|
on Forecasting |
| By: | Matteo Mogliani; Florens Odendahl |
| Abstract: | The popular choice of using a direct forecasting scheme implies that the individual predictions ignore information on their cross-horizon dependence. However, this dependence is needed if the forecaster has to construct, based on direct density forecasts, predictive objects that are functions of several horizons (e.g. when constructing annual-average growth rates from quarter-on-quarter growth rates). To address this issue, we propose to use copulas to combine the individual h-step-ahead predictive distributions into a joint predictive distribution. Our method is particularly appealing to practitioners for whom changing the direct forecasting specification is too costly. In a Monte Carlo study, we demonstrate that our approach leads to a better approximation of the true density than an approach which ignores the potential dependence. We show the superior performance of our method in several empirical examples, where we construct (i) quarterly forecasts using month-on-month direct forecasts, (ii) annual-average forecasts using monthly year-on-year direct forecasts, and (iii) annual-average forecasts using quarter-on-quarter direct forecast. |
| Keywords: | Joint Predictive Distribution, Frequency Transformation, Path Forecasts, Cross-horizon Dependence |
| JEL: | C53 C32 E37 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:bfr:banfra:1027 |
| By: | Uluc Aysun (University of Central Florida, Orlando, FL); Melanie Guldi (University of Central Florida, Orlando, FL) |
| Abstract: | We revisit the exchange-rate predictability puzzle by asking whether standard, widely used machine-learning (ML) algorithms convincingly improve exchange rate forecasting once evaluation is disciplined and implementation is made robust. Using monthly data from January 1986 to February 2025, we study US dollar to British pound as the baseline case (in both levels and monthly percent changes). We compare five ML methods -- random forests, neural networks, LASSO, gradient boosting, and linear support-vector classification -- against canonical benchmarks (random walk and ARIMA) in a rolling one-step-ahead out-of-sample forecasting design. To mitigate sensitivity to stochastic estimation, we average forecasts across multiple random seeds and assess performance using RMSE and Diebold-Mariano tests. We find that ML does not improve level forecasts and typically underperforms ARIMA. For exchange-rate changes, ML methods consistently outperform the random-walk benchmark, but only neural networks -- under a specific design -- reliably beat ARIMA. A theory-based UIP/PPP filtering approach improves accuracy for both ML and univariate methods, yet does not change the overall ranking. Extensive robustness checks across windows, currencies, frequencies, and tuning choices confirm that ML’s advantages are limited and fragile relative to conventional univariate benchmarks. |
| Keywords: | Machine learning, exchange rates, forecasting, theoretical filtering, random walk, ARIMA. |
| JEL: | C53 F31 F37 G17 |
| Date: | 2026–01 |
| URL: | https://d.repec.org/n?u=RePEc:cfl:wpaper:2026-01ua |
| By: | Andrea Bastianin (Department of Economics, Management, and Quantitative Methods, University of Milan and Fondazione Eni Enrico Mattei); Luca Rossini (Department of Economics, Management, and Quantitative Methods, University of Milan and Fondazione Eni Enrico Mattei); Lorenzo Tonni (Department of Economics, Management, and Quantitative Methods, University of Milan) |
| Abstract: | This paper develops a real-time forecasting framework for monthly real prices of four key industrial metals – aluminum, copper, nickel, and zinc – whose demand is rising due to their widespread use in manufacturing and low-carbon technologies. To replicate the information set available to forecasters in real time, we construct a new dataset combining daily financial variables with first-release macroeconomic indicators and use nowcasting techniques to address publication lags. Within this real-time environment, we evaluate the predictive accuracy of a broad set of univariate, multivariate, and factor-augmented models, comparing their performance with two industry benchmarks: survey expectations and futures-spot spread models. Results show that although short-run metal price movements remain difficult to predict, medium-term horizons display substantial forecastability. Indicators of manufacturing activity tied to primary metals — such as new orders and capacity utilization — significantly improve forecasting accuracy for aluminum and copper, with more moderate gains for zinc and limited improvements for nickel. Futures and survey forecasts generally underperform the real-time econometric models. These findings highlight the value of incorporating timely macroeconomic information into forecasting frameworks for industrial metal markets. |
| Keywords: | First-Release Data, Energy Transition, Forecasting, Metals, Critical Raw materials |
| JEL: | C32 Q02 Q41 Q43 Q48 |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:fem:femwpa:2025.34 |
| By: | Mokinski, Frieder; Roth, Markus |
| Abstract: | This note explores conditional forecasting under conditions on annual growth rates, where variables enter a (possibly structural) vector autoregressive (VAR) model in logarithms or logarithmic first differences. For example, imposing conditions on the annual growth rate of quarterly real GDP modeled in logarithms is challenging be- cause annual growth rates are nonlinear functions of the log variables. We address this by approximating the annual growth rate with a linear function of the model variables, enabling the use of standard conditional forecasting methods. An approximation error arises since the condition is not imposed directly; to mitigate this, we iteratively adjust the condition until the error is acceptable. We provide MATLAB companion code that also accepts other types of conditions: (1) conditions on the path of variables entering the VAR, (2) conditions on the path of structural shocks, and (3) conditions on sums of successive variable observations. |
| Keywords: | conditional forecasting, annual growth rate constraints, log-linear approximation, structural vector autoregression |
| JEL: | C32 C53 E17 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:bubdps:334532 |
| By: | Dinggao Liu; Robert \'Slepaczuk; Zhenpeng Tang |
| Abstract: | Accurately forecasting daily exchange rate returns represents a longstanding challenge in international finance, as the exchange rate returns are driven by a multitude of correlated market factors and exhibit high-frequency fluctuations. This paper proposes EXFormer, a novel Transformer-based architecture specifically designed for forecasting the daily exchange rate returns. We introduce a multi-scale trend-aware self-attention mechanism that employs parallel convolutional branches with differing receptive fields to align observations on the basis of local slopes, preserving long-range dependencies while remaining sensitive to regime shifts. A dynamic variable selector assigns time-varying importance weights to 28 exogenous covariates related to exchange rate returns, providing pre-hoc interpretability. An embedded squeeze-and-excitation block recalibrates channel responses to emphasize informative features and depress noise in the forecasting. Using the daily data for EUR/USD, USD/JPY, and GBP/USD, we conduct out-of-sample evaluations across five different sliding windows. EXFormer consistently outperforms the random walk and other baselines, improving directional accuracy by a statistically significant margin of up to 8.5--22.8%. In nearly one year of trading backtests, the model converts these gains into cumulative returns of 18%, 25%, and 18% for the three pairs, with Sharpe ratios exceeding 1.8. When conservative transaction costs and slippage are accounted for, EXFormer retains cumulative returns of 7%, 19%, and 9%, while other baselines achieve negative. The robustness checks further confirm the model's superiority under high-volatility and bear-market regimes. EXFormer furnishes both economically valuable forecasts and transparent, time-varying insights into the drivers of exchange rate dynamics for international investors, corporations, and central bank practitioners. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.12727 |
| By: | Charles Shaw |
| Abstract: | We introduce srvar-toolkit, an open-source Python package for Bayesian vector autoregression with shadow-rate constraints and stochastic volatility. The toolkit implements the methodology of Grammatikopoulos (2025, Journal of Forecasting) for forecasting macroeconomic variables when interest rates hit the effective lower bound. We provide conjugate Normal-Inverse-Wishart priors with Minnesota-style shrinkage, latent shadow-rate data augmentation via Gibbs sampling, diagonal stochastic volatility using the Kim-Shephard-Chib mixture approximation, and stochastic search variable selection. Core dependencies are NumPy, SciPy, and Pandas, with optional extras for plotting and a configuration-driven command-line interface. We release the software under the MIT licence at https://github.com/shawcharles/srvar-too lkit. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.19589 |
| By: | Ilias Aarab |
| Abstract: | A growing empirical literature suggests that equity-premium predictability is state dependent, with much of the forecasting power concentrated around recessionary periods \parencite{Henkel2011, DanglHalling2012, Devpura2018}. I study U.S. stock return predictability across economic regimes and document strong evidence of time-varying expected returns across both expansionary and contractionary states. I contribute in two ways. First, I introduce a state-switching predictive regression in which the market state is defined in real time using the slope of the yield curve. Relative to the standard one-state predictive regression, the state-switching specification increases both in-sample and out-of-sample performance for the set of popular predictors considered by \textcite{WelchGoyal2008}, improving the out-of-sample performance of most predictors in economically meaningful ways. Second, I propose a new aggregate predictor, the Aligned Economic Index, constructed via partial least squares (PLS). Under the state-switching model, the Aligned Economic Index exhibits statistically and economically significant predictive power in sample and out of sample, and it outperforms widely used benchmark predictors and alternative predictor-combination methods. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.20460 |
| By: | MacLachlan, Matthew J.; Thompson, Jada; Dai, Bingyan |
| Abstract: | Shifting seasonal patterns have arisen in food markets due to changing supply chains, consumer preferences, and infectious disease prevalence. Persistent infections of H5N1 avian influenza among U.S. poultry and egg-laying bird populations have altered the seasonal patterns in corresponding market dynamics, particularly prices. While the geographic distribution of the precise timing of cases remains difficult, the broad pattern of higher prevalence in Winter and lower prevalence in summer typically leads to price spikes early each year. This pattern represents a shift from historical seasonality, which typically saw mild price spikes around the winter holidays and Easter. At the same time, the imposition of desirable model features may enhance forecast performance when historical data do not yet capture these phenomena. However, such ad hoc modifications should be done carefully, as the addition, potentially intuitively appealing, of a model structure often increases forecast errors. We find that simpler forecasting models typically yield the lowest forecast error if they are allowed to adapt over time. More accurate predictions facilitate better planning among producers, consumers, and entities providing food assistance to low-income households. |
| Keywords: | Marketing |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:ags:aaea25:360847 |
| By: | 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:bny:wpaper:0144 |
| By: | Amar, Amine; Antonio, Ronald Jeremy S.; Okou, Cyrille Guei; Pede, Valerien O. |
| Abstract: | Demand for rice in Africa has been steadily increasing due to population and economic growth and changing preferences. However, low production growth has led to major gaps between rice supply and demand leading to import reliance. Given the concerns on long term food security and availability that comes with import reliance, many studies have focused on evaluating how shocks in the global markets and exporting countries are transmitted to import reliant countries. With this development, our paper furthers this endeavour by developing an econometric predictive framework to identify dependency relationships for forecasting purposes, which provides information not only on the relationship between African countries and its import sources, but also how current prices impact future prices. The novelty of our study lies in integrating the Vine Copula into cointegration analysis to forecast future rice prices using monthly time series data from three sub-Saharan African countries between 2018 and 2024. The results demonstrate a high predictive performance from the proposed approach and finds that food insecurity dynamics are both asymmetric and path dependent. Additional insights are obtained through sensitivity and impulse response analyses. |
| Keywords: | Risk and Uncertainty |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:ags:aaea25:360696 |
| By: | Drin, Svitlana (Örebro University School of Business); Zhuravlova, Anastasiia (National University of Kyiv-Mohyla Academy) |
| Abstract: | imely assessment of regional economic activity in Ukraine is severely constrained by institutional and data-related limitations. Official regional gross regional product (GRP) statistics are available only at low frequency, are published with substantial delays, and, in the post-2022 period, are further affected by disruptions to statistical production caused by martial law. At the same time, a growing set of potentially informative regional indicators derived from administrative records and official short-term statistics is available at higher frequencies but only over short and heterogeneous time spans. These features make the direct application of standard regional nowcasting models infeasible. This paper develops a mixed-frequency factor-augmented vector autoregressive framework tailored to the Ukrainian data environment and designed to incorporate short and incomplete regional indicators into the nowcasting of regional GDP. The model explicitly exploits the hierarchical structure of Ukrainian regional statistics by combining information from quarterly and annual measures of economic activity and by linking regional dynamics to national output developments. Short regional indicators are summarised through latent regional factors extracted using missing-data factor estimation techniques that are robust to ragged edges at both the beginning and the end of the sample. The proposed framework is implemented using Ukrainian macro-regional aggregates constructed from official data published by the State Statistics Service of Ukraine. Particular attention is paid to the treatment of labour market indicators, housing price dynamics, and other short-term variables that exhibit discontinuities or limited availability. A pseudo-real-time nowcasting exercise shows that conditioning regional GDP nowcasts on factor information derived from short regional data improves predictive performance when contemporaneous national GDP estimates are not yet available. Once national aggregates are released, the marginal informational contribution of regional short-term indicators diminishes. Overall, the results demonstrate that mixed-frequency factor-augmented VAR models provide a coherent and empirically viable framework for regional GDP nowcasting in Ukraine. The approach is particularly well suited to data environments 1 characterised by short samples, publication delays, and institutional disruptions, and thus offers a valuable tool for real-time regional economic monitoring in periods of heightened uncertainty. |
| Keywords: | MF-FAVAR; FAVAR; Nowcasting; EMPCA; GRP; Google Trends |
| JEL: | C53 E37 |
| Date: | 2026–01–02 |
| URL: | https://d.repec.org/n?u=RePEc:hhs:oruesi:2026_001 |
| By: | Petr Koklev |
| Abstract: | Financial institutions face a trade-off between predictive accuracy and interpretability when deploying machine learning models for credit risk. Monotonicity constraints align model behavior with domain knowledge, but their performance cost - the price of monotonicity - is not well quantified. This paper benchmarks monotone-constrained versus unconstrained gradient boosting models for credit probability of default across five public datasets and three libraries. We define the Price of Monotonicity (PoM) as the relative change in standard performance metrics when moving from unconstrained to constrained models, estimated via paired comparisons with bootstrap uncertainty. In our experiments, PoM in AUC ranges from essentially zero to about 2.9 percent: constraints are almost costless on large datasets (typically less than 0.2 percent, often indistinguishable from zero) and most costly on smaller datasets with extensive constraint coverage (around 2-3 percent). Thus, appropriately specified monotonicity constraints can often deliver interpretability with small accuracy losses, particularly in large-scale credit portfolios. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.17945 |
| By: | Benjamin Born; Nora Lamersdorf; Jana-Lynn Schuster; Sascha Steffen |
| Abstract: | Using modern natural language processing, we construct a high-frequency inflation expectations index from German-language tweets. This index closely tracks realized inflation and aligns even more closely with household survey expectations. It also improves short-run forecasts relative to standard benchmarks. In response to monetary policy tightening, the index declines within about a week, with the effects concentrated in tweets by private individuals and during the recent period of elevated inflation. Using 117 million online transactions from German retailers, we show that higher inflation expectations are followed by lower household spending on discretionary goods. By linking these shifts in demand to stock returns, we find that, during periods of elevated inflation, firms operating in discretionary sectors experience significantly lower stock returns when inflation expectations rise. Thus, our Twitter-based index provides market participants and policymakers with a timely tool to monitor inflation sentiment and its economic consequences. |
| Keywords: | Inflation expectations, social media (Twitter/X), large language models (LLMs), NLP, household consumption, stock returns, monetary policy |
| JEL: | E31 D84 E58 C45 C81 |
| Date: | 2026–01 |
| URL: | https://d.repec.org/n?u=RePEc:bon:boncrc:crctr224_2025_724 |
| By: | Weiyang ZHAI; Yushi YOSHIDA |
| Abstract: | Inflation expectation is one of the essential components of monetary policy decisions. Upon examining Japanese inflation expectations between 1991:Q4 and 2025:Q1, we propose a modified empirical model that includes a forecast trend term in addition to the forecast revision term. We found that the forecast trend term affects the forecast errors. The full sample results indicate that people in Japan form non-rational expectations with information rigidity. However, this holds only in the recent episode of inflation following the post-COVID period. During the zero-inflation periods, people formed full-information rational expectations. In addition, we find evidence that consumption tax hikes affect the forecast errors, partly due to the uncertainty about future implementation. In both periods, the possibility of deviating from rational expectations cannot be ruled out. |
| Date: | 2026–01 |
| URL: | https://d.repec.org/n?u=RePEc:eti:dpaper:26004 |
| By: | Wojciech Charemza (University of Leicester); Christian Francq (CREST - Centre de Recherche en Économie et Statistique - ENSAI - Ecole Nationale de la Statistique et de l'Analyse de l'Information [Bruz] - GENES - Groupe des Écoles Nationales d'Économie et Statistique - X - École polytechnique - IP Paris - Institut Polytechnique de Paris - ENSAE Paris - École Nationale de la Statistique et de l'Administration Économique - GENES - Groupe des Écoles Nationales d'Économie et Statistique - IP Paris - Institut Polytechnique de Paris - CNRS - Centre National de la Recherche Scientifique, IP Paris - Institut Polytechnique de Paris); Radu Lupu (A.S.E. - The Bucharest University of Economic Studies / Academia de Studii Economice din Bucureşti); Svetlana Makarova (ANU College of Science [Canberra] - ANU - Australian National University); Jean-Michel Zakoïan (CREST - Centre de Recherche en Economie et Statistique [Bruz] - ENSAI - Ecole Nationale de la Statistique et de l'Analyse de l'Information [Bruz] - GENES - Groupe des Écoles Nationales d'Économie et Statistique) |
| Abstract: | This paper introduces a novel statistical test, the Policy Effects Lagrange Multiplier (PELM) test, to detect stabilization policy effects in the distribution of forecast errors from dynamic financial models. Traditional analyses of policy impact typically rely on explicit policy information or direct intervention data, which are often unavailable or incomplete. In contrast, the proposed PELM test infers policy footprints from the distribution of forecast errors alone. Empirically applied to sovereign bond yield data from 33 countries before the Russian financial crisis of 2014, the test identifies countries showing stabilization policy footprints. Subsequent analysis shows that significant budgetary improvements were observed for years following the crisis in the group of countries where our test statistically confirmed stabilization policies. This confirms the rationale of test foundations and also indicates its predictive properties. Robustness checks further validate these findings across various model specifications and sensitivity scenarios. The proposed PELM test offers policymakers and researchers a powerful tool for evaluating stabilization policies, facilitating better forecasting and assessing policy efficiency in diverse economic contexts without necessitating detailed policy intervention data. |
| Date: | 2025–12–01 |
| URL: | https://d.repec.org/n?u=RePEc:hal:journl:hal-05430912 |
| By: | Foltas, Alexander |
| Abstract: | This paper proposes a novel method to uncover shifting thematic priorities in textual business cycle reports and links them to macroeconomic fluctuations. To this end, I leverage qualitative business cycle forecasts published by leading German economic research institutes from 1970-2017 to estimate the proportions of latent topics. These topics are then aggregated into broader macroeconomic subjects using a supervised approach. By extracting the cyclical components of these subjects' proportions, I derive dynamic measures of forecasters' thematic priorities. Correlating the cyclic components with key macroeconomic indicators reveals consistent patterns across economic expansions and contractions. Around economic peaks, forecasters emphasize inflation-related over recession-related topics. I thus propose that forecasters' failure to predict recessions may stem from a tendency to underestimate growth risks and overestimate inflation risks during periods of contractionary monetary policy. Around troughs, forecasters prioritize investment-related topics over general growth considerations. |
| Abstract: | Diese Studie stellt eine neue Methode vor, um sich wandelnde thematische Prioritäten in textbasierten Konjunkturberichten aufzudecken und diese mit makroökonomischen Schwankungen zu verknüpfen. Zu diesem Zweck nutze ich qualitative Konjunkturprognosen führender deutscher Wirtschaftsforschungsinstitute aus den Jahren 1970 bis 2017, um die Anteile latenter Themen zu schätzen. Diese Themen werden anschließend mithilfe eines überwachten Verfahrens zu breiteren makroökonomischen Themenfeldern aggregiert. Durch die Extraktion der zyklischen Komponenten dieser Themenanteile leite ich dynamische Maße für die thematischen Prioritäten der Prognostiker ab. Die Korrelation der zyklischen Komponenten mit zentralen makroökonomischen Indikatoren zeigt konsistente Muster über Auf- und Abschwungphasen hinweg. Rund um Konjunkturhochs legen Prognostiker einen stärkeren Fokus auf inflations- als auf rezessionsbezogene Themen. Ich schlage daher vor, dass das Scheitern der Prognostiker bei der Vorhersage von Rezessionen auf eine tendenzielle Unterschätzung von Wachstums- und Überschätzung von Inflationsrisiken in Phasen restriktiver Geldpolitik zurückzuführen sein könnte. Rund um konjunkturelle Tiefpunkte priorisieren Prognostiker investitionsbezogene Themen gegenüber allgemeinen Wachstumsüberlegungen. |
| Keywords: | Macroeconomic forecasting, Evaluating forecasts, Recession forecasting, Topic Modeling, Natural language processing, Judgemental forecasting |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:hwwiwp:334495 |
| By: | Gang Li; Dandan Qiao; Mingxuan Zheng |
| Abstract: | We find that event features extracted by large language models (LLMs) are effective for text-based stock return prediction. Using a pre-trained LLM to extract event features from news articles, we propose a novel deep learning model based on structured event representation (SER) and attention mechanisms to predict stock returns in the cross-section. Our SER-based model provides superior performance compared with other existing text-driven models to forecast stock returns out of sample and offers highly interpretable feature structures to examine the mechanisms underlying the stock return predictability. We further provide various implications based on SER and highlight the crucial benefit of structured model inputs in stock return predictability. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.19484 |
| By: | Gagan Deep; Akash Deep; William Lamptey |
| Abstract: | We develop a rigorous walk-forward validation framework for algorithmic trading designed to mitigate overfitting and lookahead bias. Our methodology combines interpretable hypothesis-driven signal generation with reinforcement learning and strict out-of-sample testing. The framework enforces strict information set discipline, employs rolling window validation across 34 independent test periods, maintains complete interpretability through natural language hypothesis explanations, and incorporates realistic transaction costs and position constraints. Validating five market microstructure patterns across 100 US equities from 2015 to 2024, the system yields modest annualized returns (0.55%, Sharpe ratio 0.33) with exceptional downside protection (maximum drawdown -2.76%) and market-neutral characteristics (beta = 0.058). Performance exhibits strong regime dependence, generating positive returns during high-volatility periods (0.60% quarterly, 2020-2024) while underperforming in stable markets (-0.16%, 2015-2019). We report statistically insignificant aggregate results (p-value 0.34) to demonstrate a reproducible, honest validation protocol that prioritizes interpretability and extends naturally to advanced hypothesis generators, including large language models. The key empirical finding reveals that daily OHLCV-based microstructure signals require elevated information arrival and trading activity to function effectively. The framework provides complete mathematical specifications and open-source implementation, establishing a template for rigorous trading system evaluation that addresses the reproducibility crisis in quantitative finance research. For researchers, practitioners, and regulators, this work demonstrates that interpretable algorithmic trading strategies can be rigorously validated without sacrificing transparency or regulatory compliance. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.12924 |
| By: | Foltas, Alexander |
| Abstract: | This paper contributes to previous research on the efficient integration of forecasters' narratives into business cycle forecasts. Using a Bidirectional Encoder Representations from Transformers (BERT) model, I quantify 19, 300 paragraphs from German business cycle reports (1998-2021) and use them to predict the direction of consumption forecast errors. By testing the model on an evaluation sample, I find a highly significant correlation of modest strength between predicted and actual sign of the forecast error. The correlation coefficient is substantially higher for 12.8% of paragraphs with a predicted class probability of 85% or higher. By qualitatively reviewing 150 of such high-probability paragraphs, I find recurring narratives correlated with consumption forecast errors. Underestimations of consumption growth often mention rising employment, increasing wages and transfer payments, low inflation, decreasing taxes, crisis-related fiscal support, and reduced relevance of marginal employment. Conversely, overestimated consumption forecasts present opposing narratives. Forecasters appear to particularly underestimate these factors when they disproportionately affect low-income households. |
| Abstract: | Diese Studie leistet einen Beitrag zur bisherigen Forschung hinsichtlich der effizienten Einbindung von Prognosenarrativen in Konjunkturprognosen. Unter Verwendung eines BERT-Modells (Bidirectional Encoder Representations from Transformers) quantifiziere ich 19.300 Absätze aus deutschen Konjunkturberichten (1998-2021) und nutze diese um die Richtung von Konsumprognosefehlern vorherzusagen. Durch die Überprüfung des Modells anhand einer Evaluationsstichprobe stelle ich eine hochsignifikante Korrelation moderater Stärke zwischen dem vorhergesagten und dem tatsächlichen Vorzeichen des Prognosefehlers fest. Der Korrelationskoeffizient ist für jene 12, 8 % der Absätze wesentlich höher, die eine vorhergesagte Klassenwahrscheinlichkeit von 85 % oder mehr aufweisen. Eine qualitative Untersuchung von 150 dieser Absätze mit hoher Wahrscheinlichkeit zeigt wiederkehrende Narrative, die mit Fehlern in der Konsumprognose korrelieren. Unterschätzungen des Konsumwachstums erwähnen häufig steigende Beschäftigung, zunehmende Löhne und Transferzahlungen, niedrige Inflation, sinkende Steuern, krisenbedingte fiskalische Unterstützung sowie eine abnehmende Bedeutung geringfügiger Beschäftigung. Umgekehrt weisen überschätzte Konsumprognosen gegensätzliche Narrative auf. Prognostiker scheinen diese Faktoren insbesondere dann zu unterschätzen, wenn sie einkommensschwache Haushalte überproportional betreffen. |
| Keywords: | Macroeconomic forecasting, Evaluating forecasts, Business cycles, Consumption forecasting, Natural language processing, Language Modeling, Machine learning, Judgmental forecasting |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:hwwiwp:334496 |
| By: | Siddhartha Chib; Fei Tan |
| Abstract: | We show how state-of-the-art large language models (LLMs), seemingly inapplicable to the small samples typical of macroeconomics, can be trained to learn the language of macroeconomy. We estimate a large-scale dynamic stochastic general equilibrium (DSGE) model on an initial segment of the data and obtain a posterior distribution over structural parameters. We sample from this posterior to generate millions of theory-consistent synthetic panels that, when mixed with actual macroeconomic data, form the training corpus for a time-series transformer with attention. The trained model is then used to forecast out-of-sample through 2025. The results show that this hybrid forecaster, which combines the theoretical coherence of DSGE models with the representational power of modern LLMs, successfully learns the macroeconomic language. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.21031 |
| By: | Eden Gross; Ryan Kruger; Francois Toerien |
| Abstract: | In the last five years, expected shortfall (ES) and stressed ES (SES) have become key required regulatory measures of market risk in the banking sector, especially following events such as the global financial crisis. Thus, finding ways to optimize their estimation is of great importance. We extend the application of dynamic Bayesian networks (DBNs) to the estimation of 10-day 97.5% ES and stressed ES, building on prior work applying DBNs to value at risk. Using the S&P 500 index as a proxy for the equities trading desk of a US bank, we compare the performance of three DBN structure-learning algorithms with several traditional market risk models, using either the normal or the skewed Student's t return distributions. Backtesting shows that all models fail to produce statistically accurate ES and SES forecasts at the 2.5% level, reflecting the difficulty of modeling extreme tail behavior. For ES, the EGARCH(1, 1) model (normal) produces the most accurate forecasts, while, for SES, the GARCH(1, 1) model (normal) performs best. All distribution-dependent models deteriorate substantially when using the skewed Student's t distribution. The DBNs perform comparably to the historical simulation model, but their contribution to tail prediction is limited by the small weight assigned to their one-day-ahead forecasts within the return distribution. Future research should examine weighting schemes that enhance the influence of forward-looking DBN forecasts on tail risk estimation. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.12334 |