|
on Forecasting |
| By: | Gergely Ganics (BANCO DE ESPAÑA); Lluc Puig Codina (UNIVERSITY OF ALICANTE AND BANCO DE ESPAÑA) |
| Abstract: | We propose a simplified framework for evaluating conditional predictive densities based on the probability integral transform (PIT). The approach accommodates a wide range of estimation schemes, including expanding and rolling windows, and applies to both stationary and non-stationary processes. By treating the PIT as a primitive, our approach enables researchers to apply widely used tests in settings where their validity was previously uncertain. Monte Carlo simulations demonstrate favorable size and power properties of the tests. In an empirical application, we show that incorporating stochastic volatility into an unobserved components model is essential for generating correctly calibrated density forecasts of US industrial production growth at both monthly and quarterly frequencies. |
| Keywords: | predictive density, forecast evaluation, probability integral transform, Kolmogorov–Smirnov test, Cramér–von Mises test |
| JEL: | C22 C52 C53 |
| Date: | 2035–09 |
| URL: | https://d.repec.org/n?u=RePEc:bde:wpaper:2535 |
| By: | Xiyuan Liu (School of Economics and Management, Tshinghua University, Beijing, Beijing 100084, China); Zongwu Cai (Department of Economics, The University of Kansas, Lawrence, KS 66045, USA); Liangjun Su (School of Economics and Management, Tshinghua University, Beijing, Beijing 100084, China) |
| Abstract: | We study a novel time-varying (TV) factor-augmented (FA) forecasting model, where the forecast target is driven by a strict subset of all the latent factors driving the predictors. To consistently select the target-related factors and estimate the TV parameters simultaneously, we first obtain the unobserved common factors via the local principal component analysis. Next, we conduct a variable selection procedure via a time-varying weighted group least absolute shrinkage and selection operator to select relevant factors. The identification restrictions used in this paper permit asymptotically rotation-free estimation of both factors and loadings. The asymptotic properties, such as consistency, sparsity and the oracle property of these two-step estimators are established. Simulation studies demonstrate the excellent finite sample performance of the proposed estimators. In an empirical application to the U.S. macroeconomic dataset, we show that the penalized TV-FA forecasting model outperforms the conventional TV-FAVAR model in predicting certain key macroeconomic series |
| Keywords: | Factor-augmented forecasting models; Local-linear smoothing; Structural change; Weighted group LASSO, Time-varying modeling |
| JEL: | C13 C23 C33 C38 |
| Date: | 2025–09 |
| URL: | https://d.repec.org/n?u=RePEc:kan:wpaper:202515 |
| By: | Rehim Kılıç |
| Abstract: | This paper investigates whether the “virtue of complexity” (VoC), documented in equity return prediction, extends to exchange rate forecasting. Using nonlinear Ridge regressions with Random Fourier Features (Ridge–RFF), we compare the predictive performance of complex models against linear regression and the robust random walk benchmark. Forecasts are constructed across three sets of economic fundamentals—traditional monetary, expanded monetary and non-monetary, and Taylor-rule predictors—with nominal complexity varied through rolling training windows of 12, 60, and 120 months. Our results offer a cautionary perspective. Complexity delivers only modest, localized gains: in very small samples with rich predictor sets, Ridge–RFF can outperform linear regression. Yet these improvements never translate into systematic gains over the random walk. As training windows expand, Ridge–RFF quickly loses ground, while linear regression increasingly dominates, at times even surpassing the random walk under expanded fundamentals. Market-timing analyses reinforce these findings: complexity-based strategies yield occasional short-sample gains but are unstable and prone to sharp drawdowns, whereas simpler linear and random walk strategies provide more robust and consistent economic value. By incorporating formal forecast evaluation tests—including Clark–West and Diebold–Mariano—we show that apparent gains from complexity are fragile and rarely statistically significant. Overall, our evidence points to a limited virtue of complexity in FX forecasting: complexity may help under narrowly defined conditions, but parsimony and the random walk benchmark remain more reliable across samples, predictor sets, and economic evaluations. |
| Keywords: | Foreign exchange rate; Exchange rate disconnect puzzle; Predictability; Complexity; Machine learning; Ridge; RFF |
| JEL: | F41 C50 G11 G15 |
| Date: | 2025–09–25 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedgfe:2025-89 |
| By: | Saiz, Lorena; Magro, Manuel Medina |
| Abstract: | This study introduces a novel approach to dictionary-based sentiment analysis that extracts valuable insights from economic newspaper articles in the euro area without requiring article translation. We develop sentiment indices that accurately measure economic, labour, and inflation perceptions in Germany, France, Italy, and Spain using native-language texts. The aggregation of these country-specific sentiments provides a reliable indicator for the euro area as a whole, demonstrating the effectiveness of our approach in several nowcasting and forecasting experiments. This translation-free method significantly reduces resource requirements, facilitates easy replication across various languages, and enables daily updates. By eliminating the translation bottleneck, our approach emerges as one of the most timely and cost-effective economic measures available, offering a powerful tool for monitoring and forecasting business cycles in the multilingual context of the euro area. JEL Classification: E32, E37, C53, C82 |
| Keywords: | forecasting, inflation, output, recession, textual analysis |
| Date: | 2025–09 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbwps:20253122 |
| By: | Xu, R.; Fan, Q. |
| Abstract: | We propose a characteristics-augmented quantile factor (QCF) model in which unknown factor loading functions are linked to a large set of observed individual-level (e.g., bond- or stock-specific) covariates via a single-index projection. The single-index specification offers a parsimonious, interpretable, and statistically efficient way to nonparametrically characterize the time-varying loadings, thereby circumventing the curse of dimensionality in flexible nonparametric models. Employing a three-step sieve estimation procedure, the QCF model exhibits superior in-sample and out-of-sample performance in simulations. We derive asymptotic properties for the estimators of the latent factors, loading functions, and index parameters. In an empirical study, we analyse the dynamic distributional structure of U.S. corporate bond returns from 2003 to 2020. Our approach outperforms bench-mark models, including the quantile Fama-French five-factor model and the quantile latent factor model, especially in the tails (Ï„ = 0.05, 0.95). The model uncovers state-dependent risk exposures influenced by characteristics such as bond and equity volatility, coupon rate, and spread. Finally, we offer economic interpretations of the latent factors. |
| Keywords: | Quantile Latent Factor, Panel Nonlinear Regression, Single-Index Model, Corporate Bonds |
| JEL: | C14 C31 C32 C38 G12 |
| Date: | 2025–09–08 |
| URL: | https://d.repec.org/n?u=RePEc:cam:camdae:2562 |
| By: | Matteo Aquilina; Douglas Kiarelly Godoy de Araujo; Gaston Gelos; Taejin Park; Fernando Perez-Cruz |
| Abstract: | Predicting financial market stress has long proven to be a largely elusive goal. Advances in artificial intelligence and machine learning offer new possibilities to tackle this problem, given their ability to handle large datasets and unearth hidden nonlinear patterns. In this paper, we develop a new approach based on a combination of a recurrent neural network (RNN) and a large language model. Focusing on deviations from triangular arbitrage parity (TAP) in the Euro-Yen currency pair, our RNN produces interpretable daily forecasts of market dysfunction 60 business days ahead. To address the "black box" limitations of RNNs, our model assigns data-driven, time-varying weights to the input variables, making its decision process transparent. These weights serve a dual purpose. First, their evolution in and of itself provides early signals of latent changes in market dynamics. Second, when the network forecasts a higher probability of market dysfunction, these variable-specific weights help identify relevant market variables that we use to prompt an LLM to search for relevant information about potential market stress drivers. |
| Keywords: | market dysfunction, liquidity, arbitrage, artificial intelligence, financial stability |
| JEL: | G14 G15 G17 |
| Date: | 2025–09 |
| URL: | https://d.repec.org/n?u=RePEc:bis:biswps:1291 |
| By: | Julia Hellstrand (Max Planck Institute for Demographic Research, Rostock, Germany); Linus Andersson; Lars Dommermuth; Peter Fallesen; Ari Klængur Jónsson; Marika Jalovaara; Mikko Myrskylä (Max Planck Institute for Demographic Research, Rostock, Germany) |
| Abstract: | Recent period trends in the Nordic countries show rapid declines in first births, particularly among lower-educated men and women. This study translates these period changes into cohort patterns and analyzes observed and forecasted ultimate childlessness by education for men and women born 1970–1987/88 in Denmark, Finland, Iceland, Norway, and Sweden using register-based data. We apply three forecasting methods: freeze rates, five-year extrapolation, and a nonparametric approach based on historical first-birth probabilities. Results reveal the steepest increases in ultimate childlessness among the lowest educated, approaching as high as 40% among low-educated women and 50% among low-educated men in some of the countries. Among the higher tertiary educated, childlessness is overall lower and remains relatively stable. By contrast, men with lower tertiary education show notable increases in childlessness, in some cases reaching levels similar to or higher than those of upper-secondary-educated men. While overall childlessness in Denmark remains stable, it exhibits the fastest widening educational gap. These findings underscore a growing educational polarization in the transition to parenthood across the Nordic societies, with women’s childlessness patterns increasingly resembling those of men—a marked shift in the region’s fertility landscape. Keywords: Ultimate childlessness, educational gradients, Nordic countries, forecasting, gender convergence |
| JEL: | J1 Z0 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:dem:wpaper:wp-2025-029 |
| By: | Simons, J. R.; Chen, Y.; Brunner, E.; French, E. |
| Abstract: | This paper estimates the stochastic process of how dementia incidence evolves over time. We proceed in two steps: first, we estimate a time trend for dementia using a multi-state Cox model. The multi-state model addresses problems of both interval censoring arising from infrequent measurement and also measurement error in dementia. Second, we feed the estimated mean and variance of the time trend into a Kalman filter to infer the population level dementia process. Using data from the English Longitudinal Study of Aging (ELSA), we find that dementia incidence is no longer declining in England. Furthermore, our forecast is that future incidence remains constant, although there is considerable uncertainty in this forecast. Our twostep estimation procedure has significant computational advantages by combining a multi-state model with a time series method. To account for the short sample that is available for dementia, we derive expressions for the Kalman filter’s convergence speed, size, and power to detect changes and conclude our estimator performs well even in short samples. |
| Keywords: | Dementia Incidence, Time Trends, Forecasting |
| Date: | 2025–09–09 |
| URL: | https://d.repec.org/n?u=RePEc:cam:camdae:2563 |
| By: | Gallego-Moll, Carlos; Carrasco-Ribelles, Lucía A.; Casajuana, Marc; Maynou, Laia; Arocena, Pablo; Violán, Concepción; Zabaleta-Del-Olmo, Edurne |
| Abstract: | Objectives: To broadly map the research landscape to identify trends, gaps, and opportunities in data sets, methodologies, outcomes, and reporting standards for artificial intelligence (AI)-based healthcare utilization prediction. Methods: We conducted a scoping review following the Joanna Briggs Institute methodology. We searched 3 major international databases (from inception to January 2025) for studies applying AI in predictive healthcare utilization. Extracted data were categorized into data sets characteristics, AI methods and performance metrics, predicted outcomes, and adherence to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) + AI reporting guidelines. Results: Among 1116 records, 121 met inclusion criteria. Most were conducted in the United States (62%). No study incorporated all 6 relevant variable groups: demographic, socioeconomic, health status, perceived need, provider characteristics, and prior utilization. Only 7 studies included 5 of these groups. The main data sources were electronic health records (60%) and claims (28%). Ensemble models were the most frequently used (66.9%), whereas deep learning models were less common (16.5%). AI methods were primarily used to predict future events (90.1%), with hospitalizations (57.9%) and visits (33.1%) being the most predicted outcomes. Adherence to general reporting standards was moderate; however, compliance with AI-specific TRIPOD + AI items was limited. Conclusions: Future research should broaden predicted outcomes to include process- and logistics-oriented events, extend applications beyond prediction—such as cohort selection and matching—and explore underused AI methods, including distance-based algorithms and deep neural networks. Strengthening adherence to TRIPOD-AI reporting guidelines is also essential to enhance the reliability and impact of AI in healthcare planning and economic evaluation. |
| Keywords: | artificial intelligence; health economics; healthcare utilisation outcomes; resource allocation; review |
| JEL: | J1 |
| Date: | 2025–08–01 |
| URL: | https://d.repec.org/n?u=RePEc:ehl:lserod:129293 |