|
on Forecasting |
| By: | Andrzej Tokajuk; Jaros{\l}aw A. Chudziak |
| Abstract: | Forecasting cryptocurrency prices is hindered by extreme volatility and a methodological dilemma between information-scarce univariate models and noise-prone full-multivariate models. This paper investigates a partial-multivariate approach to balance this trade-off, hypothesizing that a strategic subset of features offers superior predictive power. We apply the Partial-Multivariate Transformer (PMformer) to forecast daily returns for BTCUSDT and ETHUSDT, benchmarking it against eleven classical and deep learning models. Our empirical results yield two primary contributions. First, we demonstrate that the partial-multivariate strategy achieves significant statistical accuracy, effectively balancing informative signals with noise. Second, we experiment and discuss an observable disconnect between this statistical performance and practical trading utility; lower prediction error did not consistently translate to higher financial returns in simulations. This finding challenges the reliance on traditional error metrics and highlights the need to develop evaluation criteria more aligned with real-world financial objectives. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.04099 |
| By: | Pablo Guerron-Quintana (Boston College; Boston College); Amazon Web Services |
| Abstract: | Pretrained time series models have enabled inference-only forecasting systems that produce accurate predictions without task-specific training. However, existing approaches largely focus on univariate forecasting, limiting their applicability in real-world scenarios where multivariate data and covariates play a crucial role. We present Chronos-2, a pretrained model capable of handling univariate, multivariate, and covariate-informed forecasting tasks in a zero-shot manner. Chronos-2 employs a group attention mechanism that facilitates in-context learning (ICL) through efficient information sharing across multiple time series within a group, which may represent sets of related series, variates of a multivariate series, or targets and covariates in a forecasting task. These general capabilities are achieved through training on synthetic datasets that impose diverse multivariate structures on univariate series. Chronos-2 delivers state-of-the-art performance across three comprehensive benchmarks: fev-bench, GIFT-Eval, and Chronos Benchmark II. On fev-bench, which emphasizes multivariate and covariate-informed forecasting, Chronos-2’s universal ICL capabilities lead to substantial improvements over existing models. On tasks involving covariates, it consistently outperforms baselines by a wide margin. Case studies in the energy and retail domains further highlight its practical advantages. The in-context learning capabilities of Chronos-2 establish it as a general-purpose forecasting model that can be used “as is” in real-world forecasting pipelines. |
| Keywords: | forecasting, time series |
| Date: | 2025–12–10 |
| URL: | https://d.repec.org/n?u=RePEc:boc:bocoec:1105 |
| By: | César Anzola Bravo; Paola Poveda |
| Abstract: | Food prices have consistently been one of the leading contributors to Colombia’s inflation rate. They are particularly sensitive to exogenous factors such as extreme weather events, supply chain disruptions, and global commodity price shocks, often resulting in sharp and unpredictable price fluctuations. This document pursues two main objectives. First, it aims to estimate and evaluate methods for forecasting 33 homogeneous food inflation baskets, which together constitute the total food Consumer Price Index (Food CPI), offering tools that can assist policymakers in anticipating the drivers of future inflation. This includes both traditional time series models and modern machine learning approaches. Second, it seeks to enhance the interpretability of model predictions through explainable AI techniques. To achieve this, we propose a variable lag selection algorithm to identify optimal feature-lag pairs, and employ SHAP (SHapley Additive exPlanations) values to quantify the contribution of each feature to the model’s forecast. Our findings indicate that machine learning models outperform traditional approaches in forecasting food inflation, delivering improved accuracy across most individual baskets as well as for aggregated food inflation. *****RESUMEN: Los precios de los alimentos han sido uno de los principales factores que contribuyen a la inflación en Colombia. Estos son particularmente sensibles a factores externos como choques climáticos, interrupciones en las cadenas globales de valor y choques en los precios de los productos básicos a nivel global, lo que resulta en fluctuaciones impredecibles de precios. Este documento tiene dos objetivos. En primer lugar, busca estimar y evaluar métodos para pronosticar 33 canastas homogéneas de inflación de alimentos, ofreciendo herramientas que puedan ayudar a los hacedores de política anticipar los factores que afectan la inflación de alimentos futura. Esto incluye tanto modelos tradicionales de series de tiempo como enfoques modernos de machine learning. En segundo lugar, se propone mejorar la interpretabilidad de las predicciones de los modelos mediante técnicas de explainableAI. Para ello, proponemos un algoritmo de selección de variables que identifique las variables explicativas más relevantes, y utilizamos valores SHAP (SHapley Additive exPlanations) para cuantificar la contribución de cada variable explicativa en las predicciones del modelo. Nuestros hallazgos indican que los modelos de machine learning superan a los enfoques tradicionales en el pronóstico de la inflación de alimentos, logrando una mayor precisión tanto en la mayoría de las canastas individuales como en la inflación de alimentos agregada. |
| Keywords: | Macroeconomic Forecasts, Food Prices, Machine learning, Pronóstico Macroeconómico, Inflación de alimentos |
| JEL: | C53 E31 E37 |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:bdr:borrec:1335 |
| By: | Bullock, David W.; Okoto, Edna M. |
| Abstract: | The predictive ability of two alternative forward price distribution forecasting methods based upon the full range of option premiums was developed and tested using 10 years of price and premium history for five traded commodities. The two models were a best-fit parametric distribution and a non-parametric linear interpolation fit. These were compared to two traditional approaches: historical time series and Black-76 option implied volatility. The forecast horizons ranged from 6 months to 1 week in duration. A modification of the theoretical results of King and Fackler (1985) nonparametric option pricing model was presented to justify the fitting of a price probability density function to the option premiums with the intrinsic value removed. Time series fits to the historical futures price indicted that the integrated ARCH (1) and GARCH (1, 1) models were the most prevalent best fit to the data. For parametric fits to the option premiums, the Burr Type XII and Dagum distributions were the most prevalent best fits. Predictive ability was measured using 10-percent value-at-risk portfolio models for simple short and long futures positions where the number of actual exceptions was compared to the theoretical values. The predictive results indicated that the parametric and non-parametric distribution fits performed best on the short futures portfolios over the longer-term forecast horizons (6- and 3-months) while the Black-76 performed best over the same time horizon. For the shorter time horizons (1-month or less), the Black-76 and time series methods performed best. These results point to the possibility that a hybrid Black-76 and premium distribution fit approach (via a splice) might perform best for longer-term projections. |
| Keywords: | Agricultural and Food Policy, Demand and Price Analysis |
| Date: | 2024 |
| URL: | https://d.repec.org/n?u=RePEc:ags:nccc24:379004 |
| By: | Roberto Pinheiro; Rory G. Quinlan |
| Abstract: | We investigate the behavior of BLS monthly revisions to payroll growth at turning points. We find some evidence corroborating claims by former BLS commissioners and market analysts that revisions around turning points tend to be procyclical and more serially correlated. Furthermore, we do see large revisions before turning points. However, the ability to use revisions to forecast business cycles' turning points seems limited. First, we do see lots of false positives: large revisions occur without a subsequent recession. Second, even within-sample, other indicators, such as initial jobless claims, the Chicago Fed National Activity Index, and the Aruoba-Diebold-Scotti Index, do a better job at detecting recessions. Finally, out-of-sample forecasting performance of revisions is poor. |
| Keywords: | BLS monthly revisions; business cycles; forecast |
| JEL: | E32 E37 C53 |
| Date: | 2025–12–18 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedcwq:102239 |
| By: | DeLong, Karen L.; Trejo-Pech, Carlos O.; Johansson, Robert |
| Abstract: | The United States (US) Department of Agriculture (USDA) World Agricultural Supply and Demand Estimates (WASDE) provides forecasts of domestic sugar production and consumption as well as Mexican sugar production. These forecasts are used to assist the USDA in the implementation of US sugar policy. Therefore, this study evaluates the accuracy, bias, and efficiency properties of the USDA WASDE sugar forecasts. Results indicate USDA WASDE domestic sugar production and consumption forecasts, and Mexican sugar production forecasts, are accurate, unbiased, and efficient. US sugar policy helps to ensure the predictability of sugarrelated forecasts, which may then generate positive economic effects for sugar-using firms (SUFs) that rely on reliable knowledge and ability to hedge supplies of an important production input. We postulate that forecast predictability, in turn, reduces SUFs’ risks compared to other agribusinesses. We further postulate that a lower risk environment leads to a superior economic environment for SUFs in which they can financially outperform other agribusiness firms. |
| Keywords: | Agricultural and Food Policy, Demand and Price Analysis |
| Date: | 2024 |
| URL: | https://d.repec.org/n?u=RePEc:ags:nccc24:379005 |
| By: | Sotirios D. Nikolopoulos |
| Abstract: | Evaluating rare-event forecasts is challenging because standard metrics collapse as event prevalence declines. Measures such as F1-score, AUPRC, MCC, and accuracy induce degenerate thresholds -- converging to zero or one -- and their values become dominated by class imbalance rather than tail discrimination. We develop a family of rare-event-stable (RES) metrics whose optimal thresholds remain strictly interior as the event probability approaches zero, ensuring coherent decision rules under extreme rarity. Simulations spanning event probabilities from 0.01 down to one in a million show that RES metrics maintain stable thresholds, consistent model rankings, and near-complete prevalence invariance, whereas traditional metrics exhibit statistically significant threshold drift and structural collapse. A credit-default application confirms these results: RES metrics yield interpretable probability-of-default cutoffs (4-9%) and remain robust under subsampling, while classical metrics fail operationally. The RES framework provides a principled, prevalence-invariant basis for evaluating extreme-risk forecasts. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.00916 |
| By: | Nicolas Hardy; Dimitris Korobilis |
| Abstract: | We revisit macroeconomic time-varying parameter vector autoregressions (TVP-VARs), whose persistent coefficients may adapt too slowly to large, abrupt shifts such as those during major crises. We explore the performance of an adaptively-varying parameter (AVP) VAR that incorporates deterministic adjustments driven by observable exogenous variables, replacing latent state innovations with linear combinations of macroeconomic and financial indicators. This reformulation collapses the state equation into the measurement equation, enabling simple linear estimation of the model. Simulations show that adaptive parameters are substantially more parsimonious than conventional TVPs, effectively disciplining parameter dynamics without sacrificing flexibility. Using macroeconomic datasets for both the U.S. and the euro area, we demonstrate that AVP-VAR consistently improves out-of-sample forecasts, especially during periods of heightened volatility. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.03763 |
| By: | Lovisa Reiche; Gabriele Galati; Richhild Moessner |
| Abstract: | We study how consumers form and revise inflation expectations using a unique, highly balanced monthly panel of Dutch households. We develop a Bayesian frame-work that nests Full-Information Rational Expectations (FIRE) alongside common forecasting heuristics and test it by recovering person-specific belief-updating rules from individual time-series regressions. Our novel individual-level design reveals sub-stantial heterogeneity in how households process information over time. On average, consumers systematically overreact to current inflation, echoing patterns found for pro-fessional forecasters. Only 2.5 percent, predominantly wealthier, more educated men, behave consistently with FIRE. Most consumers rely on simple heuristics, especially adaptive expectations. Our results show that heuristic learning, not FIRE, character-izes expectation formation for the vast majority of households. Crucially, heterogeneity in belief updating is both large and systematic. |
| Keywords: | Household Inflation Expectations; FIRE; Heuristic Learning |
| JEL: | E31 E37 E70 |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:dnb:dnbwpp:852 |
| By: | Juan C. King; Jose M. Amigo |
| Abstract: | The aim of this paper is the analysis and selection of stock trading systems that combine different models with data of different nature, such as financial and microeconomic information. Specifically, based on previous work by the authors and applying advanced techniques of Machine Learning and Deep Learning, our objective is to formulate trading algorithms for the stock market with empirically tested statistical advantages, thus improving results published in the literature. Our approach integrates Long Short-Term Memory (LSTM) networks with algorithms based on decision trees, such as Random Forest and Gradient Boosting. While the former analyze price patterns of financial assets, the latter are fed with economic data of companies. Numerical simulations of algorithmic trading with data from international companies and 10-weekday predictions confirm that an approach based on both fundamental and technical variables can outperform the usual approaches, which do not combine those two types of variables. In doing so, Random Forest turned out to be the best performer among the decision trees. We also discuss how the prediction performance of such a hybrid approach can be boosted by selecting the technical variables. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.02036 |