nep-for New Economics Papers
on Forecasting
Issue of 2026–03–16
fifteen papers chosen by
Malte Knüppel, Deutsche Bundesbank


  1. Direct Gaussian Process Predictive Regressions with Mixed Frequency Data By Niko Hauzenberger Massimiliano Marcellino Michael Pfarrhofer Anna Stelzer
  2. Beyond Polarity: Multi-Dimensional LLM Sentiment Signals for WTI Crude Oil Futures Return Prediction By Dehao Dai; Ding Ma; Dou Liu; Kerui Geng; Yiqing Wang
  3. A Bipartite Graph Approach to U.S.-China Cross-Market Return Forecasting By Jing Liu; Maria Grith; Xiaowen Dong; Mihai Cucuringu
  4. Deep Learning for Financial Time Series: A Large-Scale Benchmark of Risk-Adjusted Performance By Adir Saly-Kaufmann; Kieran Wood; Jan Peter-Calliess; Stefan Zohren
  5. A robust approach to tilting: parametric relative entropy By Montes-Galdón, Carlos; Paredes, Joan; Wolf, Elias
  6. Adaptive Window Selection for Financial Risk Forecasting By Yinhuan Li; Chenxin Lyu; Ruodu Wang
  7. An Empirical Evaluation of Some Long-Horizon Macroeconomic Forecasts By Kurt G. Lunsford; Kenneth D. West
  8. On the Wisdom of Crowds (of Economists) By Francis X. Diebold; Aaron Mora; Minchul Shin
  9. A robust approach to tilting: parametric relative entropy By Carlos Montes-Galdón; Joan Paredes; Elias Wolf
  10. DatedGPT: Preventing Lookahead Bias in Large Language Models with Time-Aware Pretraining By Yutong Yan; Raphael Tang; Zhenyu Gao; Wenxi Jiang; Yao Lu
  11. Forecasting and Manipulating the Forecasts of Others By Sam Babichenko
  12. Expertise and Prediction Accuracy By Elisabeth Grewenig; Klaus Gründler; Philipp Lergetporer; Niklas Potrafke; Katharina Werner; Helen Zeidler
  13. Short-Term Stock Price Prediction Based on Single and Stacking Machine Learning Models By Chia Yean Lim
  14. A rotated Dynamic Factor Model for the yield curve: squeezing out information when it matters By Chiara Casoli; Riccardo Lucchetti
  15. Прогнозирование ВВП Казахстана на основе динамической факторной модели с регуляризацией // Forecasting Kazakhstan’s GDP Based on a Dynamic Factor Model with Regularization By Ахмет Алишер // Alisher Akhmet

  1. By: Niko Hauzenberger Massimiliano Marcellino Michael Pfarrhofer Anna Stelzer
    Abstract: We develop Bayesian machine learning methods for mixed frequency data. This involves handling frequency mismatches and specifying functional relationships between (possibly many) predictors and low frequency dependent variables. We use Gaussian Processes (GPs) in direct nonlinear predictive regressions, and compress higher frequency variables in a structured way. This yields a set of kernels for GPs with distinct properties and implications. We evaluate the proposed framework in an out-of-sample exercise focusing on quarterly US GDP growth and inflation. Our approach leverages high-dimensional mixed frequency data in a computationally efficient way, and offers robustness and gains in predictive accuracy along several dimensions.
    Keywords: Bayesian nonparametrics, direct forecasting, nowcasting, dimension reduction, MIDAS
    JEL: C11 C22 C53 E31 E37
    Date: 2026
    URL: https://d.repec.org/n?u=RePEc:baf:cbafwp:cbafwp26265
  2. By: Dehao Dai; Ding Ma; Dou Liu; Kerui Geng; Yiqing Wang
    Abstract: Forecasting crude oil prices remains challenging because market-relevant information is embedded in large volumes of unstructured news and is not fully captured by traditional polarity-based sentiment measures. This paper examines whether multi-dimensional sentiment signals extracted by large language models improve the prediction of weekly WTI crude oil futures returns. Using energy-sector news articles from 2020 to 2025, we construct five sentiment dimensions covering relevance, polarity, intensity, uncertainty, and forwardness based on GPT-4o, Llama 3.2-3b, and two benchmark models, FinBERT and AlphaVantage. We aggregate article-level signals to the weekly level and evaluate their predictive performance in a classification framework. The best results are achieved by combining GPT-4o and FinBERT, suggesting that LLM-based and conventional financial sentiment models provide complementary predictive information. SHAP analysis further shows that intensity- and uncertainty-related features are among the most important predictors, indicating that the predictive value of news sentiment extends beyond simple polarity. Overall, the results suggest that multi-dimensional LLM-based sentiment measures can improve commodity return forecasting and support energy-market risk monitoring.
    Date: 2026–03
    URL: https://d.repec.org/n?u=RePEc:arx:papers:2603.11408
  3. By: Jing Liu; Maria Grith; Xiaowen Dong; Mihai Cucuringu
    Abstract: This paper studies cross-market return predictability through a machine learning framework that preserves economic structure. Exploiting the non-overlapping trading hours of the U.S. and Chinese equity markets, we construct a directed bipartite graph that captures time-ordered predictive linkages between stocks across markets. Edges are selected via rolling-window hypothesis testing, and the resulting graph serves as a sparse, economically interpretable feature-selection layer for downstream machine learning models. We apply a range of regularized and ensemble methods to forecast open-to-close returns using lagged foreign-market information. Our results reveal a pronounced directional asymmetry: U.S. previous-close-to-close returns contain substantial predictive information for Chinese intraday returns, whereas the reverse effect is limited. This informational asymmetry translates into economically meaningful performance differences and highlights how structured machine learning frameworks can uncover cross-market dependencies while maintaining interpretability.
    Date: 2026–03
    URL: https://d.repec.org/n?u=RePEc:arx:papers:2603.10559
  4. By: Adir Saly-Kaufmann; Kieran Wood; Jan Peter-Calliess; Stefan Zohren
    Abstract: We present a large scale benchmark of modern deep learning architectures for a financial time series prediction and position sizing task, with a primary focus on Sharpe ratio optimization. Evaluating linear models, recurrent networks, transformer based architectures, state space models, and recent sequence representation approaches, we assess out of sample performance on a daily futures dataset spanning commodities, equity indices, bonds, and FX spanning 2010 to 2025. Our evaluation goes beyond average returns and includes statistical significance, downside and tail risk measures, breakeven transaction cost analysis, robustness to random seed selection, and computational efficiency. We find that models explicitly designed to learn rich temporal representations consistently outperform linear benchmarks and generic deep learning models, which often lead the ranking in standard time series benchmarks. Hybrid models such as VSN with LSTM, a combination of Variable Selection Networks (VSN) and LSTMs, achieves the highest overall Sharpe ratio, while VSN with xLSTM and LSTM with PatchTST exhibit superior downside adjusted characteristics. xLSTM demonstrates the largest breakeven transaction cost buffer, indicating improved robustness to trading frictions.
    Date: 2026–03
    URL: https://d.repec.org/n?u=RePEc:arx:papers:2603.01820
  5. By: Montes-Galdón, Carlos; Paredes, Joan; Wolf, Elias
    Abstract: We introduce a novel methodology, ”parametric tilting, ” for incorporating external information into econometric model-based density forecasts. Unlike traditional entropic tilting, which can generate unrealistic or unstable distributions under certain conditions, parametric tilting ensures more reliable and numerically stable results. Our approach leverages the flexibility of the skew-T distribution, which captures key moments of macroeconomic time series, and minimizes the Kullback-Leibler divergence between the target and model-based distributions. This method overcomes limitations of entropic tilting, such as multimodal or degenerate distributions, providing a robust alternative for policymakers and researchers aiming to integrate external views into probabilistic forecasting frameworks. JEL Classification: C14, C53, E52
    Keywords: entropic tilting, forecasting, Kullback-Leibler information criterion
    Date: 2026–03
    URL: https://d.repec.org/n?u=RePEc:ecb:ecbwps:20263200
  6. By: Yinhuan Li; Chenxin Lyu; Ruodu Wang
    Abstract: Risk forecasts in financial regulation and internal management are calculated through historical data. The unknown structural changes of financial data poses a substantial challenge in selecting an appropriate look-back window for risk modeling and forecasting. We develop a data-driven online learning method, called the bootstrap-based adaptive window selection (BAWS), that adaptively determines the window size in a sequential manner. A central component of BAWS is to compare the realized scores against a data-dependent threshold, which is evaluate based on an idea of bootstrap. The proposed method is applicable to the forecast of risk measures that are elicitable individually or jointly, such as the Value-at-Risk (VaR) and the pair of the VaR and the corresponding Expected Shortfall. Through simulation studies and empirical analyses, we demonstrate that BAWS generally outperforms the standard rolling window approach and the recently developed method of stability-based adaptive window selection, especially when there are structural changes in the data-generating process.
    Date: 2026–03
    URL: https://d.repec.org/n?u=RePEc:arx:papers:2603.01157
  7. By: Kurt G. Lunsford; Kenneth D. West
    Abstract: We use long-run annual cross-country data for 10 macroeconomic variables to evaluate the long-horizon forecast distributions of six forecasting models. The variables we use range from ones having little serial correlation to ones having persistence consistent with unit roots. Our forecasting models include simple time series models and frequency domain models developed in Müller and Watson (2016). For plausibly stationary variables, an AR(1) model and a frequency domain model that does not require the user to take a stand on the order of integration appear reasonably well calibrated for forecast horizons of 10 and 25 years. For plausibly non-stationary variables, a random walk model appears reasonably well calibrated for forecast horizons of 10 and 25 years.
    JEL: C22 C53 E17
    Date: 2026–02
    URL: https://d.repec.org/n?u=RePEc:nbr:nberwo:34904
  8. By: Francis X. Diebold; Aaron Mora; Minchul Shin
    Abstract: We study the properties of macroeconomic survey forecast response averages as the number of survey respondents grows. Such averages are “portfolios” of forecasts. We characterize the speed and pattern of the gains from diversification as a function of portfolio size (the number of survey respondents) in both (1) the key real-world data-based environment of the U.S. Survey of Professional Forecasters, and (2) the theoretical model-based environment of equicorrelated forecast errors. We proceed by proposing and comparing various direct and model-based “crowd size signature plots, ” which summarize the forecasting performance of k-average forecasts as a function of k, where k is the number of forecasts in the average. We then estimate the equicorrelation model for growth and inflation forecast errors by choosing model parameters to minimize the divergence between direct and model-based signature plots. The results indicate near-perfect equicorrelation model fit for both growth and inflation, which we explicate by showing analytically that, under very weak conditions, the direct and fitted equicorrelation model-based signature plots are identical at a particular model parameter configuration. That parameter configuration immediately suggests an analytic closed-form estimator for the direct signature plot, so that equicorrelation ultimately emerges as a device for convenient calculation of direct signature plots, rather than a separate “model” producing separate signature plots. Finally, we find that the gains from survey diversification are greater for inflation forecasts than for growth forecasts, and that they are largely exhausted with inclusion of 5–10 representative forecasters.
    Keywords: Survey of professional forecasters; forecast combination; model averaging; equicorrelation
    JEL: C5 C8 E3 E6
    Date: 2026–03–09
    URL: https://d.repec.org/n?u=RePEc:fip:fedpwp:102886
  9. By: Carlos Montes-Galdón (European Central Bank); Joan Paredes (National Bank of Slovakia); Elias Wolf (European Stability Mechanism Bank and Universitat Bonn)
    Abstract: We introduce a novel methodology, †parametric tilting, †for incorporating external information into econometric model-based density forecasts. Unlike traditional entropic tilting, which can generate unrealistic or unstable distributions under certain conditions, parametric tilting ensures more reliable and numerically stable results. Our approach leverages the flexibility of the skew-T distribution, which captures key moments of macroeconomic time series, and minimizes the Kullback-Leibler divergence between the target and model-based distributions. This method overcomes limitations of entropic tilting, such as multimodal or degenerate distributions, providing a robust alternative for policymakers and researchers aiming to integrate external views into probabilistic forecasting frameworks.
    JEL: C14 C53 E52
    Date: 2026–02
    URL: https://d.repec.org/n?u=RePEc:svk:wpaper:1136
  10. By: Yutong Yan; Raphael Tang; Zhenyu Gao; Wenxi Jiang; Yao Lu
    Abstract: In financial backtesting, large language models pretrained on internet-scale data risk introducing lookahead bias that undermines their forecasting validity, as they may have already seen the true outcome during training. To address this, we present DatedGPT, a family of twelve 1.3B-parameter language models, each trained from scratch on approximately 100 billion tokens of temporally partitioned data with strict annual cutoffs spanning 2013 to 2024. We further enhance each model with instruction fine-tuning on both general-domain and finance-specific datasets curated to respect the same temporal boundaries. Perplexity-based probing confirms that each model's knowledge is effectively bounded by its data cutoff year, while evaluation on standard benchmarks shows competitive performance with existing models of similar scale. We provide an interactive web demo that allows users to query and compare responses from models across different cutoff years.
    Date: 2026–03
    URL: https://d.repec.org/n?u=RePEc:arx:papers:2603.11838
  11. By: Sam Babichenko
    Abstract: In strategic environments with private information, evaluating a change in policy requires predicting how the equilibrium responds -- but when actions reshape opponents' signals, each agent's optimal response depends on an infinite hierarchy of beliefs about beliefs that has resisted exact analysis for four decades. We provide the first exact equilibrium characterization of finite-player continuous-time LQG games with endogenous signals. Conditioning on primitive Brownian shocks rather than the physical state -- a dynamic analogue of Harsanyi's common-prior construction -- collapses the belief hierarchy onto deterministic two-time kernels, reducing Nash equilibrium to a deterministic fixed point with no truncation and no large-population limit. The characterization yields an explicit information wedge $\mathcal{V}^i_t$ -- a deterministic Volterra process -- that prices the marginal value of shifting opponents' posteriors. The wedge vanishes precisely when signals are exogenous to controls, formally delineating the boundary where strategic belief manipulation matters, and provides a closed-form mapping from information primitives to equilibrium outcomes.
    Date: 2026–03
    URL: https://d.repec.org/n?u=RePEc:arx:papers:2603.12140
  12. By: Elisabeth Grewenig; Klaus Gründler; Philipp Lergetporer; Niklas Potrafke; Katharina Werner; Helen Zeidler
    Abstract: Public support for policy interventions depends on citizens’ beliefs about their likely ef-fects. We examine how individuals form such beliefs by studying their predictions of experimental outcomes in a policy-relevant setting, and why their predictions differ from expert benchmarks. We elicit forecasts from 127 professional economists and a repre-sentative sample of 6, 200 German households about a large-scale behavioral experi-ment on education policy (N = 3, 133). Non-experts predict both average outcomes and treatment effects far less accurately than experts. Prediction accuracy improves with calibrated priors, self-reported effort, and the use of structured reasoning, but remains well below expert levels. We show that scalable design features, including the provision of well-calibrated numerical anchors and monetary incentives to rise effort, improve non-expert predictions, with effects comparable in magnitude to tertiary education or structured reasoning. Our findings have important implications for bridging the ‘expertise gap’ in public discourse.
    Keywords: expert forecasts, lay predictions, belief formation, expertise gap, policy support, behavioral experiments
    JEL: A11 D83 H52 I22
    Date: 2026
    URL: https://d.repec.org/n?u=RePEc:ces:ceswps:_12522
  13. By: Chia Yean Lim (School of Computer Sciences, Universiti Sains Malaysia, 11800, Minden, Malaysia Author-2-Name: Wenchuan Sun Author-2-Workplace-Name: School of Computer Sciences, Universiti Sains Malaysia, 11800, Minden, Malaysia Author-3-Name: Fengqi Guo Author-3-Workplace-Name: CITIC Securities, 150000, Harbin, China Author-4-Name: Sau Loong Ang Author-4-Workplace-Name: Department of Computing and Information Technology, Tunku Abdul Rahman University of Management and Technology, Penang Branch, 11200, Tanjung Bungah, Malaysia Author-5-Name: Author-5-Workplace-Name: Author-6-Name: Author-6-Workplace-Name: Author-7-Name: Author-7-Workplace-Name: Author-8-Name: Author-8-Workplace-Name:)
    Abstract: " Objective - As the investment environment improves, individuals are increasingly eager to invest their idle funds. Securities companies have become the preferred choice for buying financial products. The current accuracy of stock predictions relies on the comprehensive models used by each securities company, including stock market trading, data, and stock pricing models. However, securities companies have not adequately explored a single suitable model for stock predictions and have rarely assessed the effectiveness of stacking and ensemble methods in improving these predictions. Methodology - This research first explored and proposed the best single-stock prediction model. Next, it combined four individual prediction models to create a stacking model. Findings - The comparison between the single and stacking models demonstrated that the stacking model's prediction accuracy exceeded that of the single model. Therefore, it is recommended that securities companies adopt a stacking-type prediction model to forecast share prices for their investment customers. Novelty - Using a stacking model could improve the accuracy of stock price predictions for investment managers, help users make better decisions, and ultimately enhance the company's earnings by delivering more accurate investment outcomes. Type of Paper - Empirical"
    Keywords: Long short-term memory, random forest model, stacking model, stock prediction, support vector machine, XGBoost model.
    JEL: F17 F47
    Date: 2026–03–31
    URL: https://d.repec.org/n?u=RePEc:gtr:gatrjs:gjbssr674
  14. By: Chiara Casoli (InsIDE Lab, DiEco, Università degli Studi dell’Insubria and Fondazione Eni Enrico Mattei); Riccardo Lucchetti (DiSES, Università Politecnica delle Marche)
    Abstract: The yield curve is widely regarded as a powerful descriptor of the economy and market expectations. A common approach to its statistical representation relies on a small number of factors summarizing the curve, which can then be used to forecast real economic activity.We argue that optimal factor extraction is crucial for retrieving information when considering an approximate factor model. By introducing a rotation of the model including cointegration, we reduce cross-sectional dependence in the idiosyncratic components. This leads to improved forecasts of key macroeconomic variables during periods of economic and financial instability, both in the US and the euro area.
    Keywords: Yield curve, Nelson-Siegel model, Dynamic Factor Model, cointegration, forecasting
    JEL: C32 C53 E43 E44
    Date: 2026–01
    URL: https://d.repec.org/n?u=RePEc:fem:femwpa:2026.03
  15. By: Ахмет Алишер // Alisher Akhmet (National Bank of Kazakhstan)
    Abstract: В работе предлагается подход к краткосрочному прогнозированию валового внутреннего продукта (ВВП) Казахстана на основе динамической факторной модели (DFM), оцененной по широкой панели макроэкономических и отраслевых показателей. Модель позволяет извлечь латентные факторы, отражающие основные источники совместной динамики в экономике, и использовать их для прогнозирования ВВП в условиях неполной и асинхронной информации. Оценка факторов осуществляется в пространстве состояний с применением фильтра Калмана, что обеспечивает корректную обработку пропусков и различной периодичности данных. Для повышения устойчивости прогнозов и учета изменяющихся во времени взаимосвязей используется регуляризированная регрессионная спецификация с экспоненциальным затуханием весов наблюдений. Прогнозная точность модели оценивается в рамках расширяющегося окна, что позволяет имитировать условия реального прогнозного раунда и исключить использование будущей информации. Сравнение с наивным прогнозом и авторегрессионной моделью ВВП показывает, что факторная структура с регуляризацией обеспечивает существенное снижение прогнозной неопределенности и демонстрирует высокую информативность на краткосрочном горизонте. // This paper develops an approach to short-term forecasting of Kazakhstan’s gross domestic product (GDP) based on a dynamic factor model (DFM) estimated using a broad panel of macroeconomic and sectoral indicators. The model enables the extraction of latent factors that represent the main sources of common variation in the economy and their use for forecasting GDP under conditions of incomplete and asynchronous information. The factors are estimated within a state-space framework using the Kalman filter, which allows for a consistent treatment of missing observations and mixed data frequencies. To enhance forecast robustness and accommodate time variation in economic relationships, a regularized regression specification with exponential decay of observation weights is applied. Forecast performance is assessed using an expanding-window evaluation scheme that replicates real-time forecasting conditions and precludes the use of future information. A comparison with a naïve benchmark and an autoregressive model of GDP indicates that the regularized factor-based specification substantially reduces forecast uncertainty and yields more informative short-term forecasts.
    Keywords: динамические факторные модели, прогнозирование ВВП, nowcasting, регуляризация, dynamic factor models, GDP forecasting, nowcasting, regularization
    JEL: C32 C38 C51 C53 E32 O47
    Date: 2026
    URL: https://d.repec.org/n?u=RePEc:aob:wpaper:69

This nep-for issue is ©2026 by Malte Knüppel. It is provided as is without any express or implied warranty. It may be freely redistributed in whole or in part for any purpose. If distributed in part, please include this notice.
General information on the NEP project can be found at https://nep.repec.org. For comments please write to the director of NEP, Marco Novarese at <director@nep.repec.org>. Put “NEP” in the subject, otherwise your mail may be rejected.
NEP’s infrastructure is sponsored by the Griffith Business School of Griffith University in Australia.