|
on Forecasting |
| By: | Imad Talhartit (Université Hassan 1er [Settat], Ecole Nationale de Commerce et Gestion - Settat, Laboratory of Finance, Audit and Organizational Governance Research); Sanae Ait Jillali (Université Hassan 1er [Settat], Ecole Nationale de Commerce et Gestion - Settat); Mounime El Kabbouri |
| Abstract: | Capital markets play a fundamental role in the economy by facilitating the flow of funds between investors with capital surpluses and those with financing needs. However, these markets' inherent complexity and high volatility-amplified by economic crises and geopolitical events-make decision-making particularly challenging. In this context, artificial intelligence (AI), especially machine learning (ML) and deep learning (DL), has become increasingly relevant for modeling complex financial time series such as stock prices. Among various learning approaches, Long Short-Term Memory (LSTM) networks stand out for their ability to capture long-term dependencies in sequential data. This study compares the predictive performance of LSTM and Artificial Neural Networks (ANN) models, on ten stocks comprising the MADEX index of the Casablanca Stock Exchange, across three forecasting horizons (10, 20, and 30 days). Results demonstrate that the LSTM model consistently outperforms the ANN model in terms of accuracy and trend detection. For instance, over a 30-day horizon, the LSTM correctly predicted 8 out of 10 stocks, compared to only 4 for the ANN. This work is part of a broader research effort aimed at identifying the most effective model for stock price forecasting. Building on the results of this and previous studies, particularly those involving LSTM models optimized using genetic algorithms, future research will explore other models such as Gated Recurrent Units (GRU) and Support Vector Machines (SVM) to further enhance prediction accuracy and robustness. |
| Keywords: | Stock price forecasting, Casablanca Stock Exchange, Long Short-Term Memory (LSTM), Artificial Neural Networks (ANN), Prediction accuracy |
| Date: | 2025–05–09 |
| URL: | https://d.repec.org/n?u=RePEc:hal:journl:hal-05063012 |
| By: | Emmanouil SOFIANOS; Thierry BETTI; Emmanouil Theophilos PAPADIMITRIOU; Amélie BARBIER-GAUCHARD; Periklis GOGAS |
| Abstract: | Forecasting public debt is essential for effective policymaking and economic stability, yet traditional approaches face challenges due to data scarcity. While machine learning (ML) has demonstrated success in financial forecasting, its application to macroeconomic forecasting remains underexplored, hindered by short historical time series and low-frequency (e.g., quarterly/annual) data availability. This study proposes a novel hybrid framework integrating Dynamic Stochastic General Equilibrium (DSGE) modeling with ML techniques to address these limitations, focusing on the evolution of France’s public debt. We first generate a large synthetic macroeconomic dataset using an estimated DSGE model for France, which allows for efficient training of ML algorithms. These trained models are then applied to actual historical data for directional debt forecasting. The results show that the best machine learning model is an XGBoost achieving 90% accuracy. Our results highlight the viability of combining structural economic models with data-driven techniques to improve macroeconomic forecasting. |
| Keywords: | DSGE, Machine Learning, Public Debt, Forecasting, France. |
| JEL: | C53 E27 E37 H63 H68 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:ulp:sbbeta:2025-18 |
| By: | Felipe Roldán-Ferrín; Julián A. Parra-Polania |
| Abstract: | This paper evaluates the predictive capacity of a machine learning model based on Random Forests (RF), combined with Google Trends (GT) data, for nowcasting monthly inflation in Colombia. The proposed RF-GT model is trained using historical inflation data, macroeconomic indicators, and internet search activity. After optimizing the model’s hyperparameters through time series cross-validation, we assess its out-of-sample performance over the period 2023–2024. The results are benchmarked against traditional approaches, including SARIMA, Ridge, and Lasso regressions, as well as professional forecasts from the Banco de la República’s monthly survey of financial analysts (MES). In terms of forecast accuracy, the RF-GT model consistently outperforms the statistical models and performs comparably to the analysts’ median forecast, while offering the additional advantage of producing predictions approximately one and a half weeks earlier. These findings highlight the practical value of integrating alternative data sources and machine learning techniques into the inflation monitoring toolkit of emerging economies. *****RESUMEN: Este artículo evalúa la capacidad predictiva de un modelo de aprendizaje automático basado en Random Forest (RF), combinado con datos de Google Trends (GT), para realizar nowcasting de la inflación mensual en Colombia. El modelo propuesto, denominado RF-GT, se entrena utilizando datos históricos de inflación, indicadores macroeconómicos y actividad de búsqueda en internet. Tras la optimización de los hiperparámetros mediante validación cruzada para series de tiempo, se evalúa su desempeño fuera de muestra durante el periodo 2023–2024. Los resultados se comparan con enfoques tradicionales, incluidos los modelos SARIMA, regresiones Ridge y Lasso, así como con los pronósticos profesionales de la Encuesta Mensual de Expectativas (EME) del Banco de la República. En términos de precisión predictiva, el modelo RF-GT supera de forma consistente a los modelos estadísticos y muestra un desempeño comparable al pronóstico mediano de los analistas, con la ventaja adicional de generar predicciones aproximadamente semana y media antes. Estos hallazgos destacan el valor práctico de integrar fuentes de datos alternativas y técnicas de aprendizaje automático en los sistemas de monitoreo de inflación de economías emergentes. |
| Keywords: | Inflation, Nowcasting, Forecasting, Random Forest, Google Trends, Machine Learning, Inflación, Pronóstico en Tiempo Real, Pronóstico, Bosques Aleatorios, Tendencias de Google, aprendizaje automático |
| JEL: | C14 C53 E17 E31 E37 |
| Date: | 2025–07 |
| URL: | https://d.repec.org/n?u=RePEc:bdr:borrec:1318 |
| By: | Frank Schorfheide; Zhiheng You |
| Abstract: | Econometricians invest substantial effort in constructing standard errors that yield valid inference under a hypothetical data-generating process. This paper asks a fundamental question: Are the uncertainty statements reported by applied researchers consistent with empirical frequencies? The short answer is no. Drawing on the forecasting literature, we predict estimates from “new” studies using estimates from corresponding baseline studies. By doing this across a large number of study groups and linking parameters through a hierarchical model, we compare stated probabilities to observed empirical frequencies. Alignment occurs only under limited external validity, namely, that the studies estimate different parameters. |
| JEL: | C11 C18 C21 |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:33962 |
| By: | Laura Capera Romero (Vrije Universiteit Amsterdam and Tinbergen Institute); Anne Opschoor (Vrije Universiteit Amsterdam and Tinbergen Institute) |
| Abstract: | This paper compares the statistical and economic performance of state-of-the-art highfrequency based multivariate volatility models with a simpler, widely used alternative - the Exponentially Weighted Moving Average (EWMA) filter. Using over two decades of 100 U.S. stock returns (2002–2023), we assess model performance through a Global Minimum Variance portfolio optimization exercise across various forecast horizons. We find that the EWMA model consistently outperforms more complex HF-based volatility models, delivering significant utility gains when including transaction costs, due in part to its lower turnover. Even in the absence of transaction costs, the EWMA filter cannot be beaten in most cases. Our results are robust to various dimensions, including no-short-selling constraints, varying portfolio sizes, and alternative parameter choices, highlighting the continued relevance of the EWMA model in high-frequency-based portfolio allocation. |
| Date: | 2025–06–26 |
| URL: | https://d.repec.org/n?u=RePEc:tin:wpaper:20250041 |
| By: | Sun, Yiqiao; de Bondt, Gabe |
| Abstract: | This study involves tasking ChatGPT with classifying an “activity sentiment score” based on PMI news releases. It explores the predictive power of this score for euro area GDP nowcasting. We find that the PMI text scores enhance GDP nowcasts beyond what is embedded in ECB/Eurosystem Staff projections and Eurostat’s first GDP estimate. The ChatGPT-derived activity score retains its significance in regressions that also include the composite output PMI diffusion index. GDP nowcasts are significantly enhanced with PMI text scores even when accounting for methodological variations, excluding extraordinary economic events like the pandemic, and for different GDP growth quantiles. However, the forecast gains from the enhancement of GDP nowcasts with ChatGPT scores are time dependent, varying by calendar years. Sizeable forecast gains of on average about 20% were obtained apart from the two most recent years due to exceptionally low forecast errors of the two benchmarks, especially the first GDP estimate. JEL Classification: C8, E32, C22 |
| Keywords: | chat generative pre-training transformer, nowcasting GDP, purchasing managers’ index (PMI), text analysis, zero-shot sentiment analysis |
| Date: | 2025–06 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbwps:20253063 |
| By: | Xia Zou (Vrije Universiteit Amsterdam and Tinbergen Institute); Yicong Lin (Vrije Universiteit Amsterdam and Tinbergen Institute); André Lucas (Vrije Universiteit Amsterdam and Tinbergen Institute) |
| Abstract: | Point forecasts of score-driven models have been shown to behave at par with those of state-space models under a variety of circumstances. We show, however, that density rather than point forecasts of plain-vanilla score-driven models substantially underperform their state-space counterparts in a factor model context. We uncover the origins of this phenomenon and show how a simple adjustment of the measurement density of the score-driven model can put score-driven and state-space models approximately back on an equal footing again. The score-driven models can subsequently easily be extended with non-Gaussian features to fit the data even better without complicating parameter estimation. We illustrate our findings using a factor model for the implied volatility surface of S&P500 index options data. |
| JEL: | C32 C38 |
| Date: | 2025–05–30 |
| URL: | https://d.repec.org/n?u=RePEc:tin:wpaper:20250036 |
| By: | Dercio de Assis |
| Abstract: | This paper analyzes data from urban Brazil using supervised machine-learning techniques to shed greater light on the role that childhood poverty plays in lifelong health and longevity. By examining a unique dataset collected over a 10-year period from thousands of very small, sub-neighborhood-level geographic areas, I document that child poverty measures have higher predictive power than household income, and other major socioeconomic variables, in forecasting child and adult health outcomes and lifespans. In addition, using a rich dictionary of hundreds of variables and different data-driven specification selections, the machine-learning models reveal that experiencing more severe deprivation in childhood is associated with a decrease of 4 percentage points in the probability of survival to ages 40 and 60. These predictions offer further economic insights on the importance of early life circumstances for human development. |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:not:notnic:2025-12 |
| By: | Eric Engstrom |
| Abstract: | Amid ongoing trade policy shifts and geopolitical uncertainty, concerns about stagflation have reemerged as a key macroeconomic risk. This paper develops a probabilistic framework to estimate the likelihood of stagflation versus soft landing scenarios over a four-quarter horizon. Building on Bekaert, Engstrom, and Ermolov (2025), the model integrates survey forecasts, structural shock decomposition, and a non-Gaussian BEGE-GARCH approach to capture time-varying volatility and skewness. Results suggest that the probability of stagflation was elevated at around 30 percent in late 2022, while the chance of a soft landing was below 5 percent. As inflation moderated and growth remained strong through 2024, these probabilities reversed. However, by mid-2025, renewed tariff concerns drove stagflation risk back up and the probability of a soft landing lower. These shifts highlight the potential value of distributional forecasting for policymakers and market participants navigating uncertain macroeconomic conditions. |
| Keywords: | GARCH; Inflation; Recession; Soft landing; Stagflation; Time-varying uncertainty |
| JEL: | E31 E32 E37 E60 |
| Date: | 2025–07–07 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedgfe:2025-47 |
| By: | Mr. Sam Ouliaris; Ms. Celine Rochon; Daniel Taumoepeau |
| Abstract: | This paper customizes to the Tongan economy a macroeconomic model for medium-term quarterly projections of key macro variables (QPM): output, inflation, interest rate, and exchange rate. The model is calibrated to embody the specific attributes of the Tongan economy such as the persistence of domestic output, core inflation and interest rates, as well as Tonga’s monetary policy transmission. It is then used to study three scenarios and assess their impact on the baseline. The first scenario involves shocks proxying for a bank failure. The second scenario introduces shocks that simulate the consequences of a natural disaster. Finally, the third scenario introduces shocks that represent a significant negative external shock on Tonga. These shocks were chosen to reflect the sensitivity of Tonga to adverse financial shocks, to their trading partners’ macroeconomic policies, and to extreme weather events. |
| Keywords: | Forecasting and Policy Analysis; Quarterly Projection Model; Monetary Policy; Transmission Mechanism |
| Date: | 2025–06–20 |
| URL: | https://d.repec.org/n?u=RePEc:imf:imfwpa:2025/121 |