|
on Forecasting |
| By: | Alcaráz, Alba; Capilla, Javier; Garcia-Hiernaux, Alfredo; Pérez-Amaral, Teodosio; Valarezo-Unda, Angel |
| Abstract: | In this work, a cost function is estimated for eight models from the M4 competition. The main objective of the M competitions is to evaluate the accuracy of numerous forecasting models. This study introduces metrics to measure the environmental cost associated with running different time series models during the training and forecasting phases. This approach enables the construction of an environmental cost function that depends on other explanatory variables. Interpretable models help identify key drivers of environmental impact, while more complex machine learning models are used to predict emissions without rerunning the algorithms. The findings contribute to Green AI by promoting the evaluation of forecasting models not only by forecasting precision but also by sustainability. |
| Keywords: | Green IA, M-competitions, forecasting, machine learning, sustainability |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:itse25:331246 |
| By: | Solhdoost, Mohsen (Xi'an Jiaotong-Liverpool University) |
| Abstract: | This study estimates the near-term risk of renewed Israel–Iran escalation using a modular, multi-model framework that converts expert judgments and open-source indicators into probabilistic scenarios over a 90-day horizon. We elicit LOW–CENTRAL–HIGH priors from domain experts across 18 indicators spanning air/missile, land, maritime, cyber, diplomacy, information, and proxy alignment (including Israeli standoff and air-strike activity), then fuse evidence via Bayesian updating and infer a weekly escalation state on a seven-rung ladder with a light stocks/flows scaffold for magazines, interceptor use, and repair/re-supply constraints. We triangulate forecasts across three complementary stacks (Bayesian state-space, statistical ensemble, and game-theoretic signalling). At the analysis cut-off (2025-10-22, UK), All three models agree that S1 (Managed Conflict) is modal, S2 (Northern War with Maritime Squeeze) is the principal alternative, and S3–S4 remain lower-probability tails. Sensitivity analysis reveals that the S1<->S2 margin is most responsive to air/missile defence saturation and combined launch/strike pressure together with maritime war-risk stress, with mediation activity providing the strongest stabilising counterweight. We also formalise tail-risk triggers for potential state fracture and specify how crossing them would reweight S4. The result is a transparent, updateable, and non-partisan forecast designed for decision support: it communicates where risk mass sits, what could move it, and which levers plausibly bend trajectories while avoiding operationally sensitive detail. |
| Date: | 2025–11–26 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:jkzby_v1 |
| By: | Sebastian Beer; Brian Erard; Tibor Hanappi |
| Abstract: | Corporate income tax (CIT) collections are among the most difficult revenues to forecast—even with adequate staffing, comprehensive data, and a stable tax design. In practice, forecasting units typically operate under less ideal conditions. As institutional constraints take time to ease, this Note sets out a practical toolkit of methods to strengthen forecasting capacity across a wide range of country contexts. It outlines techniques that provide unbiased forecasts even when the impact of past reforms is only partially known, introduces approaches to account for ongoing and prospective policy changes to leverage time-series approaches, and highlights the potential efficiency gains achievable through structural modeling. A simple empirical assessment of forecasting specifications shows that parsimonious regression models, when backed by sufficient data, can improve prediction accuracy, even though the benchmark of assuming CIT revenues grow in line with GDP remains difficult to beat. |
| Keywords: | Corporate Income Tax; Tax Revenue Forecasting; Fiscal Policy; Tax Policy Changes; Economic Forecasting; Tax Base Elasticity; Microsimulation Models |
| Date: | 2025–11–24 |
| URL: | https://d.repec.org/n?u=RePEc:imf:imfhtn:2025/010 |
| By: | Freek Holvoet; Christopher Blier-Wong; Katrien Antonio |
| Abstract: | Incorporating spatial information, particularly those influenced by climate, weather, and demographic factors, is crucial for improving underwriting precision and enhancing risk management in insurance. However, spatial data are often unstructured, high-dimensional, and difficult to integrate into predictive models. Embedding methods are needed to convert spatial data into meaningful representations for modelling tasks. We propose a novel multi-view contrastive learning framework for generating spatial embeddings that combine information from multiple spatial data sources. To train the model, we construct a spatial dataset that merges satellite imagery and OpenStreetMap features across Europe. The framework aligns these spatial views with coordinate-based encodings, producing low-dimensional embeddings that capture both spatial structure and contextual similarity. Once trained, the model generates embeddings directly from latitude-longitude pairs, enabling any dataset with coordinates to be enriched with meaningful spatial features without requiring access to the original spatial inputs. In a case study on French real estate prices, we compare models trained on raw coordinates against those using our spatial embeddings as inputs. The embeddings consistently improve predictive accuracy across generalised linear, additive, and boosting models, while providing interpretable spatial effects and demonstrating transferability to unseen regions. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2511.17954 |
| By: | Ciftci, Muhsin; Wieland, Elisabeth |
| Abstract: | In this paper, we evaluate a set of measures of underlying inflation for Germany using conventional measures, such as core inflation (excluding energy and food items), and alternative measures based on econometric models, machine learning, and micro-price evidence. We compare these measures through detailed in-sample and out-of-sample evaluations. The alternative measures exhibit lower volatility, minimal bias, and superior out-of-sample forecasting accuracy performance. While we find no evidence that any single measure clearly outperforms the others over time, the range of alternatives measures also reflects a somewhat earlier uptick and downturn in light of the recent inflation surge in comparison to traditional ones. In addition, all measures under consideration are highly sensitive to monetary policy shocks. |
| Keywords: | Underlying inflation, monetary policy, local projections, machine learning |
| JEL: | E31 E37 C22 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:bubtps:333424 |
| By: | Celani, Alessandro (Örebro University School of Business); Pedini, Luca (Fondazione ENI Enrico Mattei (FEEM)) |
| Abstract: | This paper proposes a parsimonious reparametrization for time-varying parameter models that captures smooth dynamics through a low-dimensional state process combined with B-spline weights. We apply this framework to TVP-VARs, yielding Moderate TVP-VARs that retain the interpretability of standard specifications while mitigating overfitting. Monte Carlo evidence shows faster estimation, lower bias, and strong robustness to knot placement. In U.S. macroeconomic data, moderate specifications recover meaningful long-run movements, produce stable impulse responses and deliver superior density forecasts and predictive marginal likelihoods relative to conventional TVP-VARs, particularly in high-dimensional settings. |
| Keywords: | Time-Varying Parameter models; High-dimensional Vector Autoregressions; Stochastic Volatility; B-splines; Macroeconomic Forecasting |
| JEL: | C11 C33 C53 |
| Date: | 2025–12–02 |
| URL: | https://d.repec.org/n?u=RePEc:hhs:oruesi:2025_016 |
| By: | Sarra Ben Yahia (CES - Centre d'économie de la Sorbonne - UP1 - Université Paris 1 Panthéon-Sorbonne - CNRS - Centre National de la Recherche Scientifique); Jose Angel Garcia Sanchez (CES - Centre d'économie de la Sorbonne - UP1 - Université Paris 1 Panthéon-Sorbonne - CNRS - Centre National de la Recherche Scientifique); Rania Hentati Kaffel (CES - Centre d'économie de la Sorbonne - UP1 - Université Paris 1 Panthéon-Sorbonne - CNRS - Centre National de la Recherche Scientifique) |
| Abstract: | Our research assesses the predictive value of LLM-based sentiment in forecasting energy stock returns. Using FinBERT-derived sentiment indicators from 415, 193 tweets spanning 2018-2024, we find statistically significant causal relationships for 80% of companies analyzed. Our VAR analysis reveals heterogeneous optimal lag structures ranging from 2 to 14 days, providing econometric evidence against semi-strong market efficiency. Our results show that the accuracy of the forecast depends critically on the quality and coverage of the data. Our contribution is twofold: (i) a scalable LLMdriven pipeline to quantify firm-level sentiment at daily frequency, and (ii) an econometric validation via VAR/Granger that uncovers economically meaningful lead-lag patterns |
| Keywords: | sentiment analysis, LLM, FinBERT, energy equity markets, Twitter/X sentiment, return forecasting, webscraping, information diffusion, information extraction, finBERT, financial NLP, VAR |
| Date: | 2025–09–30 |
| URL: | https://d.repec.org/n?u=RePEc:hal:cesptp:hal-05312326 |
| By: | Capilla, Javier; Alcaráz, Alba; Valarezo, Angel; Garcia-Hiernaux, Alfredo; Pérez-Amaral, Teodosio |
| Abstract: | Eco-RETINA is an innovative and eco-friendly algorithm explicitly designed for out-of-sample prediction. Functioning as a regression-based flexible approximator, it is linear in parameters but nonlinear in inputs, employing a selective model search to optimize performance. The algorithm adeptly manages multicollinearity, emphasizing speed, accuracy, and environmental sustainability. Its modular and transparent structure facilitates easy interpretation and modification, making it an invaluable tool for researchers in developing explicit models for out-of-sample forecasting. The algorithm generates outputs such as a list of relevant transformed inputs, coefficients, standard deviations, and confidence intervals, enhancing its interpretability. |
| JEL: | C14 C45 C51 C63 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:itse25:331255 |
| By: | Stefano DellaVigna; Eva Vivalt |
| Abstract: | Forecasts about research findings affect critical scientific decisions, such as what treatments an R&D lab invests in, or which papers a researcher decides to write. But what do we know about the accuracy of these forecasts? We analyze a unique data set of all 100 projects posted on the Social Science Prediction Platform from 2020 to 2024, which received 53, 298 forecasts in total, including 66 projects for which we also have results. We show that forecasters, on average, over-estimate treatment effects; however, the average forecast is quite predictive of the actual treatment effect. We also examine differences in accuracy across forecasters. Academics have a slightly higher accuracy than non-academics, but expertise in a field does not increase accuracy. A panel of motivated repeat forecasters has higher accuracy, but this does not extend more broadly to all repeat forecasters. Confidence in the accuracy of one's forecasts is perversely associated with lower accuracy. We also document substantial cross-study correlation in accuracy among forecasters and identify a group of "superforecasters". Finally, we relate our findings to results in the literature as well as to expert forecasts. |
| JEL: | D01 D91 O12 |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:34493 |