|
on Forecasting |
| By: | Marco Zanotti |
| Abstract: | Given the continuous increase in dataset sizes and the complexity of forecasting models, the tradeoff between forecast accuracy and computational cost is emerging as an extremely relevant topic, especially in the context of ensemble learning for time series forecasting. To asses it, we evaluated ten base models and eight ensemble configurations across two large-scale retail datasets (M5 and VN1), considering both point and probabilistic accuracy under varying retraining frequencies. We showed that ensembles consistently improve forecasting performance, particularly in probabilistic settings. However, these gains come at a substantial computational cost, especially for larger, accuracy-driven ensembles. We found that reducing retraining frequency significantly lowers costs, with minimal impact on accuracy, particularly for point forecasts. Moreover, efficiency-driven ensembles offer a strong balance, achieving competitive accuracy with considerably lower costs compared to accuracy-optimized combinations. Most importantly, small ensembles of two or three models are often sufficient to achieve near-optimal results. These findings provide practical guidelines for deploying scalable and cost-efficient forecasting systems, supporting the broader goals of sustainable AI in forecasting. Overall, this work shows that careful ensemble design and retraining strategy selection can yield accurate, robust, and cost-effective forecasts suitable for real-world applications. |
| Keywords: | Time series, Demand forecasting, Forecasting competitions, Cross-learning, Global models, Forecast combinations, Ensemble learning, Machine learning, Deep learning, Conformal predictions, Green AI. |
| JEL: | C53 C52 C55 |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:mib:wpaper:554 |
| By: | Marco Zanotti |
| Abstract: | Forecast stability, that is the consistency of predictions over time, is essential in business settings where sudden shifts in forecasts can disrupt planning and erode trust in predictive systems. Despite its importance, stability is often overlooked in favor of accuracy, particularly in global forecasting models. In this study, we evaluate the stability of point and probabilistic forecasts across different retraining frequencies and ensemble strategies using two large retail datasets (M5 and VN1). To do this, we introduce a new metric for probabilistic stability (MQC) and analyze ten different global models and four ensemble configurations. The results show that less frequent retraining not only preserves but often improves forecast stability, while ensembles, especially those combining diverse pool of models, further enhance consistency without sacrificing accuracy. These findings challenge the need for continuous retraining and highlight ensemble diversity as a key factor in reducing forecast stability. The study promotes a shift toward stability-aware forecasting practices, offering practical guidelines for building more robust and sustainable prediction systems. |
| Keywords: | Time series, Demand forecasting, Forecasting competitions, Cross-learning, Global models, Forecast stability, Vertical stability, Machine learning, Deep learning, Conformal predictions. |
| JEL: | C53 C52 C55 |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:mib:wpaper:553 |
| By: | Dhanashree Somani (Department of Statistics, University of Florida, 230 Newell Drive, Gainesville, FL, 32601, USA); Rangan Gupta (Department of Economics, University of Pretoria, Private Bag X20, Hatfield 0028, South Africa); Sayar Karmakar (Department of Statistics, University of Florida, 230 Newell Drive, Gainesville, FL, 32601, USA); Vasilios Plakandaras (Department of Economics, Democritus University of Thrace, Komotini, Greece) |
| Abstract: | The objective of this paper is to forecast volatilities of the stock returns of China, France, Germany, Italy, Spain, the United Kingdom (UK), and the United States (US) over the daily period of January 2010 to February 2025 by utilizing the information content of newspapers articles-based indexes of supply bottlenecks. We measure volatility by employing the interquantile range, estimated via an asymmetric slope autoregressive quantile regression fitted on stock returns to derive the conditional quantiles. In the process, we are also able to obtain estimates of skewness, kurtosis, lower- and upper-tail risks, and incorporate them into our linear predictive model, alongside leverage. Based on the shrinkage estimation using the Lasso estimator to control for overparameterization, we find that the model with moments outperform the benchmark model that includes both own- and cross-country volatilities, but the performance of the former, is improved further when we incorporate the role of the metrics of supply constraints for all the 7 countries simultaneously. These findings carry significant implications for investors. |
| Keywords: | Supply Bottlenecks, Stock Market Volatility, Asymmetric Autoregressive Quantile Regression, Lasso Estimator, Forecasting |
| JEL: | C22 C53 E23 G15 |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:pre:wpaper:202521 |
| By: | Daniel H. Cooper; Giovanni P. Olivei; Hannah Rhodenhiser |
| Abstract: | We provide a parsimonious setup for forecasting U.S. GDP growth and the unemployment rate based on a few fundamental drivers. This setup yields forecasts that are reasonably accurate compared with private-sector and Federal Reserve forecasts over the 1984–2019 and post COVID-19 pandemic periods. This result is achieved by jointly estimating the processes for GDP growth and the unemployment rate, with the constraint that GDP and unemployment follow Okun’s law in first differences. This setup can be easily extended to replace the variables in the information set with factors that might better capture the underlying fundamentals. |
| Keywords: | macroeconomic forecasting; small information set; forecast accuracy |
| JEL: | E27 E37 |
| Date: | 2025–06–01 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedbwp:101183 |
| By: | Carlos Segura-Rodriguez (Departamento Investigación Económica, Banco Central de Costa Rica) |
| Abstract: | This study presents a methodology for forecasting inflation in Costa Rica using a FAVAR model that combines data from 156 relevant time series. This approach consists of two stages: first, static and dynamic factors are estimated, which are then incorporated into a VAR model along with monthly inflation to project the annual variation of the Consumer Price Index. Automatic selection criteria are employed to choose which variables to include in the factors and to determine the number of factors, lags, and restrictions on the coefficients of the VAR model. Eight inflation forecasts are generated and combined using three averages: simple, inverse mean squared error weighted, and Bayesian. The results indicate that the Bayesian forecast is the most accurate for the period between 2021 and 2023, outperforming even the most accurate of traditional VAR models that consider only inflation and individually any of the 156 variables. This suggests that the FAVAR model can effectively integrate information from available variables without requiring prior knowledge of which ones are most relevant. |
| Keywords: | Inflation; Forecasting; Dynamic Factors; FAVAR |
| JEL: | E32 R10 |
| Date: | 2024–08 |
| URL: | https://d.repec.org/n?u=RePEc:apk:doctra:2403 |
| By: | Paula Bejarano Carbo; Rory MacQueen; Efthymios Xylangouras |
| Abstract: | Our predecessors at NIESR helped pioneer the estimation of monthly Gross Domestic Product (GDP) in the United Kingdom (Mitchell et al., 2005). This approach, which led to the publication of monthly GDP by the Office for National Statistics (ONS) from 2018, aggregates sectoral indices (e.g. agriculture, construction) to produce an overall estimate of monthly economic growth. Since 2018, NIESR has produced its monthly GDP 'tracker' on the ONS estimate release date, commenting on the latest data point and producing a 'bottom-up' forecast (i.e. constructed by aggregating sectoral forecasts) of economic performance up to the end of the next quarter (Kara et al., 2018). The NIESR GDP tracker forecast of 10 sub-sectors and aggregate GDP combines a 'fixed parameter' approach and forecaster judgement to produce, we find, accurate estimates one month in advance of the official first estimate. |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:nsr:niesrp:46 |
| By: | Tim de Silva; Eugene Larsen-Hallock; Adam Rej; David Thesmar |
| Abstract: | This paper studies expectations formation when the underlying process has fat tails. Using a large sample of firm sales growth expectations, we document three facts: (i) the relationship between forecast revisions and future forecast errors is strongly non-linear, (ii) the distribution of sales growth has fat tails, and (iii) extreme values of sales growth tend to mean-revert. We formally show that these three facts are consistent with a model in which the underlying process is non-Gaussian, but forecasters fail to recognize this fully. We estimate this model and show it quantitatively explains our three facts. Finally, we show the model is consistent with evidence from an online forecasting experiment where the underlying process is non-Gaussian and the non-linearity in the momentum of stock returns. |
| JEL: | D84 D91 G41 |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:33808 |
| By: | Xin Sheng (Lord Ashcroft International Business School, Anglia Ruskin University, Chelmsford, United Kingdom); Oguzhan Cepni (Ostim Technical University, Ankara, Turkiye; University of Edinburgh Business School, Centre for Business, Climate Change, and Sustainability; Department of Economics, Copenhagen Business School, Denmark); Rangan Gupta (Department of Economics, University of Pretoria, Private Bag X20, Hatfield 0028, South Africa); Minko Markovski (Department of Economics, University of Reading, Reading, United Kingdom) |
| Abstract: | We forecast quarterly growth rate of real gross fixed capital formation of the United States using the information content of a monthly metric of extreme weather conditions, while controlling for a set of principal components derived from a large data set of economic and financial indicators. In this regard, we utilize a Mixed Frequency Machine Learning framework over the period of 1974:Q1 to 2022:Q1. Our results show that incorporating monthly data on severe climatic conditions significantly, especially information contained in relatively higher (above the mean) extreme weather values, outperforms not only the benchmark autoregressive model, but also the econometric framework that includes the macro-finance factors when forecasting the growth rate of quarterly real gross fixed capital formation. |
| Keywords: | Gross fixed capital formation, Extreme weather conditions, Mixed frequency, Machine learning, Forecasting |
| JEL: | C22 C53 E22 Q54 |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:pre:wpaper:202520 |
| By: | Jorge Onrubia; Fernando Pinto; María del Carmen Rodado Ruíz |
| Abstract: | This paper examines the predictive relationship between online search behavior and international housing demand, focusing on UK citizens purchasing property in Spain from 2014 to 2024. Using Google Trends data for the search term "Spain villas" alongside official transaction records, we estimate autoregressive(AR), argumented(ARX), and interaction models to asses whether digital intent anticipates real estate purchases.Results show that search intensity significantly enhances model performance before the 2016 Brexit referendum |
| Date: | 2025–07 |
| URL: | https://d.repec.org/n?u=RePEc:fda:fdaddt:2025-08 |
| By: | Leland D. Crane; Akhil Karra; Paul E. Soto |
| Abstract: | We evaluate the ability of large language models (LLMs) to estimate historical macroeconomic variables and data release dates. We find that LLMs have precise knowledge of some recent statistics, but performance degrades as we go farther back in history. We highlight two particularly important kinds of recall errors: mixing together first print data with subsequent revisions (i.e., smoothing across vintages) and mixing data for past and future reference periods (i.e., smoothing within vintages). We also find that LLMs can often recall individual data release dates accurately, but aggregating across series shows that on any given day the LLM is likely to believe it has data in hand which has not been released. Our results indicate that while LLMs have impressively accurate recall, their errors point to some limitations when used for historical analysis or to mimic real time forecasters. |
| Keywords: | Artificial intelligence; Forecasting; Large language models; Real-time data |
| JEL: | C53 C80 E37 |
| Date: | 2025–06–25 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedgfe:2025-44 |
| By: | Kay, Benjamin (Federal Reserve Board); Lakdawala, Aeimit (Wake Forest University, Economics Department); Ryngaert, Jane (University of Notre Dame) |
| Abstract: | Using a novel dataset linking professional forecasters in the Wall Street Journal Economic Forecasting Survey to their political affiliations, we document a partisan bias in GDP growth forecasts. Republican-affiliated forecasters project 0.3--0.4 percentage points higher growth when Republicans hold the presidency, relative to Democratic-affiliated forecasters. Forecast accuracy shows a similar partisan pattern: Republican-affiliated forecasters are less accurate under Republican presidents, indicating that partisan optimism impairs predictive performance. This bias appears uniquely in GDP forecasts and does not extend to inflation, unemployment, or interest rates. We explain these findings with a model where forecasters combine noisy signals with politically-influenced priors: because GDP data are relatively more uncertain, priors carry more weight, letting ideology shape growth projections while leaving easier-to-forecast variables unaffected. Noisy information therefore amplifies, rather than substitutes for, heterogeneous political priors, implying that expectation models should account for both information rigidities and belief heterogeneity. Finally, we show that Republican forecasters become more optimistic when tax cuts are salient in public discourse, suggesting that partisan differences reflect divergent beliefs about the economic effects of fiscal policy. |
| Keywords: | partisan bias; professional forecasts; GDP growth forecasts; tax policy expectations |
| JEL: | C53 D72 D84 E37 |
| Date: | 2025–06–30 |
| URL: | https://d.repec.org/n?u=RePEc:ris:wfuewp:0127 |
| By: | Li, Mengheng; Mendieta-Munoz, Ivan |
| Abstract: | We propose a factor correlated unobserved components (FCUC) model to analyze the sticky and flexible components of U.S. inflation. The proposed FCUC framework estimates trend inflation and component cycles in a flexible stochastic environment with time-varying volatility, factor loadings, and cross-frequency (trend-cycle) correlations, thus capturing how structural heterogeneity in price adjustment shapes the evolution of aggregate trend inflation over time. Using Bayesian estimation methods, we show that the FCUC model substantially reduces the uncertainty surrounding estimates of trend inflation and improves both point and density forecast accuracy. Our findings reveal that, particularly following the Global Financial Crisis and more markedly since the COVID-19 recession, transitory price shocks originating from flexible inflation have become a major driver of trend inflation, whereas sticky inflation explains only part of the variation. These results indicate that temporary price movements can have persistent effects, highlighting important policy implications regarding the cyclical dynamics of disaggregated inflation components amid evolving macroeconomic conditions. |
| Keywords: | trend inflation, sticky inflation, flexible inflation, stochastic volatility, dynamic factor model |
| JEL: | C32 C53 E37 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:esprep:320299 |
| By: | Andrew B. Martinez; Alexander D. Schibuola; David Beckworth |
| Abstract: | Arguments for nominal income targeting are often dismissed because it is an unreliable measure. To assess these concerns, we compare the real-time performance of several nominal and real measures of economic slack. We find that the nominal GDP expectations gap - the difference between nominal GDP and average projections thereof from surveys of professional forecasters - performs well as a measure of economic slack: its historical revisions are 2-3 times smaller than other measures, it significantly improves real-time forecasts of inflation since the pandemic, and it makes monetary policy rules up to 40 percent less volatile. Overall, concerns about nominal income targets are misplaced. |
| Keywords: | Business cycles; Forecast accuracy; Phillips curve; Taylor rule |
| JEL: | C53 E32 E37 E47 |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:gwc:wpaper:2025-004 |
| By: | Emiliano Basco (Central Bank of Argentina); Diego Elías (Central Bank of Argentina); Luciana Pastore (Central Bank of Argentina); Maximiliano Gómez Aguirre (Central Bank of Argentina) |
| Abstract: | Agriculture, and especially soybean production, has a critical role in Argentina’s economy, as a major contributor to GDP and export revenue. This paper studies the impact of climate variability on soybean yields in Argentina using a novel department-level dataset spanning 1980–2023. We estimate a fixed effects spatial error model (SEM) to quantify the effects of weather shocks– measured by extreme heat, precipitation, and ENSO phases–while controlling for economic and technological factors such as seed technology and relative prices. Our results show that extreme heat significantly reduces yields, while moderate rainfall boosts them up to a nonlinear threshold. El Niño phases increase yields, whereas La Niña events are detrimental. Technological adoption and favorable price signals also enhance productivity. These findings highlight the importance of accounting for both climate dynamics and spatial distributions when estimating agricultural outcomes. Time series models provide a strong empirical basis for forecasting soybean yields and informing policy decisions under increasing climate uncertainty. These models can be employed as effective tools for anticipating yield outcomes under different climate scenarios and utilized in stress test exercises. This work provides valuable insights for policymaking decisions, contributing to prepare for potential economic impacts stemming from climate risks on Argentina’s agricultural sector. |
| Keywords: | Soybean Yields; Argentina; Forecasting; Model Selection |
| JEL: | Q10 Q12 C13 C32 C33 |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:bcr:wpaper:2025117 |
| By: | Eric Benyo, Reinhard Ellwanger, Stephen Snudden |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:wlu:lcerpa:jc0158 |
| By: | Monia Magnani; Massimo Guidolin |
| Abstract: | We study the complex, non-linear linkages between short-term policy rates and the size and expected durations of equity bubbles. We extend empirical models of periodically collapsing, rational bubbles to test whether and to what extent the long cycle of rates at the zero lower bound and of quantitative easing policies may have increased the probability of bubbles inflating and persisting, with emphasis on the US stock market. We find that the linkages between S&P returns, and ratebased indicators of monetary policies contain evidence of recurring regimes that can be characterized as one of a persisting vs. one of a collapsing bubble. Moreover, the probabilities of financial markets transitioning from a bubble to a state of (partial) collapse turns out to depend on both the initial, relative size of the bubble and on monetary policy indicators. This implies that an easier (tighter) monetary policy will inflate (deflate) a bubble through a simple, regression-style effect, but also yield a non-linear, “concave” effect by which, starting from low rates, rate hikes may at first inflate bubbles before contributing to their bursting, when rates are pushed above a critical threshold. Besides fitting the data, the resulting, parsimonious, regime switching models provide accurate and economically valuable recursive out-of-sample predictive performance, even when transaction costs are taken into account. |
| Keywords: | Rational bubbles, monetary policy, stock returns, regime switching, forecasting. |
| JEL: | G12 E52 C58 G17 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:baf:cbafwp:cbafwp25252 |