|
on Forecasting |
| By: | Tobias Lausser; Joao Eduardo Vuolo; Rudi Zagst |
| Abstract: | This paper compares different methods for forecasting the term structure of U.S. and European zero-coupon government bonds using both traditional econometric and Machine Learning (ML) approaches. We compare classical models (e.g., Dynamic Nelson-Siegel (DNS) and Principal Component Analysis (PCA)) with different Neural Network (NN) architectures, including those inspired by the classical models, on the U.S. Treasury market and bonds issued by the European Central Bank (ECB). To enhance predictive performance, macroeconomic variables are incorporated. The findings for both markets are separately analyzed and compared. To this end, we propose a robust model evaluation framework combining statistical accuracy metrics - such as RMSE, MAE, and directional accuracy - with the economic relevance of a quantitative bond trading strategy. Results show that NNs consistently outperform traditional models in both forecasting accuracy and portfolio performance. For the U.S., the most effective approach is a direct-forecasting NN that incorporates DNS factors to reduce the dimensionality of zero-rate data and an Autoencoder (AE) to extract macroeconomic features, while for Europe, the optimal model is a factor-based NN using PCA-derived zero-rate factors without the integration of macroeconomic variables. Overall, the paper demonstrates how combining traditional modeling approaches with modern ML techniques and evaluation can improve yield curve forecasts and support applications in fixed-income portfolio construction. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.26815 |
| By: | Christian König-Kersting; Yana Litovsky; Robert Böhm; Igor Grossmann; Jürgen Huber; Michael Kirchler; WoCCAP consortium |
| Abstract: | Accurate forecasts are central to decision-making in many domains. Although aggregating independent judgments can improve forecast accuracy, a phenomenon known as “Wisdom of the Crowd, ” methods for effectively combining these forecasts remain underexplored. In this preregistered many-designs study, 129 research teams independently submitted a total of 516 algorithms to aggregate forecasts over six months, across four domains: economics, politics, climate, and sports. By testing hundreds of independently developed aggregation algorithms under identical conditions, we established an empirical benchmark for forecast aggregation. Drawing on forecasts by 1, 182 people, we evaluated algorithm accuracy and variability. The share of algorithms significantly outperforming the mean and median varied by domain, with the strongest gains over simple benchmarks in politics and climate. Algorithm performance was also more persistent over time in politics and climate than in economics and sports. When asked to predict algorithm performance, the participating researchers were overconfident about the success of their own submissions and of others. We also examined potential predictors of algorithm accuracy and variability. Although researchers expected various features to predict accuracy, no algorithmic or researcher feature consistently explained why algorithms varied in performance. The main exception was the use by algorithms of previously observed realized outcomes, typically to compute past forecast error or train a model, which predicted greater accuracy in politics and climate. The null findings contrast with prior evidence that certain weighting criteria can improve forecast aggregation, and illustrate the value of preregistered many-designs studies which report the full distribution of submitted approaches, revealing when previously supported approaches fail to generalize. |
| Keywords: | Meta-science, Forecasting, Wisdom of the Crowd |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:inn:wpaper:2026-05 |
| By: | Saeed Varasteh Yazdi (EM - EMLyon Business School) |
| Abstract: | Reliable demand forecasting is crucial for effective supply chain management, where inaccurate forecasts can lead to frequent out-of-stock or overstock situations. While numerous statistical and machine learning methods have been explored for demand forecasting, reinforcement learning approaches, despite their significant potential, remain little known in this domain. In this paper, we propose a multi-agent deep reinforcement learning solution designed to accurately predict demand across multiple stores. We present empirical evidence that demonstrates the effectiveness of our model using a real-world dataset. The results confirm the practicality of our proposed approach and highlight its potential to improve demand forecasting in retail and potentially other forecasting scenarios. |
| Keywords: | reinforcement learning, multi-agent systems, demand forecasting |
| Date: | 2025–07–31 |
| URL: | https://d.repec.org/n?u=RePEc:hal:journl:hal-05656779 |
| By: | Susan Jiménez-Montero (Department of Economic Research, Central Bank of Costa Rica) |
| Abstract: | The document characterizes the Central Bank of Costa Rica’s approach to short-term inflation forecasting within the context of an inflation-targeting regime, highlighting its importance as a key input for medium term projections that inform the monetary policy decision-making process. This essay describes (i) the rationale for forecasting under inflation targeting, (ii) the transmission mechanism and the relevant horizon, (iii) the set of methodologies employed—univariate models, Bayesian techniques, factor models (FAVAR), and item-level CPI models—and (iv) the integration and validation process that transforms statistical results into a coherent economic forecast, which feeds into the macroeconomic model and policy recommendations, as well as institutional outputs such as the Monthly Economic Developments Report (IMCE) and the Monetary Policy Report (IPM). Finally, it emphasizes the dynamic nature of the forecasting system, including the exploration of machine learning techniques as a complement to traditional econometric approaches. ***Resumen: Este documento caracteriza el proceso que se implementa en el Banco Central de Costa Rica (BCCR) para la elaboración de pronósticos de inflación de corto plazo en el contexto de un régimen de metas de inflación. Se destaca su importancia como insumo fundamental para generar pronósticos de mediano plazo que informan la toma de decisiones de política monetaria. Este ensayo describe (i) la racionalidad del pronóstico bajo metas de inflación, (ii) el mecanismo de transmisión y el horizonte relevante, (iii) la batería de metodologías empleadas—modelos univariados, técnicas bayesianas, modelos de factores (FAVAR) y modelos por artículo del IPC—y (iv) el proceso de integración y validación que transforma resultados estadísticos en un pronóstico económico coherente, que alimenta el modelo macroeconómico y la recomendación de política, así como productos institucionales como el Informe Mensual de Coyuntura Económica (IMCE) y el Informe de Política Monetaria (IPM). Finalmente, se resalta el carácter dinámico del sistema de pronóstico, y la exploración de nuevos modelos y técnicas como por ejemplo de machine-learning como complemento a los enfoques econométricos tradicionales. |
| Keywords: | inflation targeting, monetary policy, inflation forecasting, expected inflation, forecasting models, persistent effects, metas de inflación, pronóstico de corto plazo, inflación esperada, gobernanza, política monetaria |
| JEL: | E52 E37 C53 C32 |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:apk:epolec:2603 |
| By: | Jason Abaluck; Kevin Bryan; Rebecca Ceppas de Castro; Basil Halperin; Todd Jones; Ezra Karger; Otto Kuusela; Dan Mayland; Ananaya Mittal; Connacher Murphy; Matt Reynolds; Josh Rosenberg; Philip Tetlock; Phil Trammell; Ria Viswanathan |
| Abstract: | We elicit forecasts of how AI will affect the U.S. economy, comparing the beliefs of five groups: academic economists, employees at AI companies, policy researchers focused on AI, highly accurate forecasters, and the general public. The median respondent in each group expects substantial advances in AI capabilities by 2030, small declines in labor force participation consistent with demographic shifts, and an annual GDP growth rate of 2.5%, which exceeds both the typical medium-run (2.0%) and long-run (1.7%) baseline forecasts from government agencies and private-sector forecasters. Conditional on a “rapid” AI progress scenario, in which AI systems surpass human performance on many cognitive and physical tasks, experts forecast substantial, though not historically unprecedented, economic shifts: annualized GDP growth rising to around 4% and the labor force participation rate falling from its current level of 62% to 55% by 2050, with roughly half of that decline—equivalent to around 10 million lost jobs—attributable to AI. A variance decomposition suggests that expert disagreement about these effects is driven primarily by different beliefs about the economic effects of highly capable AI systems rather than by disagreement about the pace of AI progress. These forecasts map onto notably different policy preferences across groups: experts strongly favor targeted measures such as worker retraining, whereas the general public supports both targeted programs and broader interventions, including a job guarantee and universal basic income. |
| Keywords: | Artificial intelligence; Economic forecasting; macroeconomic impacts |
| JEL: | O33 O38 O40 E27 J21 |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedhwp:103454 |
| By: | Miquel Noguer I Alonso; Rodolfo Pereira Franklin |
| Abstract: | Financial return forecasting is a difficult test case for time-series foundation models (TSFMs) due to low signal-to-noise ratios, structural breaks, heavy tails, and weak persistence. This paper benchmarks pretrained TSFMs against train-from-scratch neural baselines in a deliberately conservative financial setting. We evaluate TimeGPT/TimeGPT-LH, TimesFM-2.5, Moirai-2.0, Chronos, and Chronos-2 against NBEATS, NHITS, PatchTST, iTransformer, and KAN on five liquid U.S. equities (AAPL, AMZN, GOOG, JPM, META) using linear and log returns. Models are compared under an equalized context budget, a rolling-origin protocol, and against random-walk benchmarks. We provide a theoretical framing of pretraining as an inductive prior, linking PAC-Bayes transfer intuition, information-theoretic predictability limits, and attention geometry. This clarifies why strong model rankings need not imply economically meaningful predictability in noisy markets. Pragmatically, pretrained TSFMs dominate the ranking distribution, accounting for 8 of 10 task-level wins. Moirai-2.0 and TimesFM-2.5 achieve the strongest average ranks, leading tasks for AAPL, JPM, GOOG, and AMZN, while Chronos wins the remaining AMZN task. However, the iTransformer baseline wins both META tasks, showing local supervised learning can still outperform generic pretraining for specific assets. Crucially, gains over the random-walk benchmark are small and sparse. A one-sided Diebold-Mariano test rejects equal or inferior predictive accuracy only for Chronos on AMZN and Moirai-2.0 on GOOG. We conclude that TSFMs serve as useful practical priors that reduce model-development costs in low-data financial forecasting, but are not universal engines for statistically reliable alpha generation in realistic empirical deployment. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.27100 |
| By: | Li Chen (Paderborn University); Yuanhua Feng (Paderborn University) |
| Abstract: | Motivated by more and more semi- or nonparametric models applied in time series forecasting and their demonstrated superior performance in many empirical researches, this paper explores the adoption and integration of a semiparametric ARMA model in an enterprise system landscape. We begin by reviewing basic construction of the semiparametric ARMA model, the iterative plug-in algorithm for estimating the trend component of trend stationary times series, forecast techniques and quality measurements, which were well researched and published with the R package smoots. Subsequently, we showcase a novel approach to adopt the semiparametric ARMA model in a forecast application based on SAP Analytics Cloud (SAC), which leverages the platform’s strengths in system integrity, state-of-the-art user interface (UI) design as well as seamless connection to a R engine with smoots package embedded. The forecast application addresses key challenges in terms of cost efficiency, user experience, and the requirement for in-house statistical or machine learning expertise while adopting such statistical algorithms in enterprise context. Finally, we empirically evaluate the forecast quality of the integrated semiparametric ARMA model using real-world data, demonstrating promising results overall. |
| Keywords: | Time series forecasting, semiparametric algorithm, forecasting accuracy, smoots package, SAP Analytics Cloud, enterprise adoption |
| JEL: | C |
| Date: | 2025–08 |
| URL: | https://d.repec.org/n?u=RePEc:pdn:dispap:179 |
| By: | Sariola, Mikko; Viertola, Hannu |
| Abstract: | The Bank of Finland produces forecasts for the Finnish economy using advanced modelling tools. Central to this process is its Aino model, which is a dynamic stochastic general equilibrium (DSGE) model that integrates various aspects of the Finnish economy. The forecasts support policy decisions within the Eurosystem and are updated regularly. This paper describes the forecasting process, main features of the forecasting models used and their key role in economic analysis at the Bank of Finland. |
| Keywords: | Aino model, forecast, Dynamic Stochastic General Equilibrium, DSGE, monetary policy, Finnish economy |
| JEL: | E0 E3 E4 E5 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:bofecr:341669 |
| By: | Miquel Noguer i Alonso |
| Abstract: | Modeling the future requires specifying conditional laws relative to an evolving information flow and describing their movement across time. This paper provides a unified mathematical synthesis of this problem along a single spine. Filtrations encode known data; conditional expectation and regular conditional probabilities yield point and distributional forecasts; Markov kernels and semigroups propagate observables and laws; and infinitesimal generators encode local dynamics, producing Kolmogorov equations and stochastic differential equations. Along this spine, martingales isolate surprise, filtering handles partial observation, finance prices futures, stochastic control optimizes choices, and ergodic theory describes the far future. The contribution is architectural. We explicitly connect derivations that turn classical objects into a unified forecasting calculus: the tower property becomes the semigroup law; Ito's formula yields the backward equation after conditioning; integration by parts provides the forward operator; and generator perturbations become model-risk distortions. Forecasting is shown not as mere data extrapolation, but the construction of dynamically coherent conditional distributions constrained by information, geometry, and admissible models. These concepts are illustrated via Gaussian Ornstein--Uhlenbeck and non-Gaussian Cox--Ingersoll--Ross processes, demonstrating how abstract machinery produces explicit transition laws, spectral decompositions, term-structure formulae, and asymptotics in diverse geometries. We recast density evolution as a Wasserstein gradient flow, place forecasting within Hilbert, Fisher--Rao, and Wasserstein geometries, provide a discrete-time empirical dictionary, and address model-risk. The result is a compact mathematical map from information to prediction, local dynamics to global laws, and idealized models to empirical forecasting. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.20977 |
| By: | Rouven Beiner; Bernd Süssmuth |
| Abstract: | Density expansions such as the Gram-Charlier (GC) expansion allow for the modeling of time-varying higher moments. However, they can suffer from spurious multimodality, negative densities, and asymptotically light tails if truncated. This paper introduces monotonic polynomial generalized autoregressive conditional heteroskedasticity (GARCH) models. They generate conditional skewness and kurtosis via a monotonic polynomial transformation of innovations. By construction, this approach guarantees a valid, unimodal probability density without requiring truncation. It naturally accommodates heavy Weibull-type tails. We provide a theoretical framework proving strict stationarity and ergodicity. In empirical applications to financial returns, the proposed estimator outperforms both GC-based and score-driven benchmarks in out-of-sample density forecasting. It demonstrates superior structural stability and robustness against overfitting. |
| Keywords: | GARCH, observation-driven models, conditional higher moments, density forecasting, monotonic polynomials |
| JEL: | C22 C53 C58 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_12734 |
| By: | Scott A. Brave; Ben Henken; Ezra Karger; Aryan Safi |
| Abstract: | We present the Chicago Fed Labor Market Indicators (LMI): a twice-monthly release that includes the job-finding rate, the job-separation rate, and a forecast for the U.S. Bureau of Labor Statistics (BLS) unemployment rate. To overcome limitations in data availability, the LMI uses partial least squares to combine CPS-based finding and separation rates with higher-frequency alternative and traditional labor market data series—such as unemployment insurance claims, Google Trends searches, online job postings, and survey-based indicators. Our resulting flow-consistent unemployment rate (FCR) correlates strongly with the BLS unemployment rate, and can be used to characterize the current “low-hire, low-fire” nature of the U.S. labor market. We use a Bayesian linear regression centered on a no-change prior to translate changes in our FCR into a real-time forecast for the next BLS unemployment rate reading. In backtesting spanning 2018–2026, our unemployment rate forecast improves on both a random-walk benchmark and the Bloomberg consensus forecast, with the largest accuracy gains during the Covid-19 pandemic when the unemployment rate rose rapidly in a way that was well-captured by high-frequency labor market data. |
| Keywords: | unemployment; Job separation; job finding; nowcast; big data; Alternative data |
| JEL: | C01 C53 E24 E37 J64 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedhwp:103456 |
| By: | Hördahl, Peter; Kısacıkoğlu, Burçin; Xia, Fan Dora |
| Abstract: | Bond yields react to macroeconomic surprises, but the magnitude of this responsiveness depends on macroeconomic forecast disagreement and monetary policy uncertainty. Using intraday responses of US Treasury futures to surprises in macroeconomic data releases, we find that greater forecast disagreement about an economic indicator prior to its release dampens the yield curve response, while higher monetary policy uncertainty amplifies it. An exception is inflation surprises: prior to the post-COVID inflation surge, bond yield reactions to inflation surprises were not amplified by short-rate uncertainty. We use a model with Bayesian learning to rationalize these findings. Specifically, large forecast disagreement indicates a weak link between the macroeconomic variable and future monetary policy, reducing the information value of macro news to forecast monetary policy. In contrast, during periods of high monetary policy uncertainty, macro news becomes more informative. Before the post-COVID inflation surge, investors may have perceived that the Federal Reserve placed little emphasis on its price stability mandate, which could have muted the yield curve response to inflation news even when short rate uncertainty was high. The proposed model generates distinct, empirically testable effects of disagreement and monetary policy uncertainty on yield responses which, when extended to allow time-varying signal precision, accounts for the post-COVID shift in inflation sensitivity within a single unified framework. |
| Keywords: | Macroeconomic news; Forecast dispersion; Policy uncertainty; Bond yields; Bayesian learning |
| JEL: | E43 E44 G14 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:21501 |
| By: | McMahon, Michael; Petersen, Luba; Rholes, Ryan |
| Abstract: | Using over 17, 000 incentivized inflation forecasts, we provide causal evidence that environmental complexity and subjective complexity are distinct drivers of rounding in survey responses. Experimental variation in shock volatility and central-bank communication regimes shows that both channels raise forecast uncertainty and the propensity to round, with subjective complexity the dominant force — explaining 58–86% of rounding depending on horizon and specification. Survey of Consumer Expectations microdata corroborate these findings: rounding declines with survey tenure, rises with inflation volatility, and inflates measured inflation expectations by nearly 7 percentage points among inexperienced respondents. |
| Keywords: | Expectation formation; Uncertainty |
| JEL: | C91 D84 E52 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:21374 |