha. |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2505.16136 |
| By: | Siye Bae (Northwestern University); Sangyup Choi (Yonsei University); Sang-Hyun Kim (Yonsei University); Myunghwan Andrew Lee (New York University); Myungkyu Shim (Yonsei University) |
| Abstract: | We study how politically diverse households form and update macroeconomic expectations in response to public communication, using novel survey waves of Korean individuals conducted in 2022 during a historic inflation surge. The survey includes a randomized information treatment in which respondents are exposed to government forecasts about inflation stabilization, with treatments varying in messenger, framing, media source, and numerical content. We first document substantial political polarization in macroeconomic beliefs, including inflation expectations. We then find that only pro-government individuals revise their expectations downward in response to the information, while anti-government and centrist individuals remain largely unresponsive, regardless of message source, content, or presentation. These asymmetric responses are driven by differences in trust toward the policy authority, which are themselves linked to partisanship, highlighting the challenges of anchoring expectations in politically polarized environments. |
| Keywords: | Inflation expectations; Macroeconomic beliefs; Partisan bias; Central bank communication; Household survey |
| JEL: | C83 D84 E31 |
| Date: | 2025–07 |
| URL: | https://d.repec.org/n?u=RePEc:yon:wpaper:2025rwp-255 |